Aerial surveys for Antarctic minke whales (Balaenoptera bonaerensis) reveal sea ice dependent distribution patterns

Abstract This study investigates the distribution of Antarctic minke whales (AMW) in relation to sea ice concentration and variations therein. Information on AMW densities in the sea ice‐covered parts of the Southern Ocean is required to contextualize abundance estimates obtained from circumpolar shipboard surveys in open waters, suggesting a 30% decline in AMW abundance. Conventional line‐transect shipboard surveys for density estimation are impossible in ice‐covered regions, therefore we used icebreaker‐supported helicopter surveys to obtain information on AMW densities along gradients of 0%–100% of ice concentration. We conducted five helicopter surveys in the Southern Ocean, between 2006 and 2013. Distance sampling data, satellite‐derived sea‐ice data, and bathymetric parameters were used in generalized additive models (GAMs) to produce predictions on how the density of AMWs varied over space and time, and with environmental covariates. Ice concentration, distance to the ice edge and distance from the shelf break were found to describe the distribution of AMWs. Highest densities were predicted at the ice edge and through to medium ice concentrations. Medium densities were found up to 500 km into the ice edge in all concentrations of ice. Very low numbers of AMWs were found in the ice‐free waters of the West Antarctic Peninsula (WAP). A consistent relationship between AMW distribution and sea ice concentration weakens the support for the hypothesis that varying numbers of AMWs in ice‐covered waters were responsible for observed changes in estimated abundance. The potential decline in AMW abundance stresses the need for conservation measures and further studies into the AMW population status. Very low numbers of AMWs recorded in the ice‐free waters along the WAP support the hypothesis that this species is strongly dependent on sea ice and that forecasted sea ice changes have the potential of heavily impacting AMWs.


| INTRODUC TI ON
Antarctic minke whales (Balaenoptera bonaerensis, AMW) are the only baleen whale species consistently inhabiting the ice-covered parts of the Southern Ocean (Ainley, Dugger, Toniolo, & Gaffney, 2007). Highly adapted to sea-ice habitats, they are frequently observed in a range of sea ice concentrations (Ainley et al., 2007(Ainley et al., , 2011(Ainley et al., , 2017Scheidat et al., 2011;Thiele et al., 2004;Williams, Grand, et al., 2014). AMWs have been shown to feed on krill under the ice, with Friedlaender et al., (2014) suggesting that they exploit a unique niche among sympatric whale species feeding on krill in the Southern Ocean ( Figure 1). Furthermore, predator avoidance has been suggested as another reason for AMWs to use sea-ice habitats, inaccessible to Type A killer whales (Pitman & Ensor, 2003).
Despite sea ice having been recognized as an important habitat characteristic for a long time (Ainley et al., 2007(Ainley et al., , 2017Ribic, Ainley, & Fraser, 1991), little is known about the variations in densities of AMWs in sea ice-covered areas. Many Antarctic research voyages collecting information on cetacean distribution and abundance do not venture much inside the ice edge as either the research program is dedicated to mainly ice-free waters (Santora, Schroeder, & Loeb, 2014), or ships are not ice-strengthened and have to restrict their dedicated survey efforts to operationally safe, ice-free waters (Murase et al., 2003). Even from icebreakers, conventional line-transect surveys for density estimation are impossible on ships navigating or breaking through ice, constantly changing speed and direction (Ainley et al., 2011;Buckland et al., 2001). Moreover, biases in shipboard cetacean surveys in ice-covered waters may arise from the influence of the ship's presence (e.g., its noise and the noise caused by ice-breaking activities) on the detection rate of cetaceans due to behavioral reactions (Erbe & Farmer, 2000). Consequently, reliable information on densities of AMWs in sea ice is lacking. This information, however, is crucial for assessing the species' vulnerability with regard to predicted, climate-related changes in sea ice conditions (Silber et al., 2017). High-latitude marine systems are among the regions responding most rapidly to changing climatic conditions, with ice-obligate species facing some of the largest changes in their habitat (Nicol, Worby, & Leaper, 2008;Silber et al., 2017). Furthermore, knowledge about the order of magnitude of AMW numbers in sea ice-covered areas could help gauge whether the proportion of the population inside the ice edge has the potential to account for at least some of the estimated decline in circumpolar AMW abundance between the late 1980s and the late 1990s (IWC, 2013). AMW abundance was last estimated at 515,000 (95% CI: 361,000-733,000) individuals based on data from the third and most recent circumpolar shipboard sighting survey (also known as CPIII; conducted under the auspices of the International Whaling Commission (IWC) during the austral summers of 1992/93 to 2003/04;IWC, 2013). Comparison with an abundance estimate of 720,000 (95% CI: 512,000-1,012,000) obtained from the second circumpolar survey (CPII;1985/86-1990) suggests a decline of approximately 30% between the late 1980s and the late 1990s (IWC, 2013). However, there is considerable uncertainty regarding the extent to which this reflects true population decline, distributional shifts or changing survey methods (IWC, 2013). Both abundance estimates are subject to some degree of negative bias since the full range of the species' summer distribution was not covered by the surveys (Branch & Butterworth, 2001). It is unclear, whether a shift between the two surveys in the proportion of the population that was north or south of the survey boundary, and therefore unavailable during the surveys, could be responsible for the observed change in estimated abundance. To the north, the circumpolar abundance estimates from CPII and CPIII surveys were restricted to the area south of 60°S, a latitude where AMWs. Highest densities were predicted at the ice edge and through to medium ice concentrations. Medium densities were found up to 500 km into the ice edge in all concentrations of ice. Very low numbers of AMWs were found in the ice-free waters of the West Antarctic Peninsula (WAP). A consistent relationship between AMW distribution and sea ice concentration weakens the support for the hypothesis that varying numbers of AMWs in ice-covered waters were responsible for observed changes in estimated abundance. The potential decline in AMW abundance stresses the need for conservation measures and further studies into the AMW population status. Very low numbers of AMWs recorded in the ice-free waters along the WAP support the hypothesis that this species is strongly dependent on sea ice and that forecasted sea ice changes have the potential of heavily impacting AMWs.

K E Y W O R D S
Antarctic minke whale distribution, density surface models, distance sampling, marginal ice zone, ship-based helicopter surveys, Southern Ocean density of AMWs is assumed to be low and the range begins to overlap with the dwarf minke whale (B. acutorostrata; Best, 1985).
The southern boundary of both surveys was determined by the edge of the Marginal Ice Zone (MIZ), hereafter referred to as the ice edge. The proportion of the population of AMWs sampled may thus have been affected by intra-or inter-seasonal changes in the position of the ice edge changing the area sampled and, ultimately, the numbers of animals encountered. Furthermore, any overall changes in the relative proportion of AMWs within ice-covered areas throughout the years of the IWC's sighting surveys may have affected the number of whales available within the area up to the ice edge. These and other survey-related hypotheses to explain the decline in estimated abundances are summarized in Murase and Bravington (2012). Ultimately, it remains unknown whether the number of AMWs in ice-covered areas (i.e., unavailable to IWC shipboard surveys) had a large enough effect on the abundance estimates from open-water to affect the conclusion that the population declined. If the decline is real, the causes are currently unknown, and the decline may be continuing. This raises concerns for the long-term conservation of the species in the face of climate change, but also complicates the politically charged debate over whale management in the Southern Ocean, particularly Japanese special permit whaling (Brierley et al., 2016;Gales, Kasuya, Clapham, & Brownell, 2005).
Analyzing densities of AMWs in sea ice-covered areas and variation therein is one way to test some of the hypotheses trying to explain the observed decrease in abundance. While it remains impossible to produce retrospective estimates of absolute abundance of AMWs in sea ice-covered areas for the years that the IWC surveys were running, regional relative density estimates from ice-covered parts of the habitat can provide an idea of likely magnitudes of abundances of minke whales inside sea ice regions (i.e., as compared to those found in adjacent open-water areas) and potential variations therein. This may allow consideration whether the "moved-into-sea ice" hypothesis is at least tenable. Moreover, in the likely absence of absolute circumpolar abundance estimates in the near future, identifying dependencies and vulnerabilities of the species in their habitat seems to be a reasonable step toward helping to ensure protection, recognize threats, and mitigate impacts.
Aerial surveys have been proven as a means to survey ice-covered areas at constant survey speeds, with assumed minimal disturbance effects on animal distribution and effectively collecting cetacean sighting data in 0%-100% ice concentrations (Herr et al., 2016;Scheidat et al., 2011;Vacquié-Garcia et al., 2017;Williams, Kelly, et al., 2014). Williams, Kelly, et al. (2014) provided first information on relative AMW abundance on both sides of the ice edge, suggesting up to 20% of AMWs of the Weddell Sea were within icecovered waters.
In this study, we investigate AMW densities along gradients of sea ice concentration, based on five icebreaker-supported helicopter surveys, including the two surveys evaluated by Williams, Kelly, et al. (2014). The much broader temporal and spatial extent of this data set allows us to not only quantify the numbers of AMWs within the ice-covered areas but also to assess the spatial and temporal variability therein.
Here, we present the first distribution models for the density of AMWs based on the largest existing aerial survey data set from the Southern Ocean.

| Aerial surveys
We conducted aerial surveys during five expeditions of the German research icebreaker Polarstern (Alfred Wegener Institute, Helmholtz Centre for Polar and Marine Research, 2017) between 2006(ANT23-8: Gutt, 2008ANT25-2: Boebel, 2009;ANT27-2: Fahrbach, 2011;ANT28-2: Kattner, 2012;ANT29-3: Gutt, 2013;  Given logistical constraints and daily changes in sea ice conditions, it was impossible to create a systematic survey design in advance of the survey. This meant that the study was designed at the outset in anticipation of the use of model-based density/abundance estimators over conventional/design-based analyses (Buckland et al., 2007;Hedley & Buckland, 2004;Miller, Burt, Rexstad, & Thomas, 2013). We planned all flights in an ad-hoc manner when All data were entered directly into a computer continuously storing GPS data obtained by a handheld GPS device in intervals of four seconds. Environmental and sighting conditions (sea state, cloud cover, glare, ice coverage, subjective sighting condition [a compound variable which describes the overall ease of observing minke whales, dependent on weather and ambient light conditions, taking three levels: "good", "moderate" and "poor"]) were entered as assessed by the observers. Sightings were recorded noting species, distance to transect line (calculated postsurvey via declination and horizontal angle), and group size. Declination angles were measured using inclinometers. If a sighting occurred and the species could not be identified or group size could not be determined immediately, the survey was halted in order to approach the sighting for closer inspection, a procedure known as "closing mode" (Calambokidis & Barlow, 2004;Strindberg & Buckland, 2004). After identification, the helicopter returned to the transect line at the point of departure and the survey was resumed.

| Data preparation
All survey data were associated with the following parameters: dayspecific ice cover derived from satellite data and distance to the ice edge (i.e., edge of MIZ), distance to shelf break, water depth, and bathymetric slope. Therefore, required data were obtained as described in the following.

| Sea-ice cover and distance to ice edge
The daily 6.25 km resolution images of ice concentration using the ASI algorithm on data from the Advanced Microwave Scanning Radiometer for EOS (AMSR-E) and AMSR2 satellite sensors from the Institute of Environmental Physics, University of Bremen, https:// seaice.uni-bremen.de/data/amsre/ (Spreen, Kaleschke, & Heygster, 2008), were used to extract ice concentration values for each survey point. The same data source was used to estimate the position of the outer boundary of the MIZ for each survey day and subsequently calculate the distance to the ice edge for every data point. The ice edge was defined as the 15% contour of ice concentration (a typical threshold used to define the ice edge within satellite sea-ice data;  Note. Total numbers of sightings are given along with numbers of sightings that were used in the detection function modeling process. Excluded from detection function modeling were sightings that a) were made by the right-hand side observer and b) occurred under poor sighting conditions. Worby & Comiso, 2004), that is a smoothed line bounding the 15% ice concentration margin. We used built-in "Spatial Analyst" functions in ArcMap (Environmental Systems Research Institute, Inc., Redlands, California, 2009) to select the largest polygon of contiguous raster pixels featuring at least 15% ice concentration, defining the outermost edge, and then smoothing it using the "Boundary Clean" function. Finally, we calculated the Euclidean distance (in km, projection: Stereographic South Pole) to the closest ice edge position of the respective day for each recorded effort position (recorded every 4 s), cetacean sightings and all midpoints of the prediction grids, differentiating between distances inside and outside of the ice edge.

| Depth and slope
Depth and bathymetric slope (rate of change in depth over distance) values were extracted from the GEBCO 2014 grid available from the British Oceanographic Data Centre (BODC). We calculated slope values using the ArcMap "Slope" tool (Burrough & McDonell, 1998).
Depth and slope values were extracted for each survey point as well as the prediction grid.

| Distance to shelf break
The shelf break is defined as the line between the shelf and the upper continental slope. It is generally represented by a relatively abrupt increase in seabed inclination from the flat shelf (<1°) to the steeper upper continental slope (>1°). Globally, the shelf break generally is between 100 and 150 m water depth (Weatherall et al., 2015). Around Antarctica, the ice load and the resulting isostatic equilibrium (Ivins & James, 2005;Whitehouse, Bentley, Milne, King, & Thomas, 2012) and erosion (Livingstone et al., 2016) result in a deep shelf. There, the shelf break is mostly located between 400 and 600 m water depth (Arndt et al., 2013).
For this study, the shelf break ( Figure 3) was detected on the base of the International Bathymetric Chart of the Southern Ocean (IBCSO) (Arndt et al., 2013). From the IBCSO raster data set, we calculated a slope angle raster data set following the approach of Burrough and McDonell (1998). The 1° slope angle line in most places clearly indicates the shelf break. In areas where the slope angle was less distinct, also the 500 m water depth contour and the hillshade relief raster data set (an illuminated terrain F I G U R E 2 Tracks of all helicopter survey flights conducted during five expeditions of RV Polarstern between 2006 and 2013 (see Table 1 for expedition dates). The red line delineates the Antarctic shelf break as detected in this study. Depicted depth is based on the ETOPO grid (Amante & Eakins, 2009) model generating a pseudo 3-D visualization) were used to define the shelf break. North of the Antarctic Peninsula, the shelf break was extended from the Antarctic Peninsula to the South Orkney Plateau although the seabed dropped to around 1,000 m water depth in places. We considered the Bransfield Strait a tectonic inner-shelf depression and defined the shelf break west of the South Shetland Island. In the southern Weddell Sea, sounding data were sparse, and IBCSO was often based on satellite altimetry data (Arndt et al., 2013). The lower resolution of these data sometimes reduced the accuracy of the detection of the shelf break.
Distance to the shelf break was then calculated (in Stereographic South Polar projection) for each data point, discriminating between positions on versus off the shelf.

| Detection function
We used distance sampling methods to estimate the probability of detection as a function of distance from the track line (Buckland et al., 2001). As mentioned above, we only used data from the left side (collected by the left and front observers) for modeling a detection function. In order to ensure adequate sample size, we pooled the sightings, and all their associated covariates, from all survey years, to estimate a single detection function. All minke whale sightings were used, including B. bonaerensis, B. acuturostrata, and sightings that could not be clearly identified as either of them or that were defined as "minke-like". As the front observer had an unobstructed view of the track line, no left truncation was applied to the perpendicular distances going into the fit of the detection functions.
Half-normal and hazard-rate detection function models were tested (Buckland et al., 2001) using the software package "mrds" (Laake, Borchers, Thomas, Miller, & Bishop, 2013)  Beaufort sea state, group size and a measure of local sea ice concentration (ice concentration as judged by the observers), which was classified as 0%-9% ice coverage = "no ice" and 10%-100% = "ice".
Ice concentration was included as a factor to test for evidence that increasing complexity in the visual field may decrease the probability of detection. Perpendicular distances were truncated to exclude the furthest ~10% of detections, which is recommended to avoid overfitting the tail of the distribution (Buckland et al., 2001). The best detection function model was selected based on Akaike's information criterion (AIC; Akaike, 1974).

| Density surface model
A density surface modeling approach was used to produce predictions on how the density of AMWs varied over space and time, and with environmental covariates. Note that these density estimates could not be corrected for availability bias (sensu Marsh & Sinclair, 1989) and it was thus assumed that g(0) = 1 (Buckland et al., 2001).
This means, that estimated densities are minimum densities and most likely an underestimate. We used the count method, as described by Hedley and Buckland (2004) and Miller et al. (2013), in combination with generalized additive models (GAMs; Wood, 2006), to predict densities of AMWs across the study area. Restricted maximum likelihood was used as the criterion for estimating smoothing parameters (Wood, 2011). The response variable was the number of individuals per "segment" of transect, where the segment length was selected to ensure relative homogeneity in sighting conditions within a single segment. Based on analyses of varying segment lengths by exploration of the dispersion patterns of whale detections and of the spatial autocorrelation of environmental covariates across the segments, a segment length of 30 km was selected in order to balance between over-dispersion in the number of animals sighted and having too much heterogeneity in sighting conditions within a single segment.
Effort data with sighting conditions less than "moderate" (i.e., poor) in quality were removed from the analysis. An offset variable was incorporated in the model to account for changes in estimated probabilities of detection within each segment, which ultimately manifest in differences of areas effectively searched within the segments . The log-transformed effective search area, estimated using the MCDS model described above, was used as the offset in the GAM-based density surface model.
The modeling for the density surfaces was undertaken in R (R Development Core Team, 2014) using the mgcv package (mgcv v1.7-28; Wood, 2006Wood, , 2011. A Tweedie distribution (Jørgensen, 1987) was used to model the count data (where the Tweedie parameter was estimated during the GAM fit). Variance from the detection function fitting was propagated through the GAMs at the segment level, to the final abundance estimates via a random effects method described in Williams et al. (2011) and Miller et al. (2013). Variables tested for inclusion in the spatio-environmental models included smooths of x and y (projected longitude and latitude values; Lambert azimuthal equal area projection), water depth, bathymetric slope, distance to the shelf break, ASI sea ice concentration (corresponding to dates of particular survey effort), and distance to the ice edge (derived from the satellite data for each survey day as described above).
Survey season was included in the density surface models as a factor term to account for any inter-seasonal variations that may occur in AMW densities across the survey areas.

| Prediction
We predicted and compared AMW densities between four strata: Neumayer (

| RE SULTS
In total, 40,695 km of track lines were covered by helicopter surveys ( Figure 2, Table 2). Flights were conducted on 106 survey days, with concentration of effort in December and January, across various austral summer seasons (Appendix Table A1) and varying sighting conditions (Appendix Tables A2, A3 and A4). 25,987 km of survey effort were allocated to EAP, WS and NM, covering areas of 0%-100% of ice concentration in good and moderate conditions. Almost half of the survey effort (18,077 km) in EAP, WS and NM was allocated to areas inside the ice edge. In WAP, 13,631 km of survey effort was completed in mainly ice-free waters (Appendix Table A5).
A total of 155 minke whale sightings, comprising 286 individuals including one calf, were recorded (Tables 1 and 2, Figure 4).
AMWs (B. bonaerensis) represented the majority, with 150 sightings of 275 individuals. One sighting of two individuals was identified as B. acuturostrata, one sighting of six individuals as either B. bonaerensis or B. acuturostrata, and three sightings of single animals were recorded as "minke-like" (Figure 4). We encountered AMWs all over the study area. However, encounter rates differed greatly between strata. The lowest encounter rate (0.001 groups/km) was found in WAP (west coast of the Antarctic Peninsula), the highest encounter rate (0.010 groups/km) was found in EAP (east coast of the Antarctic Peninsula). East of EAP, a longitudinal gradient was observed with encounter rates generally decreasing from west to east (Table 2).
Antarctic minke whales were encountered in sea-ice covers of up to 99% of the visual field of the observers / 100% ice concentration as indicated by the ASI sea-ice data and up to 981 km inside the ice edge. The maximum distance inside the ice edge covered by aerial surveys was 1,192 km.

| Detection function
Of all recorded sightings, 105 sightings of 199 individuals (104 sightings of B. bonaerensis, 1 sighting of B. acuturostrata) were eligible for detection function modeling (Table 1). Detection distances were right-truncated at 1,300 m for fitting the detection function. After comparing the AIC from all permutations of the detection functions and potential scaling variables, the model with lowest AIC was selected, which was a half-normal detection function with a distance sampling model scaled by sighting condition score (1 = moderate, 2 = good) as the most parsimonious model ( Figure 5). The estimated mean detection probability between track line and the truncation distance was 0.61 (CV = 0.10).
Moving between a sighting condition score of moderate to good nearly doubled the effective strip half-width (i.e., from 450 to 887 m). Details on the distribution of effort over sighting conditions is provided in the Appendix (Tables A2-A4). Of the total effort, 39,526 km (97%) were surveyed under "good" or "moderate" sighting conditions.

| Density surface model
A density surface model composed of a combination of a tensor product of ASI sea ice concentration and distance to ice edge and a smoother of distance to shelf break was selected as the best GAM to Based on the best model, we predicted AMW densities (uncorrected for availability) for separate strata and specific dates (to reflect respective ice conditions). No predictions using this model were possible for WAP, as no ice edge could be identified in this stratum for any year. Low ice concentrations in this area during all five survey years resulted in no MIZ; geographically, the closest ice edge was located east of and, thereby, "behind" the Antarctic Peninsula. We therefore considered distances to this ice edge as not ecologically meaningful for use in the model. The sample size (i.e., the number of sightings) in WAP, however, was too small to create an independent GAM solely for WAP, based on the remaining variables. In order to obtain density predictions for this stratum in absence of a model-based density estimate, we used conventional distance sampling (i.e., a simple detection function with no sighting covariates in the scale parameter, and a design-based approach to estimating animal density), using the software package "Distance" (Miller, 2015) in R (R Core Team, 2015). WAP provided reasonably representative coverage by surveyed track lines for conventional distance sampling analyses (WAP stratum boundary is given in Figure 3 and realized effort is given in Figure 4), so that this approach seemed justifiable to obtain at least a rough density estimate for comparison with the other strata. The density estimated for WAP was 0.0006 ind./km 2 (CV 0.39, 95% CI: 0.0003-0.0012).
Predictions of AMW densities in EAP, WS, and NM showed a high variability over space and time (Figures 7-9). Predicted densities varied strongly between strata, with overall highest densities found in stratum EAP and overall lowest in NM. Within each stratum, predicted densities differed greatly between dates chosen for prediction, depending on the respective ice concentration therein. In EAP, densities ranged from 0.0082 (CV 0.25) to 0.0131 (CV 0.22) ind./km 2 , in NM from 0.0024 (CV 0.33) to 0.0040 (CV 0.24) ind./km 2 (Table 3).
In Table 4  Apart from a strong association with the sea ice edge, we also found a relationship of AMW densities in relation to the distance to the shelf break, with higher densities on the shelf and decreasing densities with increasing distance from the shelf. This result is in line with previous studies who found distance to the Antarctic shelf break front (ASBF, Ainley et al., 2011); known also as the Antarctic slope front (Heywood et al., 2014)) or the position of the shelf break (Thiele et al., 2004) important factors describing AMW occurrence. A common observation is higher densities of Antarctic krill-a dominant prey item for AMWs (Friedlaender et al., 2006)-in the vicinity of the shelf break area in various locations throughout the Southern Ocean (Siegel & Watkins, 2016).

| D ISCUSS I ON
These studies, however, had some difficulties separating the influence of the position of the shelf break from the ice edge, for example, in the Amundsen and Bellingshausen Seas, where these two factors mostly coincide during summer (Ainley et al., 2011).
Our surveys, covering the Weddell Sea, where even in summer the sea-ice extent largely exceeds the area of the shelf, suggest both parameters go some way to explain variation in AMW distribution.
Highest densities of AMWs were predicted in areas where the ice of densities of all other strata (based on the assumption that availability bias for AMWs remains relatively constant throughout the strata). Our encounter rate (whale per surveyed km) is also substantially lower than those reported by Thiele et al. (2004) and Friedlaender et al. (2006). This may partly be attributed to lower availability during aerial surveys compared to shipboard surveys which observations by Thiele et al., (2004) and Friedlaender et al. (2006) were based on. Furthermore, minke whales are known to be encountered particularly in the embayments of the Antarctic Peninsula close to shore (Friedlaender et al., 2006;Williams, Hedley, & Hammond, 2006). These were covered by our surveys only to a limited extent, owing to an even survey coverage across the area. However, low encounter rates may also be related to the absence of ice in the area covered by our surveys, located further north around the tip of the Antarctic Peninsula compared to the other studies. In all our survey years, very low ice concentrations were encountered in stratum WAP. Observations from the Bellingshausen and Amundsen Sea have shown that more AMWs were observed in years with more ice compared to years with less ice along the Antarctic Peninsula (Kasamatsu, Ensor, Joyce, & Kimura, 2000;Thiele et al., 2004). Warmer sea-surface temperatures, fewer cold water intrusions, and the smaller extents of sea ice have been suggested be related to the difference in distribution of minke whales along the Antarctic Peninsula (Kasamatsu et al., 2000). With sea ice providing habitat for the overwintering of phytoplankton and krill, seasons of high winter ice extension typically support greater primary production than seasons with less ice-covered habitat in winter (Loeb et al., 1997;Meyer et al., 2017;Nicol et al., 2008). Overall winter sea-ice extent has recently been suggested to directly affect the body condition of krill predators after the feeding seasons (Bengtson Nash et al., 2017;Braithwaite, Meeuwig, Letessier, Jenner, & Brierley, 2015;Seyboth et al., 2016).

| Error discussion
There are several sources of potential error associated with the results of this study. It has to be noted that encounter rates and densities in this study are minimum estimates, because no correction for availability or g(0) during the surveys could be undertaken. in the field-of-view of observers, compared to shipboard surveys.
Another concern, that ice cover could increase complexity in the visual field and thus influence detection probability of whales, was controlled by testing ice cover as a factor during detection function modeling. Ice cover dropped out as an explanatory variable, thus providing no evidence for ice cover impairing or enhancing the detectability of whales to the observers during these surveys.
Unfortunately, there is no evidence to indicate how availability differs across the entire study region. Differences in availability between open water and ice-covered areas may also apply, but again, there is no strong evidence to indicate the directions and magnitudes of such bias. We therefore decided not to apply a generic correction factor, but instead to use minimum density estimates as a baseline to reduce additional sources of error and assump- The position of the sea ice edge, as well as sea ice concentration, were derived from ASI satellite data. Herein lies another source of potential error in this study, due to the inaccuracy associated with these data sets (Zhao, Su, Stein, & Pang, 2015). For example, AMWs could not have been spotted in 100% of ice concentration (i.e., not visible to an observer). The ASI product does not discriminate between different thicknesses of ice. Very thin ice-physically not a hindrance for breathing for AMWs-could be identified as solid ice cover in a similar way as might a thick floe of ice. Moreover, the relatively coarse resolution of the used ASI product of 6.25 km may lead to generalizations. Also, the spatial and temporal variability of the ice concentration may introduce additional uncertainties. Ice drift can easily be in the order of a 6.25 km grid cell per day (Heil & Allison, 1999), while a daily product of satellite-based sea ice concentration estimates can originate from overflights up to 24 hr apart. As no exact observation time is associated to each individual grid cell in the ASI sea ice concentration product, the surveys may originate from spatially and temporal shifted sea ice conditions compared to the ASI ice concentration product.
Finally, the results of this study apply to only one sector of the Southern Ocean. The observed longitudinal gradient across the Weddell Sea and the Antarctic Peninsula alone points to potential differences in AMW densities around Antarctica. In addition, sea ice conditions differ spatially as well, in addition to substantial variations throughout the summer season. While the WAP is seeing dramatic decreases in sea ice, other areas of the Antarctic are experiencing increases in winter sea-ice cover (Parkinson & DiGirolamo, 2016). Therefore, information from other sectors of the Southern Ocean is necessary to complete the picture and to fully assess AMW relationships with sea ice.

| CON CLUS ION
This study increases and extends our knowledge about relative densities of AMWs in ice-covered as well as open waters of the Southern Ocean. The multiyear data set provides indication for a consistent relationship of AMW distribution with the position of the sea ice edge and across a range of ice concentrations. Our results provide no indication for numbers of AMWs in ice-covered waters highly variable in space and time. This weakens the support for the hypothesis that varying numbers of AMWs in ice-covered waters were responsible for observed changes in AMW abundance between the two IWC circumpolar abundance surveys in the late 1980s and 1990s, respectively. The potential decline in AMW abundance stresses the need for conservation measures and further studies into the AMW population status.
Lowest densities of AMW observed in the areas with lowest ice concentration surveyed in this study (i.e., in WAP), are consistent with the hypothesis that this species is dependent on sea ice. The nature and magnitude of future responses of AMWs as a species are uncertain in novel environments as projected under climate change. An overall decrease of sea-ice cover would narrow available habitat for AMWs. Our results suggest that the forecasted sea ice changes (Böning, Dispert, Visbeck, Rintoul, & Schwarzkopf, 2008;Rignot et al., 2008;Silber et al., 2017) have the potential of heavily impacting AMWs.

ACK N OWLED G M ENT
We would like to thank Captains Pahl, Schwarze and Wunderlich and