The quandary with QPF in hydrologic forecasting

Results from an on-going operational analysis of Quantitative Precipitation Forecast (QPF) based hydrologic forecasts for 42 locations in the Ohio River Forecast Center (OHRFC) area of responsibility are presented. NOAA/NWS Hydrometeorological Prediction Center (HPC) QPF for 6-, 12-, 24-, 36-, 48-, and 72-hour periods are used as input to the OHRFC operational hydrologic forecast system to generate independent 6-hourly forecasts out to 5 days lead-time each day. The period of study spans 01/23/2009 to 02/27/2010. Verification results show that when all (both flood and non-flood) forecasts are grouped together, longer (that is, 72-hour) duration QPFs, show lower forecast error and reduced bias than with lesser QPF durations. However, when flood and non-flood river stage forecasts are stratified, error statistics show considerably greater error and increased error at increasingly greater lead-times for flood forecast events, whereas non-flood events show reduced error with the use of longer duration QPF. For both flood and non-flood forecasts, forecast error increase monotonically with increased lead-times, except for fast responding basins (less than 48-hours), which exhibit a local relative maximum error between 24 to 48 hours lead-time and then increases again at lead-times exceeding about 60 to 66 hours.

Verification results further show very small differences between 6- and 12-hour river stage forecast errors for flood events, but errors increase measurably with the use of 24-hour QPF and significantly for 48-hour QPF and greater. The quandary is that the use of 72-hour (or possibly more) QPF is a benefit for non-flood events, but for flood events, errors increase dramatically with the use of QPF in excess of 24 hours duration.