Mapeamento de solos e uso de algoritmos de aprendizagem em Lavras (MG)

• Autor(a) principal: Silva, Elidiane da
• Data de Publicação: 2017
• Tipo de documento: Tese
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/28826
• topic: Levantamento pedológico; Mapeamento digital de solos; Aprendizado de máquina; Pedological survey; Digital Soil Mapping; Machine learning; Ciência do Solo

The knowledge of soil classes and their attributes in a region is an important factor for more assertive decisions to be made regarding the use and soil and water management.The soil mapping is an important planning and management tool for the proper land use. This work aims to contribute with the technical and scientific knowledge regarding the soil classes that occur in the Lavras municipality and the use of modern techniques that allow the prediction of soil classes and properties. It is shown a Lavras municipality pedological survey, and a study comparing two techniques (Support Vector Machine - SVM and Artificial Neural Networks - ANN) of digital soil mapping for predicting soil classes and properties on the campus of the Federal University of Lavras. The soils mapping of the Lavras municipality used 27 modal profiles evenly distributed throughout the study area. From the elaboration of the soil map it was possible to evaluate also the land agricultural suitability and the adequability of the current use. In the study of digital soil mapping techniques, different sets of environmental covariates were evaluated in different sizes of training sets. The pedological survey at the medium intensity recconaissance level of the Lavras municipality registered Latosols, Argisols, Cambisols, and Neosols, with predominance of the Latosols, followed by Argisols, that together represent 72% of the total area of the municipality. Typic Dystrophic RedYellow Argisol was the subgroup level class with the greatest geographical expression in the municipality, occurring in 34% of the total area. The land use suitability showed 89% of the area have adequate uses and 7%, inadequate. In the prediction of soil classes on the campus of UFLA, ANN pre sented better performance when a training set was used consisting of the largest number of environmental covariates. In the prediction of soil properties ANN also performance well. In both cases increased training data increased the performance of the techniques. The pedological survey of the municipality of Lavras stratified the occurrence of soils in the different landscapes, which could base future research, extension work and also serve as a subsidy to politicaladministrative decisions. The technique of SVM and ANN can be used as a tool of digital mapping of classes and properties of soils in similarconditions.