Evaluation on SO₂ Emission Inventory Optimizing Applied to RMAPS_Chem V1.0 System

Air pollution emission inventory is an important input data of air quality model. The uncertainty of emission inventory is a primary source of error in air quality forecasts and it also affects the regulation of air pollution sources. RMAPS_Chem V1.0 is an operational forecasting system for haze and atmospheric pollution in North China. It is established based on an online coupled regional chemical transport model WRF_Chem. In order to reduce the large deviation of forecasted SO₂ concentration, through the test of model accuracy on weather condition, a conclusion is drawn that the simulated error of SO₂ concentration mainly comes from the deviation of emission. An optimized SO₂ emission inventory is established, first inversed by ensemble square root Kalman filter (EnKF) approach, and then revised by using statistical error correction method. Comparison indicates that the optimized emission has obvious advantages to improve the prediction accuracy of ground SO₂ concentration. Distribution of surface SO₂ concentration over North China in October 2017 is simulated using initial emission inventory (MEIC_2012) and the optimized emission inventory. Simulated results are compared with observations at 616 stations from China National Environmental Monitoring Center (CNEMC), and the difference between simulated results using two emission inventories is analyzed. Results show that the above emission inventory optimizing method is applicable for the correcting of regional deviation in SO₂ emission, which is very effective on improving SO₂ forecast accuracy in main regions and urban areas. Simulated results using optimized emissions are closer to the observed value in focus areas of RMAPS_Chem V1.0 system. The largest forecast deviation areas concentrate in south region of Hebei, west region of Shandong and Beijing, which is consistent with the distribution of SO₂ emissions deviation. Optimizing of the emission inventory brought significant reduction in forecast deviation in these regions, with root mean square error and normalized mean absolute error reduced obviously. The simulation error show normal distribution characteristics. The probability of error distribution range, the maximum range are significantly narrowed, and the biggest error probability value rises significantly, indicating errors are reduced.