Abstract
Biomass production in ecosystems is a complex process regulated by several facets of biodiversity and species identity, but also species interactions such as competition or complementarity between species. For studying these different facets separately, ecosystem biomass is generally partitioned in two biodiversity effects. The composition effect is a simple, linear effect, and the interaction effect is a more subtle, nonlinear effect. Here we used a clustering approach (1) to separately and comprehensively capture all linear and nonlinear effects induced by both biodiversity effects on ecosystem functioning, and (2) to determine the functional composition at the origin of each biodiversity effect. We used data from the long-term Cedar Creek BioDIV experiment carried out over 22 yr, and we partitioned multiplicatively the biomass in composition and interaction effects. Both biodiversity effects were weakly correlated. Our clustering approach accurately explains and predicts each diversity effect over time: each one is modeled by a different functional composition. Even if environmental conditions and the strength of interaction effect strongly varied over time, the functional clusters of species that govern the interaction effect do not change over the 22 yr of the experiment. The functional composition governing the interaction effect is therefore very robust. In contrast, the functional clusters of species that govern the composition effect are less robust and change with environmental conditions. Understanding ecosystem functioning therefore requires that ecological properties are first partitioned by type, then each type of property is analyzed and modeled separately. Approaches without a priori groupings of species, such as functional clustering, appear particularly efficient and robust to unravel the web of species interactions, and identify the role played by species on biodiversity effects.