02020nam a2200193 a 450000100080000000500110000800800410001910000190006024501590007926002060023852012250044465300250166965300170169465300270171170000180173870000310175670000170178770000220180418735782023-03-13       2010    bl uuuu     u00u1 u    #d1 aDART, R. de O.  aDigital soil mapping at Parque Estadual da Mata Seca, Minas Gerais state, Brazilbapplying regression tree to predict soil classes.h[electronic resource]  aIn: INTERNATIONAL WORKSHOP ON DIGITAL SOIL MAPPING, 4., 2010, Rome. From digital soil mapping to digital soil assessment: identifying key gaps from fields to continents: proceedings... Rome: IUSSc2010  aThe use of Digital Soil Mapping (DSM) to predict soil classes is an important issue to decrease costs and subjectivity of soil maps. The main objective of this study was to use DSM to produce soil maps of a relatively small area (about 100 km2) and compare it to a preliminary soil map made by traditional techniques. The study area is located at north of Minas Gerais State, southwest of Brazil. In this study we used decision tree classifier, See5, and 278 soil samples to predict soil class at order level of the Brazilian System of Soil Classification. We also did use ancillary data as Landsat ratios and variables of the topography. DSM didn?t show a good performance of soil prediction because basically three factors: (a) taxonomic similarity between Argissolos and Latossolos, (b) great spatial and attributes variability of Cambissolos that occurred in different landscapes types, and (c) low accuracy of soil prediction to Gleissolos, Neossolos and Cambissolos of the river plain domain because its shows great environment complexity. Following works will make a better selection of environmental covariates, predict the soil classes in higher categorical level and assessment of quality of digital soil maps.  aDigital Soil Mapping  aSoil classes  aTraditional techniques1 aCOELHO, M. R.1 aMENDONÇA-SANTOS, M. de L.1 aPARES, J. G.1 aBERBARA, R. L. L.