WRF-Hydro Streamflow Prediction Driven by West-WRF Ensemble Precipitation: A Case Study in Californias Lake Mendocino Basin

Skillful hydrologic forecasts are critical during extreme precipitation events, such as the atmospheric river (AR) events. However, such forecasts carry significant uncertainties from the precipitation forecast from a weather model and from the hydrological model. Consequently, ensemble forecasting is needed to capture a range of forecast uncertainties. We develop hourly WRF-Hydro ensemble streamflow forecasts driven by 64-member ensemble West-WRF meteorological forecasts, with 9-km resolution downscaled to 1-km land surface model resolution. We present a case study to evaluate WRF-Hydro ensemble forecasts performance at 24- to 120-hr lead times during a series of AR events in January-February of water years 2017 and 2019. This study focuses on evaluating the coupled meteorological-hydrological model ensemble behavior and its error over the Lake Mendocino basin in Northern California, specifically: 1) if the ensemble sufficiently captures a range of forecast uncertainties related to different categories of AR meteorology, and 2) how much the uncertainties of the ensemble precipitation input affect WRF-Hydro streamflow forecasts. The West-WRF ensemble precipitation shows underestimations throughout the Lake Mendocino domain. The ensemble spread and the underestimation are accentuated at longer lead times and with larger precipitation events. For example, during the 7-11 January 2017 AR event, the underestimation for the ensemble mean amounts up to 30 mm/day at 24-hr lead time and 50 mm/day at 120-hr lead time. The underestimation similarly occurs in the WRF-Hydro ensemble streamflow forecasts, where the peak flow underestimation amounts up to 92 m3/s (24-hr lead) and 138 m3/s (120-hr lead) at USGS Calpella gauge. The ensemble streamflows also exhibit 6- to 12-hr delays in the peak flow timing and flow overestimations during recession periods, relative to the observation. This evaluation is ultimately pivotal to our effort to develop this coupled-ensemble modeling framework into a near-real time system.