Soil grids (SOIL_GRIDS)

SoilGrids is (i) a system for automated mapping of soil properties based on global soil profile and covariate data and  machine learning algorithms and (ii) a collection of updatable soil property and class maps of the world at two spatial resolutions: 1 km and 250 m. The statistical models used are global models, which means that they have been calibrated to make unbiased predictions at any location within the global soil mask.

SoilGrids provides at above spatial resolutions predictions of site characteristics (depth to bedrock), physical and chemical soil properties (bulk density, clay, silt and sand percentages, coarse fragments, soil organic carbon,  CEC, pH), all for different depths up to 2 m, and soil classifications (predicted most probable class and predicted probabilities for each WRB soil unit and Soil Taxonomy suborder). The actual mapping accuracy of each targeted soil property and classes is still limited: the amount of variation explained by the models ranges between 30 and 70%. On the other hand, and in comparison to other previous global soil databases, SoilGrids provides an objective estimate of the uncertainty of mapping (average overall error of estimated prediction error). The confidence limits can be used to assess the impact of uncertainty in soil predictions on scenario / model testing through e.g. error propagation techniques.

The initial SoilGrids1km outputs were produced using a compilation of national and regional soil profile databases. SoilGrids invites potential contributors, organizations or individuals, to improve local predictions  by providing soil profile data.