Identificação de áreas vulneráveis a movimentos de massa com base em aprendizado de máquina, Rio Ligeiro (Pato Branco – PR)

• Ano de defesa: 2023
• Autor(a) principal: Pontes, Priscila da Silva Victorino
• Orientador(a): Paisani, Julio Cesar
• Banca de defesa: Paisani , Julio Cesar , Pontelli, Marga Eliz , Fujita, Rafaela Harumi , Sordi, Michael Vinicius de , Lima, Vanderlei Aparecido de
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Estadual do Oeste do Paraná; Francisco Beltrão
• Programa de Pós-Graduação: Programa de Pós-Graduação em Geografia
• Departamento: Centro de Ciências Humanas
• País: Brasil
• Palavras-chave em Português: Estabilidade de encosta; Aprendizado de máquina; Movimento de massa
• Palavras-chave em Inglês: Slope stability; Machine learning; Landslides
• Área do conhecimento CNPq: CIÊNCIAS HUMANAS:GEOGRAFIA
• Link de acesso: https://tede.unioeste.br/handle/tede/6920

Studies involving the occurrence of landslides are still infrequent in municipalities in the Southwest of Paraná, which had a low population rate until the 1970s. However, this region has demonstrated a significant urbanization growth in recent decades, mainly in its hubs, such as Francisco Beltrão, Palmas, and Pato Branco. The relief of the Southwest mesoregion is marked by morphological homogeneity resulting from the predominance of flat, wavy features and a portion of strongly undulating areas. Thus, given the recurrent landslides in places of urban growth and the particular characteristics of the Southwest of Paraná relief , this work seeks to apply a methodology for identifying susceptible and vulnerable areas to landslides. A pilot experiment was carried out in the municipality of Pato Branco – PR at the limit of the Ligeiro River basin. Possible natural and anthropogenic conditions causing landslides were identified and the machine learning technique was applied to discover patterns of occurrence between these conditions. Random Forest (FR), LibSVM, and LazyKStar algorithms were used for data modeling and validation. From the results obtained by the machine learning technique, criterionmaps were generated representing the possible areas susceptible to landslides in the region of interest