Selected soil properties were predicted from 232 topsoil samples subjected to spectral analysis (MIRS). A subset of samples were also subjected to wet chemistry analysis, and results were used to calibrate a machine-learning algorithm developed by the International Centre for Research in Agroforestry (ICRAF) in Kenya. Coordinates were truncated to protect farmer's privacy.
Unless specified, all properties were predicted. This dataset can be linked with yield data (coming soon) and previous soil analysis data (https://hdl.handle.net/1152…) through the unique farm identifer "fid".
A link is provided to match terms used in the "terminag" GitHub (https://github.com/reagro/t…) as of June 2025.