Introduction to data assimilation for parameter estimation

TL;DR

This paper introduces statistical and variational data assimilation methods, compares Kalman Filter algorithms and 3D-Var, and demonstrates their application to a mass-spring system with experimental results highlighting their strengths and weaknesses.

Contribution

It provides a comparative overview of data assimilation techniques and illustrates their application to a simple dynamical system with experimental analysis.

Findings

Kalman Filter and variational methods have equivalent solutions for linear Gaussian problems.

Different data assimilation algorithms exhibit distinct advantages and disadvantages.

Experimental results demonstrate the practical differences between the methods.

Abstract

In this study, two classes of methods including statistical and variational data assimilation algorithms will be described. In statistical methods, the model state is updated sequentially based on the previous estimate. Variational methods, on the other hand, seek an estimation in space and time by minimizing a cost function. Both of these methods require estimates of background state which is the prior information of the system and its error covariances. In terms of linear and Gaussian problems, they have the same solution. In the family of Kalman Filter algorithms, the conventional Kalman Filter (KF) and Ensemble Kalman Filter (EnKF) will be implemented. A three-dimension variational method (3D-Var) will be employed to illustrate the variational approaches. A simple case of an ordinary differential equation (ODE) is coupled to highlight the difference between these algorithms. Namely, a mass-spring system governed by a second order differential equation will be examined. We also look at the situation where a periodic external force is applied to the system. Then different data assimilation algorithms will be applied to this system. Results from the experiments will be analyzed to showcase the advantages and disadvantages of each method.