Global change alters the stability of biological communities by affecting species richness and how species covary through time (i.e., synchrony). There are few large-scale empirical tests of stability-diversity-synchrony relationships and those mostly focus on the terrestrial realm. Moreover, the effect of synchrony is largely unknown when species only covary at either high or low extremes of abundance (i.e., tail-dependent synchrony), a common phenomenon in ecological communities. Here, we synthesized long-term community time-series data (20+ years of species’ abundances/biomass for 2,668 communities across 7 taxonomic groups) from both terrestrial and freshwater realms and explored how the relationships among richness, synchrony, and stability vary across realms. We also investigated the effect of tail-dependent synchrony on stability across 714 freshwater and 1,954 terrestrial communities. For terrestrial communities, we found a positive diversity-stability relationship and that the tail-dependent synchrony was a more important determinant of stability than the traditional measure of overall synchrony (i.e., based on the covariation of all species). For freshwater communities, only overall synchrony explained some variation in stability. Assessing tail-dependent synchrony can improve our ability to understand why stability varies across different ecosystems and thereby our inferences about the causes of human-mediated biodiversity loss.

Estimating the distribution of phenotypes in populations and communities is central to many questions in ecology and evolutionary biology. These distributions can be characterized by their moments: the mean, variance, skewness, and kurtosis. Typically, these moments are calculated using a community-weighted approach (e.g. community-weighted mean) which ignores intraspecific variation. As an alternative, bootstrapping approaches can incorporate intraspecific variation to improve estimates, and also quantify uncertainty in the estimate. Here, we compare the performance of different approaches for estimating the moments of trait distributions across a variety of sampling scenarios, taxa, and datasets. We introduce the traitstrap R package to facilitate inferences of trait distributions via bootstrapping. Our results suggest that randomly sampling ~9 individuals per sampling unit and species, focusing on covering all species in the community, and analysing the data using nonparametric bootstrapping generally enables reliable inference on trait distributions, including the central moments, of communities.