Classical Methods Tour of Advanced Data Assimilation using Lorenz ‘ 96 Model Final

Using Lorenz96 model with 40 variables, classical methods of advanced data assimilation are explained, implemented and examined. The classical methods include full Kalman filter (KF), extended Kalman filter (EKF), full Kalman smoother (KS), its iterative versions, and sawtooth algorithms (Johnston and Kurishnamurthy 2001). A brief explanation of the theoretical background of ensemble Kalman filter (EnKF) is also provided. The methods are tested under the perfect model assumption, and it is shown that KS clearly outperforms KF as expected thanks to the use of the future information. In addition, it is shown that iterative KF works more stably and outperforms KF especially in less dense observations both temporally and spatially. Furthermore, model errors are considered in a very simple way. It is shown that the effect of model errors was significantly reduced by increasing the variance inflation parameter.