Production of global soil data predictions using scikit-map within the EO-SoilMapper framework.
Target spatial resolution is 30 m (global coverage excluding deserts and permanent ice).
Target period of interest is 2000 to 2022+ with 5 year intervals (spacetime block predictions).
Depth intervals provided: 0–30, 30–60, 60–100 cm.
Predictions are based on using Quantile Regression Random Forest with with output predictions showing
the mean plus the lower and upper prediction intervals of 68% probability to approximate one standard deviation.
Repository contains detailed description of how were the models fitted and how were some visualizations produced.
Layers available in the current version of the OpenLandMap-soildb include:
- Soil organic carbon density [kg/m3];
- Soil organic carbon content [g/kg];
- Soil pH in H2O [-];
- Bulk density fine earth [kg/m3];
- Soil texture fraction clay-silt-sand [%];
- USDA subgroup taxa [-];
Full list of layers in available in this table. To access layers at finest resolution please use the S3 links.
Additional layers are available via https://stac.openlandmap.org. To cite layers distributed via OpenLandMap-soildb please use:
- Hengl, T., Consoli, D., Tian, X., and others, (2025??). OpenLandMap-soildb: global soil information at 30~m spatial resolution for 2000--2022+ based on spatiotemporal Machine Learning and harmonized legacy soil samples and observations. Submitted to ESSD journal,