GPU-Accelerated Urban Flood Modeling Using a Nonuniform Structured Grid and a Super Grid Scale River Channel

New remote sensing technologies, and the meter-scale geospatial data they create, now allow for detailed urban landscape characterization, thereby advancing grid-based hydrodynamic models. However, using a uniform fine grid over urban catchments generally result in dense grids and can lead to prohibitive computational costs. Moreover, an inability to see below the water surface and measure river bathymetry in most terrain remote sensing can severely impact local-scale river hydraulics calculations given the significant volume of water conveyed by the channel. This paper introduces a super grid channel model which allow river channels with any width above that of the grid resolution to be simulated in 1D manner. As an extension of a previous subgrid model, this integration facilitates a seamless transition between subgrid and super grid channels, accommodating situations where channel width may surpass or fall below the grid resolution. The key contribution is the integration of the novel 1D channel representation with a nonuniform structured 2D floodplain hydrodynamic model and then coding this for application on GPU. Compared with the previous pure 2D nonuniform structured approaches, the new model presents an efficient compromise for riverine urban flooding where we are less concerned about fine-scale details of in-channel flow. Three tests reveal that the proposed model maintains accuracy but with significantly reduced computational cost. By leveraging GPU architectures, a ∼10× speedup compared to CPU computations is achieved, and a typical 6-day urban flooding problem (domain size 1.42 km2) at 1 m resolution can be achieved within 10 hr on a single 8 GB GPU.