An Efficient Estimation of Time‐Varying Parameters of Dynamic Models by Combining Offline Batch Optimization and Online Data Assimilation

Abstract

In Earth system sciences, numerical models are important tools to understand, monitor, and predict earth systems. Many numerical models in earth system sciences have many unknown parameters which cannot be directly determined by current theory and cannot be directly measured. This uncertainty in model parameters substantially affects the skill of numerical models to simulate real phenomena. It is a grand challenge to infer model parameters by integrating observations and models. Although model parameters are often assumed to be time-invariant, there are time-varying parameters in practical applications due to incomplete parameterizations and unconsidered dynamics that control parameters (e.g., Reichert et al., 2021). It is beneficial to develop an efficient and practical parameter estimation method which allows parameters to temporally change.To estimate time-varying parameters, online parameter estimation by sequential data assimilation has been recognized as a useful method. Since ensemble data assimilation such as Ensemble Kalman filter (EnKF) and Particle Filter (PF) can sequentially adjust model parameters using real-time observations, it can be easily applied to the estimation of time-varying parameters. Ruiz et al. (2013) applied the Local Ensemble Transform Kalman Filter (LETKF; Hunt et al., 2007) to jointly estimate state variables and parameters in a simple low-resolution General