UW Aquatic & Fishery Sciences Quantitative Seminar

Devin Johnson

National Marine Mammal Laboratoy

Continuous-Time Correlated Random Walk Models for Animal Telemetry Data

Abstract:

I will present work on a continuous-time version of the correlated random walk model for animal telemetry data. The continuous-time formulation allows data that have been nonuniformly collected over time to be modeled without subsampling, interpolation or aggregation to obtain a set of locations uniformly spaced in time. The model is derived from a continuous time Ornstein-Uhlenbeck velocity process that is integrated to form a location process. The continuous-time model was placed into a state-space framework to allow parameter estimation and location predictions from observed animal locations. Two previously unpublished marine mammal telemetry data sets will be analyzed to illustrate use of the model, by-products available from the analysis, and different modifications which are possible. A harbor seal data set will illustrate a model that incorporates the proportion of each hour spent on land. Also, a northern fur seal pup data set will illustrate a random drift component to account for directed travel and ocean currents. Frequentist inference is employed in the examples, but, if time permits I will examine a currently "under investigation" method for Bayesian inference.