Study For Impact Of Data Assimilation On PM_2.5 Forecast

Fine particulate matter(PM2.5)is one of the most important air pollutants in China,which could damage not merely regional air quality,human health but also climate change,and has become the focus of government and public attention.Aerosol models are useful tools to simulate and predict the PM2.5 concentrations and aerosol-related influences,providing support of effective emission reduction scheme and emergency decisions for the government.However,aerosol modeling remains challenging with large uncertainties in model structure,aerosol emissions,and initial fields.Data assimilation(DA)is a statistically optimal approach,which can reduce the uncertainties of above factors through combining observations and background fields so as to improve the forecasting accuracy.This research is divided into two parts.The first part is focused on the impact of 3DVAR assimilation of surface PM2.5 observations on PM2.5 forecasts over China during wintertime,and North China Plain(NCP),Yangtze River Delta(YRD),and Sichuan Basin(SCB)are the areas with key analysis.We first extend and couple the Grid point Statistical Interpolation(GSI)DA system with the Weather Research and Forecast(WRF)model and Community Multiscale Air Quality(CMAQ)model.The background error covariance is calculated using "NCEP method" Then,two parallel experiments with and without DA are conducted.Cycle assimilation every 6 hours and rolling forecast every 24 hours are performed and verified using mean PM2.5 concentration of each city.The impact of DA on initial fields and forecast fields and the dominant factor affecting the duration of DA benefit are analyzed.In the second part,the impact of inversing emissions of sulfur dioxide(SO2),nitrogen oxide(NOx),and PM2.5 on forecasts is studied.Firstly,we construct a regional air pollution assimilation system based on ensemble Kalman filter(EnKF)method and WRF-CMAQ model.The assimilation window is set to one day,and then SO2,nitrogen dioxide(NO2),PM2.5 observations are simultaneously assimilated into the model in a sequential way.To avoid the influence of model random errors and actual emission uncertainties,the final optimized emission inventory is obtained by temporally averaging of the estimated emissions of each window within the whole study period,which is then used to analysis emission changes and the impact of optimized emissions on forecasts.The main results are as follows:1)DA can significantly reduce the uncertainties of the initial PM2.5 fields.The influences of DA on analysis fields are different in different areas.The aerosol analyses match PM2.5 observations much better after DA.Overall,the mean bias(BIAS)of the analysis field is reduced by more than 70%,the root-mean-square error could be reduced by at least 50%,and the correlation coefficient could be improved to more than 0.9.2)The assimilation of PM2.5 can improve the forecasts at a certain extent and within a certain time.On average,the benefits from DA could last more than 48 hours over China.Much longer DA benefits are found in SCB,Xinjiang,southern China and part of northern China.For the 0-24h PM2.5 forecasts,more than half of the cities with their daily mean hit rates increase more than 10%,and some cities could increase over 30%.3)The duration of DA benefits are dominated by weather condition and emission intensity.The areas with longer DA benefits generally have more stable weather condition and/or weaker emission intensity.The absence of heterogeneous reactions in chemical transport models may also has negative effects on the durations.In addition,we found that the assimilated observation information could transport along with the air masses,and the downwind areas generally have better DA benefits,indicating that when doing air quality forecasting using nested domains,we should conduct the DA in the largest domain rather than the innermost one.4)For SO2 emissions,the increase in emissions after assimilation mainly distributes in Northeast China and Northwest China.While in the NCP,YRD,SCB,Central China,and the Pearl River Delta,the emissions significantly reduce,of which the NCP,the YRD,and the SCB decrease by 40.3%,71.6%,and 84%,respectively.For NOx emissions,the difference between the inversed and the original emissions is relatively small.The emissions in the Mid-western Region of China slightly increase,while the Eastern China slightly decrease,of which the YRD decreases obviously,and decrease by 30.1%.The increase of PM2.5 emissions mainly distributes in North China and Northeast China and Northwest China,and the decrease mainly distributes in Henan province and Southern China.Overall,mean PM2.5 emissions increase by 12.5%nationwide,of which the NCP increase by 33.4%,while in the YRD and SCB,the PM2.5 emissions are reduce by 88.3%,41.5%,respectively.5)The optimized emissions significantly improve the prediction of SO2,NO2 and PM2.5 concentrations.For SO2,the BIAS of Mainland,NCP,YRD,and SCB decrease by 67%,71.4%,96.3%and 98.6%respectively;for NO2,since the BIAS of the Mainland before inversion is small,the improvement of BIAS over Mainland is inapparent,but in the three sub-regions,they decrease 52%,84.1%,and 59.7%respectively.For PM2.5 simulation,the BIAS decrease 49%,48.4%,53.6%,and 25.5%for the 4 regions,respectively.Overall,by using inversed emissions,the prediction of SO2,NO2,and PM2.5 of 85.8%,82.5%,and 85.4%of cities in China are improved,respectively.6)The inversed emission inventory still has large uncertainty.In order to further decrease the uncertainty,it is necessary to improve the chemistry model(e.g.,heterogeneous chemical),enhance the spatial resolution,increase the time of inversion,and improve the simulation of meteorological field through DA in the future.However,comparing the impact of 3DVAR of initial field and emission inversion on the PM2.5 forecasts,we found that the latter is far better than the former.