Three-dimensional variational combined with physical initialization for assimilation of Doppler radar data.

The three-dimensional variational (3D-Var) data assimilation system of the weather research and forecasting (WRF) model (WRF-Var) is further developed with physical initialization (PI) to assimilate Doppler radar radial velocity and reflectivity observations. In this new 3D-Var assimilation scheme, specific humidity, cloud water content and vertical velocity are first derived from reflectivity observations with PI, then the model fields of specific humidity and cloud water content are replaced by the modified ones, and finally, the estimated vertical velocity is added into the cost-function of the existing WRF-Var (version 2.0) as a new type of observation, and radial velocity observations are assimilated directly by the method afforded by WRF-Var. The new assimilation scheme is tested with a heavy convective precipitation event in the middle reach of the Yangtze River on 19 June 2002 and a meiyu front torrential rain event in the Huaihe river basin on 5 July 2003. Assimilation results show that the increments of analyzed variables correspond well with the horizontal distribution of the observed reflectivity. There are positive increments of cloud water content, specific humidity and vertical velocity in the echo region and negative increments of vertical velocity in the echo-free region where the increments of horizontal winds present an anticlockwise transition. Results of forecast experiments show that the effects of adjusting cloud water content or vertical velocity directly with PI on precipitation forecasts are not obvious. Adjusting specific humidity shows a better performance in forecasting the precipitation than directly adjusting cloud water content or vertical velocity. Significant improvements in predicting the precipitation as well as in reducing model's spin-up time are observed when radial velocity and reflectivity observations are assimilated with the new scheme. Therefore, it is an effective way to improve the short-range prediction of precipitation.