Incorporating Spatial Autocorrelation in Species Distribution Models

Abstract

Species distribution models, based on ecological niche theory and gradient analysis, require digital maps of environmental factors that influence species distributions, such as topography and climate, as well as spatial information on the species attribute of interest (for example, presence/absence, type, abundance), typically sampled directly or compiled from existing datasets such as museum records. Reflecting the fact that their use spans several disciplines, these types of models have been referred to previously as ‘predictive vegetation mapping’ (Franklin 1995), ‘predictive habitat distribution modeling’ (Guisan and Zimmermann 2000), and ‘niche modeling’ (Stockwell 2007). The terminology seems to be converging on ‘species distribution modeling’, which is used here. It should be noted that it is technically the environmental habitat suitability that is produced (mapped) from these models, which renders them appropriate for studying the distribution of communities/assemblages (Ferrier et al. 2002) as well as species, in addition to a number of related biogeographical variables, such as species richness (Rangel et al. 2006), invasive species (Richardson and Thuiller 2007), and disease transmission (Peterson 2006).