More effective methodologies to determine the soil class distribution must be evaluated in order to meet the demand for soil maps at regional and global scales. In this study, logistic regressions were used as predictive models in an application of Digital Soil Mapping. The models were derived from an existing soil map as dependent variable and terrain attributes as independent variables. The probability of finding soil classes in the landscape at the 1st and 2nd Categorical Level of the Brazilian System of Soil Classification (SiBCS) was determined. The quality of the predicted map was tested using a contingency matrix. Approximately 85 % of the Acrisols (Argissolos) were correctly predicted, in relation to the original map. Of the hydromorphic soils, 75 % were correctly predicted. The prediction was inaccurate for classes in very similar positions in the landscape. It was also found that the non-representative soil classes of the landscape were not properly spatialized, due to sensitivity of the logistic models to the relative proportion of the samples used to adjust the models.
digital soil mapping; pedometry; generalized linear models
