Probabilistic Streamflow Forecasting Based on Generalized Likelihood Uncertainty Estimation and Bayesian Model Average in Huangnizhuang Basin

Abstract:

It has been widely acknowledged that the existence of various uncertainties, both in the initial conditions and in the model physics, unavoidably result in forecast uncertainty. In this study, an ensemble prediction system of streamflow in Huangnizhuang basin is developed to capture the forecast uncertainty and improve the forecasting reliability. In this system, the generalized likelihood uncertainty estimation (GLUE) approach is firstly used to select 50 sets of model parameters of XinAnJiang (XAJ) model, which makes the simulated series matches the observation well. For the given initial conditions, the 50 models with same model structure and different model parameters are driven simultaneously to produce 50-member ensemble streamflow forecasts. Then, the Bayesian model average (BMA) method is used to post-process raw ensemble streamflow forecasts to obtain calibrated forecasts. The study shows that the ensemble streamflow prediction system provides an effective and robust way to assess the uncertainty of streamflow forecasting.