| Summary: | Landslide detection and susceptibility prediction are valuable tools for disaster prevention. Despite there being many various solutions to accomplish these tasks, they all generally depend on topographic features of the environment. However, there are not many solutions that can adapt to triggering events such as hurricanes, earthquakes, or volcanic eruptions. This lack of adaptability can greatly limit the performance of the algorithms designed to solve these problems, which, in turn, makes it difficult for emergency managers and responders in the area to prepare for these events appropriately. This work experiments with various kinds of machine learning models and analyzes the effects of incorporating dynamic features based on triggering events in the training process. Ultimately, the final versions of the best performing models produced in this thesis will be deployed as a part of a landslide monitoring system to be used in Mocoa, Colombia. This system is being adapted and developed for the Drones/UAVs for Equitable Climate Change Adaptation (DECCA) project run by MIT’s Environmental Solutions Initiative.
|
|---|