The little auk is a high Arctic colonial seabird, weighing between 130 and 190 grams and laying a single-egg clutch, generally in early June on Svalbard with hatching occurring in early/mid-July (Stempniewicz, 2001; Ramírez et al., 2017). Chicks usually fledge when they are between 3 and 4 weeks of age (Gaston and Jones, 1998). Their life history is characterized by a high adult survival (Descamps et al., 2021) and delayed age at first breeding (Del Hoyo et al., 1996; Stempniewicz, 2001). On Svalbard, the population size is uncertain but may be as high as a million breeding pairs (Anker-Nilssen et al., 2015; Keslinka et al., 2019), and may be declining in some colonies (Descamps and Strøm, 2021). The little auk is a zooplanktivorous species, primarily utilizing the Arctic copepods, with C. glacialis generally representing its main prey on Svalbard during the breeding period (Stempniewicz, 2001).

From 2005 to 2020, a sample of adult little auks have been captured annually (see details in Table 1 ) in three different colonies during the chick-rearing period (Ariekammen colony in Hornsund, Bjørndalen colony in Isfjorden and Feiringfjellet colony in Kongsfjorden; Figure 1 ). Captures started every year when the first sign of chick feeding (i.e. birds with a full gular pouch) have been observed. These birds were captured either by hand in the nest chamber (Hornsund only) and/or using passive traps (noose carpets) deployed on rocks in the colony (all three fjords) or mist-nets (Hornsund only). Such captures with noose carpets or mist-nets could not target breeding birds specifically and it is possible that a few non-breeders (or failed breeders) were also captured. All these birds were ringed, and most were measured (total head length, to the nearest 0.1 mm) and weighed (to the nearest 5 g). Adults transport the food for their chick in a sublingual gular pouch and this food can thus be easily sampled at capture for further prey identification (details below). The diet results presented in this study thus correspond to the chick diet (which is likely the same as adult diet, see Fort et al., 2010).

In Hornsund and Isfjorden, a group of nests (64 and 24 nests/year on average, with a range of [10, 155] and [9, 40], at Hornsund and Isfjorden respectively; Table 1 ) was monitored two to three times a week every season from approximately mid- (Isfjorden) or late- (Hornsund) incubation until the chicks were 15 (Isfjorden) or 21 (Hornsund) days old. At each visit, the status of the nest (egg, chick or empty) was recorded, and for a sample of nests at Hornsund, the chicks were weighed every 3 days.

Diet samples were analysed according to the procedures described in Harris et al. (2000). Organisms were enumerated and identified from subsamples until at least 300 individuals were counted. All Calanus specimens from copepodite stage (C1) to adult females (AF) were measured and identified on the basis of their prosome lengths, using size distribution tables that are established for Svalbard waters (Kwasniewski et al., 2003). Prey species were characterized as Arctic or Atlantic following the classification in Hop et al. (2019) and the total proportion of Atlantic and Arctic prey was calculated for each diet sample. A few species could not be clearly categorized as Arctic or Atlantic, but these species represented only approximately 5% of the little auk chick diet. Each diet sample was thus characterized by three proportions (i.e., % Arctic species, % Atlantic species and % species with undetermined distribution).

The Atlantic Water inflow was quantified as the percent contribution of Atlantic Water following the procedure introduced in Hop et al. (2019). Atlantic Water (AW) was defined as water with practical salinity at least 34.65 and potential temperature ≥ 3°C following the definition from Cottier et al. (2005). Temperature-salinity data collected using onboard CTDs (i.e. instruments measuring conductivity, temperature and depth) in July-September were extracted from multiple sources. The UNIS Hydrographic Database was used as the main source for data up to 2019 (described in Skogseth et al., 2020). This dataset was complemented with data from the Norwegian Polar Institute, Institute of Oceanology Polish Academy of Sciences, Joint Norwegian-Russian Ecosystem Survey for Barents Sea (Eriksen et al., 2018), the University Centre in Svalbard and UiT the Arctic University of Norway for 2019-2020. In total, these sources provided 945 (59.1 per year on average), 1156 (72.2), and 1085 (77.5) CTD casts to estimate an Atlantic Water Index (AWI) in the main basins of Kongsfjorden, Isfjorden and Hornsund respectively, without side fjords and ending at the fjord mouth. Only one CTD cast from the same position, measured using the available coordinates, per day was included in the AWI estimation. Even though little auks do not always forage inside the fjords during chick rearing, they are regularly observed there (at least in Hornsund, Stempniewicz et al., 2021a; Stempniewicz et al., 2021b) and generally do short trips to find food for their offspring and use areas located <50 km away from the colony (e.g., Welcker et al., 2009; Jakubas et al., 2020). We thus expect the within-fjord AWI to be representative of the level of “Atlantification” in the foraging areas used by breeding little auks for finding food for their chicks.

The Atlantic Water Index (AWI) was quantified as the average proportion of AW in CTD cast s between July and September as:

Annual AWI for a year y and region r was calculated as the average of all casts for the given year and fjord:

Where n is the number of stations per year (y) and fjord (r). Casts to shallower depths than 100 m were excluded to reduce the bias surface waters would cause to the AWI. Data between July and September were used because these months are oceanographically comparable in the studied fjords (Cottier et al., 2005; Skogseth et al., 2020).

All data were analysed with R software version 4.1.1 (R Development Core Team, 2021). Except when specified otherwise, visual inspection of model residual distributions indicated no heteroscedasticity and no auto-correlation for all models considered.

For analysing the inter-fjord and inter-annual variations in the proportion of Atlantic prey in the little auk chick diet (n=1096 diet samples when considering only the years where the AWI could be estimated in all the three fjords), we used beta regression mixed models (with the glmmTMB function of package glmmTMB, Brooks et al. (2017)) with a beta distribution and a logit link function. Year was included as a random factor in all models. Since the models require the dependent variable to be in the open interval (0,1), data were transformed as y × ( n − 1 ) + 0.5 n where y is the dependent variable (proportion of Atlantic prey in each little auk diet sample) and n total number of diet samples available (Smithson and Verkuilen, 2006; Cribari-Neto and Zeileis, 2010). We adopted a model comparison approach based on the Akaike’s Information Criterion or AIC (Burnham and Anderson, 2002) to determine whether or not the % of Atlantic prey varied among fjords and/or varied (linearly) with the AWI. We also considered the Fjord × AWI interaction to test whether the relationships between % Atlantic prey and AWI varied among colonies. The most general model for this analysis was thus of the form: %Atlantic prey (in chick diet) ~ AWI + Fjord + AWI × Fjord +(1|Year).

Then, we tested for an effect of the proportion of Atlantic prey in the chick diet on adult body mass. We considered the mass of adults weighed between mid-July and mid-August, which corresponds to chick rearing period (n=738). We used a two-step approach: first, we considered the effect of the annual average proportions of Atlantic prey on individual body mass, and second, the effect of the proportion of Atlantic prey in individual diet samples on individual body mass. A significant effect of the proportion of Atlantic prey at both the inter-annual and inter-individual level would indicate that (i) seasons with high % Atlantic prey are associated with lower bird body mass on average, and that (ii) within a given year, individuals foraging more on Atlantic prey are characterized by a lower body mass. Such combined results would give support to a causal (i.e. mechanistic) relationship between the % Atlantic prey and bird body mass. To test these relationships, we used linear mixed models (lmer function of package lme4, Bates et al. (2015)) with a Gaussian error distribution and a random effect of the year. We also adopted a model comparison approach based on AIC (using maximum-likelihood estimation as all models had the same random structure) to assess if the bird body mass varied among fjords and/or varied (linearly) with the % Atlantic prey. We also considered the Fjord × % Atlantic prey interaction to test if the relationships between % Atlantic prey and body mass varied among colonies. The body size (i.e. total head length of little auks) was included in all models so that results could be interpreted in terms of body condition (i.e. body mass independent of body size) as these two traits are significantly correlated in little auks (Wojczulanis-Jakubas et al., 2014a; Wojczulanis-Jakubas et al., 2014b). Preliminary analyses indicated that the date of capture had no effect on little auk mass (between mid-July and mid-August) and was thus not included in our models. The most general model for this analysis was thus of the form: Adult body mass ~ Body size (total head length) + % Atlantic prey (annual mean or individual level data) + Fjord + %Atlantic prey × Fjord + (1|Year).

We then tested for an effect of the proportion of Atlantic prey on little auk chick survival (Hornsund and Isfjorden) and growth rate (Hornsund). Data on chick survival and growth rate could not be linked to the specific diet brought by their parents and we thus only tested for a relationship at the inter-annual level (i.e. tested for an effect of the annual average proportion of Atlantic prey on chick survival and growth rate). For analysing the chick survival, we considered mixed models with a binomial error distribution (the dependent variable following a Bernoulli distribution) and a logit link function using the glmer function of package lme4 (Bates et al., 2015). In Hornsund, 39% of the monitored nests and 24% in Isfjorden had chick survival data for multiple years. However, this dependency (i.e. repeated measures of chick survival for some nests) did not affect our results; results were identical whatever the method used (e.g. randomly sampling only one measure per nest or including the nest identity, nested within fjord, as a random effect). Therefore, we only present results from simple mixed models with year as a random factor). We first considered the chick survival until 15 days of age (both fjords; n=1203) and then until 21 days of age (Hornsund only, n=695). For 15 day-survival analyses, we used AIC to compare models and to assess if the chick survival was affected by the chick diet and if this relationship was the same in both fjords. The most general model for this analysis was thus of the form: Chick survival to 15 days ~ % Atlantic prey (annual mean) + Fjord + %Atlantic prey × Fjord + (1|Year). For the 21 day-survival analysis, data were only available for Hornsund and we thus considered a single model Chick survival ~ %Atlantic prey+(1|Year).

The chick growth rates were calculated as the average gain in body mass per day. We only considered the first 16 days after hatching when the growth was linear (see Figure SM-1 ) and only chicks weighed at least twice during this period were included (n=405). Chick growth rates were analysed with a linear mixed model and a Gaussian error distribution. We used the lmer function (from the lme4 package), and included the year as a random factor. Data were available for one fjord only and we considered a single model Chick growth rate ~ %Atlantic prey + (1|Year).

For each analysis, we assessed the proportion of variance in the data explained by the different models as the squared correlation between the observed and predicted values (Zheng and Agresti, 2000). To represent the uncertainty around the predicted relationships between the different variables considered, we calculated the 95% bootstrapped confidence intervals of the predicted values using the percentile method (i.e., using the 0.025 and 0.975 quantiles of 1000 replicates).

The average proportion of Atlantic prey in the little auk chick diet was 0.17 (± 0.18 SD), 0.23 (± 0.19 SD) and 0.23 (± 0.21 SD) in Hornsund, Kongsfjorden and Isfjorden, respectively, when considering only the years where data were available for all three fjords (2006, 2007, 2011-2020). The Atlantic prey was mainly dominated by C. finmarchicus (stage CV; Table 2 ) while the Arctic prey was dominated by C. glacialis (stage CV; Table 2 ). The differences in Atlantic prey among fjords were significant (even though they explained only a small fraction of the variance in diet; see Table 3 ) and Atlantic prey species were less present in Hornsund little auk diets while the difference between Isfjorden and Kongsfjorden was not significant ( Table 3 ; results from model %Atlantic prey ~ Fjord with Isfjorden as the reference level: estimate of the difference between Kongsfjorden and Isfjorden on a logit scale: 0.11 ± 0.08 SE, z=1.39, p=0.16, estimate of the difference between Hornsund and Isfjorden on a logit scale: -0.20 ± 0.07 SE, z=-2.91, p=0.004).

The proportion of Atlantic prey species showed large inter-annual fluctuations and these fluctuations were highly synchronous between Isfjorden and Kongsfjorden (Pearson’s r=0.93, p<0.001), but not between Hornsund and Isfjorden or between Hornsund and Kongsfjorden (Pearson’s r<0.4 and p>0.2 in both cases; Figure 2 ). Even though inter-annual changes in the proportion of Atlantic prey were not synchronized between Hornsund and the other fjords, some years were characterized by very high (e.g. 2014) or very low (e.g. 2018) proportions of Atlantic prey in all three fjords ( Figure 2 ). The proportion of Atlantic species in the little auk chick diet has increased in all three fjords during the study period with no obvious difference in trend among fjords ( Figure SM-2 ; Table SM-1 ).

The inter-annual changes in the proportion of Atlantic prey in the little auk chick diet were positively associated with the AWI ( Table 3 ). This relationship varied among fjords and the slope of the AWI effect was steepest in Hornsund ( Table 3 ; Figure 2 ). These relationships were independent of the long-term trends in AWI and in the % Atlantic prey (i.e. they remained the same, whether or not we kept a Trend × Fjord interaction in the models; Table SM-2 ; FigureSM-2 ).

Body mass of adult little auks (controlled for body size) was similar among fjords but showed inter-annual variations that were negatively associated with the inter-annual fluctuations in the average proportion of Atlantic prey (slope of the effect: -6.29± 2.71 SE; Table 4 ; Figure 3A ). This relationship was also significant when an individual body mass was linked to the proportion of Atlantic prey in its own chick diet (slope of the effect: -5.61 ± 1.85 SE; Table 5 and Figure 3B ). However, even though the effect of the proportion of Atlantic prey was significant, it explained only a negligible proportion of the inter-annual variations in little auk adult body mass ( Tables 4 , 5 ) and had a small effect size (i.e., an increase of 50% in the proportion of Atlantic prey only led to a decline in body mass of approx. 3 grams meaning ~2% of little auk body mass; Figure 3 ). The effect of the proportion of Atlantic prey was similar in all three fjords (i.e., the best models did not include a %Atlantic prey × Fjord interaction; Tables 4 , 5 ) but appeared weaker in Hornsund ( Tables 4 , 5 ; Figure 3 ). The relationship between the proportion of Atlantic prey and little auk body mass was not confounded by any potential linear trend common in both variables (i.e. the effect of the proportion of Atlantic prey remained significant even if we included a Trend × Fjord interaction in the model; see Table SM-3 ).

The chick survival rate (until 15 days of age) averaged 0.92 ± 0.07 SD in Hornsund (2006-2020) and 0.84 ± 0.10 SD in Isfjorden (2005-2020) and the difference was significant ( Table 6 ). Overall, there was no clear effect of the proportion of Atlantic prey in the diet on chick survival in either of the fjords, but this effect was much stronger (and marginally significant) in Hornsund than in Isfjorden ( Table 6 ; Figure 4A ).

When considering the chick survival until 21 days (Hornsund only), the average chick survival dropped to 0.81 ± 0.25 SD and the effect of the proportion of Atlantic prey became highly significant even though it explained a small amount of variation in chick survival (z=-3.51, p=0.04, R² = 0.04, slope ± SE = -9.92 ± 4.95; Figure 4B ).

During the first 16 days of life (linear phase of growth), chicks gained an average of 4.6 grams per day at Hornsund in 2006-2020. This growth rate was negatively affected by the average proportion of Atlantic prey in the chick diet; the effect was only marginally significant (n=405, t=-1.89, p=0.08, R^{2 =} 0.25, slope ± SE = -11.53 ± 6.09; Figure 5 ).

The effects of the proportion of Atlantic prey on chick survival or growth rate at Hornsund were not confounded by any potential linear trend common in both variables (i.e. the effect of the proportion of Atlantic prey remained the same even if we include a Trend covariate in the model; see Table SM-4 ).

Our results confirm previous observations made in Svalbard fjords that were generally based on the comparison of two types of years (“warm” versus “cold”), or two types of fjords (one strongly influenced by the Atlantic Water inflow and one not). These earlier observations have shown that the proportion of C. glacialis in the chick diet is higher and the chick meals have higher energy content in Hornsund compared to Kongsfjorden (Karnovsky et al., 2011) and Magdalenefjorden, a fjord north of Kongsfjorden, largely influenced by Atlantic Water (Kwasniewski et al., 2010; Kidawa et al., 2015). Furthermore, the food loads delivered to the chicks are generally less energy-rich in Hornsund in warmer seasons (Jakubas et al., 2007; Kidawa et al., 2015; but see Wojczulanis-Jakubas et al., 2010) which in turn may be associated with a lower chick survival (Kidawa et al., 2015). All these results indicate that the little auk chick diet is more energy-rich when C. glacialis is more abundant (Kwasniewski et al., 2012). Chick diet composition of little auks from the main breeding colonies in Greenland and Svalbard shows that Atlantic prey, and more specifically C. finmarchicus, serve as a sub-optimal prey for little auks (Karnovsky et al., 2003; Frandsen et al., 2014) requiring a higher foraging effort from the parents (Karnovsky et al., 2010; Karnovsky et al., 2011). Our results support these findings and indicate that the adult little auks were in poorer condition when the Atlantic Water inflow was high, and that this may be a consequence of a higher foraging effort. Searching for the preferred prey, C. glacialis, among abundant but less favoured C. finmarchicus, may indeed require more time and energy (Jakubas et al., 2011; Karnovsky et al., 2011). Alternatively, the apparent negative effect of the Atlantic inflow on little auk body mass could be driven, at least partly, by changes in the proportion of non-breeders (including failed breeders) in our samples of captured birds. This would, however, assume that these birds have a lower body mass that breeders and that the proportion of non-breeders captured increased with a higher Atlantic inflow. We have no evidence for this and additional data on non-breeder body mass and on the probability to breed as a function of environmental conditions would be needed.

The potential increase in the little auk foraging effort when Atlantic waters, and thus Atlantic prey, are more prevalent is supported by the negative association between adult body mass and the proportion of Atlantic prey in the chick diet. However, even though this relationship was significant, the effect size was small and a large increase in the proportion of Atlantic prey in the chick diet only slightly decreased adult body mass. This indicates that, when foraging conditions deteriorate, even if there is a slight cost in terms of lower body condition, breeding little auks may prioritize their self-maintenance, and thus own condition, which could be at the cost of a lower chick growth and survival. This would represent the expected resource allocation strategy for such a long-lived species (Stearns, 1992) and fits with previous observations (Welcker et al., 2009; Kidawa et al., 2015). Therefore, little auks may increase their foraging effort to adjust to deteriorating environmental conditions (i.e. when the inflow of Atlantic Waters increases) but they would do it only until a certain extent that is unlikely to jeopardize their own survival. The fitness costs of the Atlantification of Svalbard may thus be mostly paid by the chicks. Our results from Hornsund fit such interpretation but additional data would be needed to confirm such potential costs, and to better understand the shapes of such relationships between Atlantic inflow, foraging behaviour and reproductive investment, that are likely complex.

The Atlantification is not only influencing Svalbard, but similar changes are occurring in the other parts of the Arctic (Ferguson et al., 2010; Møller and Nielsen, 2020; Polyakov et al., 2020; Ingvaldsen et al., 2021). Even though inter-annual variations have been very large, there has been a positive trend in the inflow of Atlantic Water into the Arctic in the last decades (Ingvaldsen et al., 2021). The little auk, an endemic Arctic species, is negatively affected by such Atlantification and even if they can be plastic and to some extent adapt their behaviour to new environmental conditions (e.g., Grémillet et al., 2012; Gremillet et al., 2015; Jakubas et al., 2017), they may still pay important fitness costs when the marine environment and associated prey become more “boreal” (i.e. north temperate). This may ultimately lead to population decline and local extinctions. The southernmost little auk population breeding in Iceland was extirpated at the end of the 20th century, possibly as a consequence of both ocean warming (that has affected the abundance and distribution of little auk prey) and human harvest (Jakubas et al., 2022). However, our work also showed that a large part of the inter-annual variations in little auk chick diet, body mass or chick growth/survival could not be accounted for by our Atlantic Water inflow index. Two explanations can be proposed to explain this low explanatory power.

First, this index was defined in our study as the proportion of Atlantic waters inside the fjords. However, little auk may also use areas outside the fjords to forage where oceanographic conditions and the influence of Atlantic water masses may be different. Creating an AWI index based on the actual foraging areas of the little auks for each specific colony would require identifying these foraging areas every year (through GPS-tracking for example). Although possible, this would be a very challenging study.

Second, the Atlantic Water inflow is only one among many factors that have the potential to affect little auk foraging and life-history. Other environmental parameters like the sea-ice (Amélineau et al., 2016; Ramírez et al., 2017; Amélineau et al., 2019; Jakubas et al., 2020), or the weather conditions (Konarzewski and Taylor, 1989) also affect their foraging behaviour or chick growth rate. All these parameters may interact with each other and also have non-linear and complex effects. Disentangling their respective role should be a priority in future research, and this can only be achieved through the use of long-term data. Such data are essential to improve our understanding of the ongoing Atlantification, or borealization, and its consequences on Arctic endemic species like the little auk.