Toward a Reliable Prediction of Seasonal Forecast Uncertainty: Addressing Model and Initial Condition Uncertainty with Ensemble Data Assimilation and Sequential Bayesian Combination

Uncertainties are an unfortunate yet inevitable part of any forecasting system. Within the context of seasonal hydrologic predictions, these uncertainties can be attributed to three causes: imperfect characterization of initial conditions, an incomplete knowledge of future climate and errors within computational models. This study proposes a method to account for all threes sources of uncertainty, providing a framework to reduce uncertainty and accurately convey persistent predictive uncertainty. In currently available forecast products, only a partial accounting of uncertainty is performed, with the focus primarily on meteorological forcing. For example, the Ensemble Streamflow Prediction (ESP) technique uses meteorological climatology to estimate total uncertainty, thus ignoring initial condition and modeling uncertainty. In order to manage all three sources of uncertainty, this study combines ESP with Ensemble Data Assimilation, to quantify initial condition uncertainty, and Sequential Bayesian Combination, to quantify model errors. This gives a more complete description of seasonal hydrologic forecasting uncertainty. Results from this experiment suggest that the proposed method increases the reliability of probabilistic forecasts, particularly with respect to the tails of the predictive distribution