Geomorphometric segmentation associated with soils types via geotechnologies

Abstract

The knowledge of soil-landscape relationships has been traditionally used as a guide for soil mapping. They have been applied using the tacit knowledge of field experienced pedologists. Thus, they are not explicit and difficult to quantify. Their explicitness and quantification, especially the segmentation of relief units, can be useful in soils mappings. The goals of this research are (a) to segment the hillslope elements in detailed level using one decision tree, (b) analyze the relation of these elements with soils types and (c) evaluate the use of a similarity algorithm to group soils classes based on their spectral and conventional analyses and an association of them. Parameters such as slope, curvature profile and relative elevation were hierarchized in a decision tree in order to identify and map hillslope elements. Soil samples were collected in the total area and in two toposequences based from this map associated with a preexisting geological map, in an area in Piracicaba (SP). They were analyzed in the laboratory and in a spectroradiometer (350-2500 nm). Their results were used to classify soils and to evaluate the similarity of their samples, based on three sets of variables (spectral and conventional analyses and an association of them). The segmentation of hillslope elements in detailed level (minimum mappable area of 0.05 ha) presented high agreement with the field observations, with high Kappa and low error. The summit, backslope and toeslope are among the elements that were best identified by the digital model. There were discrepancies for the shoulder and footslope, possibly associated with errors in the digital elevation model, generated by the contour lines obtained from planialtimetric maps. The toposequence with a more recent geomorphic surface presented smaller variations in soil attributes and types instead of the oldest one. The similarity analysis based on spectral analysis associated to conventional ones grouped similar soil classes better than the other sets of variables when analyzed in toposequences. For soil samples from the total area, all sets of variables grouped distinct soil classes. The digital classification of hillslope elements can be useful for digital or conventional 1st and 2nd order soil surveys.