準地衡流モデルへの決定論的アンサンブルデータ同化

Abstract

Ensemble data assimilation experiments for a quasi-geostrophic model are conducted with the deterministic ensemble Kalman filter under an idealized mid-latitude ocean with double gyres. The model uses the Arakawa Jacobian for the nonlinear term and the multigrid method to solve the Helmholtz equation. To reduce pseudo correlation and increase the apparent sample size, the forecast error covariance matrix is localized by Schur product of a Gaussian function of the distance between two grid points. Localization can be applied to the blocks of the forecast error covariance separately by variable decomposition. The innovations of a diagnostic variable (streamfunction) correct the prognostic variable (potential vorticity) through the forecast error covariance between the two, indicating that the analysis of the potential vorticity is essential. By contrast, the analysis of the streamfunction only smoothes vortices in a few cycles and results in an ensemble collapse.