Volume 76, Issue 3 p. 448-458
Free Access

Evaluation of reproductive costs for Weddell seals in Erebus Bay, Antarctica



Department of Ecology, Montana State University, Bozeman, Montana 59717, USA

Search for more papers by this author


Department of Ecology, Montana State University, Bozeman, Montana 59717, USA

Search for more papers by this author


Department of Ecology, Montana State University, Bozeman, Montana 59717, USA

Search for more papers by this author
First published: 16 March 2007
Citations: 69
Gillian L. Hadley, 27820 Southside Centennial Road, Lima, MT 59739, USA. E-mail: [email protected]


  • 1

    Organisms balance current reproduction against future survival and reproduction, which results in life-history trade-offs. These trade-offs are also known as reproductive costs and may represent significant factors shaping life-history strategy for many species.

  • 2

    Using multistate mark–resight models and 26 years of mark–resight data (1979–2004), we estimated the costs of reproduction to survival and reproductive probabilities for Weddell seals in Erebus Bay, Antarctica and evaluated whether this species either conformed to the ‘prudent parent’ reproductive strategy predicted by life-history theory for long-lived mammals or alternatively, incurred costs to survival in order to reproduce in a variable environment (flexible-strategy hypothesis).

  • 3

    Results strongly supported the presence of reproductive costs to survival (mean annual survival probability was 0·91 for breeders vs. 0·94 for nonbreeders), a notable difference for a long-lived mammal, demonstrating that investment in reproduction does result in a cost to survival for Weddell seals, contrary to the prudent parent hypothesis.

  • 4

    Reproductive costs to subsequent reproductive probabilities were also present for first-time breeders (mean probability of breeding the next year was 31·3% lower for first-time breeders than for experienced breeders), thus supporting our prediction of the influence of breeding experience.

  • 5

    We detected substantial annual variation in survival and breeding probabilities. Breeding probabilities were negatively influenced by summer sea-ice extent, whereas weak evidence suggested that survival probabilities were affected more by winter sea-ice extent, and the direction of this effect was negative. However, a model with annual variation unrelated to any of our climate or sea-ice covariates performed best, indicating that further study will be needed to determine the appropriate mechanism or combination of mechanisms underlying this annual variation.


An organism's evolutionary success is determined by its lifetime reproductive output. Producing and caring for offspring may decrease energy available for subsequent reproductive opportunities, and reduce longevity. Thus, organisms balance current reproduction against future survival and reproduction (Williams 1966). This trade-off is known as reproductive cost and is hypothesized to be an important factor shaping life-history strategy for many species (Roff 1992; Stearns 1992). Diverse factors such as age, age at first reproduction, previous breeding experience, population density, and environmental conditions may affect the occurrence and size of reproductive costs (e.g. Clutton-Brock 1984; Weimerskirch 1990; Reiter & LeBoeuf 1991; Stearns 1992; Viallefont, Cooke & Lebreton 1995; Pistorius et al. 2004; Tavecchia et al. 2005; Beauplet et al. 2006).

In species for which the age at first reproduction and reproductive effort vary among individuals, it is essential to study animals of known age to gain insights into reproductive costs. Long-term studies are especially valuable to demonstrate how reproductive costs vary with changing ecological circumstances (Festa-Bianchet 1989). The Erebus Bay population of Weddell seals Leptonychotes weddellii (Lesson) is a model system for investigating costs of reproduction in a long-lived mammal inhabiting a variable environment. This population of seals has been the subject of an ongoing mark–resight research programme since 1969 (Stirling 1969; Siniff et al. 1977; Testa & Siniff 1987; Cameron & Siniff 2004) and contains a large number and proportion of animals of known age for whom reproductive histories are known. Further, the population, which is located at approximately 77°S in McMurdo Sound in the Ross Sea region of Antarctica, is the southernmost breeding mammal population in the world and data from this population provide researchers with a thorough, long-term data set for a long-lived marine mammal. Weddell seals are intermittent breeders, and some individuals may breed annually for many consecutive years, while others may remain nonbreeders for several years in between breeding years (Testa 1987; Hadley et al. unpublished data). Nonbreeding has been shown to have both positive and negative consequences for survival rate in various species (Orell et al. 1994; Harris & Wanless 1995). The variation in breeding propensity displayed by Weddell seals makes the species an excellent subject for learning about life-history decisions and reproductive costs in long-lived animals.

In this study, we evaluated both costs to survival and costs to future reproduction for female Weddell seals in Erebus Bay, Antarctica. Contrasting predictions have been made regarding reproductive costs for long-lived species in variable environments. Life-history theory predicts that costs to survival should be strongly selected against in species with long reproductive life spans. Reproductive output for long-lived species is determined by the outcomes of numerous breeding attempts throughout life, whereas species with short life spans may benefit from maximal investment in one reproductive bout. Therefore, surviving to future reproductive opportunities is crucial for maximizing lifetime reproductive output in long-lived species (Williams 1966; Goodman 1974; Charlesworth 1980) and individuals should be less prone to trade their own survival for that of their offspring because any reduction in adult survival would greatly lower lifetime reproductive success (‘prudent parent’ reproductive strategy; Drent & Daan 1980; Cam et al. 1998). However, numerous studies of long-lived seabirds have demonstrated that large variation in breeding conditions favours flexibility of reproductive effort (i.e. occasional investment in reproduction at a cost to survival, or ‘flexible-strategy hypothesis’) (Reid 1987; Jacobsen, Erikstad & Sæther 1995; Erikstad et al. 1997, 1998; Golet, Irons & Estes 1998). Despite the lower reproductive costs to survival generally predicted for long-lived species, the variability of the polar environment may require that seals occasionally invest in reproduction at a cost to survival (the flexible-strategy hypothesis). Therefore, we expected that both types of costs would be apparent.

Life-history strategy and lifetime reproductive output are also influenced by the manner in which reproductive costs vary during an individual's lifetime. Numerous studies have demonstrated that younger animals conceive, implant, or give birth later, and have lower natality rates than do older animals (Sæther & Haagenrud 1983; Lunn, Boyd & Croxall 1994). Furthermore, first-year survival of offspring has been shown to increase with increasing maternal age for southern elephant seals Mirounga leonina (McMahon & Bradshaw 2004). For Weddell seals, we therefore predicted that costs to both survival and reproduction would vary with age, with young seals exhibiting larger costs.

Animals breeding for the first time are also thought to experience higher reproductive costs than are experienced breeders, regardless of age (Weimerskirch 1990; Viallefont et al. 1995). However, separating the effects of age and breeding experience may be difficult in practice. Many investigators lack detailed information about individual reproductive histories or age of animals in their study population (Barbraud & Weimerskirch 2005), whereas others may not have adequate sample sizes or variation in their data sets to examine animals of equal age with varying levels of experience, or animals of various ages with equivalent experience. Our data set included seals with and without breeding experience for every age from 4 to 14, and we were thus able to investigate separately the effects of age and breeding experience. Several studies have found higher reproductive costs for inexperienced breeders while controlling for age (Lunn et al. 1994; Viallefont et al. 1995), and we therefore predicted higher reproductive costs for inexperienced seals in our study.

In accordance with Pistorius et al.'s (2004) findings for southern elephant seals, we predicted that age of first reproduction would not affect reproductive costs. However, based on variation in age at first reproduction for Weddell seals reported in a previous analysis (Hadley et al. 2006), we expected that substantial heterogeneity in individual quality was present in this population. Numerous life-history studies have demonstrated the importance of considering individual variation in quality when assessing reproductive costs (van Noordwijk & deJong 1986; Weladji et al. 2006). Therefore, we also proposed an alternative prediction that if reproductive costs did vary with age at first reproduction, highest costs would be experienced by those individuals that delayed primiparity. We hypothesized that such individuals were of lower quality and thus expected them to suffer higher costs of reproduction relative to high-quality seals of the same age that had begun reproduction at the mean age or even earlier (Curio 1983; Cam & Monnat 2000; Cam et al. 2002; Beauplet et al. 2006).

For some species, reproductive costs may only be present under scenarios of population density or resource scarcity that lower survival rates beyond some threshold (Tuomi, Hakala & Haukioja 1983; Festa-Bianchet 1989; Stearns 1992; Reznick, Nunney & Tessier 2000; Tavecchia et al. 2005). Therefore, we expected reproductive costs to both survival and reproduction of Weddell seals to vary annually due to changing environmental conditions. The sensitivity of marine ecosystems, and especially upper trophic level predators, to climate change has been noted in numerous recent studies (e.g. Croxall, Trathan & Murphy 2002; Beauplet et al. 2005; Jenouvrier et al. 2005; McMahon & Burton 2005). During El-Niño Southern Oscillation (ENSO) events, lower pressure is evident in the Southern Ocean, leading to cooler sea surface temperatures (SSTs) and greater sea-ice extent (SIE) (Kwok & Comiso 2002). In the Ross Sea, phytoplankton blooms occurred later and were less extensive following winters with high maximum sea-ice extent (Arrigo & van Dijken 2004). Summer sea-ice extent determines the amount of open water available for phytoplankton blooms and therefore may determine foraging success of Weddell seals during the summer. Because of this relationship with marine primary productivity, annual measures of ENSO strength, SST, or seasonal measures of sea-ice extent in the Ross Sea region may explain annual variability in the magnitude of reproductive costs experienced by female Weddell seals in Erebus Bay. We expected that heavy sea-ice years (high ENSO index, low SST, and high sea-ice extent) would lead to decreased foraging success for female Weddell seals and would therefore be correlated with increased reproductive costs. Based on the principles of the prudent parent hypothesis (that long-lived species should incur costs to reproduction before they incur costs to their own survival) (Drent & Daan 1980; we predicted that climate and sea-ice fluctuations would induce greater variability in costs to reproductive probabilities than in survival costs.


study area and population

The Erebus Bay study area is located at the southern end of McMurdo Sound, Antarctica (−77·62° to −77·87°S, 166·3° to 167·0°E; see Cameron & Siniff 2004 for description and map of study area). Eight to 14 Weddell seal breeding colonies are located within this study area (Stirling 1969). Colonies are associated with tidal cracks that form in the sea ice each austral spring, creating areas where adult female seals haul out to have pups. Most of these cracks form along the coast of Ross Island or smaller offshore islands. Colony size ranges from a few animals up to 250 animals and varies among years. Marking and resighting of this population has occurred for 35 years. Over this period, both the proportion of the population that is marked, and the proportion of marked animals that are of known age have gradually increased (Cameron & Siniff 2004). Currently, approximately 80% of the seals (males and females) in this population are marked and over 80% of these marked individuals are of known age. Each year, 300–600 pups are born at colonies in Erebus Bay, and most females surviving to reproductive age return to breed in Erebus Bay (Cameron & Siniff 2004).

data collection

Each year from 1969 until the present, Weddell seal pups born within the Erebus Bay study area have been individually marked (usually within about 5 days of birth) with plastic livestock tags attached to the interdigital webbing of each rear flipper. From 1969 to 1981, the proportion of pups that were tagged varied, and from 1982 through the present all pups in the study area have been tagged. In addition, any seal with a broken or missing tag was retagged and untagged adults were tagged opportunistically. The majority of the tagging effort occurred from approximately 15 October to 15 November each year, during the peak of parturition when colonies were visited every 2–3 days to tag newborn pups (Stirling 1969). Beginning in early November, six to eight resighting surveys of the study area were carried out, usually separated by intervals of 3–5 days. Weddell seal pups do not go into the water until they are at least 8–10 days old (Stirling 1969), and no predators exist in the study area that can remove pups from the vicinity. Thus, pups that die at early ages remain available for detection and no breeding information is missed. For mother–pup pairs, the presence and tag number of a relative (mother or pup) was also recorded.

Annual climate and sea-ice covariates were obtained online through various resources. The annual ENSO value was an estimate of ENSO strength obtained from Golden Gate Weather Services (http://ggweather.com/enso/years.htm). Data from various sources [Western Regional Climate Center, National Ocean and Atmospheric Administration (NOAA) Climate Diagnostics Center, and NOAA Climate Prediction Center] were used to derive the ENSO score, which represented a consensus on the strength of El Niño or La Niña events each year. Monthly mean SSTs for 110 locations covering the region from 156° to 174°E and 60° to 80°S were obtained from NOAA (http://iridl.ldeo.columbia.edu/). The mean of these values was calculated for each year from 1984 to 2003, generating annual mean SST for the Eastern McMurdo Sound region. Monthly SIE for the Ross Sea Region was provided by the National Snow and Ice Data Center (http://nsidc.org) (Comiso 1990), and monthly values for September (sepSIE) and February (febSIE) were used in models to represent winter and summer (roughly maximum and minimum) sea-ice extent for each year from 1984 to 2003.

statistical methods

Seals included in this study were females tagged as pups in the Erebus Bay study area since 1979 and were thus of known age. Based on subsequent observations made from 1979 through 2004, we built encounter histories for all known-age females that were known to have begun reproduction (sighted with a pup) by 2004. To reduce the complexity of our analysis and eliminate the necessity for modelling age-specific juvenile survival rates, we included observations of seals only after first reproduction. The minimum breeding age for individuals in our data set was 4 (one seal was detected breeding at age 4). Because a seal born in 1980 began reproduction in 1984, the first year included in the encounter histories was 1984 and encounter histories ended in 2004. Seals were partitioned into three groups based on age at first reproduction. Estimated mean age at first reproduction for female Weddell seals was 7·6 (Hadley et al. 2006), thus seals first breeding at ages 6–8 were considered ‘average breeders’, while ‘early breeders’ were those that first reproduced at ages 4–5 and ‘late breeders’ at ages 9–14.

Multistate modelling (Arnason 1972, 1973; Hestbeck, Nichols & Malecki 1991; Brownie et al. 1993) allows investigators to account for variation in detection probability associated with breeding state and permits the estimation of survival and breeding probabilities specific to yearly breeding state. Here, we used multistate mark–resight models to examine differences in apparent survival (hereafter, survival) for adult female seals in breeding (B) and nonbreeding (N) states, and variation in transition rates between breeding and nonbreeding states for adult female seals. We required that a female be seen with a pup more than once during the breeding season in order to be classified as a breeder (state B) for that year. This criterion was chosen to avoid misclassification of nonbreeding seals with an unrelated pup lying nearby during a survey. For convenience and consistency with other literature, we used the term breeder to denote that an individual produced a pup, and thus, as used here, the term does not simply indicate that the seal engaged in breeding activity. We also evaluated whether it was necessary to consider an unobservable state to avoid possible estimation bias associated with any temporary emigration that may have occurred (see below).

To evaluate our predictions about reproductive costs, we developed a set of a priori models that included three types of parameters: survival (φ), sighting (p) and transition (ψ) probabilities (Supplementary material, Tables S1 and S2). Transition probabilities were conditioned on survival and thus defined as follows: ψbb represented the probability of a breeder in year t becoming a breeder again in year t + 1, given that the seal survived to year t + 1, and ψnb represented the probability of a nonbreeder in year t becoming a breeder in year t + 1, assuming survival to year t + 1. Presence of costs to survival would be indicated by lowered breeder vs. nonbreeder survival rates (φb < φn), and presence of costs to future reproduction would be suggested by lowered breeding probability for breeders in the subsequent year relative to nonbreeders (ψbb < ψnb). In both cases we estimated immediate costs of reproduction in the form of reductions to survival or breeding probability for the year following reproduction. These parameters (φr, pr and ψrs, where r is breeding state at time t and s is breeding state at time t + 1) were modelled as functions of various covariates to represent our hypotheses about reproductive costs (Supplementary material, Tables S1 and S2).

To determine the combination of covariates that best explained variation in φr, pr and ψrs, we fit a series of effects to each parameter sequentially while constraints on remaining parameters were held constant (Supplementary material, Tables S1 and S2). This process began with modelling various effects on p while a relatively complex structure (breeding state, group and year effects with interactions) was applied to φ and ψ. Models of p included five possible combinations of breeding state, age and year effects, and a quadratic effect of age. The use of the quadratic effect was based on the concept that presence in the study area may be determined by age, and evidence from a prior study showing that sighting probability increased with age to a point, then decreased (Cameron & Siniff 2004). Once the most appropriate structure for p was selected (in this case, only the effect of breeding state), we repeated the process for φ using the selected structure for p determined in the process above, and the relatively complex structure (breeding state, group and year effects with interactions) for ψ. Finally, various models of ψ were evaluated using the structures for p and φ selected in previous steps (Supplementary material, Table S1). To avoid bias resulting from the order in which we modelled φ and ψ, we conducted a second model selection process where p was modelled first, followed by ψ and finally φ (Supplementary material, Table S2).

Our specific predictions regarding reproductive costs were evaluated with models that used various combinations of breeding state, age, age at first reproduction, breeding experience, and year to explain variation in φ and ψ (Supplementary material, Tables S1 and S2). Higher estimates of φ and ψ for nonbreeders than for breeders would support our prediction of reproductive costs to both survival and future reproduction. Support for our prediction regarding the influence of breeding experience was evaluated by examining models that included the effect of breeding experience, thus partitioning seals into three groups: first-time breeders, experienced breeders and nonbreeders. We predicted larger differences between breeder and nonbreeder φ and ψ for inexperienced seals.

As the marked animals in our data set (only known-age seals, and therefore marked as pups) have aged, the mean age of seals in our sample increased with time. Thus, we evaluated models that included an interactive effect of age and year but did not consider models with additive effects of age and year because the two effects were strongly correlated. We did evaluate models that combined age effects with annual climate and sea-ice covariates in an attempt to separate the effects of age and environmental variation. Additionally, for top-ranked models that included a year effect, we created a subset of models that replaced the year effect with a set of annual climate and sea-ice covariates (ENSO, SST, sepSIE, febSIE) in an attempt to improve model fit and learn the mechanism underlying the observed annual variation. The covariates used were for the year prior to the year for which survival and breeding probabilities were estimated.

Multistate Cormack–Jolly–Seber (CJS) models were constructed and estimates of model parameters generated using program mark (White & Burnham 1999). In order to compare ψbb with ψnb and be able to constrain both of these parameters in our models, we changed the parameter definition setting in program mark to allow direct estimation of ψbb and ψnb, with ψnb and ψnn obtained by subtraction. We used the Akaike's Information Criteria (AICc; Akaike 1973), corrected for small sample bias (AICc; Hurvich & Tsai 1989), associated with each model to obtain the difference in AICc (ΔAICc) between the model in question and the model with the minimum AICc (Burnham & Anderson 1998). We also used normalized Akaike weights (wi, for each model i) as an index of relative plausibility of each model, and to obtain model-averaged estimates of state-specific probabilities (Burnham & Anderson 1998). To determine how well a general model fit the data, we implemented the recently developed goodness-of-fit (GOF) test for multistate models available in program u-care (Choquet et al. 2003). This program tests the fit of the JollyMove (JMV) model (Brownie et al. 1993), which assumes that survival, transition and encounter probabilities are solely time- and state-dependent (Choquet et al. 2003). The sum of multisite tests was performed for each of our three groups (early, average, and late breeders) independently and the highest (most conservative) estimated overdispersion coefficient (ĉ) was used to adjust model selection results in program mark and convert AICc values to quasi-AICc (QAICc) values (Burnham & Anderson 1998).

For our study population, it is known that tag loss occurs at a low rate (probability that a female seal retained at least one of two tags for 1 year ranged from 0·977 to 0·998; Cameron & Siniff 2004). Therefore, we used estimates of annual tag-retention rate (inline image) that were generated for this population by Cameron & Siniff (2004) to adjust inline image; Arnason & Mills 1981). This was necessary because if some animals lost both tags, φ̂i represented the product of the underlying survival rate and the tag retention rate (Nichols et al. 1992). Transition probabilities did not need to be corrected for tag loss as they are conditional on survival (Williams, Nichols & Conroy 2002). That is, they are the estimated rates of movement from one state to another, assuming that the animal had survived to the current year and was available in the study area to make the transition.

Some seals may have temporarily emigrated during our study: specifically, some marked females went undetected for one or more years and were subsequently seen in a later year. Further, estimates of survival rates obtained from the multistate modelling described above, which ignore temporary emigration, can be biased in the presence of some forms of temporary emigration (Kendall, Nichols & Hines 1997; Fujiwara & Caswell 2002; Kendall & Nichols 2002; Schaub et al. 2004). In particular, estimates of survival can be biased if the probability of temporary emigration depends on an animal's state during the previous occasion, i.e. Markovian (or nonrandom) temporary emigration (Kendall et al. 1997).

To test for potential bias in our estimates and the adequacy of our modelling approach, we evaluated a subset of multistate models that incorporated temporary emigration (TE) and contained three states: breeders present in the study area (B), nonbreeders present in the study area (N), and seals not present in the study area during the breeding season and therefore, unobservable (U). Such an approach was successfully implemented by Beauplet et al. (2006) in a similar modelling effort on multistate data for sub-Antarctic fur seals Arctocephalus tropicalis. Given that Weddell seals are strongly philopatric (Stirling 1969, 1974; Croxall & Hiby 1983; Cameron & Siniff 2004), we assumed that seals in the unobservable state were nonbreeders.

To estimate all parameters in a TE model, additional constraints on parameters are necessary beyond those required for CJS multistate analogues (Kendall & Nichols 2002). Thus, in all TE models, we constrained survival rates for seals in state U to be the same as those in state N. Sighting probabilities for animals in state U were fixed at 0. The specific TE models evaluated were structured to approximate the best-supported CJS multistate models (see Results) as closely as possible. Estimates from the TE models and CJS multistate models were then compared with emphasis on differences in survival rates and transition probabilities. Estimates obtained from TE models and CJS multistate models were remarkably similar: the average per cent relative bias of time-varying survival estimates from the CJS approach was 0·05%. Given that CJS multistate models provided estimates with little bias, and required fewer assumptions and parameter constraints than did TE models, we evaluated our predictions based on results of CJS multistate models.


goodness of fit and model selection

Between 1979 and 2004, 5051 females were tagged as pups and therefore of known age. Of these, 607 returned to breed at least once between 1984 and 2004 and were included in our analysis of reproductive costs. Forty females were considered early breeders (age 4 or 5), 462 were average breeders (first breeding at age 6, 7 or 8), and 105 were late breeders (first breeding at age 9 or older). State-specific sighting probabilities were estimated using the top-ranked model (Table 1) and were high for breeders (p̂ > 0·99, SE < 0·01) and nonbreeders (p̂ = 0·76, SE = 0·01) (Table 2). We were able to estimate all parameters of interest for a 17-year time period (1986–2003) (Table 2).

Table 1. Model selection results for models representing hypotheses about reproductive costs to survival and breeding probabilities of female Weddell seals in Erebus Bay, Antarctica
Model k Dev QAICc w i ΔQAICc
inline image 47 2605·27 2700·75 0·47 0·00
inline image 49 2603·15 2702·76 0·17 2·01
inline image 48 2605·23 2702·77 0·17 2·02
inline image 49 2604·53 2704·14 0·09 3·39
inline image 51 2602·21 2705·95 0·04 5·21
inline image 50 2604·51 2706·18 0·03 5·43
inline image 51 2604·16 2707·90 0·02 7·15
inline image 52 2604·13 2709·94 0·01 9·19
  • Note: Results are presented for all models < 10 QAICc units from top-ranked model after combining results from two model suites (Supplementary material, Tables S1 and S2). Model parameters are state-dependent survival (φr), sighting (pr), and transition (ψrs) probabilities where r and s are breeding states at times t and t + 1. Covariates are indicated with model subscripts and include a (age), be (breeding experience; modelled by constraining survival and breeding probabilities following first breeding to differ from all subsequent breeding events), g (group; determined by age at first reproduction: 4–5, 6–8 or 9–14), and t (year; representing annual variation). Annual environmental covariates included ENSO (El-Niño Southern Oscillation score), SST (sea-surface temperature), sepSIE and febSIE (winter and summer sea-ice extent). Results from model selection are included: k (number of parameters), Dev (model deviance), QAICc, wi (‘weight of evidence’ in favour of each model i), and ΔQAICc (difference in QAICc value from top model). *Indicates an interaction between designated variable and breeding state.
Table 2. Model averaged estimates of state-specific apparent survival (φb and φn), sighting probability (pb and pn), and breeding probability (ψbb and ψnb) for female Weddell seals in Erebus Bay, Antarctica (Table 1)
Year φb φn p b pn inline image inline image ψnb
1986 0·84 (0·17) 0·90 (0·19) > 0·99 (< 0·01) 0·76 (0·01) 0·36 (0·35) 0·57 (0·37) 0·56 (0·37)
1987 0·88 (0·09) 0·93 (0·10) > 0·99 (< 0·01) 0·76 (0·01) 0·38 (0·18) 0·60 (0·19) 0·59 (0·19)
1988 0·87 (0·06) 0·92 (0·07) > 0·99 (< 0·01) 0·76 (0·01) 0·44 (0·13) 0·65 (0·12) 0·64 (0·12)
1989 0·96 (0·04) 0·98 (0·04) > 0·99 (< 0·01) 0·76 (0·01) 0·57 (0·12) 0·76 (0·09) 0·75 (0·09)
1990 0·91 (0·04) 0·95 (0·04) > 0·99 (< 0·01) 0·76 (0·01) 0·51 (0·10) 0·72 (0·08) 0·71 (0·08)
1991 0·97 (0·03) 0·99 (0·02) > 0·99 (< 0·01) 0·76 (0·01) 0·45 (0·08) 0·66 (0·07) 0·65 (0·08)
1992 0·91 (0·03) 0·94 (0·04) > 0·99 (< 0·01) 0·76 (0·01) 0·59 (0·09) 0·78 (0·06) 0·77 (0·06)
1993 0·90 (0·03) 0·94 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·44 (0·08) 0·65 (0·07) 0·64 (0·07)
1994 0·92 (0·03) 0·95 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·34 (0·07) 0·56 (0·07) 0·55 (0·07)
1995 0·88 (0·03) 0·92 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·35 (0·06) 0·56 (0·06) 0·55 (0·07)
1996 0·90 (0·03) 0·94 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·60 (0·08) 0·78 (0·05) 0·77 (0·05)
1997 0·94 (0·02) 0·96 (0·02) > 0·99 (< 0·01) 0·76 (0·01) 0·59 (0·07) 0·78 (0·04) 0·77 (0·05)
1998 0·89 (0·02) 0·93 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·52 (0·07) 0·72 (0·05) 0·71 (0·05)
1999 0·91 (0·02) 0·94 (0·02) > 0·99 (< 0·01) 0·76 (0·01) 0·57 (0·06) 0·76 (0·04) 0·75 (0·05)
2000 0·93 (0·02) 0·95 (0·02) > 0·99 (< 0·01) 0·76 (0·01) 0·35 (0·07) 0·57 (0·05) 0·56 (0·05)
2001 0·87 (0·03) 0·92 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·47 (0·07) 0·69 (0·05) 0·68 (0·05)
2002 0·92 (0·03) 0·95 (0·03) > 0·99 (< 0·01) 0·76 (0·01) 0·28 (0·05) 0·49 (0·05) 0·48 (0·06)
Mean 0·91 (0·04) 0·94 (0·04) > 0·99 (0·00) 0·76 (0·00) 0·46 (0·10) 0·67 (0·09) 0·65 (0·09)
  • Note: Sample sizes for the first two occasions (1984, 1985) were too low for accurate estimation, and sighting and survival probabilities for the final interval (2003–04) were not individually identifiable. Standard errors associated with each estimate are in parentheses, except for means where standard deviations are presented in parentheses. Estimates of survival probability and standard errors were corrected for tag loss. Survival probabilities were the same for the various ages of first reproduction and levels of breeding experience, thus we have presented only the survival probabilities associated with average, experienced breeders. Breeding probabilities varied with breeding experience, but not with age of first reproduction, and estimates are presented for first-time and experienced breeders with average age of first reproduction.

Results of the GOF test for the JMV model, applied to each of the three groups independently, provided evidence of overdispersion for the average and late breeding groups (average breeders: ĉ = 2·63, χ2 = 242·03, d.f. = 92, P < 0·01; late breeders: ĉ = 1·22, χ2 = 47·76, d.f. = 39, P < 0·01), and underdispersion for early breeders (ĉ = 0·73, χ2 = 21·87, d.f. = 30, P = 0·86). To adjust conservatively for overdispersion and avoid over-fitting our models, we applied the highest ĉ value (2·63) to inflate variances on estimated parameters and to adjust AICc values prior to model selection and hypothesis evaluation.

Of the five different parameterizations for sighting probability that were considered, the model in which p varied with breeding state was most strongly supported by the data (Supplementary material, Table S1). No other model was within 1·75 QAICc units, thus we chose to apply only breeding state effects to p for all subsequent modelling. The only other model for sighting probability that was moderately well supported allowed p to vary with age as well as breeding state. However, sighting probabilities for breeders were still estimated to be p̂ > 0·99 (SE < 0·01) across all ages, so detectability of breeders was minimally influenced by our choice to use the simpler, more well-supported model without age variation for all subsequent modelling. Twenty-seven different combinations of covariates were used to model both φ and ψ, and when results from both model suites were combined, the data most strongly supported a model that included the effects of breeding state and year on φ, and effects of breeding state, breeding experience, and year on ψ (Table 1). This model (wi = 0·47, Table 1) was ranked highest in both the ‘survival-first’ and ‘breeding-probability-first’ model suites (Supplementary material, Tables S1 and S2), and was not improved by the substitution of environmental covariates for year effects (Table 3). However, we presented results from other models when they were < 10 QAICc units from the top model and were the only models addressing a specific prediction, and additionally presented model-averaged estimates of parameters (from the set of models without environmental covariates) in order to account for model selection uncertainty (Table 2).

Table 3. Model selection results for models representing hypotheses about influence of climate and sea-ice covariates on reproductive costs to survival and breeding probabilities of female Weddell seals in Erebus Bay, Antarctica
Model k Dev QAICc w i ΔQAICc
inline image 47 2605·27 2700·75 0·39 0·00
inline image 28 2644·45 2700·98 0·34 0·24
inline image 29 2643·57 2702·13 0·19 1·39
inline image 28 2648·71 2705·24 0·04 4·50
inline image 29 2648·69 2707·26 0·02 6·51
inline image 29 2650·47 2709·04 0·01 8·29
inline image 28 2652·57 2709·10 0·01 8·36
inline image 28 2653·14 2709·67 0·00 8·92
  • Results are presented for all models < 10 QAICc units from top-ranked model. Covariates are indicated with model subscripts and include; SST (sea-surface temperature); sepSIE and febSIE (winter and summer sea-ice extent). Results from model selection are included: k (number of parameters); Dev (model deviance); QAICc, wi (‘weight of evidence’ in favour of each model i); and ΔQAICc (difference in QAICc value from top model). *Indicates an interaction between designated variable and breeding state.

We were able to implement several TE models that closely mimicked the best-supported CJS multistate model to test for potential bias in our estimates. However, to avoid convergence problems with the TE model (Kendall & Nichols 2002), we had to constrain transition probabilities to be constant across years. The TE model that most closely resembled the best CJS model was structured as inline image, where the states were B, N and U. Based on results from this TE model, temporary emigration did occur and the rate did depend on an animal's state: first-time breeders had an estimated temporary emigration rate of 0·15 (SE = 0·02), whereas rates were (1) lower for experienced breeders (0·07, SE = 0·01) and nonbreeders present in the study area (0·04, SE = 0·01), and (2) higher for nonbreeders in the unobservable state (0·26, SE = 0·03). However, despite the presence of Markovian temporary emigration, estimates of survival rates and transition probabilities between the observable states were remarkably similar for the TE and CJS models. The average difference between TE and CJS estimates of φi for breeders was 0·001 (SE = 0·007) and for nonbreeders was < 0·001 (SE = 0·005). Further details of the comparisons of TE and CJS estimates are presented below.

evidence for reproductive costs to survival

The results strongly supported our prediction that breeding imposes an immediate cost to survival probability. In the top-ranked model, the coefficient estimating the effect of being a breeder on φ was negative and the confidence interval did not include zero (inline image= −0·48, SE = 0·15, CI = −0·79 to −0·19). For breeding females, mean annual survival probability (CJS model-averaging: 0·905, SD = 0·03; TE model: 0·904, SD = 0·03) was lower than mean annual survival probability for nonbreeders (CJS model-averaging: 0·942, SD = 0·02; TE model: 0·943, SD = 0·02) (Table 2). This important result was further supported by the fact that, in all models within 7 QAICc units from the top model, the effect of being a breeder on φ̂ was negative and the confidence interval did not include zero.

Our prediction that breeding experience would influence costs to survival was moderately supported by the third-ranked model in our set (Table 1). The point estimate for the effect of first-time breeding was negative (inline image = −0·06), suggesting a higher cost of reproduction to first-time breeders, but the confidence interval included zero (CI = −0·39–0·28) and thus, strong inference could not be drawn from this result. Furthermore, model-averaged estimates of survival probability did not differ between levels of breeding experience (Table 2). Our prediction that age of first reproduction would not strongly influence reproductive costs to survival was supported by the absence of an interaction between age at first reproduction and breeding state in models < 7 QAICc from the top-ranked model. After model-averaging, survival probabilities were the same across ages of first reproduction (Table 2).

We found that survival probabilities did vary substantially by year, but the magnitude of reproductive costs did not. For breeding females, model-averaged annual survival probability estimates (φ̂i, for each year i from 1986 to 2002) ranged from 0·84 (SE = 0·17) to 0·98 (SE = 0·03) and in nonbreeders φ̂i ranged from 0·90 (SE = 0·19) to 0·99 (SE = 0·02) (Table 2). In the subset of models substituting environmental covariates for year effects in the top-ranked model (Table 3), there was no evidence that the covariates we used explained the annual variation in survival probabilities.

evidence for reproductive costs to breeding probability

Two of our predictions regarding costs to future reproduction were strongly supported by results from the top-ranked models, providing evidence that costs were present and that they varied with breeding experience. Following model averaging, mean annual probability of breeding (inline image) for surviving first-time breeders from the previous year [inline image = 0·46, SD = 0·10 (TE model: inline image = 0·46, SE = 0·02)] was much lower than for experienced breeders [inline image = 0·67, SD = 0·09 (TE model: inline image = 0·66, SE = 0·01)]. Experienced breeders that had a pup in the previous year had breeding probabilities that were comparable with those of seals that did not produce a pup the previous year (inline image = 0·65, SD = 0·09) (Table 2). These results provided evidence that there was an immediate cost of reproduction to the subsequent year's breeding probability, but that this cost was only observed for first-time breeders.

As with survival, we had predicted that age at first reproduction would not influence the size of the reproductive costs to breeding probability. This prediction was supported by the absence of this covariate in the most-supported model. However, there was extremely weak support for the influence of age at first reproduction in a lower-ranked model, which included an interaction between group (age at first reproduction) and breeding state (Model 5, wi = 0·04, Table 1). According to results from this model, seals that were late breeders (began reproduction at age 9 or older) experienced a 39·1% reduction in breeding probability after first-time breeding, relative to average and early breeders, whose breeding probabilities lowered by 28·8% and 20·3%, respectively. These results offered weak support for our prediction that late breeders experience higher reproductive costs due to their inferiority relative to other individuals in the population, but confidence intervals on the effect sizes for age at first reproduction in this model included zero and thus, strong inference could not be drawn from this result. After model averaging, breeding probabilities were similar across ages of first reproduction.

We detected annual variation in breeding probabilities (Table 2) and all models < 10 QAICc units from the top model included an effect of year on breeding probabilities (Table 1). However, the amount of annual variation in breeding probabilities was not different for breeders and nonbreeders, thus reproductive costs to breeding probability did not appear to vary substantially year to year. After substituting environmental covariates for year effects in the top-ranked model, a model indicating that summer sea-ice extent influenced breeding probabilities received similar support (wi= 0·34 vs. wi = 0·39 for top-ranked model, Table 3). Breeding probabilities increased following years with low febSIE (inline image = −0·25, SE = 0·04, CI = −0·33 to −0·16). There was also moderate support (wi = 0·19) for the influence of febSIE on the magnitude of reproductive costs. Contrary to our prediction, reproductive costs to breeding probability did not increase following years with high summer sea-ice conditions. Breeding probabilities for both breeders and nonbreeders were negatively affected by high summer sea-ice extent, but surprisingly, breeders were less affected than nonbreeders. However, the coefficient for the interaction term describing this relationship was estimated with low precision (inline image= 0·14, SE = 0·09, CI = −0·04–0·32), and thus our ability for inference from this result is limited.


Our results for female Weddell seals strongly indicate the presence of reproductive costs to both survival and breeding probability. Although survival probabilities varied annually, nonbreeder survival was consistently approximately 3% higher than breeder survival (Table 2). This seemingly small difference in survival probability translates into a substantial difference in mean life span. The extreme case of a female that produced a pup every year would yield a mean life span of approximately 10·0 years (Mean Life Span = −1/ln(φ̂), where φ̂ = 0·905). In contrast, a female that was a nonbreeder every year had a mean survival probability of 0·942, and as a result, a longer mean life span of approximately 16·7 years. The reproductive cost to survival experienced by female Weddell seals may therefore have meaningful consequences for life span, and consequently for lifetime reproductive output. Weddell seals evidently follow the ‘flexible-strategy’ hypothesis (Reid 1987; Erikstad et al. 1997, 1998), investing in reproduction at an ongoing cost to their own survival, rather than maximizing survival rate and restricting breeding effort, as is usually predicted by life-history theory for long-lived animals (Charlesworth 1980). This may be partly attributed to the fact that they are capital breeders. Successfully producing a pup requires transferring a large proportion of body reserves to the pup while feeding only opportunistically during the lactation period, which typically lasts 6–7 weeks (Hill 1987; Testa, Hill & Siniff 1989). The extreme reduction in body mass resulting from fasting may be a primary reason for lowered survival probability following reproduction.

Reproductive costs to a Weddell seal female's breeding probability were only apparent for first-time breeders. Testa (1987) found a cost of pupping for Weddell seals equivalent to a 0·05 drop in the probability of pupping the following year but found that this cost was not evident for females over 7 years old. Because many seals breeding before age 7 are first-time breeders, our results lend support to Testa's (1987) finding and additionally suggest a mechanism explaining the absence of costs beyond age 7. Our finding of costs to breeding probability only for first-time breeders again contrasts with the ‘prudent parent’ strategy expected for long-lived species. Rather than minimizing reproductive costs to survival and experiencing costs chiefly to future reproduction, Weddell seals seem to endure ongoing costs of reproduction to their own survival, but only suffer reduced breeding probability after first reproduction. Once a seal was experienced, her annual breeding probability was similar whether or not she produced a pup the previous year. The fact that first-time breeders are 20–40% less likely to breed the subsequent year than experienced breeders (Table 2) may in part be attributed to breeding strategy, as with survival costs. As capital breeders, Weddell seal females must rely on fat reserves to produce milk for their pups. First-time breeders are primarily young, relatively small seals, and thus the reduction in body mass following successful weaning of a pup is likely to take a larger toll than for larger, experienced seals. As a result, first-time breeders may be much less likely to regain their own body mass and additionally store enough energy to produce a pup again in the following year. Similar cases of inexperienced individuals requiring a longer ‘recovery time’ following a breeding season have been found for numerous species (Wooller & Coulson 1977; Ollason & Dunnet 1988; Weimerskirch 1990; Viallefont et al. 1995).

An alternate explanation for apparent delayed reproduction is that females breeding for the first time in our study area have previously reproduced outside the study area. However, female Weddell seals are strongly philopatric (Stirling 1969, 1974; Croxall & Hiby 1983; Cameron & Siniff 2004) and seals born in Erebus Bay generally return there to breed. Estimates from unobservable-states modelling did indicate that some seals did temporarily emigrate. However, anecdotal evidence exists to support our assumption of minimal breeding outside the study area by females originally tagged within the study area. During extensive surveys from 1997 to 2000, between 368 and 567 adult females were sighted each year within 100 km of Erebus Bay (see Cameron & Siniff 2004, fig. 1 for survey locations). Of all females that were encountered off the study area, an average of only 0·4% were breeding females that were originally tagged inside the study area (M.F. Cameron, personal communication). We emphasize that this percentage is based on less-intensive survey work conducted throughout a larger area via aerial surveys and ground-truthing during 1997–2000. Given that female seals limit their movements away from their natal sites and therefore our study area, we expect that the occurrence of reproduction outside the study area has minimal impact on the estimates of survival probability, breeding probability and reproductive costs presented here.

The annual variation in reproductive costs that we expected to observe was not evident from results of our top-ranked model. However, models that included unique coefficients for the individual effects of each year on each breeding state required the addition of 20 parameters, and thus only substantial levels of year-to-year variation in reproductive costs could be detected in reduced parameter models. Models that used annual climate and sea-ice covariates required fewer parameters and provided evidence that sea-ice conditions may determine magnitude of reproductive costs to both survival and future reproduction. Although not as well supported as the model with unique coefficients for each year, these models revealed an interesting relationship between the amount of sea ice in the Ross Sea sector of the Southern Ocean and the magnitude of reproductive costs for Weddell seals, and supported our prediction that costs to breeding probability would be more tied to environmental variation than survival costs. We found strong evidence that breeding probabilities depended on extent of sea ice during the previous summer, supporting our prediction that heavy sea-ice conditions in summer would lead to decreased foraging success. A probable explanation is the reduced amount of open water available for phytoplankton blooms and the consequent reduction in maternal foraging success. In support of this explanation, Proffitt et al. (unpublished manuscript) demonstrated that foraging success of pregnant Weddell seals (reflected by weaning mass in the following year) increased during summers characterized by reduced sea-ice cover.

Reproductive costs have been measured in various ways. Reznick (1985) argues that unambiguously detecting a cost of reproduction requires genetic correlation or experimental manipulations to demonstrate a genetic basis for the inverse relationship or trade-off between two life-history traits. Our study, like other recent efforts (e.g. Beauplet et al. 2006), utilized phenotypic correlation (statistical association between life-history traits) and thus may not have accurately reflected genetic costs of reproduction. This is because of naturally occurring variation in reproductive effort and the possibility that environmental heterogeneity in the study system will lead to variation in life-history traits unrelated to reproductive effort. However, determining genetic costs and/or conducting experimental manipulations is often not possible or practical with large, wild animals. Thus, phenotypic correlations may often be the only method available to assess the ecological costs of reproduction (reproductive costs that take into account interaction with the environment instead of solely evolutionary trade-offs) as well as the consequences of these costs for population dynamics. Although a genetic basis for reproductive costs has not been established, studies such as ours may still provide novel results by estimating the overall consequences of current reproduction on survival and future reproduction (Reznick 1992; Festa-Bianchet, Gaillard & Jorgenson 1998).

In conclusion, our results show that female Weddell seals pay a cost for reproduction, in the form of reduced survival during years following a reproductive event, and reduced breeding probability following first reproduction. These costs may depend on age at first reproduction. Seals primiparous at younger ages may be high-quality individuals and as a result, may not experience reduced breeding probability following reproduction as do the presumably lower-quality seals primiparous at older ages. It will be revealing to explore further the possible role of individual heterogeneity suggested by the differential costs we observed for various ages at first reproduction. Mass data collected during an ongoing study of mass dynamics in the Erebus Bay Weddell seal population may be useful as a surrogate variable representing individual quality and could reveal the extent that reproductive costs are influenced by heterogeneity among individuals. This study detected moderate support for an interesting relationship between sea-ice extent and magnitude of reproductive costs to breeding probability. Stronger relationships between reproductive costs and environmental condition may exist but will require more study to determine the relevant resolution for annual environmental covariates. Moreover, Weddell seal vital rates may not be linked to a single climate variable, and may instead respond to some combination of factors. More detailed exploration of appropriate climate and sea-ice indices may elucidate such linkages.


We thank the many individuals who have worked on projects associated with the Erebus Bay Weddell seal population since the 1960s. The project was supported by the National Science Foundation, Division of Polar Programs (grant no. OPP-0225110 to R. A. Garrott, J. J. Rotella, and D. B. Siniff) and prior NSF grants to D. B. Siniff and J. W. Testa. Comments and insights from J. D. Nichols, D. B. Siniff, and two anonymous reviewers improved the analyses and earlier drafts of this manuscript. Sea-ice extent data was extracted by K. Proffitt. Animal handling protocol was approved by Montana State University's Animal Care and Use Committee (Protocol no. 1093).

      Journal list menu