Pre- and postprocessing flood forecasts using Bayesian model averaging

Abstract In this study, pre- and postprocessing of hydrological ensemble forecasts are evaluated with a special focus on floods for 119 Norwegian catchments. Two years ECMWF temperature precipitation lead time up to 9 days were used force the operational HBV model establish streamflow forecasts. A Bayesian averaging processing approach was applied preprocess Ensemble generated eight schemes based combinations raw, preprocessed, postprocessed datasets evaluate forecasts: (i) all (ii) flood events above mean annual flood. Evaluations data showed that improved only 2–3 days, whereas preprocessing 50–90% catchments beyond 3 days' time. We found large differences in ability issue warnings between spring autumn floods. Spring had predictability many catchments, predict marginal.