Risk Factor Detection and Landslide Susceptibility Mapping Using Geo-Detector and Random Forest Models: The 2018 Hokkaido Eastern Iburi Earthquake

Landslide susceptibility mapping is an effective approach for landslide risk prevention and assessments. The occurrence of slope instability highly correlated with intrinsic variables that contribute to the landslides, such as geology, geomorphology, climate, hydrology, etc. However, feature selection those conditioning factors constitute datasets optimal predictive capability effectively accurately still open question. present study aims examine further integration selected Q-statistic in Geo-detector determining stratification analysis ultimately optimize model prediction. location chosen was Atsuma Town, which suffered from landslides following Eastern Iburi Earthquake 2018 Hokkaido, Japan. A total 13 were obtained different sources belonging six categories: seismology, land cover/use human activity; these generate mapping. original analyzed their explanatory powers regarding landslides. Random Forest (RF) adopted Subsequently, four subsets, including Manually delineated Points 9 features Dataset (MPD9), Randomly (RPD9), (MPD13), (RPD13), by training validating Geo-detector-RF- integrated model. Overall, using dataset MPD9, Geo-detector-RF-integrated yielded highest prediction accuracy (89.90%), followed MPD13 (89.53%), RPD13 (88.63%) RPD9 (87.07%), implied optimized can improve