Impact of Radar Data Assimilation on the Simulation of Typhoon Morakot (2009)'s Rainband Using the Ensemble Kalman Filter

Morakot became the deadliest typhoon in Taiwan due to a long-lasting rainband extending from the Taiwan Strait to southwestern Taiwan, dumping extreme rainfall in mountainous southern Taiwan from 7 to 9 August 2009. A lot of works have been done about Morakot mainly on examining the affecting factors of the extremely large accumulated rainfall. However, very few efforts have been made on the detailed simulation of the quasi-stationary rainband as well as its organization and evolution features. This study examines the impact of radar data assimilation on the simulation of this rainband through a WRF-based ensemble Kalman filter (EnKF).

The CNTL experiment without data assimilation was initiated at 0000 UTC 7 August 2009 for a 60-h integration with WRF V3.4.1. Two model domains were employed with grid spacings of 13.5 and 4.5 km. The Kain-Fritsch cumulus, the WSM6 microphysics, and the YSU PBL parameterization schemes were used. The initial and boundary conditions were provided by the NCEP/FNL data. Radar radial velocity data were assimilated using the EnKF to explore to what extent the simulation of this rainband may be improved. The EnKF used an ensemble size of 40, initiated through WRF/3DVAR at 0000 UTC 7 August 2009, and integrated for 12 h to 1200 UTC 7 August to assimilate the first radar data.

Impact of single- versus multi-time radar data assimilation was first examined. “DA_ST” was performed by assimilating the radial velocity of four radars in Taiwan Island at single time of 1200 UTC 7 August, followed by a 48-h deterministic forecast initiated from the posterior ensemble mean. “DA_MT” was performed to assimilate the four radars from 1200 to 1800 UTC 7 August at 1-h intervals followed by a 42-h deterministic forecast. Result shows that both DA_ST and DA_MT produced more realistic typhoon circulation and rainband structures in a longer time than the CNTL. Results of using multiple data assimilation cycles apparently outperformed that of single-time data assimilation. The duration of well-simulated rainband was extended by 1 and 2 hours in DA_ST and DA_MT, respectively, and realistic rainband echoes were further obtained at several discontinuous moments in DA_MT. In addition, the average track error was reduced by 31% and 9% in DA_MT and DA_ST compared with CNTL, respectively. As a result, more realistic pattern of the accumulated precipitation in DA_MT was obtained.

Impact of assimilating data from single versus multiple radars was then examined. Sensitivity experiments by assimilating each one of the four radars individually at 1-h intervals from 1200 to 1800 UTC 7 August were conducted. The radar which has a good coverage of the quasi-stationary rainband region performed better than other radars in terms of the rainband structures. However, the benefits of assimilating any one of the four radars for 7 times were all less than those of either single-time or multiple-time 4-radar data assimilation.

Overall, the more radar data in the temporal and spatial dimensions were assimilated, the better the rainband simulation would be. Further data assimilation experiments are underway to achieve a better simulation of the rainband, based on which the organization and evolution mechanism of the rainband will be explored.