Evaluating Enhanced Streamflow Forecasting Techniques in the context of Reclamation and U.S. Army Corps of Engineers Water Management
The first effort evaluates traditional versus enhanced forecasting methods to reduce streamflow forecasting uncertainties, using hindcast evaluations within a large-sample of basins across the contiguous U.S. (CONUS) for various forecast situations (i.e. different types of forecasts (e.g., 1-day stage forecasts, 3 month volume forecasts) initiated throughout the water year and in different hydroclimate regions). Methodological contrasts are considered at various workflow stages, including the use of single-model forecast vs. multi-model, deterministic vs. probabilistic historical forcing data development, forecast initialization based on manual vs. different automated data assimilation techniques, which gives rise to a variety of method combinations. We summarize key findings from this CONUS-wide evaluation.
The second effort builds from the first, extending the evaluation from metrics of forecast quality to an assessment of water management impact. The research focuses on three reservoir system watersheds (the Boise Project, the Colorado-Big Thompson Project, and a Pacific Northwest flood control system), working with Reclamation and USACE reservoir operators to retrospectively develop operations outlooks informed by hindcast information in order to evaluate potential water management impact. The presentation will summarize the status and preliminary results of these case study activities.
The third effort involves applying lessons learned from the hindcasting efforts that informed preference among enhanced forecasting techniques, and further evaluating those preferred techniques in a more challenging real-time forecasting and forecast-use environment. To that end, collaborators are beginning a new three-year, real-time, experimental forecasting demonstration featuring both traditional and preferred advanced techniques. Techniques will be evaluated for impact on forecast quality for a large-sample of basins and also for potential water management impact in the same case study basins considered in the second effort (among potentially others). Also, the demonstration will rely on an automated forecasting workflow that is supervised by a forecaster "over-the-loop" rather than "in-the-loop," as in the current operational approach. As such, the demonstration affords the opportunity to evaluate "over-the-loop" workflow merits and limitations in the context of Reclamation and USACE water management support. The presentation will summarize experimental design and 2015 platform development activities leading up to forecast system launch in water year 2016.