동해지역모델에 대한 앙상블 칼만필터의 적용 연구

Abstract

The Kalman Filter is a powerful framework of a sequential data assimilation, in which the background error covariance is evolved by a forecast from the previous analysis. The expensive computation, however, has kept from the practical implementation. Though the ensemble Kalman Filter (EnKF) was introduced by Evensen (1994) and has been applied successfully in meteorological applications, there are few oceanographic applications. The EnKF was employed for the Regional Ocean Modeling System (ROMS) in this study. The technical approaches are based on the algorithm suggested by Evensen (1994) and modified by Houtekamer and Mitchell (1998). In spite of merits of the EnKF, there can be some issues due to the finite ensemble size. The EnKF can underestimate the rank of the error covariance since the forecast error covariance is estimated, which can make the estimated forecast error covariance noisy and cause the Kalman Filter to diverge. To deal with these issues in the EnKF, covariance localization has been suggested by Gaspari and Cohn (1999) and covariance inflation by Wang and Bishop (2003). To discuss these issues to implement the EnKF to the ROMS, sensitivity test to the covariance localization and inflation has been conducted using twin experiments in the simple case of the ROMS. Analysis error and its uncertainty with or without the covariance localization and inflation will be discussed. In addition, the multivariate forecast error covariance has been applied and will be discussed. Finally, a simple example of the East Sea Regional Ocean Modeling System equipped with the EnKF will be presented.