The Quest for Model Uncertainty Quantification: A Hybrid Ensemble and Variational Data Assimilation Framework.

Area of Science:

• Environmental Science
• Hydrology
• Data Assimilation

Background:

• Accurate hydrologic predictions require robust data assimilation methods to handle model and parameter uncertainties.
• Existing methods often struggle with computational cost and characterizing diverse error sources.

Purpose of the Study:

• To develop a novel hybrid data assimilation framework, HEAVEN, for dual-state parameter estimation.
• To improve the characterization of model structural uncertainty alongside parameter and input uncertainties.
• To provide a computationally efficient and robust approach for hydrologic predictions.

Main Methods:

• Coupling a deterministic four-dimensional variational (4DVAR) method with a particle filter (PF) ensemble system.
• Formulating a sequential PF within the 4DVAR system for efficient feedback.
• Utilizing 4DVAR for initial state estimation and PF for posterior distribution, updating error covariance to account for structural uncertainty.

Main Results:

• The HEAVEN framework successfully estimates dual states and parameters in a nonlinear hydrologic model.
• Demonstrated effectiveness, robustness, and reliability across multiple U.S. river basins.
• The method accounts for all uncertainty sources, uses small ensemble sizes, and avoids particle degeneracy.

Conclusions:

• The HEAVEN framework offers a significant advancement in hydrologic data assimilation.
• It provides a reliable method for dual-state parameter estimation by comprehensively addressing uncertainties.
• The approach is computationally efficient and applicable to real-world hydrologic systems.