Flow-like landslides are a common type of natural hazards that may impose a great risk to people and their properties in mountainous areas. The landslide occurring in India in 2013 killed more than 6000 people and damaged at least 30 hydropower plants. To mitigate the impact of these hazardous events, it is necessary to develop robust modelling tools to predict the process of flow-like landslides and support the development of risk management strategies.
A new model is developed for simulating flow-like landslides that harnesses the advantages of both the discrete element method (DEM) and depth-averaged model (DAM). Implemented through a longitudinal coupling approach, the new model employed a DEM to simulate the complex granular collapsing process in the source area and a DAM to predict the predominantly convective movement of the landslide in the runout and deposition zones. As a result, the newly developed coupled model is able to efficiently and accurately simulate the fully dynamic process of flow-like landslides from geomaterial collapsing to depositing. To achieve high-performance computing and further improve its performance, the coupled model is implemented on Graphic Processing Units (GPUs) through the NVIDIA CUDA modelling platform.
The GPU-accelerated coupled model is then validated by successfully reproducing experimental cases of granular flows before being applied to reproduce the 2019 Shuicheng landslide in Guizhou, China. Satisfactory results demonstrate the good performance of the coupled model in terms of both of the computational efficiency and numerical accuracy, confirming that the coupled model has the potential to predict the dynamic process of large-scale flow-like landslide. Additionally, it is found that the basal friction significantly influences the landslide dynamics may need to be carefully calibrated for reliable predictions. The terrain resolution also has a great effect on the landslide propagation by affecting the energy dissipation, and landslide mass move faster on the coarse digital terrain than that on the fine digital terrain.