Comparing 3DVAR, the Ensemble Kalman Filter (EnKF), and Hybrid Ensemble-variational Data Assimilation Methods Based on WRF for East Asia over a Two-month Period

A hybrid data assimilation method which couples an ensemble-based one with a variational one has been widely used at many meteorological organizations and research centers over the world. This is because a hybrid method can exploit advantages of each one: an ensemble-based method can represent uncertainties of the day based on ensemble forecasts and a variational one can produce a balanced analysis. However, in East Asia, there are no studies to compare ensemble-based, variational, and hybrid methods at the same time, and an impact of a hybrid assimilation method has not been thoroughly explored over a month-long period. Thus, in this study, the performances of 3DVAR, the ensemble Kalman filter (EnKF), and E3DVAR methods are compared over East Asia for a two-month period.

In this study, 3DVAR, EnKF, and E3DVAR are developed based on the Weather Research and Forecasting (WRF, v3.7.1) model over East Asia. The horizontal resolution is 12-km and 50 vertical levels up to 5 hPa from the surface are used. The 5th ECMWF ReAnalysis (ERA5) data are used as initial and lateral boundary conditions. For each method, analysis fields are produced every 6-h (00, 06, 12, 18 UTC) by assimilating conventional observation data with a 6-h assimilation window for January and July in 2016. Every analysis field is integrated every 6-h for forecast fields up to 36-h. For ensemble-based assimilation methods, 40 ensemble members are used.

Based on the verification results, overall, a hybrid method is found to be the most exact among three methods for two seasons. While the EnKF RMSE is generally the largest among three methods in the winter, the 3DVAR RMSE is the largest in the summer. Regarding the ensemble spread, although the spread in the summer season is a bit large compared to that in the winter season, the spreads of two seasons are found to be sufficient over East Asia. The more detailed evaluation results will be presented at the meeting.

Acknowledgments

This work was supported by a National Research Foundation of Korea (NRF) grant funded by the South Korean government (Ministry of Science and ICT) (Grant 2017R1E1A1A03070968) and the Korea Polar Research Institute (KOPRI, PN19081). The authors gratefully acknowledge the late Dr. Fuqing Zhang for providing the resources and discussions at the earlier stages of this study.