Predicting soil properties over a region using sample information from a mapped reference area and digital elevation data: a conditional probability approach

In a previous paper [Voltz, M., Lagacherie, P., Louchart, X., 1997, Predicting soil properties over a region using sample information from a mapped reference area. Eur. J. Soil Sci. 48, 19–30] a method for mapping soil properties with acceptable precision and cost was proposed. It combined soil classification and interpolation by kriging, and used sample information from a reference area and simple soil observations over the mapping region. In this paper a new version of the method was developed in which soil patterns are modelled by a conditional probability approach and are used to improve interpolation. The mapping method consists of three stages. First is the classification of a set of sites covering the region according to the soil classification of the reference area. Second is the determination over the reference area of the probability of occurrence of soil classes at a site from the knowledge of the relief and of the soil classes at neighbouring sites. Third is the prediction of the soil properties at unvisited sites by a weighted average of the values of the soil properties at the representative profiles of the soil classes, with the weights being taken proportional to the conditional probability of occurrence of each soil class. The performance of the procedure was evaluated for mapping water content at wilting point in an area of 1736 ha in a physiographic region of Southern France. The method was compared with its previous version, namely classification with kriging, and with ordinary kriging and prediction from a soil map at a scale of 1:100 000. It was clearly more precise than the predictions from the 1:100 000 soil map and close to that of ordinary kriging with measured data. It performed similarly to classification with kriging when prediction points were close to observation points, but provided better results than classification with kriging when prediction points were far from the observation points. This latter result illustrates the benefits of modelling soil patterns as related to topographical features for interpolation purposes.