Using digital soil maps to infer edaphic affinities of plant species in Amazonia: Problems and prospects.

Amazonia combines semi-continental size with difficult access, so both current ranges of species and their ability to cope with environmental change have to be inferred from sparse field data. Although efficient techniques for modeling species distributions on the basis of a small number of species occurrences exist, their success depends on the availability of relevant environmental data layers. Soil data are important in this context, because soil properties have been found to determine plant occurrence patterns in Amazonian lowlands at all spatial scales. Here we evaluate the potential for this purpose of three digital soil maps that are freely available online: SOTERLAC, HWSD, and SoilGrids. We first tested how well they reflect local soil cation concentration as documented with 1,500 widely distributed soil samples. We found that measured soil cation concentration differed by up to two orders of magnitude between sites mapped into the same soil class. The best map-based predictor of local soil cation concentration was obtained with a regression model combining soil classes from HWSD with cation exchange capacity (CEC) from SoilGrids. Next, we evaluated to what degree the known edaphic affinities of thirteen plant species (as documented with field data from 1,200 of the soil sample sites) can be inferred from the soil maps. The species segregated clearly along the soil cation concentration gradient in the field, but only partially along the model-estimated cation concentration gradient, and hardly at all along the mapped CEC gradient. The main problems reducing the predictive ability of the soil maps were insufficient spatial resolution and/or georeferencing errors combined with thematic inaccuracy and absence of the most relevant edaphic variables. Addressing these problems would provide better models of the edaphic environment for ecological studies in Amazonia.