Accounting for unobserved spatial variation in step selection analyses of animal movement via spatial random effects

Abstract

Step selection analysis (SSA) is a common framework for understanding animal movement and resource selection using telemetry data. Such data are, however, inherently autocorrelated in space, a complication that could impact SSA-based inference if left unaddressed. Accounting for spatial correlation is standard statistical practice when analysing spatial data, and its importance is increasingly recognized in ecological models (e.g. species distribution models). Nonetheless, no framework yet exists to account for such correlation when analysing animal movement using SSA. Here, we extend the popular method integrated step selection analysis (iSSA) by including a Gaussian field (GF) in the linear predictor to account for spatial correlation. For this, we use the Bayesian framework R-INLA and the stochastic partial differential equations (SPDE) technique. We show through a simulation study that our method provides accurate fixed effects estimates, quantifies their uncertainty well and improves the predictions. In addition, we demonstrate the practical utility of our method by applying it to three wolverine (Gulo gulo) tracks. Our method solves the problems of assuming spatially independent residuals in the SSA framework. In addition, it offers new possibilities for making long-term predictions of habitat usage.