Data Assimilation Application Experiment On Initial Condition Of Air Quality Model In Pearl River Delta

In recent years, the atmospheric pollution problem has been widely concerned. The air quality model is an important tool for early warning and forecasting, however there are still a lot of room for improvement. The application of data assimilation methods in atmospheric chemistry model is a hot research topic in the world. It provides an effective method for improving the accuracy of prediction. With the development of technology, the method develops from the simple optimal interpolation method whose background error field is constant to Ensemble Kalman Filter considering the background field error developed with time. Data assimilation methods are widely used to adjust various uncertain factors of models. In this paper,the SO2 in the Pearl River Delta in December 2013 was simulated by WRF-CMAQ models, and the initial field assimilation prediction experiment was carried out by using the Optimal Interpolation method (OI) and the Ensemble Square Root Filter(EnSRF). Compared the assimilation effect between different methods, analysis the data assimilation effect on the initial fields, and preliminarily discuss of the effect of the adjusted initial fields on the model forecast. The main scientific understanding and conclusions are:(1) The simulation results of WRF-CMAQ show that the simulation of meteorological elements is well, but the simulation of concentration of SO2 is higher.The analysis of the background field shows that the high error field mainly located in the Jiangmen area, and the simulation uncertainty of the SO2 in the boundary layer especially under 400?m. The trend of background error is consistent with the change trend of concentration field.(2) The results of the sensitivity experiments on the number of assimilation sites and related scales show that the mean square error of the assimilation site increases with the number of assimilation sites, and smaller compared with the assimilation. The error of the analysis field increases with the increase of the correlation scale. The optimization scale in the experimental period is 20km. Under the same correlation,the EnSRF method has better effect on the initial field than the OI method.(3) Compared with the OI and EnSRF methods, the mean absolute gross error and root mean square error of the site decreased after the assimilation. The root mean square error decreasing percentage between the assimilation sites and the verification sites reached 73% and 39%, respectively. Assimilation adjusts the distribution pattern of pollutant concentration field, which is more consistent with the observation field,providing the model with the initial field closer to the actual. The effect of EnSRF method is better than the OI method during the pollution process, while the result of the cleaning process is opposite. On the whole, the EnSRF method has better performance on the initial field improvement than the OI method.(4) Compared the effect of different analysis fields from OI and EnSRF on the forecast field,the error of forecast filed with assimilation is smaller than that without assimilation, and the correlation is improved. The results of the two methods are similar, and the error of the prediction field in the EnSRF method is relative small.(5) Using the 01 method to carry out the 6-hour assimilation and prediction experiments with different number of sites, it was found that the initial value of assimilation alleviates the higher simulation of SO2. The reduction in the mean square error of 1 - 6 hours is reduced from 80% to 4%. As the forecast time increases,the difference between the different schemes with different numbers of sites is decreasing.(6) The influence of the perturbation on the initial value to the atmospheric chemistry model is relative small. And the ensemble spread decreases rapidly with the assimilation and simulation. It leads to the problem of filtering divergence in EnSRF method so that we cannot continue to assimilate observations. It is planned to further explore the effect of various kinds of uncertainty factor to the assimilation result and filter divergence.