Model-based clustering reveals patterns in central place use of a marine top predator

Satellite telemetry data are commonly used to quantify habitat selection, examine animal movements, and delineate home ranges. These data also contain valuable information concerning dens, nests, roosts, and other central places that are often associated with important life history events and may exhibit unique characteristics; however, using satellite telemetry data to study central places is complicated by common nuances like locational error and animal movement. We coupled a novel modeling framework that accounts for these nuances with an Argos satellite telemetry dataset to examine the spatiotemporal behavior associated with harbor seal haul-out sites on Kodiak Island, Alaska, USA. The methodology incorporates an observation model that accommodates multiple sources of uncertainty in telemetry data and a flexible Bayesian nonparametric model to uncover latent clustering in the telemetry locations. We also contribute extensions to examine the effect of covariates on site selection and to obtain population-level inference concerning central place use. Harbor seal haul-out sites generally occurred in inlets and bays, areas that are isolated from the open water of the Gulf of Alaska. Most individuals selected haul-out sites that were protected from wave exposure. The effects of bathymetry and shoreline complexity on haul-out site selection were variable among individual seals, as were the effects of time of day, time since low tide, and day of year on temporal patterns of haul-out use. As repositories of satellite telemetry data on a wide variety of species accumulate, so do opportunities for using this information to learn about the locations of central places, as well as the temporal patterns in their use. The model-based approach we describe offers a practical and rigorous means for gaining insight concerning these sensitive locations, knowledge of which is important for the effective management and conservation of many species.