At the Junction Between Deep Learning and Statistics of Extremes: Formalizing the Landslide Hazard Definition

The most adopted definition of landslide hazard combines spatial informationabout landslide location (susceptibility), threat (intensity), and frequency(return period). Only the first two elements are usually considered andestimated when working over vast areas. Even then, separate models constitutethe standard, with frequency being rarely investigated. Frequency and intensityare intertwined and depend on each other because larger events occur lessfrequently and vice versa. However, due to the lack of multi-temporalinventories and joint statistical models, modelling such properties via aunified hazard model has always been challenging and has yet to be attempted.Here, we develop a unified model to estimate landslide hazard at the slope unitlevel to address such gaps. We employed deep learning, combined with a modelmotivated by extreme-value theory to analyse an inventory of 30 years ofobserved rainfall-triggered landslides in Nepal and assess landslide hazard formultiple return periods. We also use our model to further explore landslidehazard for the same return periods under different climate change scenarios upto the end of the century. Our results show that the proposed model performsexcellently and can be used to model landslide hazard in a unified manner.Geomorphologically, we find that under both climate change scenarios (SSP245and SSP885), landslide hazard is likely to increase up to two times on averagein the lower Himalayan regions while remaining the same in the middle Himalayanregion whilst decreasing slightly in the upper Himalayan region areas.