Forecasts of tropical cyclones have seen rapid improvements in recent years as expanding computational capacity permits more runs of finer resolution meteorological models with increasing representation of physical processes.
However, the utilization of a hydrodynamic component in these models is often neglected, meaning flood forecasts typically output point water levels that give little indication of a projected inundation extent on the ground. Here, we append this critical component to the forecast cascade by coupling Fathom-US, a continental-scale hydraulic model which employs the LISFLOOD-FP numerical scheme, to forecasts of streamflow, rainfall and coastal surge height from the National Oceanic and Atmospheric Administration (NOAA). Medium-term (2–15 days) flood inundation forecasts, as well as hindcasts driven by real-time observations, were executed for Hurricane Harvey by rapidly simulating pluvial and coastal flood hazard and extracting fluvial flood maps from an existing US-wide simulation library.
Flood inundation forecast results
The resultant ~30 m resolution depth grids were then validated against post-event observations collated by the US Geological Survey. Across the disaster zone, the hindcast (forecast) model captured, on average, 78% (75%) of the benchmark flood extent, obtained a Critical Success Index of 0.66 (0.57) and deviated from observed high water marks by ~1 m (~1.2 m). When compared to a simpler GIS-based approach, the hydraulic model exhibited much higher skill in replicating observations. This study shows that fully hydrodynamic approaches can be practicably employed in large-scale forecast frameworks at high resolution to produce skillful projections of inundation extent without significantly affecting the forecast lead time.
Hurricane forecasts typically neglect the simulation of flood inundation.
A 2D hydrodynamic model is coupled to NOAA forecasts of Hurricane Harvey.
In considering fluvial, pluvial and coastal flooding, it matches observations well.
Simple, zero-physics approaches exhibit relatively low performance in comparison.
Representing the physics of flooding is both possible and necessary in forecasts.