Inferring ecological selection from multidimensional community trait distributions along environmental gradients

Understanding the drivers of community assembly is critical for predicting the future of biodiversity and ecosystem services. Ecological selection ubiquitously shapes communities by selecting for individuals with most suitable trait combinations. Detecting selection types on key traits across environmental gradients and over time has the potential to reveal underlying abiotic and biotic drivers of community dynamics. Here we present a model-based predictive framework to quantify multidimensional trait distributions of communities (community trait niches), which we use to identify ecological selection types shaping communities along environmental gradients. We apply the framework to over 3600 boreal forest understory plant communities with results indicating that directional, stabilizing, and divergent selection all modify community trait niches and that the selection type acting on individual traits may change over time. Our results provide novel and rare empirical evidence for divergent selection within a natural system. Our approach provides a framework for identifying key traits under selection and facilitates the detection of processes underlying community dynamics.

Methods
Boreal forest understory species trait data, applied trait analysis code, and associated model outputs for reproducing the results in the manuscript "Inferring ecological selection from multidimensional community trait distributions along environmental gradients." Trait data include species-level mean trait values for ten functional traits (plant height, specific leaf area, leaf C:N ratio, leaf dry matter content, leaf P, leaf N, leaf N:P ratio, mycorrhizal status, clonality, and dispersal mechanism) for 39 vascular plant species located found in the understory of Finnish boreal forests. Model output from a multinomial Bayesian joint species distribution model is also provided yielding predictions of species relative abundance across 3,602 vegetation survey sites sampled in 1985-1986, 1995, and 2006 in conjunction with the 8th Finnish National Forest Inventory. Code is provided to process trait data, map predicted species relative abundance to traits, and construct multivariate trait distributions along enviromental gradients (soil fertility, growing degree days, and forest overstory density).