Mapeamento digital de solos da quadrícula de Ribeirão Preto - SP pelo método Random Forest

• Autor(a) principal: Oliveira, Matheus Felipe [UNESP]
• Data de Publicação: 2015
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://hdl.handle.net/11449/154733; http://www.athena.biblioteca.unesp.br/exlibris/bd/cathedra/09-03-2017/000881014.pdf
• topic: Soil mapping; Geoprocessamento; Mapeamento digital; Solos; Morfometria; Mapeamento do solo

This study aimed to develop a model to understand the soil-landscape relationships to predict soil classes of topographic sheets of IBGE from Ribeirão Preto, Serrana, Cravinhos and Bonfim Paulista, constituting the grid Ribeirão Preto. For this, we used information included in a conventional semi-detailed soil map at 1:100,000 scale, a Digital Elevation Model (DEM) with a spatial resolution of 30 meters, in addition to the geological map at 1: 50,000 scale. From geological map was obtained lithology and from MDE were obtained the geomorphometric variables through geoprocessing techniques. All this information was linked in a matrix, from which they were selected three stratified sampling according to the area of classes, extracting data for training and testing, which were used for use in models of Random Forest method and evaluation of accuracy. Adjustments were tested with application of models in classes on the second and third categorical level. With a sample comprising only 0.43% of the total area, the model for the second categorical level had an overall accuracy of 62.5%, with the digital soil map showing a persistence of 70.63% of classes from original map, higher values than those presented for the third categorical level, with an overall accuracy of 57.1% and persistence of 44.24%. The most important variables in understanding the soil-landscape relationships were Lithology, Elevation, Slope Distance and drainage network. The study showed that the method is able to contribute to the creation of soil maps, with the possibility of being employed in areas where there is no pre-existing soil information quickly and less costly way, assisting the work of soil scientists