Comparative analysis of flood wave modelling methodologies for reservoir failure based on structure from motion

Abstract Accurate prediction of flood waves from reservoir failures and their propagation is critical to effective flood hazard assessment and risk management. Flood waves are sensitive to floodplain topography, channel geometry, and hydraulic structures located along flow paths in downstream areas, and thus the accuracy of flood wave modelling is dependent on the precision and accuracy of the representation of those features. This study introduces a novel approach to flood wave modelling by accurately representing 3D objects in downstream areas using the Structure-from-Motion (SfM) technique. Through the use of Unmanned Aerial Vehicle (UAV), this method captures topographic complexities, accounting for ground objects like bridges and trees that impact flood propagation. The 3D model offers enhanced representation of turbulent flow dynamics and computational efficiency, especially handling large topography datasets using the volume of fluid (VOF) method. Predictions from this new 3D approach were validated against recent reservoir failure observations and contrasted with traditional 2D models. The results revealed that the 3D model displayed a significant 84.4% reproducibility when juxtaposed with actual inundation traces. It was 35.5%p more accurate than the 2D diffusion wave equation (DWE) and 17.1%p more than the 2D shallow water equation (SWE) methods in predicting flood waves. The DWE was the least accurate among the results, whereas the SWE fared better but still struggled with intricate floodplains. In conclusion, the 3D method, combined with the structure-from-motion technique, emerges as a promising alternative to traditional modelling methods.