Regional Forecasting of Street-Scale Flood Inundation with a Variable-Fidelity Model: Accuracies and Runtimes for Hurricane Harvey

Abstract:

High-fidelity models for flood inundation dynamics based on shallow-water theory offer the accuracy and precision needed to support many decision-making needs related to urban flooding, but two key issues have restrained the advancement of flood inundation forecasting systems for emergency management: high rainfall forecast uncertainties and high computational costs. Here we present a variable-fidelity flood inundation model (PRIMo) of the Houston metropolitan area that ingests streamflow, storm surge, and rainfall forecasts to provide flood inundation forecasts at 3 m resolution for time horizons lasting hours to days. Variable fidelity is achieved by adjusting a single parameter in PRIMo, the upscale factor, set at runtime to balance accuracy and compute costs (Sanders and Schubert 2019). We present a PRIMo hindcast of Hurricane Harvey and we evaluate accuracy using several indicators including stream gage records, high water marks, 311 flooding calls, damages estimated by Harris County Appraisal Department, and road closure information. Accuracy and compute costs are reported as a function of the upscale factor, and considering both forecasted rainfall (NWS QPF) and observed rainfall (NCEP Stage IV). We find that a 5-day PRIMo forecast requires roughly one hour of wall-clock time, while a 6-hr forecast can be completed in a matter of minutes. Multiple accuracy indicators point to an inundation forecast that is actionable for emergency management, and we demonstrate the importance of hourly (or finer) temporal resolution rainfall data.

Sanders, B.F. and Schubert, J.E., 2019. PRIMo: Parallel raster inundation model. Advances in Water Resources, 126, pp.79-95.