Evaluating the effects of meteorology and emission changes on ozone in different regions over China based on machine learning

Abstract

Systematically understanding the impact of meteorological conditions on regional ozone pollution helps to retrieve ozone dataset and evaluate emission changes on ozone variation. Here, more air-quality observation sites were collected, and Random Forest algorithm was applied to retrieve daily maximum 8 h average (MDA8) O3 concentrations at 1 km resolution in 2019-2020 in China. The region-season model and whole-retrieved model were established to compare the contributions of meteorological variables to ozone variations on spatiotemporal scale. The former model outperformed the latter for retrieval capability, but the predictive ability of the latter was slight stronger. This may be associated with the spatiotemporal heterogeneity of meteorological influence. Daily ozone variability in China was mainly influenced by meteorology, especially in North China Plain in autumn. The key factors were temperature, ultraviolet radiation and relative humidity over the whole country, but varied significantly in different regions and seasons. The effects of meteorology and emission sources on ozone were separated by weather normalization technique. Meteorological conditions were particularly favorable for increasing ozone concentrations in spring, but particularly unfavorable in winter. During the COVID-19 lockdown, the increases in ozone in spring and winter of 2020 were the contribution of the combination of meteorology and emissions; while the decreases in ozone in summer and autumn of 2020 were mainly due to the changes of emission sources, although meteorological conditions were unfavorable to ozone mitigation in heavily polluted areas. Our findings provide scientific basis for the prevention and control of ozone pollution for the regional scale in China.