Reclamation Information Sharing Environment (RISE)

We investigated risk-informed decision making (RIDM) methods for reservoir operations and developed an approach for seasonal RIDM using RiverWare and ensemble forecasts. The RiverWare model utilizes two-stage stochastic programming with recourse to recommend operations decisions. This method performs trade-offs between reservoir uses including water supply, hydropower, flood control, fisheries, and recreation. Two metrics represented hindcast skill of ensembles generated with the Structure for Unifying Multiple Modeling Alternatives (SUMMA) Model: Continuous Ranked Probability Skill Score (CRPSS), and rank histograms. SUMMA hindcasts were also compared to PyForecast seasonal statistical forecasts disaggregated using a nearest neighbor approach. We focused on operations during the snowmelt runoff period from April through July at Buffalo Bill Reservoir. We valued reservoir benefits in dollars but valuations were highly uncertain due to poor economic data and are likely not useful for this trade-off. Because of uncertainty associated with project benefits estimates, we subjectively examined the RIDM approach utility using SUMMA hindcasts for the period 1999–2019. The RIDM approach performed worse than both deterministic operations using the median forecast and historical operations. We believe the RIDM approach is valid and suggest several improvements to the method for future efforts including improving forecast skill and simplifying reservoir use trade-offs.