Abstract
A routine air quality data assimilation (DA) system was established in the China National Environmental Monitoring Center (CNEMC) based on the optimal interpolation (OI) method. The surface observations from more than 1,400 stations over China were assimilated into a real-time air quality forecast system with three nested domains. The initial conditions of NO2, SO2 and PM2.5 in the three domains were optimized by the data assimilation system. The impact of the data assimilation on the real-time PM2.5 forecast over the Beijing-Tianjin-Hebei (BTH) region during the heavy haze season of 2015 was evaluated. The results show that the DA can significantly improve real-time PM2.5 forecasts with the root mean square error (RMSE) reduced by 23%, 8.2%, 4.8% for the forecasts of the first day, second day and the third day respectively. The mean fractional bias and the mean fractional error of the forecast were reduced from 50.9% and 70.67% to 40% and 62.3% respectively, and the performance was changed from "criteria" to approach "goal" (defined by Boylan and Russell, 2006). It is also found that increasing the assimilation frequency can improve the DA system performance for real-time forecasts. As can be seen from the various case studied here, the improvement of data assimilation is more significant when the bias of the model is higher, and there is still a lot of room for correction. The results also show a rapid decay of the DA effects on the PM2.5 forecast, which highlights the limitations of the current routine data assimilation system in which only initial conditions are optimized. Further improvements of the data assimilation system with meteorological data assimilation and chemical parameter optimization are needed.
