A boosting framework for slope unit based earthquake-induced landslide susceptibility mapping

Abstract

In mountainous areas, landslides brought on by earthquakes are a major threat to infrastructure, human life, and sustainable development. Thousands of landslides were caused by the 2015 Mw 7.8 Gorkha earthquake in Nepal, underscoring the critical need for trustworthy techniques to pinpoint regions most susceptible to seismically induced slope failures. In the Sindhupalchowk District, one of the areas most severely impacted by earthquakes, this study suggests a boosting-based framework for mapping landslide susceptibility using slope units. Twelve geophysical covariates representing terrain, hydrological, geological, and seismic conditions were integrated with a thorough landslide inventory comprising 7,159 earthquake-induced landslides. In contrast to traditional grid-based methods, slope units were employed as mapping units to more accurately depict geomorphological processes. The boosting model was trained using 70% of the dataset and validated with the remaining 30%. The receiver operating characteristic curve was one of several statistical metrics used to assess the model's performance. With an area under the curve value of 0.83, the results demonstrate strong predictive capability and good discrimination between slope units that are prone to landslides and those that are stable. High-risk areas are concentrated along steep slopes, deeply carved valleys, and regions that experience severe ground shaking, according to the resulting susceptibility map. Overall, in seismically active mountainous areas, the suggested framework offers a reliable and comprehensible method for evaluating landslide susceptibility to earthquakes and offers insightful information to support hazard mitigation, land-use planning, and disaster risk reduction.