Digital Soil Mapping of Soil Organic Carbon for Central Chile.

Here we present soil organic carbon (SOC) maps for central Chile (147,959 km²) using the data of 484 pedons and a range of environmental covariates and prediction techniques. The SOC maps were created for 6 different depths corresponding to the GlobalSoilMap specifications. This study used the SCORPAN model with soil pedon data and Soil Taxonomy maps, and environmental covariates: temperature and rainfall, land use, elevation and its derivatives; rock composition and rock age. Four different methods - Multilinear regression, Regression Trees, Rule-based Regression and Random Forest - were evaluated for the prediction of SOC at six depths. Our results suggest that Random Forests performed best in terms of R² and RMSE indices for the predicting of SOC in Chile.