Assessing the applicability of binary land-cover variables to species distribution models across multiple grains

Abstract

ContextSpecies distribution models are widely used in ecology. The selection of environmental variables is a critical step in SDMs, nowadays compounded by the increasing availability of environmental data.ObjectivesTo evaluate the interaction between the grain size and the binary (presence or absence of water) or proportional (proportion of water within the cell) representation of the water cover variable when modeling water bird species distribution.MethodseBird occurrence data with an average number of records of 880,270 per species across the North American continent were used for analysis. Models (via Random Forest) were fitted for 57 water bird species, for two seasons (breeding vs. non-breeding), at four grains (1 km2 to 2500 km2) and using water cover as a proportional or binary variable.ResultsThe models’ performances were not affected by the type of the adopted water cover variable (proportional or binary) but a significant decrease was observed in the importance of the water cover variable when used in a binary form. This was especially pronounced at coarser grains and during the breeding season. Binary representation of water cover is useful at finer grain sizes (i.e., 1 km2).ConclusionsAt more detailed grains (i.e., 1 km2), the simple presence or absence of a certain land-cover type can be a realistic descriptor of species occurrence. This is particularly advantageous when collecting habitat data in the field as simply recording the presence of a habitat is significantly less time-consuming than recording its total area. For models using coarser grains, we recommend using proportional land-cover variables.