P-17

An Ensemble-Based Variational Data Assimilation System Using Observation Localization

S. Yokota, S. Origuchi, M. Kunii, and K. Aonashi (Meteorological Research Institute)

In the data assimilation with Local Ensemble Transform Kalman Filter (LETKF, Hunt et al. 2007), the flow-dependent forecast error covariance can be used to make the analysis. However, the calculation of the error covariance including the non-linear forecast model and the observation operator must be approximated using ensemble perturbations in LETKF. To avoid this approximation, it is better that the iteration is used to minimize the cost function. Therefore, an Ensemble-Based Variational Data Assimilation (EnVar) system was developed in this study. In some previous EnVar systems, spatial localization was applied to the forecast error covariance (e.g., Liu et al. 2009) to remove the sampling error related to limited ensemble members. However, since this localization is largely different from LETKF, adding EnVar to the LETKF system or comparing between EnVar and LETKF is not simple. Therefore, we designed EnVar system using the observation localization (Hunt et al. 2007) which is adopted in LETKF. In the assimilation experiment using this system with the SPEEDY model (Molteni 2003), EnVar analyses were closer to the true value than those of LETKF in the same number of ensemble members and the localization radius, although the computation time took several times as long as LETKF.