Noise covariance estimation using dual estimation for disturbance storm time index application

The disturbance storm time (D_st) index is used for predicting the geomagnetic storm that can affect many systems on earth. The application of the dual unscented Kalman filter (DUKF) to improve the quality of the D_st index prediction by simultaneously estimating the process noise covariance is set forth in this paper. In DUKF, two unscented Kalman filters (UKFs) are run in parallel. The UKF applied to a model-based D_st index prediction is so called a state estimator; while the other, a parameter estimator, is for identifying and recursively updating the process noise covariance. The performance comparison between the traditional UKF with fixed constant values of the process noise covariance, and the DUKF are examined. The actual all D_st and the D_st data during the storm (below -80 nT) are used to assess the quality of the predictions. It is found that root mean square error (RMSE) of D_st index prediction using the DUKF is lower than that of the UKF with fixed constant process noise covariances. Specifically, RMSEs of the DUKF are 6.5816 for all D_st and 18.0615 for D_st below -80 nT, whereas, the prediction using a fixed constant process noise covariance yield RMSEs of at least 6.6678 and 19.3954 for all D_st and D_st below -80 nT, respectively. Hence, the DUKF outperforms the traditional UKF with fixed constant process noise covariances in this study.