Black soils in Argentina

Spatial prediction was done following a digital soil mapping framework. Soil profiles were classified into black soil (833) or non black soil (3007) according to the BS definitions. We used soil horizons up to 25 cm. Soil organic carbon (obtained by wet combustion), Munsell colour and horizon boundaries were used for the analysis. We filtered out organic horizons as well as those profiles with missing data. Model calibration was carried out using a classification random forest model. We predicted the probability black soil occurrence and took 0,5 as the splitting threshold. Validation was done by 10-fold 20 times repeated cross-validation.

We obtained a 75% global accuracy and a kappa of 0,52.