Flexible discrete space models of animal movement

Movement drives the spread of infectious disease, gene flow, and other critical ecological processes. To study these processes we need models for movement that capture complex behavior that changes over time and space in response to biotic and abiotic factors. Penalized likelihood approaches, such as penalized semiparametric spline expansions and LASSO regression, allow inference on complex models without overfitting. Continuous-time Markov chains (CTMCs) have been recently introduced as a flexible discrete-space model for animal movement. Modeling with CTMCs involves discretizing an animal's path to the resolution of a raster grid. The resulting stochastic process model can easily incorporate environmental and other covariates, represented as raster layers, that affect directional bias and overall movement rate. We introduce a weighted likelihood approach that allows for modeling movement using CTMCs, with path uncertainty due to missing data modeled by imputing continuous-time paths between telemetry locations. The framework we introduce allows for inference on CTMC movement models using existing software for fitting Poisson regression models, including penalized versions of Poisson regression. The result is a flexible, powerful, and accessible framework for modeling a wide range of animal movement behavior.