Assessment of Streamflow Forecast Skill in the Truckee River Basin

Abstract:

Water supply reliability in the Truckee River basin stands to substantially benefit from forecast-informed reservoir operations (FIRO), especially given expected increases in rain:snow ratios and a transition to earlier runoff under warming climate that current infrastructure and operational rules were not designed for. However, in its position on the lee side of the Sierra Nevada mountains (CA/NV, USA), several unique forecast uncertainties exist that must be considered to mitigate against increased flood risk potential. Both water supply and floods are strongly linked to wintertime atmospheric rivers (AR) but despite improvements in forecasting these events at long lead times, the timing and amount of spillover precipitation onto the lee side remains a key uncertainty. In addition, storm runoff volumes in this basin are highly sensitive to rain-snow elevation, which is also difficult to forecast. Finally, antecedent snowpack and soil conditions have the potential to modulate runoff volumes but factors controlling the strength of these modulations are incompletely understood and monitored. In this study, we assess streamflow forecast skill in the Truckee River to provide a preliminary understanding of potential forecast-related challenges and opportunities for FIRO. To accomplish this, we used an archive of available short-range Hydrologic Ensemble Forecast System winter (Oct-Apr) streamflow forecasts for water years 2015-2020 and compared these to observed daily flows at lead times of 0 to 15 days. R² values between observed and forecast ensemble median daily flows show visible improvement at lead times less than 6 days for all winter days and for days on which ARs occurred, though AR days tend to have lower R²s at longer lead times. Average R²s across 12 sites at a 1-day lead time are 0.75 for all winter days and 0.68 for winter AR days and are 0.56 (all days) and 0.37 (AR days) at a 5-day lead. Average probabilities of at least one ensemble member correctly detecting the exceedance of a 90^th percentile historical daily flow threshold are 74% and 58% for 1- and 5-day lead times respectively. There is a bias toward under-forecasting larger streamflow volumes, even at short lead times. The influences of antecedent conditions and AR attributes on streamflow forecast skill will also be analyzed and reported.