To determine the importance of soil variables relative to more commonly used topo-climatic or remotely sensed variables in species distribution models (SDMs) for understory plants.

 

White Mountain National Forest, New Hampshire, U.S.A.

 

We fit models for presence of 41 forest understory plant species across 158 plots using soil, topographic, and spectral predictors to determine the relative contribution of different predictor types. We determined (a) if the potential importance of soil variables is greater than generally described in SDM literature, (b) which predictors are most important, and (c) if a standard subset of predictors can be used to effectively model all species.

 

Models containing all three predictor types performed best. Soil and topographic variables had comparable importance; spectral variables were of lesser importance. The best predictor variable was B horizon carbon to nitrogen ratio (B C:N), followed by topographic position index, elevation, and B horizon exchangeable calcium (B Ca). No standard subset effectively modeled all species.

 

 Our results and those of other SDMs that include in-situ soil geochemical data suggest that soil variables are increasingly important with more detailed descriptions of soils. Soil fertility data, such as B C:N and B Ca, are particularly important in acidic, forest soils where pH is a poor indicator of fertility. Commonly used topo-climatic variables provide meaningful predictions but are limited by their use of indirect predictor variables, inhibiting transferability and interpretability. The poor performance of models created using standard subsets of variables highlights the uniqueness of each species’ niche and the need to combine flexible model building techniques with a variety of predictor variables.