Soil Property and Class 100-Meter Grid Maps of the Conterminous United States: Soilgrids+.

Three national US soil point datasets — the National Cooperative Soil Survey (NCSS) Characterization Database, National Soil Information System (NASIS), and the Rapid Carbon Assessment (RaCA) dataset — were combined with a stack of over 200 environmental datasets to generate complete coverage gridded soil map predictions at 100 m spatial resolution. Soil properties (percent organic C, total N, bulk density, pH, and percent sand and clay) and US soil taxonomic classes (291 great groups and 78 modified particle size classes) were predicted for the conterminous US. Soil property predictions were generated at seven standard soil depths (0, 5, 15, 30, 60, 100 and 200 cm). Model validation results indicate an out-of-bag classification accuracy of 60% for great groups, and 66% for particle size classes; for soil properties cross-validated R-squares ranged from 62% for total N to 87% for pH. Nine independent validation datasets were used to assess prediction accuracies for soil class models and results ranged between 24-58% and 24-93% for great group and modified particle size class prediction accuracies respectively. This approach also allows for cell-by-cell predictions to be assessed by metrics of uncertainty to communicate a level of confidence in each part of a map. This hybrid "SoilGrids+" modeling system incorporates remote sensing data, terrain models, climate grids, traditional soil polygon maps, and machine learning. SoilGrids+ opens up new possibilities for combining traditional (soil survey data) with machine learning technology to make soil data more accessible and easier to update.