The relative importance of niche‐based (e.g., competitive or stress‐based) and stochastic (e.g., random dispersal) processes in structuring ecological communities is frequently analyzed by studying trait distributions of co‐occurring species. While filtering processes, such as the exclusion of stress‐intolerant species from particular habitats, increase the trait similarity between co‐occurring species, other processes, such as resource competition, can limit the similarity of co‐occurring species. Comparing the observed trait distribution patterns in communities to null expectations from randomized communities (e.g., a draw of the same observed richness from the regional pool) therefore gives a first indication of the dominant process driving community assembly.
However, such comparisons do not inform us about the relative contribution of these different processes in shaping community compositions in case of their joint operation (a likely scenario). Using an Approximate Bayesian Computation approach, we develop a new method that allows inference of the relative importance of dispersal, filtering, and limiting similarity processes for the assembly of observed communities with known species and trait composition. We applied this approach to a tree community data set, collected across 20 plots along strong rainfall and fire gradients in a South African savanna.
Based on comparisons with simulations, we find that our new approach is powerful in identifying which community assembly scenario has the highest probability to generate the observed trait distribution patterns, while traditional null model comparisons perform poorly in detecting signs of limiting similarity. For the studied savanna tree communities, our analysis yields that dispersal processes are most important in shaping the functional trait distribution patterns. Furthermore, our models indicate that filtering processes were relatively most important in areas with high fire frequencies, while limiting similarity processes were relatively most important in areas with low fire frequency and high rainfall. We conclude that our new method is a promising improvement on current approaches to estimate the relative importance of community assembly processes across different species groups, ecosystems, and biomes. Future model modifications (e.g., the inclusion of individual‐based processes) could provide further steps in uncovering the underlying assembly processes behind observed community patterns.