Inferences on Spatial Data

Ecological data typically are spatially structured, or autocorrelated. Autocorrelation might be due to spatially structured environmental constraints (e.g., topographic controls on soil moisture). Autocorrelation can also arise from local interspecific interactions (e.g., predator/prey relations, competition), or spatial processes such as dispersal. Spatial structure might also be a legacy of past spatial events (e.g., disturbances); this history is often unobserved. There are two approaches to dealing with autocorrelation in inferential models. The first is to avoid it, by sampling deliberately (Chap. 1 ). Here we adopt the alternative approach, of embracing autocorrelation as a feature of interest in ecological data. The focus here is on multivariate regression, in which we model species abundances in terms of multivariate as well as spatial predictors. The workflow includes pre-processing of spatial data, the analysis itself, and post-processing the results for interpretation and communication. A general summary of the results partitions the variation in species abundances into that explained by the environmental variables as compared to the spatial predictors while also accounting for the spatial structure in the environmental variables themselves.