Revealing landslide exposure of informal settlements in Medellín using Deep Learning
Michael Wurm, Raphael Tubbesing, Thomas H. Stark, Marlene Kühnl, Marta Sapena, Wolfgang Sulzer, Hannes Taubenböck · 2023
Large areas of informal settlements on the slopes of Medellín are exposed to landslide risk, but there exists no accurate and up-to date data set on the location and size of informal areas. It is thus difficult to develop mitigation strategies to reduce the risk for the local population. Here, we tackle the issue of inaccurate geodata and apply a CNN for the extraction of individual building footprints from orthophotos. With it we achieve a more reliable data base for a more precise estimation of the amount of exposed population in informal areas towards landslides.