Towards a large-scale locally relevant flood modelling using adaptive mesh generation and an integrated sub-/super grid channel solver

Developing reliable and efficient flood modelling systems on a large scale is crucial for addressing errors and inconsistencies in both observations and modelling. However, the computational demands of hydrodynamic models have constrained their widespread application to coarse resolutions (30m-1km), compromising accuracy by neglecting the local and small-scale features that may significantly influence flooding, especially in urban areas. Furthermore, traditional models struggle to effectively incorporate river bathymetry, especially given the significant flood volume conveyed by the river channel during floods. These models often rely on surveyed cross-sections for river channel representation, leading to missing topography between cross-sections and hindering the resolution of complex floodplain flow paths. To resolve small-scale effects in limited areas while simulating large domains, grid adaptation methodologies are implemented in this project to locally adjust the resolution of the computation in a static or a dynamic way. A hybrid 1D-2D flood model is developed, incorporating the static/dynamic adaptive mesh generation and an integrated sub-/super grid channel model. The sub-/super grid channel is applied to accommodate situations where river channel width exceeds or fall below the grid resolution. Parallelized with GPU architecture, the performance of hybrid 1D-2D with either static or dynamics nonuniform structured grid was thoroughly evaluated, benchmarked with the full resolution CPU solver, shedding light on their effectiveness in enhancing flood modelling approach.