New analytical tools for studying habitat selection in terrestrial mammals

The study of habitat selection is a foundational component of basic and applied animal ecology. Today, habitat selection in mammals is primarily studied using resource selection functions, a class of models that uses logistic regression to compare “used” to “available” habitat. However, these models have several statistical problems, including rampant pseudoreplication from failing to account for autocorrelation in modern animal movement data, no clear guidelines for sampling available habitat, and large amounts of numerical error from sampling too few available points. These problems are widely acknowledged but have no generally accepted solutions, so we propose three new methods for addressing them: likelihood weighting, Gaussian availability sampling, and numerical convergence checks. We demonstrate the practical advantages of these methods over conventional approaches using simulations and empirical data on a water mongoose (Atilax paludinosus), a caracal (Caracal caracal), and a serval (Leptailurus serval), and briefly demonstrate how to apply our methods to animal tracking data using the ‘ctmm’ R package. Broad uptake of these methods could substantially improve our estimates of habitat selection in mammals.