著者

TSUYUKI Tadashi

出版者

Meteorological Society of Japan

雑誌

気象集誌. 第2輯 (ISSN:00261165)

巻号頁・発行日

pp.2019-067, (Released:2019-09-24)

被引用文献数

2

---

A multi-scale data assimilation method for the ensemble Kalman filter (EnKF) is proposed for atmospheric models in cases with insufficient observations of fast variables. This method is based on the conservation and invertibility of potential vorticity (PV). The dynamical state variables in the free atmosphere of forecast ensemble members are decomposed into balanced and unbalanced parts by applying PV inversion to the PV anomalies computed from spatially smoothed state variables. The mass variables of the two parts are adjusted to remove additional sampling errors introduced by this decomposition. The forecast error covariances between those parts are ignored in the Kalman gain to suppress spurious error correlations. This approximation makes it possible to apply different covariance localizations to each part. The Kalman gain thus obtained is used to assimilate observations. The performance of the proposed method is demonstrated with a shallow water model through twin experiments in a perfect model scenario. Results using the same localization radius for the two parts show that the proposed EnKF is superior in the accuracy of the analysis to a conventional EnKF unless the ensemble size is sufficiently large. It is found that the adjustment of mass variables is necessary to outperform the conventional EnKF. The benefits of the PV inversion using the Bolin-Charney balance over the quasi-geostrophic inversion are marginal in the experiments.

著者

TSUYUKI Tadashi TAMURA Ryosuke

出版者

Meteorological Society of Japan

雑誌

気象集誌. 第2輯 (ISSN:00261165)

巻号頁・発行日

pp.2022-027, (Released:2022-02-22)

被引用文献数

4

---

Recent progress in the particle filter has made it possible to use it for nonlinear or non-Gaussian data assimilation in high-dimensional systems, but a relatively large ensemble is still needed to outperform the ensemble Kalman filter (EnKF) in terms of accuracy. An alternative ensemble data assimilation method based on deep learning is presented, in which deep neural networks are locally embedded in the EnKF. This method is named the deep learning-ensemble Kalman filter (DL-EnKF). The DL-EnKF analysis ensemble is generated from the DL-EnKF analysis and the EnKF analysis deviation ensemble. The performance of the DL-EnKF is investigated through data assimilation experiments in both perfect and imperfect model scenarios using three versions of the Lorenz 96 model and a deterministic EnKF with an ensemble size of 10. Nonlinearity in data assimilation is controlled by changing the time interval between observations. Results demonstrate that despite such a small ensemble the DL-EnKF is superior to the EnKF in terms of accuracy in strongly nonlinear regimes and that the DL-EnKF analysis is more accurate than the output of deep learning due to positive feedback in assimilation cycles. Even if the target of training is an EnKF analysis with a large ensemble or a simulation by an imperfect model, the improvement introduced by the DL-EnKF is not very different from the case where the target of training is the true state.