Fine Scale Soil Texture Estimation Using Soil Maps and Profile Descriptions

Abstract

Soil scientists have often to face problems of accurate data scarcity. More than the amount of available data, limitations may occur from the estimation algorithms that are used. Classical methods like Simple Kriging (SK) are not able to take into account the soft information provided by a soil map without making approximations. An example of such approximation would be to take as hard value the middle of the intervals derived from the texture class, yielding very poor results. In this paper, the advantages of using a new approach called the Bayesian Maximum Entropy (BME) are investigated. In order to take into account the complex definition of the texture classes that are intimately linked to the soil mapping context, a modification of the method (called BME/MC) has been implemented. Based on a realistic simulated example and using standard statistical criteria (e.g., ME, RMSE, spatial correlation of the errors, etc.), BME/MC is found to be the most efficient method compared to SK, SKMI and BME in the soil texture estimation context.