Author(s): Yicong Tong; Lige Cao; Xuan Wang

Linked Author(s):

Keywords: Data Assimilation; Parameter optimization; Lorenz63; A-4DEnVar

Abstract: A significant source of numerical model error arises from uncertainty in the parameters. Optimizing these parameters can enhance model predictions. Data assimilation is a method that integrates observational information with numerical models, enabling updates to both model states and parameters. The Four-Dimensional Variational (4D-Var) and the Ensemble Kalman Filter (EnKF) are two prominent techniques in the field of data assimilation, frequently employed for parameter updating. However, the 4D-Var method encounters challenges related to tangent linear models and adjoint models, resulting in poor portability and high computational costs. Conversely, EnKF, as a sequential assimilation technique, struggles to achieve the global optimum parameter values for the entire simulation period, particularly in high-dimensional non-linear systems. In this study, we employ a hybrid data assimilation method known as Analytical Four-Dimensional Ensemble Variational (A-4DEnVar). This approach approximates the background error covariance through matrix decomposition and an iterative procedure, which not only avoids matrix inversion but also reduces computational demands to the order of generated ensembles. The proposed method was conducted and tested based on the Lorenz63 model to optimize its three parameters. The results demonstrate that the A-4DEnVar method can derive global optimum parameter values with reduced computational demands, indicating its potential for widespread application in strongly non-linear and multivariable models.

DOI: https://doi.org/10.64697/978-90-835589-7-4_41WC-P2040-cd

Year: 2025