Sexual segregation in a highly pagophilic and sexually dimorphic marine predator

Sexual segregation is common in many species and has been attributed to intra-specific competition, sex-specific differences in foraging efficiency or in activity budgets and habitat choice. However, very few studies have simultaneously quantified sex-specific foraging strategies, at sea distribution, habitat use, and trophic ecology. Moreover, these studies come from low latitude areas reflecting a lack of evidence for polar species. We investigated sexual segregation in snow petrels Pagodroma nivea and combined movement, foraging trip efficiency, stable isotope and oceanographic data to test whether sexual segregation results from sex-specific habitat use. Breeding birds foraging in the Dumont d’Urville sea, Antarctica, were tracked during incubation. Space-use sharing and utilization distribution were similar between males and females indicating no spatial segregation. Males and females foraged more in waters ≈400 m deep and less in waters deeper than ≈1000 m. There was no difference in δ13C values between males and females. Females foraged less than males in areas with higher sea ice concentration (SIC >70%) and had lower δ15N values in plasma, blood cells and feathers. Male and female foraging trip performances (trip duration, length, speed and directions, mass gain, proportion mass gain) were similar, but foraging efficiency (proportionate daily mass gain while foraging), was greater for females than for males, and was greater for larger females with deeper bills. Females were more efficient than males during short (<2 days) foraging trips. For females, but not for males, mass gain, proportion mass gain and body condition at return from a foraging trip were positively correlated to SIC of the foraging areas. Together, these results indicate that sexual segregation in snow petrels during incubation is mainly driven by habitat segregation between high (>70%) more profitable SIC and low SIC areas, probably driven by intra-specific competition.

latitude of the foraging habitat (Cherel & Hobson 2007, Jaeger et al. 2010). Plasma has a half-200 life of about 3 days (Hobson & Clark 1993), a shorter period than the average duration of 201 foraging trip during incubation (7 days, Barbraud et al. 1999), and represents prey ingestion 202 and trophic ecology during the last trip before sampling (Cherel et al. 2005a). Blood cells 203 have a half-life of about 30 days (Hobson & Clark 1993) and represent dietary information 204 integrated over a few months. Feathers contain dietary information at the time they were 205 grown, because keratin is inert after synthesis (Hobson & Clark 1992, 1993 2002). In snow petrels body moult is a gradual process extending over at least 4 months in 207 summer and autumn. It begins during incubation, but most body feathers grow in the weeks 208 following completion of breeding, i.e. from February to April (Maher 1962, Beck 1969. 209 Therefore, isotopic values of body feathers contain information about diet near the end of the 210 previous breeding season and the beginning of the previous non-breeding season.

211
Feathers (one single feather per bird) were cleaned to remove surface contaminants using 212 a 2:1 chloroform:methanol solution followed by two methanol rinses. They were then oven 213 dried for 48 h at 50°C and cut into small pieces using stainless steel scissors. Blood cells and 214 plasma samples were freeze-dried and powdered. Since avian plasma, unlike blood cells, 215 contains a high and variable lipid content that affect its δ 13 C values, lipids were removed from 216 plasma samples using chloroform/methanol ( Cherel et al. 2005a, Cherel et al. 2005b. Then, 217 tissue sub-samples were weighed with a microbalance (aliquots mass:  0.3 mg dw), packed 218 in tin containers, and nitrogen and carbon isotope ratios were subsequently determined at the 219 laboratory LIENSs by a continuous flow mass spectrometer (Thermo Scientific Delta V 220 Advantage) coupled to an elemental analyser (Thermo Scientific Flash EA 1112). Results are 221 presented in the usual  notation relative to Vienna PeeDee Belemnite and atmospheric N 2 for 222 δ 13 C and δ 15 N, respectively. Replicate measurements of internal laboratory standards (acetanilide and peptone) indicate measurement errors <0.15 ‰ for both  13 C and  15 N 224 values. 225 226 2.4 Foraging analysis and spatial usage 227 Spatial and statistical analyses were performed using R 3.2.1 using the "stats" package (R 228 Development Core Team 2015) and "adehabitatLT" package (Calenge 2006, Calenge et al. 229 2009). From the GPS recorded data, foraging trips were reconstructed and data were 230 rediscretized to have one location each 40 min. Some of the trips were largely incomplete 231 (return journey not initiated; n = 15 corresponding to the 15 min intervals) because of battery 232 limitations and were removed from the analysis. For each complete (n = 40) and incomplete 233 (return journey initiated; n = 7) foraging trip, we computed the following foraging indices: 234 maximum distance to the colony (Dmax, km), average movement speed (MS, km h -1 ) and Using these metrics we performed a randomization procedure to test the null hypothesis that 249 there was no difference in the spatial distribution of males and females at the population level 250 (Breed et al. 2006). The sex of each bird was randomly assigned using the observed sex ratio 251 in our data set and the overlap metric between males and females was calculated for 25%, 252 50%, 75% and 95% kernels. We performed 1000 randomizations of our dataset from which 253 the probability of accepting the null hypothesis was calculated as the proportion of random 254 overlaps that were smaller than the observed overlap. Since we were testing only if the 255 observed overlap was smaller than random overlap, we considered this as a one-tailed test.

256
Second, we tested the null hypothesis that there was no difference in the extent of overlap in 257 spatial distribution of males and females at the individual level.

258
For each foraging trip we also calculated the following metrics from the phenotypic data: 259 the body mass change (m, in g) between departure and arrival of a foraging trip, the daily 260 mass gain (Mday, in g.day -1 ) calculated as the ratio between m and the trip duration, the 261 proportion mass gain calculated as the ratio between m and mass at departure for a foraging 262 trip, and the proportion daily mass gain calculated as the ratio between Daym and mass at 263 departure for a foraging trip. A body condition index before departure and after return from a 264 foraging trip was also calculated. To estimate the body condition we used the body 265 measurements to calculate the scale mass index (SMI) as recommended by Peig and Green 12 where M i and L i are, respectively, the body mass and the PC1 score of the individual i, L 0 , is 272 the value of PC1 for the whole studied population and b the slope estimate of the RMA 273 (Reduced Major Axis) regression of log-transformed body mass on log-transformed PC1.  Similarly, the total variance in d col /D max occurring after t/T was plotted against t/T and the t/T 292 value from which a monotonic decrease of the variance began was recorded. Tracking 293 locations recorded after this point were classified as those corresponding to the return trip, and 294 locations between both points were considered as foraging locations.
Previous studies have shown that the snow petrel is a sea ice obligate species and remains 296 highly associated with sea ice year round (Griffiths 1983, Ainley et al. 1984, 1986, 1992, 297 1993, Delord et al. 2016. We therefore used sea ice concentration (SIC) to describe the 298 foraging habitat of snow petrels. Although sea surface temperature is commonly used to 299 describe foraging habitats in seabirds, there are very few sea surface temperature observations 300 in regions covered by sea ice, especially in the Southern Ocean (Rayner et al. 2003).

301
Therefore this covariate could not be used. We used passive-microwave estimates of daily sea 302 ice concentration from the Special Sensor Microwave Imager (SSMI/I) brightness 303 temperatures (12.5  12.5 km resolution) from the Institut Français de Recherche pour 304 l'Exploitation de la Mer (Ifremer, ftp://ftp.ifremer.fr/ifremer/cersat/products/gridded/psi-305 concentration/data/antarctic). We also used bathymetry data (ocean depth at one-minute 306 horizontal spatial resolution) obtained from NOAA's ETOPO 307 (https://sos.noaa.gov/datasets/etopo1-topography-and-bathymetry/) as an additional habitat 308 variable. Daily sea ice concentration and depth values were extracted for each foraging 309 location (therefore excluding the commuting part of the trips at sea) on each track using 310 bilinear interpolation from the native ice and depth grids using "raster" package in R 311 (Hijmans 2018). Since snow petrels are highly associated with the sea ice region (as defined 312 by the region within >15% sea ice concentration isocline, Cavalieri et al. 1991), the SIC data 313 were filtered to retain SIC values >15%. was the tracking location, which was coded as 1 for a foraging location and as 0 for a 343 commuting location, and explanatory variables were sea ice concentration and bathymetry.
to interpret in complex nonlinear models, separate models were developed for male and 346 female birds. Models included sea ice concentration and bathymetry as fixed factors, and bird 347 identify as a random term to account for pseudoreplication issues. The smoothing parameter 348 was chosen automatically using generalized cross-validation. To model spatial auto-349 correlation an isotropic thin plate spline was included, set up as a two dimensional smoother 350 based on both x and y coordinates (Cleasby et al. 2015). To ascertain whether collinearity 351 between covariates may have occurred we examined the correlations between environmental 352 variables using a Spearman correlation coefficient since covariates were not normally 353 distributed. We assumed that a correlation of greater than r s 0.4 was problematic, but the 354 correlation was below this threshold (r s = 0.14).

366
Male snow petrels were structurally larger than females, particularly for bill length and bill 367 depth, and were 10% heavier than females (Table 1). Bill length, bill depth and body mass 368 were the most sexually dimorphic phenotypic traits.

Spatial distribution of males and females and habitat differences 371
Males and females foraged in offshore waters to the east and to the west of the colony in 372 equal proportions (²=0.03, p=0.86, Figure 1). Space-use sharing was similar between males 373 and females as the UDOI was not significantly lower than the null expectation for 25%, 50%, 374 75% or 95% UDs ( Table 2). The 95% UDOI was > 1, indicating a higher than normal overlap 375 between male and female UDs relative to uniform space use, i.e. male and female UDs were 376 non-uniformly distributed and had a high degree of overlap. By contrast, the 25% UDOI was 377 relatively close to 0 indicating less overlap between male and female UDs relative to uniform 378 space use. Males and females UDs were also similar whatever the UDs considered since BA 379 were not significantly lower than the null expectation for 25%, 50%, 75% or 95% UDs (Table   380 2).

381
In average males foraged in areas with higher SIC than females (Table 3). Fitted models 382 on foraging probability contained sex-specific smoothers for bathymetry and SIC (Table 4).

383
For females, the GAMM model explained 10% of the deviance of foraging probability. All 384 smoothers for SIC and bathymetry were significant (Table 4). Foraging probability increased 385 sharply with increasing SIC up to 30% and more smoothly for high SIC (Figure 2). Foraging 386 probability showed a first peak at depth of 600 m and a second and high peak at depth of 387 1600 m. Foraging probability sharply increased at depths >2500 m but sample size was 388 small and there was high uncertainty. Both the random intercept for bird identity and the 389 spatial smoother were significant.

390
For males, the model explained 4.6% of the deviance of foraging probability. All 391 smoothers for SIC and bathymetry, the random intercept for bird identity and the spatial 392 smoother were significant (Table 4). Male foraging probability varied non-linearly with SIC 393 and bathymetry. It increased smoothly with increasing SIC, and was higher when SIC was higher than 90% (Figure 2). Foraging probability also increased with bathymetry up to 600

538
Second, females may feed on Antarctic silverfish in similar proportions than males but on 539 smaller sized individuals (i.e. younger). It is known that  15 N values increase with body 540 length (and age) in Antarctic silverfish from 7-8‰ in larvae (10-20 mm standard length) to 541 10-11‰ in juvenile and adult fish (Giraldo et al. 2011, Pinkerton et al. 2013. It is currently 542 unknown whether sea ice concentration and characteristics differentially affect the spatial distribution of Antarctic silverfish age-classes. However, it is likely that females fed more on 544 crustaceans than on young silverfish since crustaceans have much lower  15 N values than 545 young silverfish (Cherel 2008). Thus, our results suggest that males ate more silverfish in 546 areas with higher sea ice concentration.

547
The strong positive relationship between plasma  15 N and blood  15 N indicates short term 548 (over weeks) consistency in trophic level between successive foraging trips during incubation.

549
Values of  15 N in plasma and feathers did not differ in both sexes (Appendix 1), but blood 550  15 N were smaller than feather and plasma  15 N in both sexes, suggesting that males and 551 females fed on lower trophic level prey prior to incubation than during the breeding season.

552
Short and long term consistency in foraging water masses was also low as indicated by the 553 lack of relationship between plasma and blood  13 C, and between feather and blood  13 C, 554 respectively. Indeed tracking data indicated that birds foraged on the continental shelf, 555 continental margin, and to a lesser extent in oceanic waters. Values of  13 C in feathers were 556 higher than those in blood and plasma for both sexes (Appendix 1), suggesting that during the 557 latter part of the breeding season and the beginning of the non-breeding season snow petrels 558 foraged in more oceanic waters (snow petrels start molting during the chick rearing period 559 and until early May (Beck 1969, 1970, Delord et al. 2016. This period coincides with the sea 560 ice growth and its northward extension.

561
The negative relationship between mass gain (and proportion daily mass gain) during a 562 foraging trip and body condition at departure for a foraging trip (i.e. at the end of fasting 563 while incubating the egg), indicated that males and females were able to regulate their body 564 reserves as found in other Procellariiformes species (Chaurand & Weimerskirch 1994, 565 Gonzales- Solis et al. 2000). Although both sexes regulated body condition, this ability seemed 566 greater for females than for males. Indeed, body condition at departure for a foraging trip was 567 lower in females than in males, but similar for both sexes at return from a foraging trip despite similar trip durations. This is further supported by the fact that females had higher daily mass