Downscaling coarse grid hydrodynamic model simulations over large domains

• Efficient downscaling of coarse hydrodynamic simulations to a fine resolution.
• Accuracies were >30 cm in water depth and >90% in inundated area for accurate DEMs.
• For SRTM, accuracies were 0.5–1.5 m for depth but for inundated area were <90%.
• Simulating the worlds major rivers at 90 m resolution could take 3+ years.
• Downscaling a 3 km resolution model for those rivers at only a fraction of the time.

SummaryIt is evident in recent literature that hydrodynamic modelling efforts have moved to increasing spatial coverage while trying to preserve simulation accuracies at computationally efficient coarse grids (100 m to several km). However, it is clear that there is a need to retain fine spatial resolutions at large scales wherever possible in order to still retrieve meaningful information from models or indeed observations, such as identifying individual assets at risk from flooding for instance. Since it is currently rather impractical to model hydrodynamics across areas larger than a couple of thousand km2 at a fine spatial resolution (finer than 100 m), this paper proposes a method to downscale coarse model simulations (model grid size of 100 m to several km) to a fine spatial resolution. The method is mass conservative and uses a hydraulic 1D approach within the channel and a pseudo region-growing algorithm on the floodplain. Comparison to a high resolution reference model over a domain size much larger than those traditionally modelled showed that downscaling a 600 m grid resolution hydrodynamic LISFLOOD-FP model to 30 m leads to average accuracies greater than 30 cm in water depth and above 90% in inundation area for a high accuracy digital elevation model (DEM). When employing a SRTM DEM accuracies were still between 0.5 m and 1.5 m for water depth but agreements in inundated area were much lower than 90%. We speculate that for simulating the world’s major rivers and their floodplains at a resolution of 90 m, even a speed-efficient model could take over three years to simulate inundation patterns at that resolution for a one-year hydrograph. However, it is expected that the proposed downscaling method could be used to downscale LISFLOOD-FP model simulations run at a 3 km resolution with reasonably similar accuracies and at only a fraction of the computational time required by the 90 m model.