A wavelet-based method to remove spatial autocorrelation in the analysis of species distributional data

Abstract

Species distributional data based on lattice data often display spatial autocorrelation. In such cases, the assumption of independently and identically distributed errors can be violated in standard regression models. Based on a recently published review on methods to account for spatial autocorrelation, we describe here a new statistical approach which relies on the theory of wavelets. It provides a powerful tool for removing spatial autocorrelation without any prior knowledge of the underlying correlation structure. Our wavelet-revised model (WRM) is applied to artificial datasets of species’ distributions, for both presence/absence (binary response) and species abundance data (Poisson or normally distributed response). Making use of these published data enables us to compare WRM to other recently tested models and to recommend it as an attractive option for effective and computationally efficient autocorrelation removal.