Landslides triggered by intense rainfall pose persistent threats to infrastructure and communities, particularly in data-scarce mountainous regions. Predicting such hazards at a regional scale remains challenging due to sparse dated landslide inventories and limited rainfall monitoring networks. This study presents an integrated spatiotemporal framework for landslide hazard prediction in the Muzaffarabad region, NW Himalayas, Pakistan. Landslide records (2006–2023) and satellite-based IMERG rainfall data were used to derive empirical cumulative event–duration (ED) thresholds using a frequentist approach to distinguish triggering from non-triggering rainfall conditions. Poisson-based modeling of threshold exceedance frequencies enabled estimation of temporal landslide probabilities for 1- to 20-year return periods. In parallel, spatial landslide susceptibility was modeled using machine learning techniques (Random Forest and XGBoost), with Random Forest outperforming XGBoost (AUC-ROC = 0.94; accuracy = 89%). Temporal probabilities were then integrated with spatial susceptibility to produce regional-scale landslide hazard index maps. This study is an early application of satellite-derived rainfall threshold exceedance in a probabilistic landslide hazard framework for data-scarce, mountainous regions. Results indicate a clear intensification of relative hazard across return periods: mean hazard increased from 0.13 to 0.18. The mean relative temporal factor (Pt) per IMERG pixel ranged from 0.19 to 1 across return periods, reflecting higher potential triggering during extreme rainfall events. From 1-year to 20-year return periods, the relative area of very high, high, and moderate hazard zones increased by 83.1%, 33.7%, and 11.5%, respectively, while low and very low hazard zones decreased by 13.4% and 7.4%. These trends underscore a spatial shift toward higher hazard zones, with significant implications for risk management, land-use planning, and infrastructure resilience. The proposed framework offers a scalable and transferable solution for landslide hazard assessment in mountainous, data-scarce regions, contributing to improved prediction, proactive mitigation, and safer engineering outcomes.
Landslide hazard prediction in data-scarce mountainous region using satellite-based frequentist rainfall thresholds
Brunetti M. T.;Rehman A.;
2026
Abstract
Landslides triggered by intense rainfall pose persistent threats to infrastructure and communities, particularly in data-scarce mountainous regions. Predicting such hazards at a regional scale remains challenging due to sparse dated landslide inventories and limited rainfall monitoring networks. This study presents an integrated spatiotemporal framework for landslide hazard prediction in the Muzaffarabad region, NW Himalayas, Pakistan. Landslide records (2006–2023) and satellite-based IMERG rainfall data were used to derive empirical cumulative event–duration (ED) thresholds using a frequentist approach to distinguish triggering from non-triggering rainfall conditions. Poisson-based modeling of threshold exceedance frequencies enabled estimation of temporal landslide probabilities for 1- to 20-year return periods. In parallel, spatial landslide susceptibility was modeled using machine learning techniques (Random Forest and XGBoost), with Random Forest outperforming XGBoost (AUC-ROC = 0.94; accuracy = 89%). Temporal probabilities were then integrated with spatial susceptibility to produce regional-scale landslide hazard index maps. This study is an early application of satellite-derived rainfall threshold exceedance in a probabilistic landslide hazard framework for data-scarce, mountainous regions. Results indicate a clear intensification of relative hazard across return periods: mean hazard increased from 0.13 to 0.18. The mean relative temporal factor (Pt) per IMERG pixel ranged from 0.19 to 1 across return periods, reflecting higher potential triggering during extreme rainfall events. From 1-year to 20-year return periods, the relative area of very high, high, and moderate hazard zones increased by 83.1%, 33.7%, and 11.5%, respectively, while low and very low hazard zones decreased by 13.4% and 7.4%. These trends underscore a spatial shift toward higher hazard zones, with significant implications for risk management, land-use planning, and infrastructure resilience. The proposed framework offers a scalable and transferable solution for landslide hazard assessment in mountainous, data-scarce regions, contributing to improved prediction, proactive mitigation, and safer engineering outcomes.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.


