The production of soil maps through digital soil mapping (DSM) techniques may be hampered due to the lack of traditional reference soil maps. In these situations, the tacit knowledge of the field soil scientist can be used for manual delineation of soil mapping units (MUs) based on generation of a map of occurrence of soil types predicted by DSM. The objective of this study was to evaluate and to compare soil maps generated by two methods. One method, called “direct DSM”, generates a map predicting soil MUs based on a model established with information from a traditional pedological reference map. The other established a predicting model through examination of morphological properties of 193 soil profiles for identification of soil types, generating a map that indicates the occurrence of soil types performed through manual delineation of MUs (based on changes in land surface features). Predictions were made using Simple Cart classification trees, correlating eight terrain variables with the occurrence of MUs identified by soil class names from the Brazilian Soil Classification System (Sistema Brasileiro de Classificação de Solos). The accuracy of the maps was evaluated based on “field truth” (field verification of the soil type and comparison with that predicted on the map) and by agreement between the prediction maps generated and the reference map. When evaluated by “field truth”, the accuracy of the map generated by the direct DSM method was 74 %, whereas the accuracy of the map generated by DSM with manual delineation was 79 %. Both methods showed satisfactory results, and the method with manual delineation and identification of soil types in some locations in the field had the advantage of not requiring reference soil maps for training prediction models.
soil maps; soil surveys; soil types; decision tree; soil profiles
