Sensitivity evaluation and prediction of extreme rainfall-induced landslides in Xinning, Hunan China, using Gumbel extreme value theory and random forest model

Landslide sensitivity research prioritizes current assessments over predictive risk modeling Although rainfall is a well-documented landslide trigger, predictive frameworks for rainfall-induced landslides remain underdeveloped due to limited case studies This study addresses these gaps in Xinning County, China through: (1) identifying rainfall as the key external driver among 10 factors via certainty factor modeling; (2) predicting maximum rainfall intensities for 5-, 10-, 30-, 50-, and 100-year return periods using CHIRPS data, Google Earth Engine, and Gumbel extreme value theory; (3) assessing spatiotemporal sensitivity with a Random Forest Model Key results show: persistent high/very high sensitivity (HVHS) clusters in western (SM, WT Towns) and northeastern (XT Town) regions; escalating sensitivity in GQ, MTQ, HLS, and YDS (transitioning from low/ moderate levels to high/ very high levels); and decadal-scale expansion of very high sensitivity (VHS) areas in GQ/MTQ/HLS Towns, with 10-year return period extents doubling other intervals Quantitatively, the proportion of VHS zones in the county ranged between 32% and 129%, with the lowest value recorded in 2023 and the peak observed under the 10-year return period; GQ Town is the only area where VHS zones account for over 40% of the total spatial coverage under a 10-year return period; WT and SM Towns collectively exceed 80% HVHS coverage in 2023 Spatially, HVHS zones in central/northern lowlands align with median maximum daily rainfall, contrasting with low/very low sensitivity (LVLS) zones in southeastern/southwestern highlands linked to higher values Proactive mitigation in GQ/MTQ/HLS Towns is urged before the 10-year return period.