Assessing the Performance of Short Range Quantitative Precipitation Forecasts Used in the New NWS National Water Model (NWM)

The National Water Model (NWM) will be implemented into National Weather Service (NWS) operations in FYQ4 2016 through a partnership among National Water Center (NWC), National Center for Atmospheric Research (NCAR) and National Centers for Environment Protection (NCEP) to provide nationwide high-resolution water forecasts over multiple space-time scales with multiple forecast horizons. This study aims to assess the accuracy of quantitative precipitation forecasts (QPFs), one of the most important forcing variables for the NWM hydrologic forecasting system, to gain insight on how QPF skill influences the NWM’s performance in predicting hydrologic and hydrometeorological processes. QPF input for short range forecast configuration of NWM comes from the operational versions of the Rapid Refresh (RAP) and High Resolution Rapid Refresh (HRRR) numerical weather prediction models. To ensure the highest degree of relevance, this study focuses on the experimental versions of these models (RAPx and HRRRx) that were scheduled to become operational at the same time as the NWM.  

This study differs from prior HRRR and RAP investigations in that it focuses on the spatial (1km) and temporal (hourly) scales at which the NWM operates. The results show that the precipitation forecasts from HRRRx appear to be more skillful than those from the RAPx in depicting the precipitation occurrence and amount compared to the Stage IV analysis across the Contiguous United States (CONUS) during the warm season of May-October 2015. Both QPF sources exhibit the highest skill in the northeast and the lowest quality over the southwest United States, and QPF skill decreases with increasing lead times and accumulation thresholds. Additionally, the RAPx and HRRRx models using all 24 forecast issuances perform better than using 4 issuances at 00 UTC, 06 UTC, 12 UTC and 18 UTC. When averaging QPFs over different hydrologic unit codes (HUCs), overall the analysis suggests that precipitation skill decreases with the size of HUC, although more details are seen at the finer HUC scale. The study also evaluates the ability of each QPF source to capture the historic heavy rainfall that resulted in catastrophic flooding over South Carolina during 1-5 October 2015 and finds that the HRRRx accurately forecasted the timing, location and amount of extreme rainfall during the flooding event.

This evaluation is not only helpful for identifying the strengths and weaknesses of the precipitation forecasts—which is important for atmospheric model development and improvement—but is also valuable for forecasters and other end users to better understand the reliability of NWM hydrologic forecasts in aiding their decision-making process.