During 2014-2017, the number of haze days and air pollution days declined year by year obviously in Beijing. The average mass concentrations of PM2.5, PM10, SO2, and NO2 also decreased with the alleviated pollution level. These decreases were more obvious during the heating period, especially in November and December. In order to analyze the reasons for the improvement of air quality, changes of the meteorological factors and emission-reduction have been discussed and quantified in this study. This work was based on the numerical simulation model WRF-CHEM and the large data mining technologies of k-nearest neighbor (KNN) and support vector machines (SVM). Meteorological observations indicated that the mean wind speed of 2017 increased by 7.9% compared with the last three years. The frequency of hourly wind speed higher than 3.4 m·s-1 was the highest (10.6%), and frequency of daily relative humidity higher than 70% was lowest (25.1%), in 2017. Meanwhile, the number of low wind days (daily wind speed lower than 2 m·s-1), environmental capacity, ventilation index, and height of the boundary layer showed that the diffusion conditions were better in the heating period of 2017 than those of 2014~2016, especially in November and December. The accumulated precipitation during the non-heating period was 558.3 mm in 2017, which is conducive to pollutant removal and wet deposition. Inter-annual changes of meteorological conditions are important to the air quality. A simulation for December 1~19 by WRF-CHEM during 2014-2017 was performed, and the results demonstrated that changes of meteorological conditions led to a reduction of the PM2.5 concentration of 2017 by 5%, 38%, and 25% compared with that of 2014-2016, respectively. However, it was not possible to quantify the specific contributions of meteorology conditions because of the lack of real emission reduction options. The KNN and SVM models are applied in this study based on the observed meteorology factors, haze days, and pollution days, and it was found that for the reduced haze days and heavy pollution days in 2017, 65.0% could be attributed to emission reduction and 35.0% was caused by improvement of the meteorological conditions.
Keywords: K-nearest neighbor(KNN); air pollution; contribution; emission; meteorological conditions; support vector machines(SVM).