The relationship between primary productivity and species diversity is one of the most widely debated in ecology. Disagreement exists about the shape of the relationship, the direction of causality, or whether a general relationship exists at all. Theory has predicted a unimodal hump between productivity and diversity, driven by resource scarcity at low productivity and competitive dominance at high productivity. Recent analyses from a series of globally distributed grassland sites, the nutrient network (NUTNET), demonstrate conspicuous variability in the relationship between productivity and diversity. At present, we are left without a coherent theory for how these two key ecological processes are causally related. Here we argue that lack of a clear bivariate relationship between productivity and diversity results because they are under joint control by other factors, such as resource availability, abiotic growing conditions, disturbance and the size of the species pool, and bivariate analyses cannot tease apart their causal connections. We developed a multivariate model incorporating environmental factors and the reciprocal effects between productivity and diversity and evaluated it with NUTNET data using structural equation modeling. To allow for scale-dependence, analyses were conducted at two spatial scales: averaged by site (across site-scale) and replicated 1-m2 plots within sites (within-site scale).
At the across-site scale, variation in plant biomass was directly associated with soil pH, total phosphorous and texture (r2=0.46). Site-level plant species richness was positively related to rainfall and also directly associated with soil pH and total phosphorus (r2=0.45), but in a direction opposite to plant biomass. There was no evidence at the across-site scale that biomass directly or indirectly, via effects on light availability, influenced species richness or that species richness influenced biomass. At the within-site scale, soil variables had strong direct and consistent associations with plant biomass (r2=0.28) and plant species richness (r2=0.33). Moreover, biomass had a strong indirect negative association with plant species richness, mediated by changes in light availability. Additionally, a weak direct path from biomass to richness remained in the model even after controlling for environmental influences. Our analysis demonstrates that biomass has a negative effect on plant richness at small scales (1-m2 plots) driven by aboveground light competition and that no discernable effect of natural variations in richness on biomass exists at either large or small scales in natural grasslands worldwide. Our model provides a multivariate alternative to bivariate approaches that have failed to incorporate system-level interactions.