A neural network model applied to landslide susceptibility analysis (Capitanejo, Colombia)

Abstract

The generation of landslide present in the municipality of Capitanejo (Santander-Colombia) is conditioned to lithological, structural and morphometric factors that are related to the climatic activity. These factors describe a pattern of instability in the rocks, which generates falls, slides and flows. These conditions are seen in the results of a neural network model generated with the backpropagation algorithm. This model presented the conditions of the study area evaluated by the degree of susceptibility to landslide. The model predicted 92.86% of the data that were not entered in the learning module, which represents 50% of the landslides mapped in the region. This simulation generates different levels of susceptibility to landslide from very low to very high. In the study area, a tendency for moderate to very high susceptibility to landslides was found for the margins of the Chicamocha river valley. In turn, this area presents the largest number of active processes (falls and flows), which generates a problem that affects several levels within the sectors of production, communication and infrastructure of the community of Capitanejo.