Improving the efficient and robust uncertainty quantification in real-time flood forecasting using Polynomial chaos expansions and ensemble Kalman filter

no research has been done in case of real-time forecasting. Furthermore, many uncertainties are exist influencing to forecasting result. EnKF is the useful technique to address that issue by updating the model state and parameter during the real-time. The proposed framework significantly enhances the efficiency and accuracy of hydrological application in real-time forecasting, which plays an important role in the flood planning, management, and mitigating the flood risk during the golden time. In real-time process, the forecast rainfall and the flow at the current time are updated, the unified framework will be capable to automatically upgrade the ensemble model states and model parameters through Dual EnKF (dual states-parameters estimation); and the ensemble hydrologic predictions are estimated a seamlessly through PCE. Besides, to maximize the efficient in forecasting, the approach of GLUE (Generalized likelihood uncertainty estimation) is used to determine the ensemble size of model states and parameters sets, and the perturbed observation. The proposed approach is applied to the Vu Gia watershed in Vietnam to demonstrate its validity and applicability. A detailed comparison with the NAM hydrologic model shows that analyzing results with surrogate model are as good as those given by NAM model, while the forecasting results are significantly improved through automatic updating of states and parameters by EnKF; not only the good accuracy, but also the model can run nearly 10 times faster than the hydrologic model. Overall, the results indicate that the uncertainty propagation in real-time flood forecasting can be effectively characterized and robust through the proposed unified framework.