The 10,240-member ensemble Kalman filtering with an intermediate AGCM without localization

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Tuesday, 6 January 2015: 4:00 PM
131AB (Phoenix Convention Center - West and North Buildings)
Keiichi Kondo, RIKEN Advanced Institute for Computational Science, Kobe, Hyogo, Japan; and T. Miyoshi

Covariance localization plays an essential role in the ensemble Kalman filter (EnKF) with a limited ensemble size. Localization limits the influence of observations and reduces the impact of sampling errors. In this study, we increase the ensemble size up to 10,240, much larger than the typical choice of about 100, and remove the localization completely in EnKF. The 10,240-memeber EnKF is implemented with an intermediate AGCM known as the SPEEDY. A 1-month data assimilation experiment under the perfect model scenario is performed, and longer-range error correlations and non-Gaussianity are investigated. The results show that the 10240-member EnKF without localization provides very accurate analysis, and if we apply 2000-km localization (more than 6000-km radius of influence), the analysis is significantly degraded. Namely, the long-range error correlations seen in the 10240-member EnKF actually help improve the analysis accuracy.