There is a need to evaluate the importance of soil relief together with soil spectral attributes as the basis on soil mapping. The aim of this study was to test a method for detecting soil boundaries through the interaction of spectral data and relief features. Fourteen toposequences were used, representing an area of 13,000 ha near the municipalities of São Carlos and Araraquara, SP, Brazil. The samples were described by the conventional method of chemical and particle size analysis, such as pH (H2O and KCl), size (coarse and fine sand, silt, and clay), iron content, and color. Spectral information from 400 to 2,500 nm was subsequently obtained. Relief information was obtained by geotechnics, such as the Digital Elevation Model of the terrain, slope map, Compound Topographic Index, curvature, and Drainage Density Potential. In addition, the point and spatial methods proposed were validated. In the first validation, the points classified in the toposequences were taken as true and compared to information contained in the existing soil map, as well as the relief data and spectral data, separated by cluster analysis. Validation on the spatial level sought to assess in which locations the different methods indicated changes in the soil boundaries and compare this with real observations. It was seen that cluster analysis proved to be effective in differentiating soil classes in toposequences when soil spectral attributes were used. However, the set of relief attributes alone was not very suitable.
toposequence; reflectance; cluster analysis
