Resumo

GOYES-PENAFIEL, Paul  e  HERNANDEZ-ROJAS, Alejandra. Double landslide susceptibility assessment based on artificial neural networks and weights of evidence. Bol. geol. [online]. 2021, vol.43, n.1, pp.173-191.  Epub 01-Jan-2021. ISSN 0120-0283.  https://doi.org/10.18273/revbol.v43n1-2021009.

Landslides are the most frequent natural hazards in tropical regions. They cause serious damages to road infrastructure, human losses and effects on economy. Therefore, a quantitative and reliable evaluation of landslide susceptibility is important for territorial planning and development. In this work, the susceptibility calculation is studied with a low uncertain level through the methodological integration of Weights of Evidence method with Artificial Neural Networks. The first one was used to extract the weighted values from the association of variables and the landslide inventory, and the second one to establish the non-linear relation between the conditioning factors and the punctual landslide inventory obtained through the geologic and geomorphologic study of the Popayan municipality. This produces a double verification allowing to extract the characteristics of categorical and continuous variables to produce more accurate susceptibility relations, avoiding multicollinearity and non-significant factors through the Principal Component Analysis. For studying the influence of variables, two methodological proposals were analyzed, the first one with two variables and the second one with five explanatory variables. For each one, it was applied Logistic Regression, Multilayer Perceptron, and Deep Neural Network quantitative methods as elements of double verification. The results of each model were assessed by the Receiver Operating Characteristics curves. The Deep Neural Networks got an Area Under the Curve with values of 0.902 and 0.969 for proposals 1 and 2, respectively, overcoming Weights of Evidence and Logistic Regression as quantitative methods.

Palavras-chave : Landslide Susceptibility; Deep Learning; Logistic Regression; Weights of Evidence; Principal Component Analysis; Artificial Neural Networks.

        · resumo em Espanhol     · texto em Espanhol     · Espanhol ( pdf )