Improved usefulness of continental soil databases for agricultural management through local adaptation

Abstract

The usefulness of a continental digital soil map (AfSoilGrids250) was investigated at two levels relevant for agricultural fertiliser recommendations in Rwanda (point locations and administrative sector unit averages). Analysis of 900 soil samples for pH and soil organic carbon (SOC) revealed that fine spatial resolution in digital soil maps can be misleading. In order to promote accurate use of such maps, adequate information must be provided to allow users to assess whether a map is suitable for an intended use. In tests to improve the continental database locally at the levels examined in this study, regression kriging of SOC and pH from the continental data set with only 100 soil samples was found to reduce errors, and the predictions better resembled those in the ‘ground truth’ data set. The improved map was better for administrative sector averages than point locations and was more pronounced for SOC than for pH. As the use of digital soil maps becomes more widespread, it is important to clearly communicate map accuracy for a range of extents and application levels. A list of basic guidelines to facilitate usage and understanding of digital soil maps is presented in which we emphasise the importance to assess uncertainty and accuracy.