Using a Continuous Time Correlated Random Walk and Bayesian Inference to Examine Spatial Use Patterns of Harbor Seals in Cook Inlet, Alaska, USA

Seventy-one harbor seals (Phoca vitulina) were tagged with satellite-linked geo-locating depth recorders in Cook Inlet, Alaska between 2004 and 2006. The recorders were deployed in the fall, after molt, and in the spring before pupping and breeding. With an average deployment length of 188 days, the data included over 62,000 ARGOS locations across most months (lower in July and August). We used a continuous-time correlated random walk ('crawl' package in R) movement model to estimate seal locations at hourly intervals based on the locations and haul-out status. Spatial use patterns were determined by drawing (n=1000) from the posterior distribution of the predicted tracks. We examined the effects of season (fall, early winter, late winter, spring, and summer) on the spatial use patterns of the tagged seals. Use maps were created on a raster grid with cell size of 100 square kilometers. Two values were calculated for each cell for each simulation: the number of hours all seals spent in the cell, the number of seals that spent any hour in the cell. Separate maps were made for each season. A large portion of seal spatial use was within 5 km of a haul-out and there was a seasonal movement of seals out of Cook Inlet into the western Gulf of Alaska in late fall and winter. For pupping, breeding and molting, seals moved back into Cook Inlet in the spring and early summer. Although these results are similar to those from other harbor seal studies, our Bayesian approach for estimating spatial use patterns accounts for, and provides an estimate of uncertainty due to location error and irregular location times. This method is broadly applicable to other marine mammals tracked by ARGOS.