Search

In digital soil mapping, machine learning (ML) techniques are being used to infer a relationship between a soil property and the covariates. The information derived from this process is often translated into pedological knowledge. This mechanism is referred to as knowledge discovery. This study shows that knowledge discovery based on ML must be treated with caution. We show how pseudo‐covariates can be used to accurately predict soil organic carbon in a hypothetical case study. We demonstrate that ML methods can find relevant patterns even when the covariates are meaningless and not related to soil‐forming factors and processes. We argue that pattern recognition for prediction should not be equated with knowledge discovery. Knowledge discovery requires more than the recognition of patterns and successful prediction. It requires the pre‐selection and preprocessing of pedologically relevant environmental covariates and the posterior interpretation and evaluation of the recognized patterns. We argue that important ML covariates could serve the purpose of providing elements to postulate hypotheses about soil processes that, once validated through experiments, could result in new pedological knowledge. Highlights: We discuss the rationale of knowledge discovery based on the most important machine learning covariatesWe use pseudo‐covariates to predict topsoil organic carbon with random forestSoil organic carbon was accurately predicted in a hypothetical case studyPattern recognition by random forest should not be equated to knowledge discovery [ABSTRACT FROM AUTHOR]

Background: Soils are widely recognized as a non-renewable natural resource and as biophysical carbon sinks. As such, there is a growing requirement for global soil information. Although several global soil information systems already exist, these tend to suffer from inconsistencies and limited spatial detail. Methodology/Principal Findings: We present SoilGrids1km — a global 3D soil information system at 1 km resolution — containing spatial predictions for a selection of soil properties (at six standard depths): soil organic carbon (g kg−1), soil pH, sand, silt and clay fractions (%), bulk density (kg m−3), cation-exchange capacity (cmol+/kg), coarse fragments (%), soil organic carbon stock (t ha−1), depth to bedrock (cm), World Reference Base soil groups, and USDA Soil Taxonomy suborders. Our predictions are based on global spatial prediction models which we fitted, per soil variable, using a compilation of major international soil profile databases (ca. 110,000 soil profiles), and a selection of ca. 75 global environmental covariates representing soil forming factors. Results of regression modeling indicate that the most useful covariates for modeling soils at the global scale are climatic and biomass indices (based on MODIS images), lithology, and taxonomic mapping units derived from conventional soil survey (Harmonized World Soil Database). Prediction accuracies assessed using 5–fold cross-validation were between 23–51%. Conclusions/Significance: SoilGrids1km provide an initial set of examples of soil spatial data for input into global models at a resolution and consistency not previously available. Some of the main limitations of the current version of SoilGrids1km are: (1) weak relationships between soil properties/classes and explanatory variables due to scale mismatches, (2) difficulty to obtain covariates that capture soil forming factors, (3) low sampling density and spatial clustering of soil profile locations. However, as the SoilGrids system is highly automated and flexible, increasingly accurate predictions can be generated as new input data become available. SoilGrids1km are available for download via http://soilgrids.org under a Creative Commons Non Commercial license. [ABSTRACT FROM AUTHOR]