Contrasting effects of climatic variability on the demography of a trans-equatorial migratory seabird



  1. Large-scale seasonal climatic indices, such as the North Atlantic Oscillation (NAO) index or the Southern Oscillation Index (SOI), account for major variations in weather and climate around the world and may influence population dynamics in many organisms. However, assessing the extent of climate impacts on species and their life-history traits requires reliable quantitative statistical approaches.

  2. We used a new analytical tool in mark–recapture, the multi-event modelling, to simultaneously assess the influence of climatic variation on multiple demographic parameters (i.e. adult survival, transient probability, reproductive skipping and nest dispersal) at two Mediterranean colonies of the Cory's shearwater Calonectris diomedea, a trans-equatorial migratory long-lived seabird. We also analysed the impact of climate in the breeding success at the two colonies.

  3. We found a clear temporal variation of survival for Cory's shearwaters, strongly associated to the large-scale SOI especially in one of the colonies (up to 66% of variance explained). Atlantic hurricane season is modulated by the SOI and coincides with shearwater migration to their wintering areas, directly affecting survival probabilities. However, the SOI was a better predictor of survival probabilities than the frequency of hurricanes; thus, we cannot discard an indirect additive effect of SOI via food availability. Accordingly, the proportion of transients was also correlated with SOI values, indicating higher costs of first reproduction (resulting in either mortality or permanent dispersal) when bad environmental conditions occurred during winter before reproduction.

  4. Breeding success was also affected by climatic factors, the NAO explaining c. 41% of variance, probably as a result of its effect in the timing of peak abundance of squid and small pelagics, the main prey for shearwaters. No climatic effect was found either on reproductive skipping or on nest dispersal.

  5. Contrarily to what we expect for a long-lived organism, large-scale climatic indexes had a more pronounced effect on survival and transient probabilities than on less sensitive fitness parameters such reproductive skipping or nest dispersal probabilities. The potential increase in hurricane frequency because of global warming may interact with other global change agents (such as incidental bycatch and predation by alien species) nowadays impacting shearwaters, affecting future viability of populations.


Climatic variations have an important role on the population dynamics of vertebrates through their influence on their demographic parameters such as survival and breeding success of individuals (Coulson et al. 2001; Stenseth et al. 2002). Large-scale seasonal indexes, such as the NAO (North Atlantic Oscillation) index or the SOI (Southern Oscillation Index), account for major variations in weather and climate around the world and have been shown to affect ecological patterns (e.g. geographical ranges, abundance, diversity) and processes (e.g. birth and death rates) in both marine and terrestrial ecosystems (e.g. Stenseth et al. 2002, 2003, 2004). The NAO was one of the first weather index known to be associated with multiple ecological processes in the Northern Hemisphere and affects environmental conditions in the North Atlantic and the European continent (Hurrell 1995). On the other hand, the Southern Oscillation is the atmospheric component of a single large-scale coupled interaction called the El Niño–Southern Oscillation (ENSO). The ENSO is the most prominent known source of inter-annual variability in weather and climate, showing stronger effects in the South Pacific regions, but also affecting weather variability all around the world (Stenseth et al. 2003). This is because of teleconnections, that is, the physical relationships that result from the dynamics of atmospheric and oceanic waves (e.g. Murphy et al. 2007). One of the stronger ENSO teleconnections is with global tropical cyclone activity in the North Atlantic, with more and fewer hurricanes than average in La Niña and El Niño years, respectively (Holland 2009). These climatic variations may also affect environmental conditions for the whole food chain by changing food quality or availability at all trophic levels (Planque & Taylor 1998; O'Brien et al. 2000). For example, there is a linkage between the SOI in the Pacific and coastal upwellings off West Africa in the Atlantic through an atmospheric teleconnection. While El Niño suppresses upwelling along the coast, La Niña increases it (Roy & Reason 2001; Rouault, Pohl & Penven 2010). All these environmental changes may specially influence population dynamics of marine organisms, such as seabirds, for which oceanographic conditions may have stronger effects than on terrestrial species (e.g. Frederiksen et al. 2008; Barbraud et al. 2011). Seabirds spend most of the time at sea, and they travel long distances in one season being subjected to potentially different conditions during the breeding and the non-breeding periods (both during migration and wintering areas).

Here, we exploit an analytical tool recently developed in mark–recapture modelling, namely the multi-event approach (Pradel 2005), to analyse whether climatic variation influences several demographic parameters of the seabird Cory's shearwater Calonectris diomedea at two colonies in the western Mediterranean. Such methodological approach allowed us not only to assess the potential association between climatic indexes and adult survival (e.g. Sandvik et al. 2005; Oro et al. 2010) or transience (Tavecchia et al. 2007) but also to analyse simultaneously their effects on other parameters usually difficult to estimate, such as the probabilities of reproductive skipping (i.e. to take a sabbatical year) and those of nest dispersal (Barbraud & Weimerskirch 2010; Sanz-Aguilar et al. 2011). The fact that Cory's shearwater is a seabird that carries out trans-equatorial Atlantic migration from their breeding colonies to their wintering sites (e.g. González-Solís et al. 2007; Dias et al. 2010) made such species a good candidate to test several hypotheses about the influence of global climatic indexes on most of its demographic parameters.

Biological hypotheses

Seabirds, as long-lived species, show extreme values in life-history traits, with very high adult survival, low fecundity and delayed sexual maturity (Weimerskirch 2002) and individual fitness depending primarily on adult survival (Roff 1992; Crone 2001). So we tested the life-history theory prediction that climate should influence more the trait less associated with fitness, that is, breeding success, whereas the most sensitive parameter, that is, adult survival, should be more buffered against that component of environmental stochasticity (Saether & Bakke 2000; Grosbois & Thompson 2005; Sandvik et al. 2005). This is the so-called environmental canalization of fitness components, higher for adult survival in long-lived organisms (Gaillard & Yoccoz 2003). Although its frequency and environmental drivers still remain unclear for most seabird species, intermittent breeding (or skipping) and transience should have an intermediate sensitivity to climatic variability (e.g. Weimerskirch, Jouventin & Stahl 1986; Bradley, Wooller & Skira 2000; Cubaynes et al. 2011). The same sensitivity should apply for dispersal, for which a much better knowledge on the environmental drivers (e.g. predation, disturbance and massive reproductive failure) is accumulated, especially in gulls and terns (e.g. Cam et al. 2004; Braby et al. 2012). So we tested the hypothesis that under harsh climatic conditions during wintering, birds would arrive in poor physical condition to the breeding sites, which would increase the probabilities of becoming a transient, either because they die after breeding (i.e. a cost of reproduction, see Sanz-Aguilar et al. 2008) or because they permanently disperse to another colony (Tavecchia et al. 2007, 2008). After such bad wintering conditions, individuals may also opt for a more conservative strategy, that is, to skip reproduction (Cubaynes et al. 2011), or if they widow or divorce, they can also disperse to alternative sites within the colony (i.e. nest dispersal, Kim et al. 2007). In summary, we expected that climatic indexes should have a more pronounced effect on breeding success than on transience, reproductive skipping and nest dispersal probabilities, whereas their influence on adult survival should be much lower.

What should be the influence of each of the climatic indexes? Since Cory's shearwaters winter in the Atlantic Ocean and mainly in its Southern areas (see González-Solís et al. 2007), NAO but specially SOI should have a direct influence on adult survival, and secondarily on transience, reproductive skipping and nest dispersal. Survival can be specially affected by hurricanes, and the Atlantic tropical hurricane season, mostly governed by the SOI, coincides with shearwaters migration to their wintering sites. On the other hand, reproduction occurs in the Mediterranean, and thus, breeding success should be more affected by the NAO than by the SOI.

Materials and methods

Study areas and population monitoring

Data were collected yearly during the breeding period (from May to September) at two Mediterranean colonies of Cory's shearwater, Pantaleu Islet (c. 200 pairs) in the Balearic Archipelago (39°34′ N, 2º21′ E, Spain), and Congreso Island (c. 850 pairs) in the Chafarinas Archipelago (35°11′ N, 2°26′ O, Spain) (see Fig. 1). Shearwaters breed in burrows under boulders or vegetation. Accessible nests were tagged and visited during the incubation and chick rearing period to record nest occupancy and breeding success. Breeding adults, 477 at Pantaleu, from 2001 to 2008, and 386 at Chafarinas, from 2000 to 2008, were captured and marked (or recaptured) with stainless bands with a unique alphanumeric code to allow identification. Capture–recapture data were obtained from May to July. Individuals were sexed based on morphometrical measures and calls (own data). From 2001 to 2009, annual breeding success was determined at the two colonies as the proportion of chicks that fledge from all nests with an egg laid. Note that population density at both colonies was stable during the study years (own data), so the potential influence of climate on demographic parameters of Cory's shearwaters cannot be confused with density dependence mechanisms (Tavecchia et al. 2007).

Figure 1.

Left panel: tracks of all tropical cyclones in the northern Atlantic Ocean between 1980 and 2005 (data from the National Hurricane Center, Atlantic hurricane tracks can be influenced by SOI, with more frequent and stronger hurricanes in La Niña years than in El Niño years. Cory's shearwaters colonies of Pantaleu and Chafarinas are shown by a star and a dot, respectively. Right panel: eight distinct trajectories of wintering migration of Cory's shearwaters breeding at Pantaleu Islet (each colour represents a single bird, from Oro et al. 2008).

Climatic data

Global indexes seem to capture better than local indexes the complex associations between weather and ecological processes (Hallett et al. 2004). Thus, we used the global indexes NAO and SOI available at the following sources and, respectively, to investigate the association between demographic parameters of shearwaters and climatic variation. For the NAO index, we used the extended annual winter NAO by averaging the winter (December–March) values, as it is a good indicator of environmental conditions in the North Atlantic and the European continent (Hurrell 1995). Positive values of NAO lead to windy and warmer conditions in the North Atlantic and Western Europe, whereas the Mediterranean basin experiences drought. On the other hand, negative values of NAO result in colder winters in Western Europe and wetter conditions in the Mediterranean. The density of small pelagic fish (the main prey together with squid for shearwaters) in the Mediterranean has been related to freshwater input (i.e. rainfall), but with 1 year time lag (Lloret et al. 2004). Thus, we also used the extended annual winter NAO (December–March), the year before (NAO t−1) to test for this effect. On the other hand, more hurricane activity in winter is observed in positive SOI phases (i.e. La Niña) and less during a negative SOI phase (i.e. El Niño) (e.g. Knutson et al. 2008). In our analysis, we used SOI annual mean values (January–December).

To specifically test whether frequency of hurricanes affect survival in these species, we also include the number of hurricanes occurred during the study period in the two Atlantic Ocean basins. Data were obtained from, and we only selected those hurricanes occurred from October to the end of the hurricanes season as shearwaters stay at their breeding colonies until October.

Multi-event model design

Multi-event models relate the true state of the individual with the observed event through a series of conditional probabilities (Pradel 2005). Here, we combined classical information on individual presence (in its previous nest/presence in a new different nest) with additional information on the nest inspection and occupancy in the case of unencountered birds to estimate simultaneously the probabilities of survival (both for newly marked and resident individuals), reproductive skipping, nest dispersal, recapture, nest inspection, nest occupancy and capture of the first and the second occupants of the nests where their previous occupants were not recaptured (see Sanz-Aguilar et al. 2011). Multi-event models were built separately for each data base using program E-SURGE 1.4.4 (Choquet, Rouan & Pradel 2009a). Note that dispersal between the two colonies has never been recorded using the mark–recapture monitoring program. The multi-event modelling distinguishes three basic types of parameters: the initial state probabilities, the transition probabilities and the event probabilities (Pradel 2005; Choquet, Rouan & Pradel 2009a). Models included four biological states: individual alive and breeding in the same nest as in the previous year, denoted AP; individual alive and breeding in a new different nest, denoted AN; individual alive and in reproductive skipping or sabbatical (i.e. not breeding at the colony), denoted AS; and individual dead, denoted D. Note that the last two states are not observable. The initial state in our models was always AP (see also Sanz-Aguilar et al. 2011). Transitions between states were modelled in a three-step approach: survival, reproductive skipping (conditional on survival) and nest dispersal (conditional on survival and breeding) (Appendix S1) to allow testing for covariates directly. Following previous analyses on the data, we constrained the probabilities of reproductive skipping to be equal among individuals breeding in a new nest and individuals breeding in their previous nest (Sanz-Aguilar et al. 2011) (Appendix S1).

In each capture–recapture occasion ‘t’, we considered seven possible events, noted from 0 to 6: breeding individual captured for the first time or recaptured breeding in its previous known nest (noted 6); individual recaptured breeding in a new different nest (noted 5); individual not recaptured at occasion ‘t’ and its last known nest has not been controlled (noted 4); individual not recaptured and its last known nest is found empty (noted 3); individual not recaptured and its last known nest is found occupied but no occupant is captured (noted 2); individual not recaptured and its last known nest is found occupied and one occupant is captured (noted 1); and individual not recaptured and its last known nest is found occupied and the two occupants are captured (noted 0). The structure for the seven events was modelled in a stepwise procedure (see specific matrix design in Appendix S1, see also Sanz-Aguilar et al. 2011). We modelled first the individual recapture probabilities, denoted ‘P’ (step 1), and then, for the individuals not encountered in the colony, we used information from the previous known nest that was occupied by the individual, and we modelled successively the probability of nest inspection, noted ‘I’ (step 2), probability of nest occupancy, noted ‘K’ (step 3), probability of capture of the first occupant in the last known nest of an unencountered individual, noted ‘F’ (step 4), and probability of capture of the second occupant in the last known nest of an unencountered individual, noted ‘S’ (step 5).

At present, goodness-of-fit tests for multi-event models are not available. Consequently, we re-coded (transformed) observations into single-state codes (individual captured ‘1’ or not captured ‘0’, which correspond to the 5–6 and 0–4 events, respectively, in the multi-event model). We assessed the goodness-of-fit (GOF) of the Cormark–Jolly–Seber model ‘CJS’ (Lebreton et al. 1992) for each sex group using program U-CARE 2.2.2 (Choquet et al. 2009b). The two main components of GOF were tests 3.SR and 2.CT, which assessed, respectively, the existence of a transient effect and some heterogeneity in recapture probabilities, in our case potentially caused by the reproductive skipping (see Choquet et al. 2009b; Pradel & Sanz-Aguilar 2012). We incorporated the effects observed when examining the test components in our multi-event models.

Model selection was based on the Akaike Information Criterion adjusted for small sample size (c) calculated as

display math

where dev represents the model deviance, K the number of separately identifiable parameters in the model and n the effective sample size (Burnham & Anderson 2002). Additionally, for each model j, we calculated the Akaike weights, w j , as an index of its relative plausibility (Burnham & Anderson 2002).

We began our model selection in the two data sets (Pantaleu and Chafarinas) by modelling the event probabilities. For the Pantaleu data set, we selected the structure retained in a previous analysis (Sanz-Aguilar et al. 2011), and for the Chafarinas data, we tested the effect of time and effort invested in monitoring (two levels: high effort from 2000 to 2007 and moderate effort in 2008) in recapture, inspection of nest, nest occupancy and capture of nest occupants probabilities. The model structure of the event probabilities with the lowest AICc was selected to model survival and between-state transition probabilities. We tested the effects of sex, time and climatic covariates (NAO, NAO t−1 and SOI) and frequency of hurricanes on local survival probabilities (i.e. a survival that cannot distinguish mortality from permanent emigration). Then, once selected the structure of survival that minimized the AICc, we tested the effect of sex, time, climatic covariates (NAO, NAO t−1 and SOI) and breeding success on reproductive skipping probabilities and then the effects of sex, time and the same climatic covariates on nest dispersal probabilities.

As information criteria (i.e. AICc) cannot be used to test whether the variability of a certain climatic covariate is statistically significant from that explained by the time-dependent model (see Grosbois et al. 2008), we performed an analysis of deviance with a Fisher–Snedecor distribution (ANODEV; Skalski, Hoffmann & Smith 1993) that compares deviances of the constant model, of the time-dependent model and of the model with one or more covariates. It is calculated as:

display math

where Dev is the deviance estimated for the constant model (M cnt ), the model with climatic covariate (M cov ) and the time-dependent model (M t ), and np is the number of estimable parameters of constant, temporal or covariate models. The percentage of variation that was explained by a covariate (r 2) was estimated as:

display math

Based on models that estimated separately the survival probabilities of newly marked birds Φ' from those of previously encountered birds Φ, both being time and colony dependent, we also calculated the annual proportion of transients at both colonies (denoted τ) as 1- Φ'/Φ, and we tested its potential relationship with relevant climatic indices (see Results). To calculate the 95% IC of this proportion, we used the Delta method (see for instance, Morgan 2000).

As nest dispersal has been related with pair loss and widowhood, we also tested for the potential association between annual nest dispersal and annual survival probabilities. We fit several linear regression models (linear, quadratic and exponential) for testing the best association between mean estimates of nest dispersal and survival.

Breeding success

We investigated inter-annual differences in breeding success (percentage of chicks that fledged from all eggs laid) at both colonies, and we assessed its association with the global climatic covariates (NAO, NAO t−1 and SOI). Because of a strong effect of rat predation observed in Chafarinas in 2001 (Igual et al. 2009), we discarded breeding success data in this year from following analyses. Over the study years, we monitored 305 nests (annual mean = 158, range 138–171) and 398 nests (annual mean = 223, range 196–268) in Pantaleu and Chafarinas, respectively. To do so, we carried out generalized linear mixed-effects models with the package lme4 in R (, with breeding success as the response variable, colony and year as explanatory factors, and NAO, NAO t−1 and SOI as covariates. We used ‘logit’ as function link and binomial errors, and we included nest as a random effect in all analyses to avoid pseudo-replication. At some point, we considered introducing breeding success directly in the CMR models, but this would have increased the complexity of the model to a prohibitive degree.


The overall test of goodness-of-fit of the CJS model was statistically significant in the two data sets: Pantaleu (inline image = 100·26, < 0·0001, Table S1) and Chafarinas (inline image = 76·01, = 0·005, Table S1). Part of the lack of fit was because of transient birds (Igual et al. 2009; Sanz-Aguilar et al. 2011), another to a sub-set of birds systematically missed for a row of years (see more details in Sanz-Aguilar et al. 2011). We thus (i) modified our model to include an effect of transients in survival and (ii) introduced a reproductive skipping state to treat the second phenomenon. We then calculated the remaining lack of fit of the adapted model by discarding the components of tests 3.SR and 2.CT (Pradel, Gimenez & Lebreton 2005). The goodness-of-fit test was no longer statistically significant.

Multi-event model selection at Pantaleu colony

We began model selection using the best structure of the event probabilities found in a previous analysis (Sanz-Aguilar et al. 2011). This model (model 1, Table S2A) considered: the probability of burrow inspection (I) and the probability of burrow occupancy (K) constant over time; the probabilities of recapture and capture of the first and the second occupants of burrows where their previous owner was not captured (noted by R, F and S, respectively) time dependent; the probabilities of reproductive skipping after breeding, reproductive skipping after reproductive skipping and nest dispersal constant; and constant survival probabilities for newly marked and resident birds. We tested the effect of transience (Pradel et al. 1997; model 2, Table S2A) and the effects of sex, time, climatic covariates and frequency of hurricanes in survival probabilities of both newly marked and resident birds (models 3–10, Table S2A). The model that minimized AICc value was model 9 (Table S2A), which included an additive effect of SOI on survival of newly marked and resident birds. SOI explained 66% of the temporal survival variability at Pantaleu (F1,6 = 9·63, = 0·02, see Fig. 2). We retained the structure of the best model for survival (model 9, Table S2A) to test the effects of sex, time, previous breeding success and climatic covariates on reproductive skipping (models 11–16, Table S2A). The model in which reproductive skipping probabilities were constant over time (model 9, Table S2A) showed a lower AICc value and was retained to test the effects of time, sex and climatic covariates on nest dispersal (models 17–20, Table S2A). The model in which nest dispersal probabilities varied between sexes (model 17, Table S2A) showed a slightly lower AICc value. None of the effects of the climatic covariates tested on reproductive skipping and nest dispersal were retained in the best models (Table S2A): NAO, NAO t−1, SOI and breeding success covariates explained only 9%, 1%, 6% and 2% of the temporal variability of reproductive skipping, respectively, while NAO and SOI covariates explained 5% and 3% of the temporal variability of nest dispersal probabilities, respectively (Table S2A).

Figure 2.

Annual survival probabilities (and 95% confidence intervals) for Cory's shearwaters in Pantaleu (open dots) and Chafarinas Islands (solid dots) colonies since 2002 to 2008 from the time-dependent models (Tables S2A and S2B, respectively). Annual SOI values for the same years are also shown in the inner graph to illustrate its negative association with adult survival.

Multi-event model selection at Chafarinas colony

We began our model selection by modelling the event probabilities. In this first step, to reduce the number of parameters in the model, we considered survival of newly marked and resident birds and transition probabilities to be constant over time. We tested the effects of time and monitoring effort on event probabilities (models 1–5, Table S2B). The best structure in terms of AICc included a constant parameter for the probabilities of burrow occupancy and the effect of the monitoring effort in the other event probabilities (model 4, Table S2B). We used this structure to examine the effect of transience (model 6, Table S2B), and sex, time, climatic covariates and frequency of hurricanes in survival probabilities of both transient and resident birds (models 7–14, Table S2B). The model that minimized the AICc value was model 10 (Table S2B), which included an additive annual variation of survival of both newly marked and resident birds. This model was tied in terms of AICc (ΔAICc = 0·42) with a model including the effect of SOI on survival (model 13, Table S2B, see Fig. 2), where the SOI covariate explained 26% of the temporal survival variability. We used the model structure of survival probabilities that minimized AICc value (model 10, Table S2B) to test the effects of sex, time and climatic covariates on transition probabilities (models 15–23, Table S2B). Model including sex effect on reproductive skipping (model 15, Table S2B) showed higher values of AICc than model 10 that did not include a sex effect on reproductive skipping (Table S2B). NAO and NAO t−1 covariates only explained 10% and 4% of the temporal variability of reproductive skipping probabilities, respectively (Table S2AB). SOI and breeding success covariates explained less than 1% of the temporal variability of reproductive skipping probabilities (Table S2B). We used the structure of survival and reproductive skipping probabilities that minimized AICc to test for an effect of time, sex and climatic covariates on nest dispersal probabilities (models 20–22, Table S2B). The model in which nest dispersal probabilities varied over time (model 22, Table S2B) showed a much lower AICc value and was clearly retained as the best model explaining our data.

The temporal variation in nest dispersal was negatively correlated to annual survival (R 2 = 0·862, exponential regression model F 1,6 = 38·334, = 0·001 see Fig. 3). Other alternative models (linear, quadratic) also resulted in large associations between mean nest dispersal and survival, but their fit was lower (results not shown),

Figure 3.

Influence of survival of resident adults of Cory's shearwater on the probability of nest dispersal at Chafarinas Islands during 2001–2009. Estimates are shown with their 95% CI. The best fit was obtained from the exponential regression model (see text), which is shown (solid line) together with its 95% CI (dashed lines).

Mean estimates of demographic parameters were very similar between colonies (Table 1), with slightly higher values for transients, nest dispersal and reproductive skipping probabilities at Pantaleu. The annual percentage in the number of transients at the two colonies, obtained from models 21 (Table S2A) and 25 (Table S2B), varied between 0% and 21%, and it increased with higher values of SOI especially at Pantaleu (F 1,5 = 11·733, = 0·019, Fig. 4). Nest dispersal was around 8% in Pantaleu and 6% in Chafarinas. Probability of reproductive skipping after breeding was estimated as 14% in Pantaleu and 10% in Chafarinas, and the probability of reproductive skipping after a reproductive skipping was much higher, being around 66% in Pantaleu and 54% in Chafarinas (Table 1). Thus, the probabilities of taking two consecutive sabbatical years were 0·09 and 0·05 in Pantaleu and Chafarinas, respectively.

Figure 4.

Proportion of transients (and 95% CI) estimated at Pantaleu (open dots) and Chafarinas (solid dots) over the study period and correlations (dashed and solid lines for Pantaleu and Chafarinas, respectively) with SOI annual values (

Table 1. Mean estimates (and 95% confidence intervals in parenthesis) of most demographic parameters of Cory's shearwaters at Pantaleu and Chafarinas Islands estimated simultaneously by multi-event capture–recapture modelling during the study (2000–2008). Estimates were extracted from model 2 and model 10 for Pantaleu (Table S2A), and model 4 for Chafarinas (Table S2B). Note that for parameters that change with time, we took the estimate from the constant model
Survival of newly marked birds0·83 (0·76–0·88)0·76 (0·70–0·81)
Survival of residents birds0·87 (0·83–0·91)0·88 (0·84–0·90)
Male probability of nest dispersal0·10 (0·07–0·13)0·06 (0·05–0·08)
Female probability of nest dispersal0·07 (0·05–0·10)
Reproductive skipping probability after breeding0·14 (0·10–0·19)0·10 (0·08–0·14)
Reproductive skipping probability after reproductive skipping0·66 (0·47–0·80)0·54 (0·36–0·71)

Climatic drivers of breeding success

In the absence of rats at Chafarinas, breeding success was not very variable (0·70, SE = 0·05, and 0·67, SE = 0·04, in Chafarinas and Pantaleu, respectively). The best model in terms of AIC explaining breeding success variation included NAO as climatic covariate (= −2·347, = 0·019; Table S3), with lower breeding success in years with positive NAO values equally for the two colonies. A statistical equivalent model (ΔAIC < 2) included not only a NAO effect (= −2·328, = 0·020) but also a colony effect (= −0·984, = 0·325; Table S3). Neither the model including NAO with 1 year lag nor the model with SOI as climatic covariate performed better at explaining breeding success variation (Table S3).


This study examines the effects of climatic variation on a range of demographic parameters of Cory's shearwater, a long-lived seabird that undertakes a trans-equatorial Atlantic migration. Using multi-event capture–recapture modelling, we simultaneously assessed such relationship on the probabilities of survival, transience, reproductive skipping and nest dispersal. We show that climatic indexes account for substantial variation in some demographic parameters: the SOI influenced survival and transient probability, while the NAO affected breeding success at the two colonies. An important amount of variability is still not explained in many parameters suggesting that other factors (climatic and non-climatic) are also playing an outstanding role in driving population dynamics in this species.

Temporal variation in survival

Unexpectedly, for a long-lived species, we found a clear temporal variation of survival for Cory's shearwaters; such variation was strongly associated to the large-scale SOI index in one of the study colonies (Pantaleu). Oro et al. (2010) also showed a temporal variation in survival for another long-lived seabird, the Blue-footed Booby Sula nebouxii, although they found that it was strongly associated to local climatic conditions rather than to global climatic indices. They suggested that survival is probably more associated to global climatic indices in species that migrate large distances during the non-breeding season, and this is the case of the Cory's shearwater, which travels over 25 000 km every year between the colonies and the wintering sites (González-Solís et al. 2007). Previous studies already found this association in Cory's shearwaters and other seabird migratory species (Brichetti, Foschi & Boano 2000; Nevoux, Weimerskirch & Barbraud 2007; Jenouvrier et al. 2009; Boano, Brichetti & Foschi 2010), suggesting that adult survival of long-lived species that perform long-range migrations may be more affected by weather patterns than previously thought. As Cory's shearwater is a top predator, the absence of a time lag in the SOI effect may indicate that survival may be decreased by a direct effect of bad weather during winter and not by a mediated effect through the food chain, as found by Sandvik et al. (2005) and Barbraud et al. (2008) in other seabirds species. Thus, as previously suggested (Brichetti, Foschi & Boano 2000; Boano, Brichetti & Foschi 2010), the annual differences in survival in this species can be mainly driven by different hurricane activity in the tropical Atlantic, affecting mortality during migration from breeding colonies to the wintering sites. Thus, the potential increase in hurricane frequency because of global warming, specially in the North Atlantic (e.g. Knutson et al. 2008; Mendelsohn et al. 2012) may negatively affect the dynamics and viability of Cory's shearwaters populations. However, when we directly included in the models the frequency of hurricanes, we observed that SOI effect was still a better predictor of survival probabilities in this species. This may be due either to a difficulty in selecting hurricanes that potentially affect shearwaters during migration or to a double effect of the SOI that would affect bird survival not only via direct mortality during bad climatic conditions but also via food availability. The link between SOI and survival was especially strong in one of the colonies (Pantaleu) explaining about 66% of the annual survival variability, whereas in the other colony (Chafarinas), the SOI effect was not so large. This may be due to slight different strategies adopted during migration between these two populations (Catry et al. 2011). A previous study in this species suggested that the proportion of birds that wintered in each Atlantic area may differ between breeding populations with individuals from Pantaleu being more prone to migrate via the North tropical Atlantic and wintering further north (González-Solís et al. 2007). However, considerable migratory connectivity among breeding populations has also been found in this species (González-Solís et al. 2007; Catry et al. 2011), thus more data on migratory strategies of individuals from those colonies are required to confirm this point. Additionally, some differences in survival between colonies may be due to different conditions at breeding sites (Davis, Nager & Furness 2005). The important amount of variability in survival still not explained by our models suggests that other non-climatic factors, such as long-line incidental bycatch, which is known to be an important cause of mortality during breeding (Barcelona et al. 2010; Laneri et al. 2010), are probably also driving population dynamics in this species.

Transient probabilities seem to be also affected by the SOI in this species at the two colonies. In years with high SOI values, we detected a high number of individuals (mostly first-time breeders) that never returned to the colony, either because they died or because they permanently dispersed to other breeding patches. This may be reflecting a higher cost of first reproduction in those particularly bad years (Barbraud & Weimerskirch 2005; Nevoux, Weimerskirch & Barbraud 2007; Sanz-Aguilar et al. 2008; Nevoux et al. 2010; Oro et al. 2010).

Temporal variation in reproductive skipping and nest dispersal

Although some studies propose recapture probability as a proxy of breeding propensity in birds (Jenouvrier, Barbraud & Weimerskirch 2003; Lee, Nur & Sydeman 2007), this may lead to erroneous estimations if the recapture or observation probability of breeders is not 1. Other studies have developed other approaches for dealing with this problem including secondary capture or extra information (Kendall & Nichols 1995; Converse et al. 2009; Kendall et al. 2009). To overcome this potential bias, we used here recent advances in capture–recapture modelling that allowed us to directly estimate the probability of reproductive skipping (Sanz-Aguilar et al. 2011). We found that probability of skipping in Cory's shearwater appears to be constant over time. On the contrary, a recent study in a tropical seabird species showed that local SST drove intermittent breeding (Cubaynes et al. 2011). We suggest that local conditions in the Mediterranean are not so variable as those in the tropics (affected by extreme oceanographic events such El Niño and the associated hurricanes). Their large foraging ability (e.g. Navarro & González-Solís 2009) would allow Cory's shearwaters to buffer against the relatively small variability in local conditions, and other factors such as individual intrinsic quality or age are probably more important in determining the probability of reproductive skipping (see also Bradley, Wooller & Skira 2000; Sanz-Aguilar et al. 2011). However, we cannot reject the possibility that our time series is not large enough to detect time variation in the probability of skipping, and longer time series should be required to confirm this hypothesis.

Results of nest dispersal from Chafarinas clearly showed that there was a strong temporal variation in that parameter; interestingly, such variation was highly and negatively correlated with survival (Fig. 2), indicating that most of the changes in nest may be linked to divorces or widowhood. Contrarily, in Pantaleu, we did not find a clear temporal variation in nest dispersal, and we detected a slight tendency for males to be more prone to change nest than females. However, nest dispersal model selection in Pantaleu should be taken with caution because some other models, as the temporal model or specially the constant model, were very close to the finally selected one.

Temporal variation in breeding success

As expected, breeding performance seemed to be more affected by the NAO than by the SOI (see Table S3). Annual breeding success in both colonies was influenced by winter NAO values, whereas we did not detect an effect of NAO of the year before or of the SOI. This may indicate that winter NAO would directly affect food availability, which tends to influence breeding performance of long-lived organisms (Oro, Bosch & Ruiz 1995). Previous studies on fish and cephalopods (the main food resource of Cory's shearwater, for example, Thibault, Bretagnolle & Rabouam 1997) found that the NAO has strong effects on the timing of the peak abundance of fish (Sims et al. 2004) and squid populations (Sims et al. 2001; Pierce et al. 2008). While growth and survival of cephalopods are highly sensitive to environmental conditions and individuals respond to environmental variation both ‘actively’ (e.g. by migrating to areas with more favourable environmental conditions for feeding or spawning) and ‘passively’ (passive migration with prevailing currents), the timing of fish migration may be also affected by the NAO, affecting annual migrations of marine fish to spawning grounds. However, we should keep in mind that in Cory's shearwater, a change of 38% in fecundity is needed to compensate a 2% increase in mortality (Igual et al. 2009). Thus, drivers affecting adult mortality in this species have a much greater effect on population dynamics than those driving breeding success.

In summary, we found temporal variation in all demographic parameters except in reproductive skipping that seems constant over time in both colonies. Temporal variation in survival and transience probability was mostly explained by the SOI, especially in one of the colonies (66% of variance explained by the index). Breeding success was also affected by climatic factors, the NAO explaining about 41% of the temporal variation in this parameter. The potential increase in hurricane frequency because of global warming (e.g. Knutson et al. 2008; Mendelsohn et al. 2012) may be additive to other drivers of global change nowadays acting on shearwaters (such as incidental fisheries bycatch and nest predation by invasive rats, see Igual et al. 2009) affecting the dynamics and viability of their populations.


We are very grateful to Giacomo Tavecchia for his invaluable advices. We thank three anonymous reviewers for providing helpful and constructive comments for improving the manuscript. We would also like to thank those people helping with the fieldwork over the years, particularly to Isabel Afán and Tomás Gómez. Permits were given by OAPN (Spanish Ministry of the Environment) and Dragonera NP (Balearic Regional Government). ASA was supported by a Marie Curie Fellowship (ref. MATERGLOBE). Research funds were provided by the Spanish Ministry of Science (grants ref. BOS2003-01960, CGL2006-04325/BOS and CGL2009-08298).