雷达径向风资料的四维变分同化试验

Four-Dimensional Variational Data Assimilation of Radar Radial Velocity Observations

在MM5四维变分同化系统(MM5 4D-Var)的基础上开发出雷达径向风资料的同化模块,利用改进后的同化系统对2002年7月22日湖北宜昌多普勒雷达的径向风观测资料进行四维变分同化试验,分析雷达径向风资料的同化对中尺度数值模式初始场的调整以及对中尺度强降水模拟的影响。研究结果表明,同化雷达径向风资料加强了中尺度对流系统低层100 km左右范围内的风场辐合。调整后的风场具有更明显的β中尺度特征。利用14 min的雷达径向风资料可以改进3 h之内的强降水模拟,尤其是对雷达站东南侧强降水的模拟。

Assimilation of Doppler radar radial velocity data has recently received an increased interest to improve the numerical weather forecasts all over the world. But it is still facing many challenging issues, including how to process the radar data with appropriate data quality control, how to specify the spatial interpolation and discretization errors, how to extract the meteorological information from radar observations with the accuracy needed by numerical models. Four-dimensional variational data assimilation （4D-Var） allows observations to be assimilated into the model initial fields directly, which employ the non-linear model as a dynamic constraint. And the multi-time-level information, instead of one-time-level information, is used to improve the analysis. MM5 4D-Var system, which is based on the nonhydrostatic mesoscale model MM5, is a famous four-dimensional varitaional data assimilation system. In this study, the observation operator for Doppler radial velocity has been developed and incorporated into MM5 4D-Var system. The testing results of the gradient calculation tend to confirm that the modified MM5 4D-Vat system are working correctly. The modified MM5 4D-Vat system is used to perform the assimilation of Doppler radar radial velocity data. The radar radial velocity data used in this study were collected from the Doppler radar at Yichang in Hubei Province on 22 July 2002. The preprocessing of the Doppler radar data includes generation of super-observatlons in model grid points as well as data quality control. To assess the impact of radar data on the mesoscale heavy rainfall forecast, the MM5 model is used to perform the simulation experiments. The model domains consist of a 45-km grid （1301, grid size of 81×81）, and a 15-km grid （1302, grid size of 61;〈61）. There are a total of 23 layers in the vertical. First, a control simulation experiment （CTRL） is performed prior assimilation of Doppler radar datm The National Center for Environmental Prediction （NCEP） re-analysis data are used as the initial fields in the two model domains in the CTRL experiment. Then the assimilation experiment is conducted in domain D02. The Doppler radar radial velocity data at 1203 UTC, 1208 UTC and 1214 UTC on 22 July 2002 are assimilated using a 14-minute assimilation windows. The assimilated results are used as the initial fields in domain D02 in the second simulation experiment. After the assimilation of radar radial velocity data, the wind fields from the assimilated results reproduce the mesoscale feature around the mesoscale convective system. The low-level convergence in the wind fields at the range of 100 km is enhanced. But the impact of radar radial velocity data on the thero- modynamic fields is limit. The reason is maybe that the background error variance matrix used in MM5 4D-Var system contains only the diagonal elements and the assimilation windows is only 14 minutes which is not enough for the adjustment in a dynamically consistent way. Generally speaking, the rainfall forecast in the first 3 hours from 4D- Var analysis with Doppler velocity data is better than the forecast without radar data. The precipitation pattern and amount are close to the observed, especially in the southeast of the radar observation station.