Fine-scale spatial segregation in a pelagic seabird driven by differential use of tidewater glacier fronts

In colonially breeding marine predators, individual movements and colonial segregation are influenced by seascape characteristics. Tidewater glacier fronts are important features of the Arctic seascape and are often described as foraging hotspots. Albeit their documented importance for wildlife, little is known about their structuring effect on Arctic predator movements and space use. In this study, we tested the hypothesis that tidewater glacier fronts can influence marine bird foraging patterns and drive spatial segregation among adjacent colonies. We analysed movements of black-legged kittiwakes (Rissa tridactyla) in a glacial fjord by tracking breeding individuals from five colonies. Although breeding kittiwakes were observed to travel up to ca. 280 km from the colony, individuals were more likely to use glacier fronts located closer to their colony and rarely used glacier fronts located farther away than 18 km. Such variation in the use of glacier fronts created fine-scale spatial segregation among the four closest (ca. 7 km distance on average) kittiwake colonies. Overall, our results support the hypothesis that spatially predictable foraging patches like glacier fronts can have strong structuring effects on predator movements and can modulate the magnitude of intercolonial spatial segregation in central-place foragers.


Effect of GPS on foraging behaviour
The GPSs were distributed in random order in the chick-rearing period of 2017. This design lets us explore the effect of a GPS's weight on kittiwake foraging behaviour. A total of 205 trips from 48 individuals were collected.
We tested whether the GPS's relative mass (% of bird mass) was modulating the distance travelled by the birds in their foraging trip. A positive relationship would indicate that birds would increase the distance flown with a heavier GPS. A negative relationship would imply that the bird remains close to the colony (and thus the fjord and associated glacier fronts) with heavier devices.
We used a mixed model with a Gaussian error structure, using the maximum distance flown per trip (km) as a function of the GPS's relative mass (%). Since the distribution of maximum distances was highly skewed, we log-transformed the variable to meet the assumption of homoscedasticity. We furthermore controlled for the sex of the bird and used the bird ID as a random factor. Models were compared using Akaike's information criterion (AIC). We used the R package NLME (Pinheiro et al. 2020) to perform the regressions and the package AICCMODAVG (Mazerolle 2020) for the AIC calculation.
Table S1.1. Estimates and approximate confidence intervals for the mixed models testing the effect of GPS relative mass (%) and sex on the (log-transformed) maximum distance flown by kittiwakes per trip. AIC estimates are based on models using the maximum likelihood (ML) procedure and model estimates by the restricted maximum likelihood (REML). 95% confidence intervals (in parentheses) for all estimates are also shown. These results suggest that the GPS's relative mass is not modulating the distance flown by the birds in their foraging trips (95% confidence interval of the predictor's estimate overlaps zero). It also indicates that the space used by the birds is likely not driven by the tracking device's weight.

Supplementary Information S2
Distribution of raw tracks for the five monitored colonies Figure

Home range sampling details using the autocorrelated kernel density estimator (AKDE)
In total, three different models have been tested for each individual, with both their isotropic (i.e., movement process independent of directions) and anisotropic (i.e., movement process dependent of directions) versions: (1) the IID, as assumed by conventional KDE, (2) Table S3.1 shows how many individuals per colony have been fitted with these models after model selection (i.e., based on Akaike's Information Criterion with correction for small sample sizes (AICc)).  Due to the low number of individuals sampled in some of the colonies (Table 1), we assessed the representativeness of each colony-level UD estimates using the bootstrap approach described in Lascelles et al. (2016) and implemented via the TRACK2KBA package (Beal et al. 2020). Briefly, the approach consists of monitoring core range (i.e., 50% isopleth) distribution changes as a function of each colony's sample size. This was achieved by calculating a colony-level inclusion index, which is computed using an iterative process: (1) by randomly selecting a subset of individuals from a colony ("selected individuals"), (2) averaging their individuals' UD, and (3) calculating the proportion of the trips of the "non-selected" individuals (i.e., inclusion index) that are overlapping the "selected individuals" mean UD. In other words, this inclusion index summarizes to which extent our sample size for each colony accounts for the variability of individual space use. We calculated the inclusion indices for every sample size (i.e., ni … ni-1) along 999 iterations. From the bootstrapped dataset, we used a nonlinear model fitted by least squares using the inclusion indices as response and the sample sizes as a predictor to estimate the horizontal asymptote of the function. The representativeness of each colony was then calculated as the average inclusion index (estimated via the nonlinear model) obtained from the maximum sample size (i.e., ni-1) divided by the asymptote. All colonies tracked in Kongsfjorden (Fuglehuken was not evaluated due to the very low sample size) had 84-95% of their core ranges covered by their relative sample sizes of ni-1 (Table S3.2, Fig. S3.1).  Figure S3.1. Nonlinear model fitted by least squares using the inclusion index as a function of the sample size and associated horizontal asymptote for the Observasjonsholmen colony in Kongsfjorden.

General proportions of glacier front use
For comparison between colonies, we calculated the seasonal average glacier front use per colony using the following steps: (1) we calculated the proportion of trips that used at least one glacier front for each individual tracked in the five colonies, (2) we calculated the colony's average use of glacier fronts using these individual proportions, but weighted values using the number of trips recorded per individual (Fig. S4.1). Averages' confidence intervals (95%) were computed by bootstrapping the individuals' proportion and associated weight along 999 iterations using the BOOT package (Canty & Ripley 2020). We reported the 2.5% and 97.5% percentiles.