3D VARIATION DATA ASSIMILATION EXPERIMENTS ON A DENSE FOG  EVENT OVER NORTH CHINA

With the development of numerical weather simulation and forecasting, the spatial and temporal resolution of numerical model become much higher and the data assimilation updating cycle much shorter. Assimilation of more meteorological data has recently received an increasing interest to improve the numerical weather forecasts all over the word. But it is still facing many challenging issues, including how to process various data with appropriate data quality control, how to specify the spatial interpolation and discretization errors, and how to extract the meteorological information from various observations and unconventional meteorological data with the accuracy needed by numerical model. Three-dimensional variational data assimilation(3D-Var) allows these data to be assimilated into the model initial fields directly, which employs the non-linear model as a dynamicconstraint to improve the analysis field. The WRF 3D-Var system, which is basedon the nonhydrostatic mesoscale model WRF and developed from MM5 3D-Var, is a famous three-dimensional variational data assimilation system. In this paper, thenonhydrostatic mesoscale WRF and its 3D-Var system were used to study a dense fog event occurred in 13-14 January 2006. Three different observation data sets include the GTS(Global Telecommunication System) data, AMDAR（Aircraft Meteorological Data Relay）data and 9210 data were assimilated into the initial analysis fields in three experiments. Also the experiments of different time interval (1 h, 3 h, 6 h) assimilation were performed to get the six different analysis fields needed by the six simulation experiments. While the control simulation experiment was performed without assimilation, and its initial fields were taken only from the National Center for Environmental Prediction (NCEP) re-analysis data. The results indicate that three assimilation experiments using 3 different kinds of data sets have to different extent corrected the analysis fields, thus showing obvious positive effect on fog simulation. Further study shows that the humidity and stratification stability of the boundary layer have been improved obviously in all assimilation experiments, although different data sets made different contribution to the analysis fields. In the assimilation experiment, the low level humidity fields were obviously improved while the stratification stability had no significant change. The AMDAR data assimilation obviously improved the stability of the boundary layer but not humidity because the humidity information is not included in AMDAR data. The 9210 data assimilation improved both the humidity and stability. The data assimilation experiments of different time intervals show that the analysis increments of multitimes cycle assimilation are better than only one time, the forecast effect of 1 h interval is better than those of 3 h and 6 h interval.