[M-GI29] Data assimilation: A fundamental approach in geosciences

[MGI29-P02] Improvement of the ocean 4D-Var scheme for accurate sea surface temperature analysis

*Norihisa Usui¹, Yosuke Fujii¹, Nariaki Hirose¹, Nadao Kohno¹ (1.Meteorological Research Institute)

Keywords:Sea Surface Temperature, 4D-Var, Data assimilation

Many ocean data assimilation systems use gridded sea surface temperatures (SSTs) for assimilation produced by statistical interpolation methods such as optimal interpolation. Such gridded SSTs are however statio-temporally smoothed. Therefore, in a high-resolution ocean data assimilation system, it is desirable to assimilate satellite Level-2 data directly, rather than the gridded data. In this study, we propose a method to analyze the SST field with high accuracy within the framework of the four-dimensional variational (4D-Var) method. In ocean data assimilation systems based on 4D-Var, the assimilation window is usually set to about 10 days, and increments to the initial condition used as the control variables are optimized by minimizing the cost function. While this setting is reasonable for estimation of ocean internal variations such as changes in ocean current and ocean internal temperature, it would not be appropriate for the SST field because the SST variability is influenced not only by the oceanic internal dynamics but also by the atmospheric forcing. In this study, we developed a new 4D-Var scheme to reproduce detailed spatio-temporal SST variations with guaranteed reproducibility of ocean internal variations. In the new scheme, daily SST increments within the assimilation window are added to control variables, which are regarded as independent from other control variables such as temperature and salinity increments to the initial condition. We implemented this scheme into the Meteorological Research Institute Multivariate Ocean Variational Estimation (MOVE) system and conducted an assimilation experiment using Himawari-SST. It was shown that assimilated SST fields are much improved compared to those analyzed by the conventional 4D-Var scheme. At the time of presentation, we would like to discuss the results more in detail as well as the setting of the background error covariance matrix.