Landslide hazard and susceptibility maps derived from satellite and remote sensing data using limit equilibrium analysis and machine learning model

Abstract

Landslide susceptibility mapping and landslide hazard mapping are approaches used to assess the potential for landslides and predict the occurrence of landslides, respectively. We evaluated and tested a limit equilibrium approach to produce a local-scale, multi-temporal geographic information system-based landslide hazard map that utilized satellite soil moisture data, soil strength and hydrologic data, and a high-resolution (1.5 m) LiDAR-derived digital elevation map. The final multi-temporal landslide hazard map was validated temporally and spatially using four study sites at known landslide locations and failure dates. The resulting product correctly indicated low factor of safety values at the study sites on the dates the landslide occurred. Also, we produced a regional-scale landslide susceptibility map using a logistic regression machine learning model using 15 variables derived from the geomorphology, soil properties, and land-cover data. The area under the curve of the receiver operating characteristic curve was used for the accuracy of the model, which yielded a success rate of 0.84. We show that using publicly available data, a multi-temporal landslide hazard map can be created that will produce a close-to-real-time landslide predictive map. The landslide hazard map provides an understanding into the evolution of landslide development temporally and spatially, whereas the landslide susceptibility map indicates the probability of landslides occurring at specific locations. When used in tandem, the two mapping models are complementary to each other. Specifically, the landslide susceptibility mapping identifies the area most susceptible to landslides, while the landslide hazard mapping predicts when landslide may occur within the identified susceptible area.