Figure 3.

Variation partitioning of the 32 metacommunities for which environmental or spatial predictors were important. In the remaining 29 metacommunities, neither environmental nor spatial predictors significantly explained observed variation. Fractions [a], [b], [c], and [d] correspond to those due to environment, shared environment-space, space, and unexplained variation, respectively. Environmental and spatial predictors were both important for nine metacommunities and all four fractions obtained from partial redundancy analysis (pRDA) models are presented. Environmental or spatial predictors were important for 19 and 4 metacommunities, respectively, and their explained fractions obtained from RDA models. In all cases, a forward selection procedure was used to select predictor variables.