Effective Methods for Increasing Model Background Error in the Ensemble Kalman Filtering in Aerosol Data Assimilation

The Ensemble Kalman Filter (EnKF) has been employed for updating the initial condition, and promising results have been reported. Unlike the variational assimilation method, the advantages of EnKF are flow-dependent background error covariance which is important in a fast-developing air quality system. However, assimilation of air quality observations often suffers from insufficient model background error due to a small ensemble spread when applying EnKF methods. This study suggests methods for effectively increasing model background error covariance (BEC) by perturbing prognostic variables and employing multiple physics parameterizations in the atmospheric chemical transport model.

This study developed an aerosol data assimilation system with the WRF-Chem model and EnKF approach. In spite of considering flow-dependent BEC, the baseline run analysis exhibits poor performance, primarily due to the small ensemble spread. This study conducted new two effective methods for increasing ensemble spread: one considering the uncertainty of model physics and the other considering the uncertainty in the prognostic variables. Both methods improved the quality of surface PM analysis substantially, compared with the baseline run. And the DA_all experiment which incorporates both uncertainty in model physics and prognostic variables, demonstrates the best performance. Physical perturbation and multiplicative perturbation have a non-linear relationship. The forecast skill is also improved. With the substantial increase of BEC, the revised EnKF system has significantly improved the PM_2.5 forecast skills.