Application of Digital Soil Mapping Techniques to Refine Soil Map of Baringo District, Rift Valley Province, Kenya

Abstract

Detailed and precise description of soil information is important for both developed and developing countries. Africa is highlighted as the most soil data-challenged land surface in the world and it is the area most in need of improved soil information. Our objective was to compile a detailed soilscape class map for the Baringo area in Kenya by using auxiliary variables (digital elevation model, satellite images, and climate maps). In the first step, we extracted landscape–soil relationships based on soil classes from KENSOTER database. We applied soil spatial prediction based on nine standardized predictor variables: x and y coordinates of the sampling points, two principal components from the seven bands of satellite images explaining 83 % of the total variance, three principal components from the 42 variables of climate database explaining 96 % of the total variance, and slope and elevation from digital elevation model. In the first phase (rule extraction), explanatory and target maps were sampled at 999 random points. In the next phase (prediction), 14 major combined soil classes were predicted based on randomly placed 10,000 points. Distances between point values and centroids of the soil classes were calculated, and the closest were scored with 1 and the others with 0. The scores were kriged to obtain continuous probability estimates. Final map was derived based upon the highest probabilities. Our approach had the clear advantage that real-world variability was represented by stacked layers of smooth probability estimates for the soil classes instead of blurred outputs where neighboring pixels can be differently allocated. Our method is suitable to update old and less detailed soil maps or predict new ones for similar environments in the presence of fine resolution auxiliary information. Validity of the prediction should be appropriately tested.