species and study site
The black-browed albatross is a medium-sized albatross (3–4 kg), inhabiting the Southern Ocean and breeding on sub-Antarctic islands during the austral summer (Marchant & Higgins 1990). This annual breeder arrives on colonies in September, laying occurs in late October, hatching in late December and fledging in April (Weimerskirch, Zotier & Jouventin 1989; Marchant & Higgins 1990). At the Kerguelen Islands – southern Indian Ocean (48·4° S, 68·4° E) the study colony of Cañon des Sourcils Noirs with 100–200 nests was monitored annually since the breeding season 1979–80 (thereafter 1980). During the breeding season, each nest was checked three times (during early incubation, at hatching and before fledging), pair members were identified and their breeding performance was determined. Each year, new individuals found in the study colony and all chicks were ringed with a stainless steel band. In summer, satellite tracking indicated that birds from this colony foraged on the northeast and southeast regions of the peri-insular Kerguelen shelf (Weimerskirch, Mougey & Hindermeyer 1997; Pinaud & Weimerskirch 2002), feeding on fish, squids (Cherel, Hobson & Weimerskirch 2000) and on offal from fisheries (Weimerskirch et al. 1988). In winter, birds migrate off southern Australia, as indicated by band recoveries (Weimerskirch et al. 1985) and stable isotopes analyses (Cherel et al. 2000).
estimation of demographic parameters
As the black-browed albatross is long-lived and start breeding late (average age at first breeding is 9·8 years, Weimerskirch, Clobert & Jouventin 1987), in order to obtain a sufficient number of sightings of birds whose age and breeding history were known, we selected the capture histories of breeders seen between 1992 and 2003 that were ringed as chicks from 1979. This represented 82 black-browed albatrosses. We did not consider any differences between sexes because the very small sexual dimorphism did not allow us to obtain enough sexed individuals.
The breeding success (BS) was estimated at the end of each reproductive season as the proportion of eggs that produced a fledgling. We did not choose to work with the age of individuals because in these long-lived species with delayed maturity the variance on the age of recruitment is very high. So, a given age-class gathers individuals with very different breeding histories. Thus, we considered experience, defined as the number of breeding events. Preliminary analyses on the definition of breeding experience (Table 1) highlighted that differentiating first-time breeders (thereafter called inexperienced birds) from individuals with at least one breeding attempt (called experienced birds) allows us to differentiate between young and prime-age adults. We considered that the first reproductive event was correctly detected, based on the high recapture probability of breeders (see Results). In addition, modelling survival for different experience-classes (Table 1, see Results) allowed us to state that the survival of first-time breeders differed from the survival of individuals breeding for at least a second time.
Table 1. Modelling experience on survival probability using different experience-classes for black-browed albatrosses at Kerguelen Islands
|Modelling experience for black-browed albatrosses|
|2||Exp (1, 2+)||3||342·444||657·481|| 0·000|
|3||Exp (1, 2, 3+)||4||341·799||658·881|| 1·400|
|4||Exp (1, 2, 3, 4+)||5||341·240||660·340|| 2·858|
|5||Exp (1, 2, 3, 4, 5+)||6||340·303||661·511|| 4·029|
|6||Exp (1, 2, 3, 4, 5, 6+)||7||339·027||662·314|| 4·833|
Adult survival probabilities were estimated with capture–mark–recapture (CMR) models, using the M-Surge software (Choquet et al. 2004). By taking into account the probability to detect birds (the probability that a ringed bird that is alive and in the study area at time t was seen at time t), this method allows us to obtain unbiased estimates of annual survival probabilities (Lebreton et al. 1992). First, a goodness-of-fit test (GOF) was performed using the U-Care software (Choquet et al. 2003) to test whether the data fitted the assumptions the models are based on. Two major violations of the assumptions are transience and trap dependence (respectively survival and recapture probability not independent of the first occasion of capture). If detected, these effects could be integrated in models by considering that survival as well as capture probability just after capture are different from all other occasions. In our case, the GOF was performed on the more complex model Φ (Exp * t) p (Exp * t), where both survival (Φ) and capture (p) probabilities are experience (Exp) and time (t) dependent. Experience was modelled using two classes where ‘Exp1’ stands for individuals that breed for the first time (i.e. inexperienced birds) and ‘Exp2’ stands for individuals that breed for the second time and more (i.e. experienced birds). Indeed, this model was an equivalent of the Cormack–Jolly–Seber model (CJS, written Φ (t) P (t)) on which we would have modelled a transient and a trap dependence effect (Lebreton et al. 1992). Thus, the GOF of the model Φ (Exp * t) P (Exp * t) was the sum of tests 3.Sm and 2.Cl (Choquet et al. 2003).
The selection among time-dependent models was done using a second-order Akaike's Information Criterion (AICc; Burnham & Anderson 1998), following the parsimony principle between the deviance and the number of parameters. The model with the smallest AICc value was considered as the best model. The ability of two models to describe the data is assumed to be not different if the difference in their AICc is below than 2 (Lebreton et al. 1992). The selection among models with climatic covariates was done using the analysis of deviance test with a Fisher–Snedecor distribution (Anodev; Skalski, Hoffmann & Smith 1993) that evaluates the impact of a covariate by comparing the amount of deviance explained by the covariate against the amount of deviance not explained by this covariate (White & Burnham 1999). It is calculated as
where Dev and np are, respectively, the deviance and the number of parameters estimated for the constant model (M.), the model with climatic covariate (Mcov) and the time-dependent model (Mt).
Once the survival and the breeding success were estimated, we compared the variances between groups using a F-test. To compare the variance of the survival probability estimates, we had to break away from the intrinsic variance of the sampling. The temporal variation in survival probability was thus assessed using variance decomposition, which allows the separation of the sampling variance from the process variance (respectively the proportion of total variance due to the sampling error or to underlying processes, (Burnham & Anderson 2002). Process variance was computed using the Mark software (White & Burnham 1999).
In order to analyse the influence of environmental fluctuations on adult survival, we selected some ecologically pertinent indices to be incorporated in models. These variables were used in log-linear models to test for correlations with demographic traits. For survival, the percentage of variation that was explained by a covariate (r2) was estimated as: , where is the process variance of the model with covariate (cov) or the time-dependent model (t).
In most cases, climatic fluctuations are suspected to affect populations through an indirect mechanism, where climate should first affect the primary production, then integrated along the trophic web up to top predators. In marine ecosystems, meteorological and oceanographic data could be used as proxies of the biotic production because they are strongly involved in the vertical mixing of the water masses, controlling the quality of the physico-chemical environment available for the primary production (Wilson & Adamec 2002). To limit the number of covariates to test in models, monthly values of indices were averaged over a season of 4 months, consistent with the presence of birds on different sites: birds are at sea off Australia in austral winter (May–August), and present at the colony in spring (September–December) and in summer (January–April). Global indices were used to characterize environmental conditions over wintering areas as their distribution is not precisely known. All indices were standardized over the study period in order to allow comparisons.
The Southern Oscillation Index (SOI), reflecting the El Niño/La Niña oscillations through changes in sea-level pressure over the tropical Pacific sector, has been selected in relation to albatross migrations. As this index reflects oceanographic conditions between Chile and Australia, we consider a direct spatio-temporal effect, with no lag, which coincides with the presence of birds during winter off southern Australia. Seasonal SOI was obtained from the Australian Bureau of Meteorology (ABM –http://www.bom.gov.au/climate/current/soihtm1.shtml). As distribution and foraging grounds are more precisely known during the summer period, local indices can be used. Pinaud et al. (2002) showed that Sea Surface Temperature Anomalies (SSTA) that occurred on foraging areas had an influence on black-browed albatrosses breeding success at Kerguelen Islands. Indeed, SSTA have been used in several studies as a proxy of the oceanographic conditions that may determine the development of the whole trophic web. Based on telemetry analyses (see ‘Species and study site’), we selected SSTA values over the eastern Kerguelen shelf defined as the main foraging ground for black-browed albatrosses (46·5–50·5° S, 69·0–72·0° E). Data were available from the Integrated Global Ocean Services System (Reynolds & Smith 1994; IGOSS –http://ingrid.ldeo.columbia.edu/(plotaxislength)540+def/SOURCES/.IGOSS/.nmc/ .Reyn_SmithOIv1/.monthly/.ssta/).