Two dominant forms of multisite similarity decline - Their origins and interpretation.

Abstract

The number of species shared by two or more sites is a fundamental measure of spatial variation in species composition. As more sites are included in the comparison of species composition, the average number of species shared across them declines, with a rate increasingly dependent on only the most widespread species. In over 80% of empirical communities, models of decline in shared species across multiple sites (multisite similarity decline) follow one of two distinct forms. An exponential form is assumed to reflect stochastic assembly and a power law form niche-based sorting, yet these explanations are largely untested, and little is known of how the two forms arise in nature. Using simulations, we first show that the distribution of the most widespread species largely differentiates the two forms, with the power law increasingly favored where such species occupy more than ~75% of sites. We reasoned the less cosmopolitan distribution of widespread species within exponential communities would manifest as differences in community biodiversity properties, specifically more aggregated within-species distributions, less even relative abundance distributions, and weaker between-species spatial associations. We tested and largely confirmed these relationships using 80 empirical datasets, suggesting that the form of multisite similarity decline offers a basis to predict how landscape-scale loss or gain of widespread species is reflected in different local-scale community structures. Such understanding could, for example, be used to predict changes in local-scale competitive interactions following shifts in widespread species' distributions. We propose multiple explanations for the origin of exponential decline, including high among-site abiotic variation, sampling of highly specialized (narrow niche width) taxa, and strong dispersal limitation. We recommend these are evaluated as alternative hypotheses to stochastic assembly.