Large-scale landslides are characterised by extreme deformation, evolving free surfaces, fragmentation, and mass entrainment. These processes present significant challenges to conventional continuum-based methods such as the finite element method (FEM). In contrast, the discrete element method (DEM) naturally captures discontinuous deformation and granular interactions, making it ideal for modelling large-scale geophysical flows with evolving material boundaries. However, DEM simulations of fully erodible terrain at a slope scale are relatively uncommon due to their computational demands.

In this study, we conduct large-scale, three-dimensional DEM simulations of the 2016 Aso Bridge landslide, triggered by the Kumamoto earthquake. For these simulations, we use an erodible slope model constructed from real topographic data. The slope is represented as a cohesive granular system, enabling both the initially released mass and the surrounding materials to deform and mobilise dynamically. Three simulations with different particle sizes were performed while keeping all other model parameters fixed. The finest model comprises over 300 million particles, demonstrating the study's high-performance computing capabilities.

The simulations reproduce large-scale runout, lateral spreading, and progressive mass entrainment under realistic three-dimensional topography. The results show that finer-grained models exhibit increased mobility and entrainment, indicating that particle size significantly influences simulated flow behaviour. Additionally, terrain-controlled flow bifurcation naturally emerges from the three-dimensional slope geometry, demonstrating the importance of realistic topographic representation in slope-scale granular flow simulations. This study demonstrates the potential of DEM for modelling deformable geomaterials with evolving boundaries, and highlights the importance of explicitly resolving the entrainment process.