A research on a new mapping method for landslide susceptibility based on SBAS-InSAR technology

Abstract

The acquisition of landslide inventory represents a pivotal challenge in landslide susceptibility mapping. Existing landslide susceptibility maps(LSMs) predominantly rely on manually obtained landslide inventories, leading to an overdependence on expert insights and susceptibilities to topographic and geomorphic influences. In regions characterized by steep terrain, obtaining a landslide inventory can be arduous or even unattainable, subsequently constraining the utility of LSMs. Addressing the limitations of conventional LSMs, this study introduces an innovative method for landslide inventory compilation and LSM creation, utilizing Small Baselines Subset Interferometry Synthetic Aperture Radar(SBAS-InSAR) technology. The study area selected for illustration is the Dongchuan district, notorious for frequent landslide occurrences. The application of SBAS-InSAR facilitated the extraction of surface deformation data, subsequently enabling the selection of landslide deformation points as samples. These samples underwent analysis through a particle swarm optimization-backpropagation neural network(PSO-BPNN) guided by deformation thresholds and the landslide developmental environment. This produced the LSM for the Dongchuan district. Subsequent validation of the LSM employed both qualitative and quantitative measures. Results elucidate that the LSM, as derived from the presented approach, primarily highlights high to very high susceptibility zones in landslide-prone areas, mirroring the spatial distribution of historical landslides. The method also achieved a commendable accuracy(ACC) of 79.59% and an area under the curve(AUC) value of 0.88. Notably, the landslide density exhibited a direct correlation with increasing susceptibility class. Such findings align with previous studies, endorsing the feasibility and reliability of the proposed approach.