On the prediction of landslide occurrences and sizes via Hierarchical Neural Networks

Abstract For more than three decades, the part of geoscientific community studying landslides through data-driven models has focused on estimating where may occur across a given landscape. This concept is widely known as landslide susceptibility. And, it seen vast improvement from old bivariate statistical techniques to modern deep learning routines. Despite all these advancements, no spatially-explicit model currently capable also predicting how large be once they trigger in specific study area. In this work, we exploit architecture that already found number applications Specifically, opt for use Neural Networks. But, instead focusing exclusively occur, extend paradigm spatially predict classes sizes. As result, keep traditional binary classification but make complement susceptibility estimates with crucial information hazard assessment. We will refer Hierarchical Network (HNN) throughout manuscript. To test analytical protocol, Nepalese area Gorkha earthquake induced tens thousands 25th April 2015. The results obtain are quite promising. component our HNN outperforms binomial Generalized Linear Model (GLM) baseline used benchmark. did GLM represents most common classifier literature. Most importantly, suitably performed entire procedure. landslide-area-class prediction returned not just single map, per tradition. produced several informative maps expected size classes. Our vision administrations consult suite outputs and better assess risk local communities infrastructure. promote diffusion HNN, sharing data codes githubsec repository hope would stimulate others replicate similar analyses.