Predicting residence time using a continuous-time discrete-space model of leatherback turtle satellite telemetry data

Abstract

The utilization and capabilities of biotelemetry are expanding enormously as technology and access rapidly improve. These large, correlated datasets pose statistical challenges requiring advanced statistical techniques to appropriately interpret and model animal movement. We used satellite telemetry data of critically endangered Eastern Pacific leatherback turtles (Dermochelys coriacea) to develop a habitat-based model of their motility (and conversely residence time) using a hierarchical Bayesian framework, which could be broadly applied across species. To account for the spatiotemporally auto-correlated, unbalanced, and presence-only telemetry observations, in combination with dynamic environmental variables, a novel modeling approach was applied. We expanded a Poisson generalized linear model in a continuous-time discrete-space (CTDS) model framework to predict individual leatherback movement based on environmental drivers, such as sea surface temperature. Population-level movement estimates were then obtained with a Bayesian approach and used to create monthly, near real-time predictions of Eastern Pacific leatherback movement in the South Pacific Ocean. This model framework will inform the development of a dynamic ocean management model, “South Pacific TurtleWatch (SPTW),” and could be applied to telemetry data from other populations and species to predict motility and residence times in dynamic environments, while accounting for statistical uncertainties arising at multiple stages of telemetry analysis.