Exploring the QPF Performance on the NWM Forecasts with and without Upstream Data Assimilation

The National Water Model (NWM) provides streamflow forecasts for up to 18 hours, using High Resolution Rapid Refresh (HRRR) and incorporating direct data assimilation of USGS discharge observations to enhance forecasts. In previous work, we explored how the quality of the HRRR quantitative precipitation forecast (QPF) compared with the Multi Radar Multi Sensor (MRMS) quantitative precipitation estimation (QPE) impacts the NWM performance for lead times up to 18 hours. According to our results, QPF quality has a more profound influence over relatively small watersheds (areas below 1,000 km²) and for lead times above 10 hours. Conditioned on the QPF performance, we delve deeper into the effects of data assimilation on the forecast performance. Specifically, we analyzed differences between watersheds with and without streamflow data assimilation, including forecasts at river segments upstream of the observation sites. Our analysis reveals that watersheds with upstream data assimilation exhibit improved forecast performance, reducing error margins and higher reliability in flood prediction. However, the forecast improvements due to data assimilation decrease quickly with the lead time. Additionally, we observed that the QPF performance and the watershed size also changed the data assimilation effect.