AimSpecies distribution models (SDMs) are an important tool for predicting species occurrences in geographic space and for understanding the drivers of these occurrences. An effect of environmental variable selection on SDM outcomes has been noted, but how the treatment of variables influences models, including model performance and predicted range area, remains largely unclear. For example, although landcover variables included in SDMs in the form of proportions, or relative cover, recent findings suggest that for species associated with uncommon habitats the simple presence or absence of a landcover feature is most informative. Here we investigate the generality of this hypothesis and determine which representation of environmental features produces the best-performing models and how this affects range area estimates. Finally, we document how outcomes are modulated by spatial grain size, which is known to influence model performance and estimated range area.LocationNorth AmericaMethodsWe fit species distribution models (via Random Forest) for 57 water bird species using proportional and binary estimates of water cover in a grid cell using occurrence data from the eBird citizen science initiative. We evaluated four different thresholds of feature prevalence (land cover representations) within the cell (1%, 10%, 20% or 50%) and fit models across both breeding and non-breeding seasons and multiple grain sizes (1, 5, 10, and 50 km cell lengths).ResultsModel performance was not significantly affected by the type of land cover representation. However, when the models were fitted using binary variables, the model-assessed importance of water bodies significantly decreased, especially at coarse grain sizes. In this binary variable-case, models relied more on other land cover variables, and over-or under-predicted the species range by 5-30%. In some cases, differences up to 70% in predicted species ranges were observed.Main conclusionsMethods for summarizing landcover features are often an afterthought in species distribution modelling. Inaccurate range areas resulting from treatment of landcover features as binary or proportional could lead to the prioritization of conservation efforts in areas where the species do not occur or cause the importance of crucial habitats to be missed. Importantly, our results suggest that at finer grain sizes, binary variables might be more useful for accurately measuring species distributions. For studies using relatively coarse grain sizes, we recommend fitting models with proportional land cover variables.