(INVITED) Measuring the performance of hydrological forecasts for hydropower production at BC Hydro and Hydro-Québec

Hydroelectric utilities, like BC Hydro and Hydro-Quebec, optimize reservoir operations using reservoir inflow forecasts. Considerable effort is made to provide accurate hydrologic forecasts for various lead-times, using various hydrologic models, weather forecasts, and both deterministic and probabilistic techniques. However, it is a big challenge to adequately assess and communicate a sense of forecast skill. Insufficient knowledge of the forecast skill eventually translates into uncertainty of the level of risk adopted into operations. In this presentation, we use statistical verification measures and skill scores to analyze the quality of deterministic and probabilistic hydrologic forecasts. Two sites in British Columbia are studied. Five-day deterministic forecasts are shown to be more skillful than naive forecasts, as well as ensemble seasonal forecasts with lead times of one to four months, issued at the end or after then end of the snow accumulation period. Ensemble seasonal forecasts issued during the winter, with lead times of five and six months, are less skillful than both climatology and the ensemble mean. This can be explained by an underestimation of the uncertainty by the ensemble hydrologic prediction system.