Chile and the Chilean soil grid: A contribution to GlobalSoilMap

Abstract

Detailed soil information for biophysical modelling at the regional/national extent in not always available. The GlobalSoilMap project aims to address this issue by producing a fine resolution digital soil map of the world, with a bottom-up approach, where each participant country develops their own maps. This work represents the first contribution of Chile to the GlobalSoilMap project. We used a CART regression method to link environmental covariates describing soil forming factors with eight selected soil properties (organic carbon, field capacity, permanent wilting point, bulk density, pH, and clay, silt, sand particle size fraction) extracted from legacy data (587 profiles), at six depth intervals, namely 0–5, 5–15, 15–30, 30–60, 60–100 and 100–200cm. Using a bootstrapping technique, we generated 100 realisations for each of the property–depth combination, generating a mean prediction map and its associated uncertainty. All maps were generated using a platform based on Google Earth Engine, which allows rapid analysis and visualisation of the results. The resulting accuracy of the maps are variable, with expected high uncertainty levels in poorly sampled areas. We illustrate the method by presenting maps of soil organic carbon and bulk density, and the use of these products with their associated uncertainties to estimate soil carbon stock for Chile.