Performance and topographic preferences of dynamic and steady models for shallow landslide prediction in a small catchment

Used to reduce damage from rainfall-induced landslides and establish disaster-warning systems, physically based stability models are efficient tools for evaluating spatial or temporal patterns of susceptibility to shallow landslides. Such comprehensive models can be classified as dynamic and steady hydrological models, which combine the simulation of dynamic or steady-state subsurface saturation generation with slope stability analysis. Although numerous dynamic and steady models have been proposed and validated against actual landslides, comprehensive comparisons of the performance and detection preferences for topographic characteristics (hereafter, topographic preferences) between the dynamic and steady models are lacking. Based on detailed observational data on topography at the soil–bedrock interface, this study compared numerical predictions from dynamic and steady models to a case of rainfall-induced shallow landslides in a small catchment. The dynamic model was based on time-varying generation of subsurface saturation using the three-dimensional Richards equation, and the steady model was based on the steady-state distribution of subsurface saturation. The results showed that both the dynamic and steady models could generally detect cells at risk of failure in the landslide areas. However, the precision of the dynamic model was approximately double that of the steady model. Rainfall patterns had significant impacts on the timing and locations of landslides, and the delayed rainfall peak pattern was most prone to landslides, which was reflected by the dynamic model but not the steady model. The steady model underestimated instability or the number of unstable locations relative to the dynamic model. This study revealed that the topographic preference at the soil–bedrock interface differs between the models. The dynamic model exhibited topographic preference in detecting landslides with deep soil depth and steep gradient, whereas the steady model was better for detecting landslides with large contribution areas. The elucidation of such topographic preferences for landslide types is crucial for comparing the performance of dynamic and steady models.