Unravelling the impact of landslide inventory on landslide susceptibility in the Indian Himalaya

Abstract

Landslide susceptibility zonation (LSZ) mapping is heavily influenced by the raster resolution and landslide inventory types. The effect of landslide inventories (polygon and point) at three raster resolutions (12.5 m, 30 m, and 90 m) on LSZ analysis is investigated in this work. The Ramban District sub-basin in Jammu and Kashmir, identified as the most vulnerable area, encompasses 302 landslides. To ensure a robust susceptibility assessment, Yule's coefficient (Yc) was utilized to examine twelve landslide conditioning factors (LCFs) for LSZ preparation. LULC (ESRI & Google) and road variables have the greatest influence at all resolutions, but lithology plays a critical role in lower-resolution polygon-based data. Aspect, geomorphology, slope, and landform exhibit moderate to low effects, which vary with resolution. LULC, roads, and lithology emerge as key influences, whereas drainage, faults, and landforms serve as secondary influences. RR and TWI demonstrate negligible influence on LSZ across all sampling and resolution. LSZ exhibits considerable variation with resolution in point-based inventory. At higher resolutions (12.5 m and 30 m), raster area coverage is below 50 % of vector coverage. Conversely, at 90 m, raster coverage roughly doubles that of vector data, potentially inflating LSZ results. AUC values are higher for point than for polygon sampling. However, for precise mapping, polygon sampling gives a more accurate picture of factors and landslide distribution. This study emphasizes the significance of using polygon sampling to delineate landslide susceptibility in the Himalayas.