Concerns over global change have increased interest in quantifying the environmental niches of species and, in particular, investigating how niches change across space and time. However, variation among statistical methods used to quantify niches and the differing assumptions underlying these methods have led to ambiguity in interpretations of the real magnitude of niche differences among species. We currently lack a statistical and theoretical basis for choosing among existing methods to quantify niche differences, making it difficult to evaluate which methods consistently produce reliable results. We ask which techniques provide the most reliable and least biased estimates of niche overlap? To address this question, we performed a systemic evaluation of existing techniques used to quantify niches and we present a new statistical framework that quantifies and compares niches in a gridded environmental space. Our method is robust to known and previously undocumented biases related to the dependence of species occurrences on the frequency of environmental conditions that occur across geographic space. We evaluate within this framework several ordination methods for measuring niche overlap between two species. To fully document uncertainty and statistical bias, we perform this evaluation using simulated species with predefined distributions and known amounts of niche overlap.
Results/Conclusions
Our results suggest that ordination approaches strongly differ in their ability to detect and accurately quantify niche differences and that failure to account for differences in occurrence density and climate availability leads to systematic bias in measurements of niche overlap. Among the most important factors explaining these differences are how the environment varies in relation to species occurrences versus the study region as a whole and how techniques select variables based on this variation. Of the techniques we considered, a Principal Component Analysis that summarizes the entire range of climatic variability found in the study area and which projects occurrences of the species in this multivariate space returned the most reliable estimates of the simulated amounts of niche overlap. Unlike other approaches that attempt to minimize or maximize differences between groups, PCA on the entire environment is less prone to artificially maximizing differences between distributions of the species that are not ecologically relevant. More broadly, quantifying niche dynamics is primary to assessing how successfully SDMs can be transferred from one region or time to another.
