An assessment of distributed flash flood forecasting accuracy using radar and rain gauge input for a physics-based distributed hydrologic model

SummaryOne approach to reducing societal impacts from flooding is to minimize the public’s exposure by closing flooded intersections, and warning stakeholders. Emergency responders must know when and where flooding is likely to occur. This article describes the real-time performance of a flash flood forecasting system for a significant flood event (September 7–8, 2010) in Austin, Texas. The system uses a physics-based distributed (PBD) hydrologic model, Vflo, together with radar rainfall input to predict stage and discharge at 222 locations in real-time. A comparison of model forecast accuracy using the operational rain gauge-adjusted radar rainfall input (GARR) is made against rain gauge only (RGO) input for a recent flash flood. A collection of calibrated hydrologic models for flash flood prone basins, within the City of Austin, is used for the comparison. A 1.9 h reduction in timing error was achieved using GARR as input rather than RGO. The RMSE of peak stage forecasts with GARR was 0.89 m, but with RGO input, the peak stage RMSE increased to 1.77 m. The use of GARR as input to the PBD model not only increases the forecast lead-time accuracy, but also the accuracy of forecast peak stage across a range of basin sizes. Rain gauge density over the forecast basins was one of the main determinants of forecast accuracy during an extreme event that resulted in significant flooding in a major metropolitan area.

► Accurate and representative rainfall measurement is essential to accurate forecasting of floods in urban areas.
► Flood inundation generated by a distributed hydrologic model was useful to emergency responders during a recent severe storm.
► Lead-time and stage forecast accuracy obtained with gauge-adjusted radar exceeded results obtained with the rain gauges alone.
► To achieve forecast accuracy equivalent to gauge-adjusted radar input, the rain gauge network would need to be five times denser.