This dataset was updated in October 2017 to reflect ICRAF's improved prediction model, known as the ensemble model. The first version of this file included data predicted purely with the Random Forest model.
Soil samples were collected from each of the selected fields. From each field, a top-soil sample (0-20cm depth) and a sub-soil sample (20-50cm depth) were analyzed. Plots with crop-cutting had a third sample tested,from the 4x4m crop-cutting subplot. Samples from the crop-cutting subplot are identified by the “cc” variable (cc=1if crop-cutting sample).