Quantitative metrics of overlaps in Grinnellian niches: advances and possible drawbacks

Abstract

Aim Studies of environmental niche shift/niche conservatism that are based on species distribution modelling require a quantification of niche purity and potential overlap. Although various metrics have been proposed for this task, no comparisons of their performance are available yet that express the linearity of range shifts and error-proneness. Herein, we assess the performance of six niche overlap metrics using three sister pairs of plethodontid salamanders as well as artificial species to test for linearity of overlap curves, impacts of varying potential distribution sizes and study area sizes. Location North America, artificial environments. Methods Species distribution models for the salamanders were performed with Maxent, and artificial species were created in the R environment. Potential distributions of species with varying range sizes and extents of the study area were compared using the Bray–Curtis distance BC, Schoener's D, two different modifications of the Hellinger distance Imod, Icor, Pianka's O and Horn's R. Niche overlaps in ecological space were compared using linear discriminant analyses based on principal components. Results Simulations of niche overlaps revealed strong variations in the performance of the niche overlap metrics. In artificial species, BC and D performed best, followed by O, R and Icor, but the modified Hellinger distance Imod showed a nonlinear slope and a truncated range. Furthermore, the simulations suggest that, in proportionally small potential distributions on large grids, an inclusion of a high proportion of grid cells with low occurrence probabilities representing background noise may bias assessments of niche overlaps. Main conclusions Both the salamander examples and simulations suggest that Schoener's D and the Bray–Curtis distance BC are best suited to compute niche overlaps from potential distributions derived from species distribution models. However, like all analysed metrics, both D and BC are seriously affected by the inclusion of high numbers of grid cells where the species are probably absent, i.e. with low occurrence probabilities. Therefore, pre-processing to eliminate background noise in the potential distribution grids is highly recommended.