Pacific Walrus Seasonal Distribution from USGS Tracking Data, Chukchi and Bering Seas, 1987-2015

We filtered Argos locations with an algorithm based on movement rates and turning angles (Freitas et al. 2008). We set the algorithm to retain locations within 15 km of the previous or subsequent location, and to retain any location based on a distance-angle-rate filter that accepted a maximum speed of 4 m/s and rejected locations at the apex of tracks with highly acute angles (less than 45°).

We estimated a daily location (at O UTC) for each animal using a continuous-time Correlated Random Walk (CRW) model developed by Johnson et al. (2008) and implemented in package crawl, version 2.0 (Johnson et al., 2016; available online at: https://cran.r-project.org/web/packages/crawl/index.html), in R (R Core Team, 2017). For the CRW location modeling, we assumed that the filtered Argos location errors were normally distributed with a mean of 0 and a standard deviation equal to that declared by Argos System operator, CLS (https://www.clsamerica.com/), for standard class locations (3, 2, and 1). For non-standard class filtered Argos locations, we allowed the CRW to estimate error distributions based on a half normal distribution with semi-informative priors. We constrained the semi-informative priors for filtered locations classes 0, A, and B at least as great as that of the standard class 1 locations declared by CLS (SD=1500 m). We then used information on non-standard class Argos location errors reported by Vincent et al. (2002) to constrain priors to have a mean error of 1500 m and a standard deviation of 5000 m for location classes 0 and A, and 7500 m for location class B. We followed the recommendations of package crawl to parameterize priors for the turning angle and step length autocorrelation CRW. We only retained CRW estimated daily locations obtained within 24 h of an actual filtered Argos location.

We assigned estimated daily locations to one of two seasons (“summer” period was defined as May–November and the “winter” period as December–April).

For each season (summer and winter), we counted the number of walrus tracking days within each 50 km by 50 km cell from a raster grid generated on a study area grid spanning the Pacific Arctic marine waters centered on an Albers equal-area projection with standard parallels of 55° and 65° north latitude, and a central meridian of 154° west longitude. These cell counts were normalized by the total number of cell counts across the study area to provide a value analogous to tracking density.