Abstract: Telemetry data can answer novel questions about animal movement and increase knowledge of wildlife ecology. To extract the information available in telemetry data, we must continue to improve statistical methods that enable estimation of where an animal went and how it moved, with associated measures of uncertainty. Bayesian continuous time models have been developed to estimate the true animal movement trajectory that underlies discretely recorded telemetry data, yet these models remain inaccessible and underutilized. These models allow for the estimation of derived quantities describing an animal’s movement (e.g., velocity, turn angle, length traveled), and the models’ Bayesian formulation allows for quantified uncertainty on the predictions of both an animal’s trajectory and associated derived quantities. However, these models do not incorporate abrupt transitions in animal movement well (e.g., an individual transitioning from sleeping to walking). To account for these transitions, we recognized a novel application of a recently developed machine learning method and incorporated it in the Bayesian continuous time modeling framework. We demonstrated this method using data from a declining grassland bird, the lesser prairie-chicken (Tympanuchus pallidicinctus). We estimated trajectories for 54 GPS-transmittered females over six years in the Mixed-Grass Prairie Ecoregion of Kansas, USA, and compared derived estimates for daily displacement between ranches subjected to patch-burn or rotationally grazed management practices. Our demonstrated method can be easily applied to any telemetry data to compare derived quantities (e.g., average velocity, rest time) across treatments using accessible R packages.
