Quantifying trait selection driving community assembly: a test in herbaceous plant communities under contrasted land use regimes

Plant traits are particularly important in determining plant community structure. However, how can one identify which traits are the most important in driving community assembly? Here we propose a method 1) to quantify the direction and strength of trait selection during community assembly and 2) to obtain parsimonious lists of traits that can predict species relative bundances in plant communities. We tested our method using floristic data from 32 plots experiencing different treatments (fertilisation and grazing) in southern France. Twelve functional traits were measured on 68 species. We determined the direction and strength of selection on these 12 traits using a metric derived from a maximum entropy model (i.e. lambda). We then determined our parsimonious list of traits using a backward selection of traits based on these lambda values (for all treatments and in each treatment separately). We finally compared our method to two other methods: one based on iterative RLQ and the other based on an entropy-based forward selection of traits. We found major differences in the direction and strength of selection across the 12 traits and treatments. From the 12 traits, plant vegetative and reproductive heights, leaf dry matter content leaf nitrogen content, specific leaf area, and leaf phosphorus content were particularly important for predicting species relative abundances when considering all treatments together. Our method yielded results similar to those produced by the entropy-based approach but differed from those produced by the iterative RLQ, whose selected traits could not significantly predict species relative abundances. Together these results suggest that the assembly of these communities is primarily driven by a small number of key functional traits. We argue that our method provides an objective way of determining a parsimonious list of traits that together accurately predict community structure and which, despite its complementarities with entropy-based method, offers significant advantages.