Object‑oriented landslide identification and susceptibility assessment for the Jiuzhaigou earthquake by integrating NDVI difference

To address challenges in identifying densely distributed Earthquake-Triggered Landslides (EqTLs) in plateau canyons, this study integrates object-oriented detection with a Random Forest susceptibility model incorporating Normalized Difference Vegetation Index difference (ΔNDVI) features. Using pre-/post-seismic Sentinel-2A imagery and ALOS PALSAR DEM data from the 2017 Jiuzhaigou Ms7.0 epicenter, we: (1) Implemented multi-scale segmentation for landslide extraction using spectral (ΔNDVI/brightness), terrain (slope/TRI), texture (GLCM/GLDV), and geometric rules; (2) Developed a Random Forest model with nine conditioning factors. Key results show: ① 841 identified landslides (11.96 km²) clustered NW-SE near the Huya Fault (0–2 km, 43%) and on 30–50° slopes (75%); ② High model accuracy (AUC = 0.98) with fault distance (0.34), PGA (0.22), and slope angle (0.17) as primary controls; ③ Very high susceptibility zones (12.5% area) contained 95% of landslides at 2.83/km² density. This framework enables effective spatial analysis and risk management for alpine valley landslides.