Progress on flash flood verification and excessive rainfall related QPF products from the HRRR model

Flash flooding is one of the leading weather-related causes of death in the United States and around the world. While significant progress has been made on improving quantitative precipitation forecasts (QPF) for heavy rainfall, many of these events remain challenging to forecast. A key outstanding question is how to define a flash flood event for the purpose of forecast verification, with significant regional variations evident in the correspondence of different available flash flood “proxy” datasets with flash flood reports or NWS-issued flash flood warnings. Official excessive rainfall outlooks, issued by the Weather Prediction Center (WPC) at NCEP, are defined in terms of the likelihood of exceeding Flash Flood Guidance (FFG; developed by River Forecast Centers) within 40 km of a point.

In this presentation, we will describe recent work focused on developing an improved ground truth dataset for flash flooding, using a combination of available observed and proxy heavy rainfall datasets. Flash flooding reports are relatively sparse in spatial coverage; thus, verification in remote areas depends upon the use of so-called proxy datasets which have regionally-varying bias characteristics related to the underlying assumptions as well as proximity to rain gauges and weather radars. We present a comparison of verification of the various proxy datasets versus flash flood reports over the 2015-2018 period, with particular attention to the spatial variations in correspondence. We also explore the possibility of creating a derived proxy flash flood event dataset consisting of a combination of the proxy datasets. The creation of a gridded “optimal” flash flood proxy dataset holds promise for the verification of probabilistic flash flood forecasts.

Finally, we briefly describe development on QPF-related post-processed fields which will be implemented in the Experimental High Resolution Rapid Refresh (HRRR). Ultimately, knowledge regarding the correspondence of proxy datasets with reported flash floods will inform their use in creating probabilistic flash flood forecasts from convection-allowing ensembles such as the Experimental HRRR ensemble.