A comparison of variational, ensemble-based, and hybrid data assimilation methods over East Asia for two one-month periods

In this study, the performances of variational (three-dimensional variational; 3DVAR), ensemble-based (ensemble Kalman filter; EnKF), and hybrid (E3DVAR) data assimilation (DA) methods based on the Advanced Research Weather Research and Forecasting (WRF) model are investigated over East Asia for two one-month period of January and July in 2016. Before a comparison between three methods for two one-month periods, a single observation experiment is conducted to tune and optimize background error covariance depending on each method, so that all methods have similar influence radius. For a comparison between three methods for two one-month period by assimilating conventional observations, the E3DVAR outperforms 3DVAR and EnKF for both two seasons. The 3DVAR outperforms EnKF in January, whereas EnKF outperforms 3DVAR in July. The root mean of difference total energy (RM-DTE) for January increases as a forecast time increases, saturating at the value less than 5 m s(-1). On the contrary, RM-DTE in July keeps increasing until 72 h forecast time reaching at the value less than 7 m s(-1). Relatively larger moisture error in initial condition for summer season can grow rapidly and change large-scale feature considerably, which can contribute to the continuous growth of RM-DTE in July. Furthermore, rank histogram and spread statistics results confirm that ensemble spreads are represented reasonably for January and July in 2016, although spreads in July are slightly overestimated compared to those in January. In conclusion, the hybrid DA method (E3DVAR) is the most appropriate among three DA methods over East Asia. In addition, for the better performance, it is necessary to tune and optimize the DA system depending on DA method for the given area.

Automated summary

This study investigated the performance of three data assimilation methods based on the WRF model over East Asia. The hybrid E3DVAR method outperformed 3DVAR and EnKF for both January and July seasons. Adjusting background error covariance can improve forecast accuracy, and each method has different strengths in different seasons.