[M-GI26] Data assimilation: A fundamental approach in geosciences

[MGI26-01] A hybrid particle filter/ensemble Kalman filter implementation with an intermediate AGCM

★Invited Papers

*Keiichi Kondo¹, Takemasa Miyoshi², Takuya Kawabata¹ (1.Meteorological Research Institute, 2.RIKEN)

Keywords:particle filter, ensemble Kalman filter, hybrid data assimilation

Kondo and Miyoshi (2016) reported big advantage of removing covariance localization for a 10240-member ensemble Kalman filter (EnKF) with an intermediate atmospheric general circulation model (AGCM) known as the SPEEDY model at the T30/L7 resolution. The analysis accuracy without localization was greatly improved in general, but the improvement in the tropical regions was relatively small. We found that the non-Gaussian PDF such as multiple peaks frequently appeared in the tropical regions, and that the spatial distribution of the non-Gaussian PDF corresponded well to that of the analysis errors (Kondo and Miyoshi 2019). To treat the non-Gaussianity from both nonlinear dynamics and observation operators, we develop a novel hybrid system to combine the EnKF and particle filter (PF). Here, the system is designed to weigh more on the PF in the region with stronger non-Gaussianity. The results show that the new hybrid system works well with 80 ensemble members, with a clear advantage for a nonlinear observation operator. This presentation will give the most recent results of our study.