A real time reservoir inflow forecast evaluation framework

Cameron Bracken , Youngjun Son, Vince Tidwell, Nathalie Voisin

Recent advances in hydrologic forecasting have produced a growing number of streamflow forecast products. Hydropower operators must determine whether they improve operations enough to justify their cost. We present a framework for comparing multiple reservoir inflow forecasts over a common period under real-time operating conditions. This design suits cases where hindcasts are infeasible, or where existing hindcasts cannot be fairly compared. In the framework operators define the variables and constraints to be evaluated, help select and revise metrics, and interpret results using the same observations and operating context for every product. We apply it to three reservoirs in the Great River Hydro system on the Connecticut River in the Northeast U.S., comparing two anonymous commercial products labeled A and B. Product B generally matched the in-house forecast and outperformed product A. Under an idealized operator-defined cumulative error stress test, B also had a lower probability of exceeding the forebay compliance band in the first 24 hours at all three dams. Its advantage persisted through the 32-hour horizon shared by all products, though exceedance probabilities were high enough that no forecast could be used without operator intervention. The real-time evaluation screens products and defines the need for a subsequent hindcast-based value assessment using a scheduling model, operating constraints, and market prices.