Location estimates were obtained via the Service Argos satellite based system, using their least squares algorithm for calculating positions. Locations were most often assigned a location quality class of 0 (28% of the location estimates) by Service Argos, while locations of class A, B and Z made up a total of 47% of all location estimates. The accuracy of classes 0 and B locations is estimated between 5 and 10 km, while the remaining classes are likely to be accurate to approximately 2 km (Boyd & Brightsmith 2013). Track data were filtered based on assumed maximum swim speeds and turning angles (Freitas et al. 2008) as detailed in McIntyre, Bornemann et al. (2011). While not quantified, we expected the filtering process to result in significantly higher accuracy locations allowing for reasonably large-scale associations of environmental data with the available behavioural data (Kuhn et al. 2009). Filtered tracks were illustrated using ArcGIS 10.1 software. A travelling speed (m/s) was assigned to each retained location, based on time and great circle distance differences between that point and the preceding location—calculated using the adehabitatLT package (Calenge 2006) in the R environment—and assuming that seals swam at consistent speeds between location points retained after applying the filter described above.

Estimates of individual dive locations were provided by the manufacturers (Sea Mammal Research Unit, University of St. Andrews) and based on interpolated locations from position estimates provided by Service Argos after removing erroneous locations based on estimated maximum swim speeds of elephant seals (McConnell et al. 1992). Individual dives were labelled as having either occurred during the day, at night or during twilight (periods within 30 min of the local sunrise and sunset times), based on local time values of individual dives and local times of sunrise and sunset, calculated using the maptools package in the R environment (Lewin-Koh & Bivand 2012).

Data of individual dives consisted of four abstracted time/depth points representing the maximum dive depth and three points of greatest inflection. These data points were identified (from high resolution time/depth profiles recorded at 4 s intervals) onboard the devices using a broken stick algorithm prior to transmission (see Fedak et al. 2002 for further details). Elephant seals are known to increase time spent during the bottom phases of dives when evidently encountering prey items, as measured by acceleration loggers (Gallon et al. 2013). In order to obtain reasonable estimates of bottom time (time spent within the deepest 20% of each dive), we generated three additional time/depth points between each transmitted time/depth point, assuming constant swimming speeds and angles of ascent/descent between transmitted points, as described by McIntyre et al. (2010). This resulted in a total number of 21 time-depth points associated with each dive. We then used multivariate linear regressions to quantify the relationship between maximum dive depth, dive duration and bottom time for each track (separately for day and night dives) (Bailleul et al. 2008). Residuals from the regression were then used to identify dives of increased “forage effort”, based on above-average amounts of time spent at the bottoms of dives for particular dive depths and durations. Bottom time residuals as an indicator of foraging effort was previously used by a number of authors studying both elephant and Weddell (Leptonychotes weddellii) seals (Bailleul et al. 2008; McIntyre, Ansorge et al. 2011; McIntyre, Bornemann et al. 2011; Heerah et al. 2013; McIntyre et al. 2013). While Dragon et al. (2012) reported inconclusive relationships between bottom time residuals and evident increased foraging (negative bottom time residuals for deeper “active” dives and positive bottom time residuals for shallower “active” dives) in southern elephant seals, the results reported by Gallon et al. (2013) and Robinson et al. (2010) lend further support to this method.

Seafloor depth estimates for each dive location were extracted from the ETOPO1 1 arc-minute global relief model from the National Oceanic and Atmospheric Administration's National Geophysical Data Centre. Sea-ice concentrations for each location estimate were obtained from daily Advanced Microwave Scanning Radiometer (AMSR-E) imagery at a spatial resolution of 6.25 km (Spreen et al. 2008). In situ temperature records were recorded and transmitted by the CTD-SRDLs. Profiles were inspected visually using the Ocean Data View software package (Schlitzer 2002) for unrealistic temperature values (≤2.5°C or >7°C). No unrealistic values were evident in the data set and all transmitted temperature profiles were used for further analyses. For each of the temperature profiles, we calculated the maximum temperature value within the profile (T _max ), the depth of T _max (T _max . _depth ) and the difference between the maximum and minimum temperatures recorded (T _diff ).

We used a series of linear mixed effects models to identify the relative influences of environmental variables on dive depth (DDEP), dive duration (DDUR), travel speed (SP) and relative amount of time spent at the bottom of dives, as indicated by bottom time residuals (BTres). Transmitted dive and temperature profiles do not necessarily correspond in time or space due to the method of temporary storage of profiles onboard the tags prior to transmission (Boehme et al. 2009), and are limited to a maximum number of four profiles per day. In situ temperature characteristics were therefore attributed to transmitted dive profiles and speed locations, based on the closest CTD profile received in time. Dive profiles and location points with no associated CTD profiles successfully retrieved within a time period of 24 hrs around the location or dive profile were discarded for the mixed model analyses (but retained for descriptive statistics; see Table 1). Time differences between dive profiles and CTD profiles averaged 319±374 min and between speed location points and CTD profiles 254±338 min.

Our starting full models were: P a r ~ J D a y + S F d e p t h + T max + T max . d e p + T d i f f + I C + r a n d o m = i s e a l , where Par = either DDEP, DDUR, SP or BTres; JDay=Julian day; SF _depth =seafloor depth (m); T _max =maximum temperature (°C); T _max . _dep =depth of T _max (m); T _diff =temperature difference between maximum and minimum temperature (°C); IC=ice concentration (%); and i _seal =individual seal (random term).

Initial models consisted of all fixed effects. Since models mostly displayed significant temporal autocorrelation, we fitted autoregressive correlation functions (Pinheiro & Bates 2004). All possible combinations of fixed variables were then compared in order to select the most parsimonious models. Model selection was undertaken based on maximum likelihood and using second-order Akaike Information Criterion (AICc) and corresponding AIC weights to select the most parsimonious models (Burnham & Anderson 2002). After model selection, final models were run using restricted maximum likelihood (Bolker et al. 2008). We used a number of different plot types to assess model fits (Pinheiro & Bates 2004). Variance component analyses were carried out on all final models to estimate variation explained by the random term.

All analyses were undertaken in the R statistical environment (version 2.15.2; R Core Team 2012). We used the nlme package (Pinheiro et al. 2012) for mixed effect model analyses. Unless otherwise stated, mean values±SD are reported. Statistical significance was set at p≤0.05.

Tracked seals mostly foraged in areas of comparatively shallow bathymetry along the Scotia shelf during the post-moult trip, before heading to South Georgia Island for the breeding season haul-out (Fig. 1). Three seals (J504, J505, J498) targeted the South Orkney Plateau before either returning to South Georgia Island (J498) for the breeding haul-out or tags ceasing transmissions (J504, J505; Fig. 2). These three seals all experienced water conditions characterized by cold surface water (ca. −2°C) and slightly warmer sub-surface temperatures, and characteristic of water masses between the Antarctic Divergence and the Antarctic Slope Front (Park et al. 1998) (Fig. 3). They predominantly performed dives to the seafloor (e.g., J505; Fig. 5) at depths in the region of 300 m (Table 1). Travelling phases between King George Island, the South Orkney Plateau and South Georgia Island were characterized by comparatively short periods of time spent during bottom phases of dives.

Two seals (J497, J501) targeted areas on different sides of the Hespérides Trough, showing increased foraging effort in these areas, before also returning to South Georgia Island for the breeding haul-out (Fig. 2). Water temperatures here were also characteristic of the area between the Antarctic Divergence and Antarctic Slope Front (e.g., J501; Fig. 3). These seals both tended to perform pelagic as well as benthic dives, and dived to depths mostly deeper than 500 m (Table 1).

Two seals, J503 and Jr674, travelled along the western side of the Antarctic Peninsula towards the Bellingshausen Sea. Jr674 first spent a brief period of time on the Antarctic Continental Shelf, as evidenced by water temperature profiles characteristic of this area south of the Antarctic Slope Front (Park et al. 1998; Bailleul, Carrassin, Monestiez et al. 2007; Fig. 3), before travelling further along the Antarctic Peninsula. Both of these seals increased their foraging effort on the southward (“outbound” from King George Island) part of their migrations, and displayed less foraging effort on the return phases. One of these animals (J503) returned along the peninsula, but slightly further offshore, to South Georgia Island. Jr674 did not travel as far along the Antarctic Peninsula before heading offshore and travelling to the Sars Bank, where the tag stopped transmitting. Both seals that travelled towards the Bellingshausen Sea performed a variety of dives that were mostly in close proximity to the seafloor at depths of approximately 550 m, but switched to pelagic dive strategies when moving away from the peninsula (e.g., Jr674; Fig. 4).

The four remaining seals all targeted various areas in close proximity to King George Island. J500 moved to an area directly north of King George Island, adjacent to the South Shetland Trough and remained in this area until transmissions ceased from this device 74 days after departure. Both Jr515 and J502 crossed the Bransfield Strait and foraged in areas adjacent to Bransfield Island, before returning to South Georgia Island for the breeding haul-out. Water temperatures here were also characteristic of Antarctic Shelf waters, and showed no vertical stratification (e.g., J502; Fig. 3). These two seals also showed clear decreases in foraging effort on the travelling phases towards South Georgia Island. J496 also crossed the Bransfield Strait, but focussed its foraging in an area closer to the edge of the Powell Basin, before also returning to South Georgia Island along the edge of the Powell Basin. Areas targeted by this seal had water properties also consistent with the inter-frontal area between the Antarctic Divergence and Antarctic Slope Front (J496; Fig. 3). Three of the four seals that foraged close to King George Island (J500, J502, Jr515) also mostly performed dives to the seafloor—J502 and Jr515 to depths in the region of 200 m and J500 to deeper seafloor depths of approximately 400 m (Table 1).

Seals displayed much variation in how much time they spent in areas of high sea-ice concentrations. Six of the seals spent most of their foraging migrations in areas of very high sea-ice concentrations—up to 100% (e.g., J505; Fig. 5), while some individuals foraged almost entirely in ice-free environments (e.g., J497; Fig. 5). There were no evident trends in relationships between length, mass or condition of the seals and sea-ice concentrations they foraged in (Supplementary Fig. S3). Also, body size and condition variables (length, mass and condition) of tracked seals did not show any obvious relationships with dive depths, dive durations or time spent at the bottoms of dives (time or percentage) (Supplementary Fig. S4).

The full model containing all environmental parameters provided the best fit to daytime dive depths (Table 2). Variables showing significant associations with dive depths included SF _depth , T _max . _dep , T _diff and IC (Table 3). The associations of these variables were mostly positive, except for IC, indicating slightly shallower dive depths in higher sea-ice concentrations. The largest coefficient associated with a fixed effect was the value of 1.2 associated with T _diff , suggesting that seals increased their dive depths in areas with greater temperature stratification in the water column. The best model for night-time dive depths excluded SF _depth , but included all other fixed effects(Table 2). Significant associations with night-time dive depths were detected from JDay, T _max . _dep ,T _diff and IC. The final model used to explain dive depths occurring during twilight periods included only SF _depth , T _max.dep and T _diff as fixed effects, with all of these showing statistically significant associations with dive depth (Table 2).

Individual variability explained 27.4% of the variance in dive depths occurring in daytime, 23.1% in night-time and 15.2% in twilight. Overall temperature difference within profiles (T _diff ) showed the strongest likely influence on both day- and night-time dive depths, dives consistently increasing in depth as T _diff increased. The influence of T _max . _dep was similar between day- and night-time dives, and indicated a slight increase in dive depths, when the T _max . _dep was located deeper. Increases in SF _depth. were associated with slightly increased dive depths for both daytime and twilight dives. Ice concentration (IC) showed statistically significant effects on daytime and night-time dive depths, and seals evidently slightly decreased their dive depths in areas with higher sea-ice concentrations.

Daytime dive durations were best explained by a full model that included all fixed effects, while the most parsimonious model for night-time dive durations was similar, but excluded T _max , and the best model for twilight dive durations included all fixed effects, except IC (Table 2). Individual variation explained 28.9% of the daytime, 19.9% of the night-time and 17.5% of twilight variance in dive durations.

Comparatively large influences were detected from T _diff for dives performed during all times of day (Table 3): dives becoming longer as T _diff increased. Dives further tended to get longer as migrations progressed, as evidenced by the positive relationship of dive durations with JDay (Table 3). We found evidence for a significant correlation between JDay and T _diff for dives performed during the day, but not during other times. The depth of T _max (T _max . _dep ) was further positively related to dive durations, and longer dives were recorded in water where the T _max . _dep was deeper. Dive durations were further positively related to SF _depth , seals tending to undertake longer dives when in deeper water, although dives in deeper water were mostly not to the seafloor (Fig. 3). Sea-ice concentrations (IC) negatively affected durations of dives performed during the day and at night, and seals performed slightly shorter dives when in areas with higher IC.

The relative amounts of time spent during the bottom phases of dives, as indicated by the bottom time residuals (BTres), during daytime dives were best explained by the full model containing all fixed effects, while night-time bottom times were best explained by a model containing JDay, SF _depth , T _diff and IC as fixed effects (Table 2). Twilight bottom time residuals were best explained by a model containing only JDay, SF _depth and T _max as fixed effects. Individual variation explained more of the twilight model variance (2.9%) than the daytime (0.9%) and night-time (0.7%) model variance. Seafloor depth consistently influenced the amount of time spent during bottom phases, and seals spent comparatively more time during the bottom phases of dives in areas with shallower seafloors (Table 3). Coefficient sizes suggest that T _diff had a greater influence on day- and night-time bottom times, with seals increasing bottom times in areas with increased temperature stratification of the water column.

The best models to explain travelling speed included all fixed effects for night-time travel, and all effects, except T _diff for daytime travel (Table 2). Twilight speed was best explained by a model that contained all fixed effects, except for T _max . Individual variation explained small proportions of model variance: 0.8% for daytime, 0.7% for night-time and 2.9% for twilight speed.

Bathymetry (SF _depth ) and sea-ice concentration (IC) showed small, but significant and consistent influences on travelling speed, indicating that seals tended to travel slightly slower through areas of shallower water, as well as areas with increased sea-ice cover (Table 3). The depth of T _max (T _max.dep ) had consistently small negative influences on dive bottom times, although this relationship was only statistically significant for night-time dives.

Model outputs suggest that seals tended to dive for longer periods of time, but spend less time during the bottom phases of dives when travelling over areas of increased depth (SF _depth )—conversely, increasing their bottom time, but diving for shorter absolute times over shallower areas. The shorter absolute dive times and longer bottom times in shallower areas suggest that dive durations were unlikely to be influenced by estimated aerobic dive limits, since similar or longer absolute dive times would be expected in areas where seals are presumably increasing forage effort and therefore likely maximizing time spent diving. Indeed, if assuming calculated aerobic dive limits (cADL) of between 40 and 50 min—based on the cADLs of adult male southern elephant seals of a similar size in Hindell et al. (1992)—it is clear that males in our study mostly dived well within their cADLs (Table 1), although all animals performed at least some dives exceeding such durations as well. Dives that exceeded 40 min (and were considered likely to approach or exceed cADLs of the tracked seals) totalled 11.8% of all dives recorded. These dives (≥40 min) generally occurred over deeper water than other dives (Wilcoxon test: W=10998717, p<0.001), lending further support to the suggestion that dives associated with increased forage effort were unlikely to be limited by aerobic dive limits in our study animals.

Furthermore, tracked seals tended to travel faster over deeper areas, than over areas of shallow bathymetry. Travelling speed is considered to be a good predictor of foraging effort in elephant seals, with faster travel speeds associated with decreased foraging and vice versa (Robinson et al. 2010). Increased bottom time is also positively related to likely prey encounters in southern elephant seals (Gallon et al. 2013). Our models showed generally good agreement of bathymetric influences on both these behavioural parameters.

Male southern elephant seals are known to often perform benthic dives in various parts of the Southern Ocean where the bathymetry is comparatively shallow (Hindell et al. 1991; Campagna et al. 1999; Campagna et al. 2007; McIntyre et al. 2012). Seafloor depth did not seem to influence many of the dive characteristics of male seals tracked from Marion Island (McIntyre, Ansorge et al. 2011) although James et al. (2012) showed that dive depths were partly explained by seafloor depth. The influence of seafloor depth on dive characteristics observed here is consistent with results reported for seals from nearby Elephant Island that also target areas of shallower bathymetry with gentler slopes (Muelbert et al. 2013).

Sea-ice concentrations were associated with day- and night-time behavioural parameters measured in our study, although these influences were not always statistically significant (Table 3). Seals tended to perform shallower and shorter dives in areas with higher ice concentrations, but increased the dive bottom times and decreased travelling speed. This may be a result of the seals foraging on the shallower continental shelf, within an area of increased sea-ice concentration. However, our models did not indicate any correlations between seafloor depth and sea-ice concentrations. Seals tracked in our study often spent substantial portions of their foraging migrations in areas with very high ice concentrations, similar to some of the male seals tracked by Tosh et al. (2009), as well as some adult females tracked by Bornemann et al. (2000). Muelbert et al. (2013) reported a relationship between the size of tracked male and female elephant seals and the time spent in areas of high sea-ice concentrations. They suggested that larger seals may be better able to cope with higher sea-ice concentrations and that younger seals are perhaps unable to exploit such environments. Similar findings were reported by Bailleul, Charrassin, Ezraty et al. (2007), suggesting that female seals avoided high sea-ice concentrations, while larger sub-adult males were able to exploit higher sea-ice concentrations. We found no clear relationship between the size (or condition) of adult male seals and the sea-ice concentrations of the areas that they foraged in (Fig. 3). However, our results suggest that areas of high sea-ice concentrations are likely to be attractive foraging grounds for southern elephant seals, given the increased foraging effort of our study animals. The lack of a relationship between size and sea-ice concentrations in our study is likely due to our sample only including comparatively large seals, all possibly capable of exploiting high ice concentration areas.

Many of the seals in our sample concentrated their foraging efforts within the Weddell–Scotia Confluence (Fig. 2). This area is characterized by weakly stratified water that is comparatively homogeneous, colder, saltier and with higher O₂ concentrations than water north and south of this region (Whitworth et al. 1994). When these seals left this area, and migrated to South Georgia Island, they encountered more stratified water masses in closer proximity to their haul-out location (Atkinson et al. 2001). Time-series plots of the water characteristics encountered by seals in our study clearly illustrate such differences in water properties encountered by seals when they moved from the Weddell–Scotia Confluence area towards South Georgia Island (e.g., J498, J501, J502; Fig. 3). While the seals encountered areas with very different water temperature properties, little seasonal changes in water temperatures at depth were apparent in the time-series profiles of seals that focussed their dive efforts within small areas for the majority of their migrations (e.g., J498, J501, J502; Fig. 3). These results agree well with those reported from a similar dataset by Meredith et al. (2011), who also illustrated a stable temperature regime at depth. Substantial individual variation was present in the water masses targeted by tracked seals, with two individuals focussing their foraging efforts in homogeneous cold waters associated with the continental shelf and others targeting areas with some stratification in temperature profiles. Interestingly, comparatively few seals in our study targeted waters associated with the continental shelf (n = 2). This is in contrast to results from Bailleul, Charrassin, Monestiez et al. (2007), who showed that male and female southern elephant seals tracked from the Kerguelen Islands mostly targeted the continental shelf waters.

Tracked seals in our study adjusted their dive depths and dive durations when swimming through areas with different temperature profiles, and also exhibited changes in the amount of time spent during the bottom phases of dives (Table 3). Seals that moved between different water masses consistently dived deeper and for longer, when foraging within water masses that showed increased differences between minimum and maximum temperatures of the dive profiles (T _diff ) (Table 3). Additionally, day- and night-time dives tended to be more “square-shaped” in terms of their time-depth profiles due to increased bottom times when seals were foraging in areas with increased T _diff . This may indicate differences in the vertical depth distribution of prey species associated with differences in the temperature gradient of the water column—possibly suggesting aggregations of suitable prey at deeper depths in more stratified water masses, and more scattered distributions of prey in weakly stratified water masses (Takahashi et al. 2008). Similar relationships between dive parameters of Weddell seals and the temperature stratification of the water column were described recently by McIntyre et al. (2013).

The maximum temperature of water profiles (T _max ) only significantly influenced daytime and twilight bottom time residuals, as well as speed of travel during the daytime (Table 3), although it was included as a variable in the best fit models for dive depth (Table 2). The directions of influence are similar to those reported by McIntyre, Ansorge et al. (2011) for seals tagged at Marion Island, where seals dive deeper and display shorter bottom times in areas of higher water temperature. The influence of T _max on behavioural parameters (in relation to other environmental parameters) was apparently smaller in this study compared to what was reported for the Marion Island elephant seals. Such differences in behavioural adjustments are possibly due to a combination of search strategy differences between seals from the two populations, as well as the actual water masses targeted for foraging. Male seals from Marion Island tend to display fewer areas of concentrated movement (although there are exceptions), spending less time in these and pursue pelagic strategies whereby they travel further distances away from the island (McIntyre et al. 2012). Seals from King George Island travel shorter distances and spend more time in a few areas of concentrated movement (Tosh et al. 2009; this study). Furthermore, much of the area utilized by seals from Marion Island is characterized by strong sub-surface temperature maxima and comparatively stronger vertical stratification in the water columns (Pollard et al. 2002), as compared with the more homogeneous water structure of the Weddell–Scotia Confluence utilized by the seals from King George Island. McIntyre, Ansorge et al. (2011) suggested that the elephant seals at Marion Island may experience negative impacts, having to dive deeper and/or shift their migrations polewards to follow prey distribution shifts as the Southern Ocean continues to warm. Our results here suggest that male elephant seals utilising the Weddell-Scotia Confluence are unlikely to face similar challenges, since they are (currently) evidently diving well within their physiological limits, and exhibiting much plasticity in foraging behaviours associated with different oceanographic conditions.

Southern elephant seals tagged at Bouvetøya (Bouvet Island) displayed dive depths that were often closely associated with the sub-surface temperature maximum depth of the water column, with seals sometimes evidently targeting prey occurring within this water depth layer (Biuw et al. 2010). While T _max.depth did show consistent and statistically significant relationships with dive depths and dive durations of seals in our sample (Table 3), time-series plots did not illustrate any clear targeting of this water depth layer (e.g., J496; Fig. 6), although some seals did display some apparent concentration of dive effort in the general depth layer associated with the T _max (e.g., J501; Fig. 6).