Thermal habitat of adult Atlantic salmon Salmo salar in a warming ocean

Funding information The study was funded by Alta Laksefiskeri Interessentskap, Research Council of Norway (project 280,308 SeaSalar) and Tromsø Research Foundation. Abstract The year-round thermal habitat at sea for adult Atlantic salmon Salmo salar (n = 49) from northern Norway was investigated using archival tags over a 10 year study period. During their ocean feeding migration, the fish spent 90% of the time in waters with temperatures from 1.6–8.4 C. Daily mean temperatures ranged from −0.5 to 12.9 C, with daily temperature variation up to 9.6 C. Fish experienced the coldest water during winter (November–March) and the greatest thermal range during the first summer at sea (July–August). Trends in sea-surface temperatures influenced the thermal habitat of salmon during late summer and autumn (August– October), with fish experiencing warmer temperatures in warmer years. This pattern was absent during winter (November–March), when daily mean temperatures ranged from 3.4–5.0 C, in both colder and warmer years. The observations of a constant thermal habitat during winter in both warmer and colder years, may suggest that the ocean distribution of salmon is flexible and that individual migration routes could shift as a response to spatiotemporal alterations of favourable prey fields and ocean temperatures.


| INTRODUCTION
Temperature has a strong influence on ectothermic fishes, as it affects their physiological performance, behaviour and spatial distribution (Fossheim et al., 2015;Graham & Harrod, 2009). For the anadromous Atlantic salmon Salmo salar L. 1758, temperature affects growth and survival both in fresh water and at sea (Nicola et al., 2018). The recent warming of the North Atlantic Ocean is linked to declines in salmon both in North America and Europe (Beaugrand & Reid, 2012;Mills et al., 2013), potentially causing a northward expansion of their ocean range (Jensen et al., 2014;Nielsen et al., 2013).
While extensive information exists of the thermal niche of salmon in freshwater (Elliott & Elliott, 2010;Jonsson et al., 2001), knowledge of their thermal habitat at sea is less known. Sampling of salmon in the North Atlantic Ocean has documented that post-smolts inhabit diverse thermal habitats during the ocean feeding migration, mainly ranging between 4 and 12 C (Holm et al., 2000;Reddin & Friedland, 1993;Sheehan et al., 2012). In studies using archival tags, maiden salmon from Iceland spent most of the time in waters ranging from 7-11 C (Guðjónsson et al., 2015), partly contrasting the thermal habitat of adults from Atlantic Canada that primarily inhabited waters with temperatures from 2-15 C during summer and autumn (Reddin et al., 2004;Strøm et al., 2017). The differences among studies suggests that the thermal habitat of salmon may depend on both geographical origin and life stage. However, as sampling at sea only provides snapshots of the thermal habitat and the data available from archival tags are either limited by tracking individuals for only parts of the marine phase or by small sample sizes, more information is needed in order to obtain a comprehensive understanding of the temperature individuals experience while at sea. The distribution of salmon also covers large geographic areas of the North Atlantic Ocean, with substantial variation in environmental conditions both between areas and between years. Hence, studies of the thermal habitat in different geographic areas and across multiple years would add to the overall understanding of the marine phase of Atlantic salmon.
Here, the thermal habitat of 49 adult salmon from northern Norway that spent c. 12 months at sea was quantified using archival tags.
Fish were tagged over a 10 year period, enabling a comparison among years. The aim was to provide a comprehensive description of the thermal habitat of adult salmon during the entire ocean migration by describing seasonal trends in experienced temperature and investigate how annual variation in thermal conditions in the Northeast Atlantic Ocean affected the thermal habitat experienced during their ocean feeding migration.

| MATERIALS AND METHODS
The care and use of experimental animals complied with Norwegian animal welfare laws, guidelines and policies as approved by Norwegian Animal Research Authority and Norwegian Food Safety Authority (permit reference number 15950).

| Tagging
Adult Atlantic salmon were sampled and tagged in the Alta River, northern Norway (70 N, 23.4 E), in May 2006 spawning the previous autumn. The Alta River discharges into the northern Norwegian Sea (Figure 1) and is a prized destination for recreational anglers due to its large-sized salmon. A total of 1044 salmon were tagged, with fork lengths (L F ) ranging from 51 to 121 cm (mean ± SD = 90 ± 13 cm). Fish were captured by angling and anaesthetised in an aqueous solution of 0.5 mL L −1 2-phenoxy ethanol prior to tagging. Tag type, tag sensors and temporal resolution of tag data varied among years (

| Data analyses
Of the 1044 tagged salmon, 49 fish were recaptured and killed with functioning tags after spending c. 1 year at sea (Table 1). At tagging, the recaptured fish had L F ranging from 76-110 cm (mean ± SD = 93.6 ± 6.6 cm). All the recaptured fish had left the river shortly

| Seasonal trends
Seasonal trends in temperature use were quantified by calculating the monthly probability density functions (PDF) of daily mean temperatures and by two sets of generalised additive mixed-effect models (GAMM) that included individual fish as random effects. In the first set of GAMMs, the seasonal change in daily mean temperature was investigated with day as the fixed smoothing term, daily mean depth and L F at tagging as continuous fixed effects and year of tagging as a factorial fixed effect (

| Oceanographic conditions
To investigate how temperature trends in the North Atlantic Ocean affected the thermal habitat of salmon, the relationship between seasurface temperature anomalies (SST anomalies) and daily mean tem- e., second order polynomial) to account for potential non-linear relationships.
T A B L E 1 Overview of the number of tagged Salmo salar (N), the number of retrieved tags (n), tag producers, attachment methods, and previous analyses of the data set All GAMMs and LMMs were corrected for temporal autocorrelation at the level of the individual fish, using a first-order autoregressive process that accounts for the immediately preceding value. Model selection was conducted using a backward selection strategy, where the optimal structure of the random effects was determined for the most complex model before fixed effects were removed in a sequential order until only explanatory model terms remained (Zuur et al., 2013). For selection of the parsimonious GAMMs, model terms were selected based on their significance, while for the LMMs Bayesian information criterion (BIC) values were used to quantify model fit.

| Seasonal trends
During the ocean migration, Atlantic salmon experienced temperatures ranging from −0.9 to 15.8 C and 90% of the recorded temperatures were between 1.6 and 8.4 C ( Figure 2). Individuals' minimum temperatures ranged from −0.9 to 3.0 C and maximum temperatures from 9.3 to 15.8 C. Daily mean temperatures experienced by the fish ranged from −0.5 to 12.9 C (mean ± SD = 5.4 ± 2.0 C). The monthly probability density functions of daily mean temperatures revealed a clear seasonal trend (Figure 3). Daily mean temperatures were highest in July, when the range from the first to the third quartile was 6.7-8.9 C, before they gradually decreased until November (Figure 3). From November until March, the daily mean temperatures were stable, ranging primarily from 3.4 to 5.0 C ( Figure 3). This seasonal pattern, with the lowest temperatures during winter, was confirmed by the GAMMs with daily mean temperatures as the response variable ( Figure 4). The parsimonious model  (Table 2).

| Oceanographic conditions
From August to March, the mean SST anomalies in the assumed distribution area ranged between −0.8 and 1.7 C (mean ± SD = 0.8 ± 0.4 C), with sea-surface temperatures varying both within and among years (Figure 1). SST anomalies had a significant effect on daily mean temperatures experienced by the fish during summer and T A B L E 2 Overview of the generalised additive mixed-effect models (GAMM) and linear mixed-effect models (LMM) used to quantify seasonal trends in thermal habitat and effects of sea-surface temperature anomalies on temperatures experienced by Salmo salar Model terms included in the most parsimonious models; R 2 , the adjusted R 2 for the GAMMs and the marginal R 2 for the LMMs, with the conditional R 2 given in the parenthesis.
autumn (August-October), with the parsimonious model including SST anomaly (β ± SE = 0.43 ± 0.08) and day (β ± SE = −0.04 ± 0.001) as fixed effects and random effects of individual fish on the model's intercept (Table 2). In contrast, an overall relationship between SST anomalies and daily mean temperatures experienced by the fish was absent during winter (November-March), with the random effect model providing the best fit to the data (Table 2).

| DISCUSSION
The ocean temperatures experienced by Atlantic salmon in this study, with most temperature recordings ranging from 1.6-8.4 C, documented residency in colder waters than indicated in previous studies, where salmon have primarily been recorded in waters with temperatures between 5 and 15 C (Table 3). For anadromous salmonids, thermal preference in the ocean has been shown to be sizedependent, with larger individuals inhabiting colder waters (Morita et al., 2010). While there was no effect of body size on temperatures experienced by individual fish in the present study, it is possible that the overall large size of the repeat ocean migrants may partially explain the lower temperatures observed here compared with previous studies of salmon post-smolts (Table 3) Mean temperature (°C) F I G U R E 3 Seasonal trends in daily mean temperatures for Salmo salar tagged with temperature-sensing archival tags (n = 49). (a)-(l) Monthly probability density functions of daily mean temperatures from July in the year of tagging to June in the year of return. , The area between the 1st and 3rd quartile (i.e., 50% of the observations); n, tagged fish sample size post-smolts to increased ocean temperatures during the first summer at sea (Friedland et al., 2009(Friedland et al., , 2014. While the current study only described the thermal habitat of adult fish, the relationship between available ocean temperatures and the experienced thermal habitat during summer and autumn may be representative also for first-time migrants. In contrast to the positive relationship between SST anomalies and daily mean temperatures experienced by the fish during summer and autumn, a correlation between temperatures in the assumed distribution range and individuals' thermal habitat was absent during winter. From November to March, the tagged salmon resided in areas with waters temperatures ranging between 3.4 and 5 C in both warmer and colder years, indicating a strong fidelity towards waters with these temperatures during large part of their ocean migration. Factors controlling habitat selection in salmon were not addressed here, but it is likely that the habitat use was influenced by prey availability instead of by water temperatures per se, or by a combination of these factors. During their ocean migration, the tagged fish experienced temperatures well within their critical limits (Elliott & Elliott, 2010), but far below temperatures that optimise growth at excess food rations at sea (Handeland et al., 2008). For ectothermic fishes, the temperatures at which somatic growth are highest gradually decline as food rations and food intake decrease (Jobling, 1997). Thus, the observed thermal habitat may be influenced by interactions between the spatiotemporal distribution of prey and ocean temperatures, with salmon inhabiting waters that enhance growth both in terms of maximising overlap with favourable prey fields and minimising metabolic cost during periods of low prey availability. Moreover, since evidence of population-specific thermal adaptations is lacking for salmon (Anttila et al., 2014;Jonsson et al., 2001), it is possible that a similar interaction affects the temperatures experienced by individuals across the North Atlantic Ocean and that differences in thermal habitat among populations utilising different areas at sea (Table 3)   Changes in ocean temperatures have led to shifts in the spatial distribution for a diversity of marine fish species (Cheung et al., 2015;Fossheim et al., 2015). Owing to the magnitude of the ongoing ocean warming, a similar trend may also present for anadromous salmonids that perform long-distant ocean feeding migrations (Nielsen et al., 2013). Based on recent catches of salmon in waters previously considered as north of the species' northernmost limit, a northward expansion of the ocean distribution has been suggested (Jensen et al., 2014). Compared with other marine fishes, where a northward displacement is well documented, detailed knowledge of the historical ocean distribution of salmon is limited, with most information originating from fisheries off the Faroe Islands and west Greenland (Jacobsen et al., 2012;Reddin et al., 2012) and sampling of post-smolts in the Norwegian and Labrador Sea (Holm et al., 2000;Reddin & Friedland, 1993).
Recent studies using archival tags have revealed novel aspects of individuals' spatial distribution (Hedger et al., 2017;Strøm et al., 2017Strøm et al., , 2018 and the suggestion of a northward expansion may therefore be an artefact due to lack of detailed historical information. However, the observations of a constant thermal habitat during winter in both warmer and colder years in the present study, may suggest that the ocean distribution of salmon is flexible and that individual migration routes could shift as a response to changes in the North Atlantic Ocean.
T A B L E 3 Overview of the thermal habitat of Salmo salar in various ocean areas derived from previously published studies and personal communication