An improved depth-averaged landslide dam breach model with modified sediment transport and bank collapse algorithms

Most landslide dams fail through overtopping, with the failure process controlled by the combined effects of longitudinal erosion and lateral bank collapse. However, existing depth-averaged models still face difficulties in simulating the evolution of longitudinal erosion and lateral collapse, mainly due to the lack of precise physical representations. To address this issue, this study proposes an improved depth-averaged landslide dam breach model that integrates enhanced sediment transport and bank collapse algorithms. A dynamic critical Shields parameter is introduced into the Meyer-Peter-Müller sediment transport formula to account for the combined influence of dam slope and material friction strength on sediment initiation. In addition, based on the traditional lateral collapse model, a new algorithm that uses the true three-dimensional slope angle of the terrain is proposed to improve the accuracy of collapse simulation. The governing equations of the improved model are solved using a Godunov-type finite volume scheme, enabling the simulation of both longitudinal erosion and lateral collapse during dam breaching. Validation against a one-dimensional overtopping erosion test, a side-bank collapse verification case with a dry-wet partition, and a sand-dike breach experiment shows that the improved model achieves high agreement with measurements in hydrodynamics, sediment transport, and lateral collapse prediction. The model is further applied to the back-analysis of the Tangjiashan landslide dam breach process, accurately reproducing the complete evolution of the breach from longitudinal incision to deepening and widening, and finally to stable drawdown. The relative errors of peak discharge, breach morphology, and water level variations are all within 10%, with time scale deviations of less than 1 h. The improved model provides a more reliable numerical tool for landslide dam breach risk assessment.