Improving Robustness of Hydrologic Ensemble Predictions Through Probabilistic Pre‐ and Post‐Processing in Sequential Data Assimilation

Abstract Data assimilation using the ensemble Kalman filter (EnKF) has been increasingly recognized as a promising tool for probabilistic hydrologic predictions. However, little effort has been made to conduct the pre‐ and post‐processing of assimilation experiments, posing a significant challenge in achieving the best performance of hydrologic predictions. This paper presents a unified data assimilation framework for improving the robustness of hydrologic ensemble predictions. Statistical pre‐processing of assimilation experiments is conducted through the factorial design and analysis to identify the best EnKF settings with maximized performance. After the data assimilation operation, statistical post‐processing analysis is also performed through the factorial polynomial chaos expansion to efficiently address uncertainties in hydrologic predictions, as well as to explicitly reveal potential interactions among model parameters and their contributions to the predictive accuracy. In addition, the Gaussian anamorphosis is used to establish a seamless bridge between data assimilation and uncertainty quantification of hydrologic predictions. Both synthetic and real data assimilation experiments are carried out to demonstrate feasibility and applicability of the proposed methodology in the Guadalupe River basin, Texas. Results suggest that statistical pre‐ and post‐processing of data assimilation experiments provide meaningful insights into the dynamic behavior of hydrologic systems and enhance robustness of hydrologic ensemble predictions. Plain Language Summary Data assimilation techniques are recognized as a promising tool for probabilistic hydrologic predictions. And the pre‐ and post‐processing of assimilation experiments play a crucial role in advancing our understanding of the nonlinear dynamic behavior of hydrologic prediction systems. This paper presents a unified computational framework that enables a systematic integration of data assimilation using the ensemble Kalman filter (EnKF) as well as statistical pre‐ and post‐processing techniques, strengthening our capability in providing probabilistic streamflow predictions. Both synthetic and real data assimilation experiments are conducted to demonstrate applicability of the proposed computational framework in the Guadalupe River basin, Texas. Results verify that the pre‐ and post‐processing of assimilation experiments provide meaningful insights into the potential interactions among the EnKF error parameters and those among hydrologic model parameters. In addition, the Gaussian anamorphosis establishes a seamless bridge between data assimilation and uncertainty quantification. Therefore, such a unified computational framework has significant potential for performing robust hydrologic forecasting. Key Points Preprocessing, data assimilation, and postprocessing experiments are conducted systematically with both synthetic and real data The precipitation error parameter has the most significant impact on the accuracy of streamflow predictions using the ensemble Kalman filter There is considerable temporal variation in the sensitivity of model parameters during the period of hydrologic predictions