Soil-landform relationships at within-field scales: an investigation using continuous classification

Continuous classification was applied to variables measuring landform, extracted from a digital elevation model (DEM) of a field. It was shown that soil physical properties (gravimetric water content at two depths and sand/clay content) and soil mineral nitrogen in the subsoil were linearly related to membership values in these classes (the memberships having been spatially smoothed and transformed from the simplex space to a real space by the log-ratio transform). Adjusted R-2 values between 0.57 and 0.84 were obtained. Stronger evidence existed for linear relationships of these soil properties to the transformed membership values than to the underlying landform variables from which the classes were derived (using the Akaike information criterion to compare regressions on different numbers of predictor variables). This suggests that the classification procedure significantly enhances the information contained in the DEM for purposes of predicting soil properties. (C) 1999 Elsevier Science B.V. All rights reserved.