A new dispersal-informed null model for community ecology shows strong performance

Abstract Null models in ecology have been developed that, by maintaining some aspects of observed communities and repeatedly randomizing others, allow researchers to test for the action of community assembly processes like habitat filtering and competitive exclusion. Such processes are often detected using phylogenetic community structure metrics. When biologically significant elements, such as the number of species per assemblage, break down during randomizations, it can lead to high error rates. Realistic dispersal probabilities are often neglected during randomization, and existing models make the oftentimes empirically unreasonable assumption that all species are equally probable of dispersing to a given site. When this assumption is unwarranted, null models need to incorporate dispersal probabilities. I do so here, and present a dispersal null model (DNM) that strictly maintains species richness, and approximately maintains species occurrence frequencies and total abundance. I tested its statistical performance when used with a wide breadth of phylogenetic community structure metrics across 3,000 simulated communities assembled according to neutral, habitat filtering, and competitive exclusion processes. The DNM performed well, exhibiting low error rates (both type I and II). I also implemented it in a re-analysis of a large empirical dataset, an abundance matrix of 696 sites and 75 species of Australian Meliphagidae. Although the overall signal from that study remained unchanged, it showed that statistically significant phylogenetic clustering could have been an artifact of dispersal limitations.