Effects of spatial scale and choice of statistical model (linear versus tree-based) on determining species–habitat relationships

Abstract

The choice of spatial scale and modelling technique used to capture species–habitat relationships needs to be considered when ascertaining environmental determinants of habitat quality for species and communities. Fish densities and environmental data were collected at three Laurentian lakes using underwater surveys by four snorkelers collecting fine spatial data acquired through geographic positioning systems. At both fine (20 m) and broad (100 m) spatial scales, tree-based approaches, which incorporated both linear and nonlinear relationships, explained more variation than their linear counterparts. At the finest spatial scale considered (20 m), local environmental conditions, such as habitat structure and heterogeneity, were important determinants of fish habitat selection. At the broadest spatial scale considered (100 m), fish tended to select habitat based on both local environmental features and riparian development. Moran’s eigenvector maps further revealed that fish–habitat associations were operating at broader spatial scales than the predefined analytical units, which can be partially attributed to the spatial structure of environmental conditions acting at spatial scales greater than 100 m. This study highlights the importance of evaluating statistical approaches at different spatial scales to identify key determinants of habitat quality for species, ultimately to assess the effects of perturbations on ecosystems.