Landslide Susceptibility Prediction from Satellite Data through an Intelligent System based on Deep Learning

Abstract

Landslides are critical natural hazards whose frequency and severity are increasing due to climate change and human activities. The consequences of landslides are severe and can lead to the destruction of homes, infrastructures and the contamination of water supplies, with severe impact also on the local ecosystems and the disruption of natural habitats. This article examines the application of an ad-hoc neural network-based intelligent system to evaluate the landslide susceptibility of the terrain on the basis of satellite data. The proposed system is validated on data from Lombardia and Abruzzo, two Italian regions that have been particularly subject to the landslide phenomenon. Results indicate that the CNN model is able to correctly identify landslide occurrences with high accuracy, demonstrating that CNNs are capable of providing accurate susceptibility mapping at a local scale and surpassing the performance of existing solutions available in the literature.