Random forest analyses identify major abiotic factors across hierarchical spatial scales shaping biodiversity patterns in riverine metacommunities.

Journal of Biogeography



Theory and experiments strongly support the importance of interactive effects of multiple factors shaping biodiversity, although their importance rarely has been investigated at biogeographically relevant scales. In particular, the importance of higher order interactions among environmental factors at such scales is largely unknown. We investigated higher order interactions of environmental factors to explain diversity patterns in a metacommunity of aquatic invertebrates at a biogeographically relevant scale and discuss the findings in an environmental management context.


All major drainage basins in Switzerland (Rhine, Rhone, Ticino and Inn; 41,285 km2).


Riverine α-diversity patterns at two taxonomic levels (family richness of all benthic macroinvertebrates and species richness of Ephemeroptera, Plecoptera and Trichoptera) were examined at 518 sites across the basins. We applied a novel machine learning technique to detect key three-way interactions of explanatory variables by comparing the relative importance of 1,140 three-way combinations for family richness and 680 three-way combinations for species richness.


Relatively few but important three-way interactions were meaningful for predicting biodiversity patterns among the numerous possible combinations. Specifically, we found that interactions among elevational gradient, prevalence of forest coverage in the upstream basin and biogeoclimatic regional classification were distinctly important.

Main conclusion

Our results indicated that a high prevalence of terrestrial forest generally sustains riverine benthic macroinvertebrate diversity, but this relationship varies considerably with biogeoclimatic and elevational conditions likely due to community composition of forests and macroinvertebrates changing along climatic and geographical gradients. An adequate management of riverine ecosystems at relevant biogeographical scales requires the identification of such interactions and a context-dependent implementation.