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

[MGI29-03] Generation of oceanic perturbations using information on the gradient of the cost function in the JMA’s new operational coupled prediction system

*Yosuke Fujii^1,2,3, Takuma Yoshida^2,1, Yutaro Kubo^2,1 (1.Meteorological Research Institute, Japan Meteorological Agency, 2.Numerical Research prediction Center, Japan Meteorological Agency, 3.Institute of the statistical Mathematics)

Keywords:Ensemble Prediction, BFGS Formula, quasi-Newton Method, Seasonal Forecast, 4-dimensional Variational Method, Data Assimilation

JMA started to use a new operational coupled prediction system, CPS3, since this February. In CPS3, oceanic perturbations calculated from the information on the gradient of the cost function obtained during the minimization process in the ocean 4-dimensional variational (4DVAR) analysis are used to generate ensemble members for performing ensemble predictions. The perturbations are represented by linear combinations of the eigen vectors of the Hessian matrix and expected to include the physical modes with potential of growing in the ocean model. The perturbations are also designed to approximate the analysis (posterior) error covariance matrix for the 4DVAR analysis. In the experiment for the evaluation, the method calculated the perturbations having large amplitude in the eastern Equatorial Pacific where large sea surface temperature anomaly is developed by El Nino-Southern Oscillation (ENSO), the 5-10°N band in the North Pacific where the tropical instability waves are active, and regions with instable western boundary currents such as the areas East of Japan, between the east coast of America, and around South Africa. The perturbations increased the spread among the ensemble members, which better approximated the root mean square prediction errors for the first month in the ENSO predictions from ends of Octobers. The root mean square errors are also slightly reduced for the lead times around one month.