- Top of page
- Material and methods
In the light of the predicted changes in climate as a consequence of global warming, it is a major concern how animal species will respond to altered meteorological and oceanographic conditions. Seabirds constitute a diverse group of marine top predators which have relatively low fecundity and high annual survival rates. In order to predict effects of climate change, it is a necessary precondition to first understand responses to naturally occurring climatic fluctuations. While the ecological effects of different large-scale climatic phenomena have received much attention in the recent past, the factors determining the responses of seabirds are still little understood. We analyze more than a hundred previously published time series of seabird offspring production and adult survival rates in the North Atlantic in order to detect climatic signals in this data base. As our analyses are phylogenetic-comparative, we are able to search for patterns across species. Using the correlation of these parameters with the North Atlantic Oscillation (NAO) as a measure of responsiveness to climatic variability, we find that effects of climate on either parameters considered are not more common than expected by chance. The magnitudes of the responsivenesses were entirely randomly distributed throughout the seabird phylogeny, but were not strongly related to the explanatory variables considered. However, some tendencies indicate that both life-history traits and feeding ecology may influence how seabirds respond to climatic variability. An explanation of those patterns based on life-history theory is given.
Seabirds are a group of marine top-predators with an extremely “slow” life history, which is to say that they have comparatively high survival and low fecundity (Bennett and Owens 2002). This makes them especially vulnerable from environmental changes because even small decreases in survival will potentially have huge effects on the life time reproductive success of individuals (Wooller et al. 1992), and on the prospects of whole populations and, ultimately, species. Seabirds furthermore offer some unique opportunities for the analysis of causal links: they share a marine environment, while at the same time exhibiting an amazing variation in trophic relations, behavioural traits, life-history tactics and geographic distributions (Weimerskirch 2002).
So far, a few studies have documented effects of NAO, which is a widely used proxy for climatic conditions in and around the North Atlantic, on breeding phenology, demographic rates and population size of seabirds (reviewed by Reid et al. 1999, Thompson and Grosbois 2002, Durant et al. 2004b). In most cases studied so far, the effect of NAO on seabirds is indirect, i.e. it is the effect of temperature on prey abundance and availability that seems to be the major link between large-scale climate indices and demographic responses. However, attempts to focus simultaneously on more than one species are still sparse. If we are to predict which species are most likely to respond to climatic variation, and what those responses will look like, we will have to generalize findings across species. Searching for lawful generalizations across species is exactly what nomothetic science is about (Hull 1999), and in biology the phylogenetic-comparative method is the prime tool for revealing such interspecific patterns (Mayr 1988, ch. 1; Martins and Hansen 1996; Sandvik 2001, ch. 6). To a certain degree, this approach involves a trade-off between generality and precision. For example, by generalizing findings across species, we do not know how well the results fit every single of them. Therefore certainly both approaches are not only legitimate but also necessary for a full understanding of, among other things, the climate response of seabirds.
Our approach was to carry out exploratory analyses (sensu Anderson et al. 2001) using existing time series on seabird demographic parameters. Our main goal was to search for the existence of patterns – or lawfulness – across species. We did this in a two-step process. First, we investigated the prevalence and magnitude of responses to climatic variability, as measured by the NAO index, in two demographic parameters, offspring production and adult survival. In the second step, we analyzed whether the responsiveness to climate was related to biological characteristics of the species, considering ecology and life-history as explanatory variables. As ecological variables we considered two measures of feeding range, viz. foraging distance and diving depth, because these parameters can be considered as proxies for the vulnerability of seabirds (Furness and Ainley 1984, Furness and Barrett 1991, Furness and Tasker 2000). The demographic variables considered were adult survival, age at maturity, clutch size and chick production. It can be derived from life-history theory (Wooller et al. 1992) that species should be the more reluctant to invest into any specific breeding attempt at the cost of future ones, the lower their fecundity and the higher their adult survival are (Erikstad et al. 1998). This is because, in long-lived species, adult survival is the life-history trait which has the highest elasticity, i.e. which is under the strongest selection pressure (Gadgil and Bossert 1970, Stearns 1992). Our expectation is therefore that the responsiveness to climate should decrease with increasing longevity.
Material and methods
- Top of page
- Material and methods
The time series used in our analyses were found in a literature search. All studies reporting chick production or adult survival of seabirds in the North Atlantic for at least four consecutive years were included. “Seabird” was defined as any species belonging to the Phalacrocoracidae (shags and cormorants), Sulidae (gannets and boobies), Procellariiformes (tubenoses), Alcidae (auks) or Lari (gulls and allies; the nomenclature follows Schreiber and Burger 2002b). Chick production was defined as the number of chicks fledged per breeding pair. The final data set contained 106 studies on 23 species, differing in length between 4 and 44 yr (see Table 1 for details). Tests were also conducted on two modified data sets, one restricted to data from the main water body of the North Atlantic Ocean (i.e. excluding data from the Baltic Sea, Kattegatt, Skagerak, Lake IJssel and White Sea), and one additionally constrained to high-quality data, i.e. to longer time series and to original studies that used more reliable sampling or estimation methods (Table 1). Results inferred from these modified data sets are not mentioned unless they differed from the standard data set. The studies used and species names are listed in Table S1 in the Appendix.
Table 1. Data sets used in the analyses and their characteristics.
| ||time series||species||min.||max.||mean|| |
|Standard data set|| || || || || ||Entire North Atlantic, irrespective of data quality|
|chick production||72||22||4||30||10|| |
|adult survival||34||17||4||44||12|| |
| || || || || || || |
|Restricted data set|| || || || || ||Excluding data from the Lake IJssel, Baltic Sea and White Sea|
|chick production||68||21||4||30||10|| |
|adult survival||33||16||4||44||11|| |
| || || || || || || |
|High quality data set|| || || || || ||As restricted data set, but time series length at least eight years; furthermore:|
|chick production||18||11||9||30||16|| – estimates based on more than one count only|
|adult survival||20||11||8||44||15|| – estimates derived from capture-recapture modeling only|
Responsiveness of chick production and adult survival to climatic variability was measured as the coefficient of determination, r2, obtained in Pearson's product–moment correlation analysis between the characters of interest and the extended winter NAO index. This index is the mean of the principal-component based NAO index values for the period December–March (Hurrell 1995), and was retrieved from Hurrell (2005). The reason to choose the extended winter index of NAO is that the signal:noise ratio of NAO is highest in winter (Barnston and Livezey 1987, Hurrell et al. 2003), and that many biological studies have found it to be the most ecologically relevant index (Ottersen et al. 2001).
Survival rates were logit-transformed before processing. In order to avoid spurious correlation, each climatic and demographic time series was individually pre-whitened prior to further analysis. “Pre-whitening” entailed substituting each time series with the residuals from its optimal autoregressive model, where the order of the autoregressive model required was inferred using Akaike's information criterion corrected for small sample sizes (AICC). As climatic effects do not have to be instantaneous, we considered time lags between zero and three years (i.e. NAO index values predating the demographic estimates by zero to three years). For comparison only, higher time lags were analysed using the same tests. The respective results are given in the Appendix only.
Because of the non-independence of biological species, combining data from more than one species requires phylogenetic-comparative analysis (Martins and Hansen 1996). We used Felsenstein's (1985) independent contrast method to accomplish that. Unfortunately, the phylogeny of seabirds is only incompletely known. Therefore, several parts of the phylogeny had to be represented by polytomies, most notably the information on relationships between the different seabird taxa within the Charadriiformes (i.e. auks, skuas, terns, gulls, and their respective allies) is partly contradictious and not yet easily reconciled (Mickevich and Parenti 1980, Sibley and Ahlquist 1990, Björklund 1994, Chu 1995, 1998, Ericson et al. 2003, Thomas et al. 2004). Rather than ignoring polytomies in our comparative analyses, we calculated all possible contrasts for each polytomy and used the mean value in subsequent analyses.
Phylogenies were available, at least with the resolution needed for our studies, for the Procellariidae (Viot et al. 1993, Nunn and Stanley 1998), the Alcidae (Strauch 1985, Moum et al. 1994, Friesen et al. 1996), the Laridae (Chu 1998, Crochet et al. 2000, 2002, Pons et al. 2005), and the basal branchings of the major seabird groups (Cracraft 1985, Sibley and Ahlquist 1990, Hedges and Sibley 1994, Siegel-Causey 1997).
As branch lengths were not reported in all studies just cited, and methodologies for estimating them were not always comparable among studies that did, we used approximations of branch lengths based upon the number of extant species belonging to each branch (Grafen 1989). Scaling by ?=0.25 the branch lengths thus obtained (cf. Grafen 1989), resulted in contrasts whose distribution was fairly close to normal (Garland et al. 1992).
Further problems arise when contrasts are zero in one variable, because the sign of the contrast in the other variable becomes arbitrary in such cases (cf. Garland et al. 1992). It has been suggested to resolve this problem by using the mean of both possible values (that is to say: zero; Jennings et al. 1999). However, we suspected that the presence of several artificially created data points in the origin could inflate the degrees of freedom, so we excluded those contrasts altogether. Regressions of independent contrasts were forced through the origin (Garland et al. 1992).
Analyses were repeated with species and populations rather than contrasts as data points. Those procedures increased the sample size, but ignored phylogenetic correlations. The results did not differ much and are not further discussed. In the analyses that were based on species and independent contrasts, each species was represented by the average climatic responsiveness of its populations, weighted by the length of the time series.
Tables reporting the results of many such test results suffer from inflated type I error rates. Bonferroni or Dunn-Šidák corrections are sometimes used to take this fact into account, however these tests are overly conservative (Moran 2003). We corrected for the occurrence of type I errors by means of a re-sampling procedure (bootstrap): 6000 NAO-like time series were created by randomly drawing values from the original series at random. The same tests were carried out with these randomised data, and the real findings were compared to the simulated ones. The overall probabilities reported are the proportion of simulated results that obtained the same effects as, or stronger effects than, the results observed. In the Appendix, both row-wise, column-wise and table-wise probabilities are reported. In the main text only the latter are given, as they fully correct for the number of tests. To account for the intercorrelation between the explanatory variables, we also considered the first principal component of the six explanatory variables plus body mass.
All statistical calculations were carried out in the R environment (R development core team 2005), using functions written by one of us (H.S.) to perform phylogenetic-comparative analyses.