This study presents a new approach to classifying types of soil based on the probability classes of the relevant set of attributes. Two key ideas are addressed in this study: (i) the use of stochastic simulations to generate a local cumulative distribution function or extreme classes of each attribute and (ii) the use of a multidimensional scaling (MDS) technique to visualize and quantify the relative importance of each attribute in the classification process. After the simulated realizations, the weighted "distances" attributes extreme values (probability classes) of each grid node are calculated and the MDS algorithm is applied for the spatial representation of the grid nodes in a new Cartesian reference frame based on the "distances" of the probability classes of attributes. This allows the classification of soil types based on the clusters in the MDS space, after expert validation. In the second step, a sensitivity analysis of the attributes is performed with MDS: each attribute is made "neutral" one at a time, by assuming the median rather than the extreme values in each grid node before the distance evaluation, and the consequent impact on the shape and centroid displacement of the clusters (soil types) in the MDS reference frame is calculated. Hence, the spatial uncertainty of the soil type/classes and the influence of various properties are evaluated in the MDS reference frame. This method is applied to soils in a region of Brazilian in which the previous classification of soil types has been a crucial tool for precision agriculture management. Using the MDS algorithm, the selected attributes (horizon, textural gradient, colors, saturation, sand content, and clay content) were represented in a two-dimensional plot and grouped into eight clusters distinguished from each other by their characteristics. A sensitivity analysis shows that the horizon and saturation attributes had the greatest influence on determination of the clusters, i.e., the soil types. © 2014 Elsevier B.V.
Year of publication: 2014