ENVIREM: An expanded set of bioclimatic and topographic variables increases flexibility and improves performance of ecological niche modeling

Abstract Species distribution modeling is a valuable tool with many applications across ecology and evolutionary biology. The selection of biologically meaningful environmental variables that determine relative habitat suitability is a crucial aspect of the modeling pipeline. The 19 bioclimatic variables from WorldClim are frequently employed, primarily because they are easily accessible and available globally for past, present and future climate scenarios. Yet, the availability of relatively few other comparable environmental datasets potentially limits our ability to select appropriate variables that will most successfully characterize a species’ distribution. We identified a set of 16 climatic and two topographic variables in the literature, which we call the envirem dataset, many of which are likely to have direct relevance to ecological or physiological processes determining species distributions. We generated this set of variables at the same resolutions as WorldClim, for the present, mid-Holocene, and Last Glacial Maximum (LGM). For 20 North American vertebrate species, we then assessed whether including the envirem variables led to improved species distribution models compared to models using only the existing WorldClim variables. We found that including the ENVIREM dataset in the pool of variables to select from led to substantial improvements in niche modeling performance in 17 out of 20 species. We also show that, when comparing models constructed with different environmental variables, differences in projected distributions were often greater in the LGM than in the present. These variables are worth consideration in species distribution modeling applications, especially as many of the variables have direct links to processes important for species ecology. We provide these variables for download at multiple resolutions and for several time periods at envirem.github.io. Furthermore, we have written the ‘envirem’ R package to facilitate the generation of these variables from other input datasets.