The Scotia Sea (SS) in the Atlantic sector of the Southern Ocean ( Figure 1 , inset) is characterized by the eastward flowing Antarctic Circumpolar Current (ACC), and four associated frontal systems (from north to south: the Subantarctic Front, the Polar front, the southern Antarctic Circumpolar Current front (sACCf), and the southern boundary of the Antarctic Circumpolar Current (sbACC); Orsi et al., 1995). The SS and greater Southwest Atlantic region is a particularly productive part of the Southern Ocean (Prézelin et al., 2000; Atkinson et al., 2001; Kahru et al., 2007). Our study area is located between longitudes 35 and 68°W, and latitudes 51 and 66°S. It consists of coastal areas, shelf waters and deep sea, crosses all fronts of the ACC, and stretches over different climate regimes ( Figure 1 ).

At-sea surveys were conducted from two cruise ships (MS Fram and MS Midnatsol, Hurtigruten AS) which ran multiple trips throughout the SS and northern AP during the austral summer of 2019-2020. Observations on MS Midnatsol were made during three consecutive trips from Ushuaia to the AP and back, from 23 November to 28 December 2019. On MS Fram, observations were made during two consecutive trips from Ushuaia, via the Falkland Islands and South Georgia, to the AP and back to Ushuaia, from 10 December 2019 to 19 January 2020. The vessels’ track lines are presented in Figure 1 .

Surveys were conducted using a continuous strip-transect count methodology (Tasker et al., 1984). On each ship, a team of two to three observers took turns identifying and counting seabirds from the bridge, approximately 15 m above sea surface. Strip-transect counts were made along transects of 300 m width (distance was estimated by eye after joint training at sea), in time intervals of 10 minutes sampling once every hour, when the ship was in transit and under satisfying weather conditions (visibility > 300 m). A strip width of 300 m has previously been identified as optimal for bird counts at sea when visibility is good (Ballance, 2007; Bolduc & Fifield, 2017). The strip-transect extended from the bow in a 90° arc to the side with better visibility (least glare), either the starboard (0° - 90°) or to the port side (270° - 360°). Observations of seabirds were done by eye, and binoculars were used for species identification when needed.

Some species-specific biases in seabird counts were expected due to size differences, species-specific behavior and methodology. The most obvious potential bias is related to inflated counts of ship-attracted species and multiple counting of individuals that follow the ship (Spear et al., 1992; Hyrenbach, 2001). Further, flying birds moving faster than the vessel or perpendicular to the strip-transect are more likely to enter the strip-transect than swimming birds or birds resting on the sea surface. This leads to flux of birds inside the strip-transect and a positive bias in flying birds in absolute density estimations, if not accounted for with methods described by Tasker et al. (1984); Spear et al. (1992), or van Franeker (1994). Typically, the snap-shot method is used, where the observer conducts instantaneous counts of all flying birds present in the observed field (Tasker et al., 1984). In this study we used a continuous counting method, and a positive bias is expected for some species.

The start time (UTC) for each strip-transect count and the observation data on species and respective abundances were recorded using the Logger 2010 software (Gillespie et al., 2010). The position of the vessel was logged continuously (on a temporal resolution of 1 minute) together with time and speed data using a Globalsat USB GPS receiver, so that observation data could be related to geographical position. Not all birds could be reliably identified to the species level, and the taxonomic groups used for data analysis are presented in Supplementary Table S1 .

We relate seabird distribution to habitat-describing environmental variables that reflect meso-scale (>100 km) processes and are generally considered as cues for biological production. These are 1) distance to coast (km), 2) bathymetric slope (degrees of inclination), 3) sea surface temperature (SST, °C) and 4) the SST gradient over space, defining meso-scale oceanographic fronts (detailed hereafter).

Data were processed and analyzed using QGIS (QGIS.org, 2019) and R (R Core Team, 2020). The ‘LoggeR’ package (Biuw, 2019) was used to import and pre-process data from the Logger Access database to R.

Vessel speed affects the flux of birds in strip-transects and only strip-transects conducted at a speed of 10-17 knots were included in data analyses. In addition, two strip-transects near South Georgia had extreme abundance of prions (>1000 individuals counted) and were excluded from the statistical analyses. Because we were interested in meso-scale patterns that act on a scale of hundreds of km, observation periods (strip-transects) were binned at a scale of 100 km for the analyses of seabird community composition and seabird density and diversity. Data aggregation on a smaller scale would result in noise from sub-mesoscale processes, while aggregation on a larger scale could cover meso-scale patterns (Fauchald et al., 2000). Binning observations every 100 km along the vessel tracks resulted in 140 groups of observations (six strip-transects at the end of vessel transects fell outside 100-km bins and were omitted) ( Figure 2 ). These groups were used as sample units in the statistical analyses. Values of total seabird density and predictor variables were calculated as means of the values in binned observations. Species richness was calculated as the total number of species observed in 100-km bins and is hence affected by the number of strip-transects within each bin.

Collinearity between predictors was explored through pairwise correlations with Pearson’s and Spearman’s correlation factors ( Supplementary Figures S1, S2 ). Collinearity of |r| < 0.7 between predictors in the same model was considered acceptable (Dormann et al., 2013), and collinearity above this threshold did not occur between our predictors. Significance was assessed at α = 0.05.

In the model fitted through CCA, the predictors of different species assemblages were SST (p=0.001) and distance to coast (p=0.001). Together they explained 13.8% of the variation in data ( Figure 4 ). Both forward selection and backward elimination resulted in the same model, and the VIFs for both model predictors were ∼1.0. The horizontal CCA1 axis mainly represents changing SST and thus a latitudinal gradient in community composition, while the vertical CCA2 axis largely represents a coastal-oceanic gradient, separating breeding coastal species from breeding oceanic species that can travel farther in search for food. Seabird communities appear in regional groups in the CCA plot, highlighting the spatial segregation of seabird species assemblages. In particular, communities in the AP and Falkland Islands stood out as substantially different from the others. The AP in the south end of the study area was characterized by ice-associated species, while the Falkland Islands which are situated north of the Subantarctic Front formed a distinct group. For the open ocean seabird communities, there was a latitudinal change in species assemblages between the northern and southern SS.

Seabird communities over temperate and Subantarctic waters north of the Polar Front were dominated by black-browed (Thalassarche melanophrys) and great albatrosses (Diomedea spp.), shearwaters (Puffinus spp.), Atlantic petrel (Pterodroma incerta), shags (Phalocrocorax spp.), and magellanic penguin (Spheniscus magellanicus), whose diet consists mainly of cephalopods and fish ( Supplementary Table S1 ). Communities over Antarctic waters south of the Polar Front, where krill is abundant, were characterized by planktivores such as Adélie (Pygoscelis adeliae) and chinstrap penguins (P. antarcticus), storm petrels (Oceanites oceanicus and Fregetta spp.), blue petrel (Halobaena caerulea), and prions (Pachyptila spp.). Snow (Pagodroma nivea) and Antarctic petrels (Thalassoica antarctica) are mixed feeders that are adapted to foraging in pack ice where they find less competition (Griffiths et al., 1982; Ainley et al., 1993; Ainley et al., 1994) and were primarily observed in the AP region. Skuas (Catharacta spp.) were observed close to penguin colonies, where they feed on penguin eggs and chicks in the austral summer (Shirihai, 2007).

The best supported GAM of species richness included the predictors SST (p=0.019), SST gradient (p<0.001), distance to coast (p=0.031), and number of observation periods within 100-km bins (p<0.001), and this model explained 43.9% of the total variation in the data ( Figure 5 ). The residual plots of the model look satisfactory ( Supplementary Figure S5 ), and residual autocorrelation was low and consequently not considered a problem ( Supplementary Figure S6 ). Generally, seabird species richness peaked at surface water temperatures around 3°C and decreased slightly with distance from coast ( Figure 5 ). The number of species observed increased with observation effort (number of observation periods within 100-km bins), as expected. The effect of the SST gradient on species richness did not show a general trend, but low and high values were associated with relatively higher species richness while medium values were associated with lower species richness ( Figure 5 ).

Latitudinal variation in species richness on the western and eastern sides of the Scotia Sea are shown in Figures 6 , 7 , respectively, together with variation in distance to coast, SST, the SST gradient, and seabird density. Except for a peak north of South Georgia, species richness did not vary considerably with latitude on the western nor the eastern side of the Scotia Sea. An elevated SST gradient was found close to land masses and at the Polar Front in the Drake Passage.

The best supported GAM of seabird density included the predictors SST (p<0.001) and bathymetric slope (p<0.001), and this model explained 21.1% of the total variation in the data ( Figure 8 ). The residual plots of the model look satisfactory ( Supplementary Figure S7 ), but there was slight autocorrelation in the residuals ( Supplementary Figure S8 ). Generally, seabird density peaked at surface water temperatures around 2-3°C, and lower seabird densities were associated with a higher bathymetric slope.

Latitudinal variation in seabird density on the western and eastern sides of the Scotia Sea are shown in Figures 6 , 7 , respectively, together with variation in distance to coast, SST, the SST gradient, and species richness. Highest seabird densities were found close to South Georgia and at the sbACC on the eastern side of the study area.

Predator communities are associated with specific ranges of geographical gradients through their diets and physiological and behavioral adaptations, and predator-habitat relationships generally reflect predator-prey relationships (Griffiths et al., 1982; Ainley et al., 1993; Ballance, 2007; Fauchald et al., 2011a). Sea surface temperature (SST) has been described as the most important predictor of marine biodiversity on a global scale (Tittensor et al., 2010) and was a strong overall predictor of seabird distribution in this study. Ocean surface temperature typically varies between water masses and gradients arise where water masses converge, creating oceanic habitats with physical characteristics occupied by different assemblages of prey species (Pocklington, 1979; Wahl et al., 1989; Jungblut et al., 2017; Chapman et al., 2020). Water characteristics such as temperature, salinity, and nutrient concentrations in the Southern Ocean show large latitudinal gradients, dividing the ocean into biogeographical zones (Grant et al., 2006). Our results show how different seabird communities inhabit waters that lie within specific ranges of a temperature gradient, corresponding to different water masses in the Scotia Sea. Seabird abundance and species richness also vary between water masses, presumably because of differences in productivity between oceanographic regions. Seabird communities over temperate and Subantarctic waters north of the Polar Front were dominated by seabirds feeding mainly on cephalopods and fish, while communities over Antarctic waters south of the Polar Front were dominated by planktivores and species adapted to foraging in pack ice. These results support earlier studies that found an association between seabird communities consisting of species with similar dietary preferences and water masses inhabited by preferred prey (Abrams, 1985; Abrams & Miller, 1986; Hunt et al., 1990; Joiris, 1991), as well as studies underlining the importance of niche segregation for the spatial distribution of seabirds. Interspecific interactions, such as competition and facilitation, and species-specific energetic constraints are factors that cause different seabird species to respond differently to the distribution of their prey (Ballance et al., 1997; Ballance, 2007; Fauchald et al., 2011a; Veit & Harrison, 2017). Interspecific competition and energetic constraints structure seabird communities along a productivity gradient (Ballance et al., 1997), while local enhancement and facilitation associate surface feeding species that have a relatively low energetic cost of flight (and therefore low cost of search behavior) with diving species that drive prey up to the surface (Veit & Harrison, 2017). Seabird species are consequently found in niches formed by the interaction between dietary preferences and interspecific interactions such as competition and facilitation. For instance, the white-chinned petrel (Procellaria aequinoctialis), and black-browed, grey-headed (T. chrysostoma), and wandering albatrosses (Diomedea exulans), have been suggested to outcompete the light-mantled sooty albatross (Phoebetria palpebrata) on foraging grounds close to breeding colonies in SG, thereby forcing it to utilize more distant areas for foraging (Phillips et al., 2005). We indeed observed light-mantled sooty albatrosses generally in more Antarctic environments and further from the breeding colonies in SG than other albatross species, possibly as a consequence of spatial segregation in foraging areas between sympatric species in the same foraging guild.

The coastal-oceanic gradient is another important factor affecting prey species present through gradients in productivity, bathymetry, and water characteristics (Doty & Oguri, 1956; Abrams & Miller, 1986), contributing to the differences in seabird assemblages between the oceanic and coastal habitats described in this study. Our survey was conducted during the early and mid-austral summer, when many seabird species breed. Breeding poses a spatial constraint on seabirds (Gaston, 2004; Amorim et al., 2009); the need to return to land to take turns with the partner for brooding duties at the nest constrains how far and for how long adult birds can forage at sea. These restrictions lead to higher densities of some bird species in coastal waters (Abrams & Miller, 1986), a pattern detected in this study as a negative relationship between seabird species richness and distance from coast, as well as a negative but not statistically significant relationship between seabird density and distance from coast. For example, large numbers of oceanic species like grey-headed albatross and white-chinned petrel were observed in the coastal waters off South Georgia.

In the open ocean, strong gradients in SST are generally indicative of oceanographic fronts, which might act in two ways in shaping seabird distribution. First, as productive areas they can be associated with an elevated abundance of seabirds and other higher-level predators (Wahl et al., 1989; Bost et al., 2009) and second, they can act as avifaunal boundaries, separating different species assemblages (Pocklington, 1979; Wahl et al., 1989), as has been described for the Polar Front and the southern boundary of the Antarctic Circumpolar Current (sbACC) (Ribic et al., 2011). Ribic et al. (2011) found increased abundances of a few species, including diving petrels and blue petrel, at the Polar Front and the sbACC, but the fronts’ effect as boundaries for differing species assemblages was more pronounced. Seabirds breeding in the South Shetland Islands have been shown to use the sbACC as an important feeding ground (Santora & Veit, 2013), pointing towards both a possible aggregating effect of the sbACC, and its possible role as a boundary for AP species in the breeding season. The role of fronts as boundaries could explain the high importance of SST for community composition and the low effect of temperature gradient in our analyses. The SST gradient did not manage to capture an effect of fronts on oceanic communities, and the effect of the SST gradient on seabird abundance did not show a general trend contrary to results from for example Serratosa et al. (2020). Instead, relatively higher densities were associated with low and high temperature gradients, while medium gradient values were associated with relatively lower seabird density. While there was no statistical support for seabird species richness increasing with an increasing temperature gradient, and its effect on seabird density was unclear, there was a general peak in density and species richness around SSTs of 3°C. These temperatures are usually associated with waters around the Polar Front and around South Georgia. A possible explanation for the lack of effect of the SST gradient is that a frontal temperature gradient might not always be detectable at the surface (Chapman et al., 2020). The role of fronts as areas of increased productivity might also be less important in summer when food is abundant in the SS (Ribic et al., 2011), or because breeding birds experience stronger constraints on their foraging range and might not be able to reach fronts. The effect of restricted foraging range likely explains the lack of effect of fronts lying far from land, notably the Polar front.

Other studies have shown a positive association of seabird abundance with areas of coastal upwelling, such as in the Eastern South Pacific Ocean (Serratosa et al., 2020). Interestingly, the density of seabirds in this study appeared to decline in relation to higher bathymetric slopes which are typically associated with coastal upwelling. Potentially, this trend may be related to the overall lower species richness in the topographically complex AP compared to the sub-Antarctic areas (Hindell et al., 2020). Seabird communities in the coastal AP region were associated with a variable bathymetry ( Figure 4 ), which likely reflects the deep basins of Bransfield Strait and the deep fjords around coasts of the AP. The highest seabird densities were found in waters around South Georgia, which lies on a large plateau. Further, a smaller number of flying seabird species in the AP might reduce the flux of birds within strip-transects compared to other areas, biasing total bird counts towards sub-Antarctic areas. Also, penguins are a dominant group in the AP and have been suggested to dive as a response to ships (Jehl, 1974, as cited in Tasker et al., 1984). On the ship transects between land masses, strip-transects situated at continental slopes were few, which may have reduced our ability to detect any effect of slope on seabird abundance in the oceanic environment.

Some biases in our data were expected due to the methodology used. Species counts are potentially inflated or deflated due to species-specific responses to the ship, mainly ship attraction or avoidance (e.g., Hyrenbach, 2001). Species such as giant petrels, cape petrel and black-browed albatross are ship-followers, and their abundances are therefore easily overestimated. The fact that many flying birds travel faster than the ship increases their chance of entering the strip-transect relative to birds sitting on the sea surface (Spear et al., 1992; van Franeker, 1994). The flux of flying birds in the strip-transect is hence a function of their speed relative to the speed of the vessel. Flux can be corrected for using a snapshot method (Tasker et al., 1984), where all flying birds are counted regularly in instantaneous counts. Here, we counted all birds, including flying birds, using a continuous method. This likely introduces bias from ship attraction and bird flux in our results. Using cruise vessels as observation platforms, we were not able to control vessel speed. This might cause variation in flux between strip-transects and we excluded strip-transects counted in low and high speeds from the analyses in order to reduce potential bias caused by vessel speed. It is also worth noting that species richness in some of our strip-transects might have been underestimated as species that were hard to identify at sea were grouped together. For example, sympatric species of storm petrels are found in the AP and sympatric species of prions are found in South Georgia and the Falkland Islands (Shirihai, 2007; Clarke et al., 2012). We recommend including snapshot counts to control for the flux of flying birds in strip-transects. To reduce variance associated with the different number of observation periods in each 100 km segment, and to increase the precision of mean counts used for data analysis, we further recommend more frequent observation periods on future surveys.

Ships of opportunity give the possibility to cover large areas and large-scale environmental gradients. This is of paramount importance for detecting seabird communities’ responses to environmental change. In the case of our study, the routes allowed a representative dataset of the Scotia Sea and northern Antarctic Peninsula to be collected by crossing all biogeographical regions in the study area (Grant et al., 2006) and covering both oceanic and coastal areas (though not the coasts of South Orkney and South Sandwich Islands). Nature-based tourist cruises usually ensure coverage of biodiversity hotspots. Coastal and hotspot areas are of particular interest for tourist cruises, and our effort was hence concentrated towards these areas. This increases data precision in coastal areas and areas with high biodiversity, which is an advantage as these are the most important areas for conservation monitoring. However, the predetermined routes of cruise vessels, strongly concentrated to specific regions, leave vast areas unsampled. Large parts of the Southern Ocean and most coasts of Antarctica are never visited by cruise ships. Areas outside the reach of cruise vessels need to be covered by research surveys with the possibility for a more representative sampling design, or possibly sampled opportunistically by observers onboard supply vessels or fishing ships going to some of these areas. We nevertheless argue that cruise vessels and other opportunistic sampling platforms provide opportunities for the collection of data (e.g., repeated surveys within and between years) that can augment our understanding of seabird ecology and ecosystem change when combined with data collected on dedicated research surveys. Johannessen et al. (2022) and Henderson et al. (2023) provide examples of the use of cruise ships in studies on baleen whales in our study area.

Data on seabird distribution are important for ecosystem monitoring, and for conservation efforts like the establishment of MPAs and spatial and temporal management of fisheries. Marine predator distribution studies have seen a shift from an area-centered approach using ship surveys towards a population-centered approach using tracking data (Hindell et al., 2020; Fauchald et al., 2021). However, tracking data typically do not represent all species, populations, demographic groups, and individuals using the studied area, and ship surveys continue to be an important method for validation and completion of tracking data (Fauchald et al., 2021). In the 1980s and 1990s, chinstrap (Pygoscelis antarcticus) and Adélie penguins (P. adeliae) were the most commonly observed penguin species in the Bransfield Strait region (Hunt et al., 1990; Hunt et al., 1994). We noticed a change in the relative abundances of pygoscelid penguins, previously described by Lynch et al. (2012), observing more gentoo penguins (P. papua) than chinstrap or Adélie penguins in the area. Studies like ours provide information on the distribution of seabirds for future comparisons assessing ecosystem change, for which community-level data are particularly useful. Models that incorporate multiple species and predictors can be effective in revealing complex relationships, since they consider multiple responses inside the same community (Reid et al., 2005; Piatt et al., 2007; Hindell et al., 2020). Considering community-level responses is important in making meaningful management decisions and networks of MPAs should be designed in a way that they consider the diversity of communities as well as future change in species distribution (Hindell et al., 2020).