Global Landslide Forecasting and Dynamic Susceptibility Mapping Using a New Dataset of Geophysical and Meteorological Features

Landslides cost billions in damage annually in the U.S. & have affected 4.8 million people over the past 2 decades. Existing landslide warning systems are ineffective in accuracy, latency, & scalability. There is a lack of publicly accessible data of landslide features on a global scale. The researchers compiled a global dataset of over 20,000 landslide & non-landslide incidences with 95 related climate and terrain features. Machine learning models trained on this dataset obtained 86.3% accuracy & 86.9% detection rates for forecasting whether landslides will occur 5 days into the future. A data-driven approach to landslide susceptibility map generation was derived.