Cold comfort: Arctic seabirds find refugia from climate change and potential competition in marginal ice zones and fjords

Climate change alters species distributions by shifting their fundamental niche in space through time. Such effects may be exacerbated by increased inter-specific competition if climate alters species dominance where competitor ranges overlap. This study used census data, telemetry and stable isotopes to examine the population and foraging ecology of a pair of Arctic and temperate congeners across an extensive zone of sympatry in Iceland, where sea temperatures varied substantially. The abundance of Arctic Brünnich’s guillemot Uria lomvia declined with sea temperature. Accessibility of refugia in cold water currents or fjords helped support higher numbers and reduce rates of population decline. Competition with temperate Common guillemots Uria aalge did not affect abundance, but similarities in foraging ecology were sufficient to cause competition when resources are limiting. Continued warming is likely to lead to further declines of Brünnich’s guillemot, with implications for conservation status and ecosystem services. Supplementary Information The online version contains supplementary material available at 10.1007/s13280-021-01650-7.

= ( 00 + 01z̄+ 10ż + + ) + ( 0 + ) Curran and Bauer (2011) where γ00 is the coefficient for the intercept (or grand mean), γ01 is a direct estimate of the between-site effect, and γ10 is a direct estimate of the withinsite effect. Linear mixed models (LMM), fitted in the R package nlme (Pinheiro et al. 2021), were used to estimate parameters and their uncertainty using an identity link and normal errors. We used normal errors rather than the Poisson errors typically used for counts as the data comprised large numbers with few zeros, and normal errors relaxed the assumption of variance equalling the mean. The proportion of BG within colonies was modelled in relation to minimum SST using the R package lmer (Bates et al. 2015) using a generalised LMM with a logit link and binomial errors to examine changes in relative abundance of the two species in relation to SST.

Adapted from Equation 12 in
We classed sites according to the sector of Iceland in which they are situated (SW, NW, N, NE and SE). Papey and Skrúður in the SE were thus combined for further analysis. These are broad regional classifications that are also associated with different water masses and SST (see Study Sites section in main article). We modelled the response variables (trip distance, SST in foraging segments and isotope ratios) using general least squares (gls, where random effects were absent from models) or LMMs (where included; Zuur et al. 2009) with an identity link and normal errors, fitted using nlme.
Explanatory fixed factors were species, colony and (for isotopes only) year, while random effects were individual (for trip distance and SST in foraging segments only). As there were missing site/species/year combinations for the stable isotope sampling, full factorial models could not be fitted, so each of the site-species-year combinations were expressed as levels of a single factor, which was fitted as a fixed effect in the model. As heteroscedasticity was evident among factor levels for all responses, we fitted identity variance structures to meet model assumptions and estimate differences in the variability among groups (the number of standard deviations relative to a reference level; SDr) according to Zuur et al. (2009). In the case of SST in foraging segments, serial autocorrelation was evident in the residuals, so an order-one auto-regressive term was fitted within individual (Zuur et al. 2009).
In all analyses, Akaike's Information Criterion, adjusted for small sample size (AICc, using the R package AICcmodavg; Mazerolle 2020), was used for model selection and diagnostic plots (of normality, kurtosis, outliers, homoscedasticity and autocorrelation) and R 2 were used to confirm model goodness of fit. For models with random effects, both marginal R 2 m (fixed effects alone) and the conditional R 2 c (fixed and random effects combined; Nakagawa and Schielzeth 2013) were calculated using the R package MuMln (Bartoń 2009). Tukey HSD was used to test differences between factor levels of interest (species, year and sites while controlling for each of the others) using the R package multComp (Hothorn et al. 2008).    Figure S1. Summary of timing and duration of individual tracks by species, site and stage of breeding season. Bars represent the period for which each individual was tracked and the numbers by the bars are the number of trips the bird made during the deployment.

Fig S2:
Derivation of SSTs for colonies according to their location, size (numbers of both guillemot species combined) and survey period for use in the models of abundance. Top figures show the relative number of birds (pie chart size) and proportion of BG (lighter segments) for all colonies in Iceland during the two census periods. Lower figures show inferred maximum foraging range from each colony estimated from population size. SST data were taken from the Copernicus Climate Change Service (see Supplement S1 for full details).