Habitat patches vary as a function of an organism's perception. The response of species to patch mosaic may depend on the scale of observation by which the spatial pattern is perceived. A hierarchical view of landscapes is proposed to define the optimal grain of landscape affecting soil fauna (Collembola). A multilevel approach is developed to quantify the landscape grain based on the concept of contrast, that is, the magnitude of difference in measures across a given boundary between adjacent patch types. An image segmentation procedure was firstly applied to satellite images to detect edges, and this defines “homogeneous” regions at multiple contrast levels. Spatial features (number of patches, diversity of patches and edge length) were then derived from segmented images to characterise the spatial pattern that surrounds Collembola sampling sites. These landscape descriptors were computed at different levels of contrast and at three spatial scales. The statistical dependence between species occurrence and landscape descriptors was assessed by means of the Mantel test. Biodiversity, estimated by species richness and the number of endemic species, was analysed by stepwise multiple linear regression.
The multilevel approach permitted a definition of the landscape units based on the response of the assemblage to landscape heterogeneity. The effect of landscape heterogeneity was especially evident on species composition and on endemic richness when patches are defined at fine grain. The heterogeneity descriptor “number of patches”, calculated at the optimal contrast level, explained 72% of the endemic richness variance. This spatial feature constituted a reliable inverse indicator of endemic richness that was extrapolated to the pixels of the image providing a spatial model.