Improving National Water Model Flood Forecast Skill over Coastal Western U.S. River Basins.

In 2016, NOAA launched the National Water Model (NWM), a hydrological modelling system intended for use across the U.S.. WRF-Hydro in stand-alone mode (with Noah-MP as the land surface model) is the hydrologic core of NWM and provides new technology for producing flood forecasts. There have been reports that WRF-Hydro-based flood forecasts are less accurate in the Western U.S. than are current methods used by the National Weather Service (NWS). Here, we perform detailed comparisons of both methods, via implementation of WRF-Hydro with best physics options and calibrated parameters for a transect of seven watersheds along the U.S. Pacific Coast. We evaluated WRF-Hydro performance for the largest flood events of the past 7-20 years (peaks over threshold that resulted in three extreme events per year on average or POT3 as well as largest three or one floods) using calibrated WRF-Hydro parameters. We used NWS Quantitative Precipitation Forecast (QPF) precipitation produced by the two coastal Western U.S. River Forecast Centers (RFCs). We compared WRF-Hydro flood reforecasts with RFC archived forecasts for the same events and found that for both POT3 floods and the largest three floods. WRF-Hydro's forecasting performance was comparable with RFC reforecasts in the three northern basins (located in Oregon and Washington) but was inferior in the four southern basins (all in California). While both sets of forecasts tended to underestimate flood peaks, WRF-Hydro's discrepancies increased more rapidly with longer lead times and typically predicted earlier peaks. For the largest floods, WRF-Hydro and RFC forecasts were largely comparable in magnitude, but WRF-Hydro often predicted earlier peaks.