Machine learning unveils the impact of anthropogenic emission changes on urban PM_2.5 and O₃: A case study in Wuhu

PM_2.5 and O₃ are two major issues hindering air quality improvement in China. However, the response of these two pollutants to anthropogenic emission variations in the real atmosphere was not yet well understood. Here, we selected the short-term epidemic lockdown in Wuhu in 2022 as a case study and evaluated the impacts of meteorology and anthropogenic emission on PM_2.5 and O₃ using field observations combined with machine learning algorithms. The results showed that NO₂ observed during the lockdown was 32.2 ± 8.1 µg/m³, 10.1 % lower than before the lockdown, and that NO₂ continued to decrease by 19.2 % after the lockdown. Notably, both PM_2.5 and O₃ concentrations were higher during the lockdown than before and after the lockdown. Random forest model revealed that meteorological conditions during the lockdown increased PM_2.5 and O₃ by 8.7 % and 24.2 %, respectively, but decreased NO₂ by 6.4 %. Atmospheric pressure and relative humidity were the main meteorological variables influencing PM_2.5 and O₃ variations, respectively. Scenario simulation analysis uncovered that anthropogenic emission reduction caused by the lockdown reduced NO₂ by 19.7 %, but increased PM_2.5 and O₃ by 6.3 % and 26.8 %, respectively. This was mainly due to the weakening titration effect of nitrogen oxides and enhanced atmospheric oxidation capacity, further increasing O₃ and secondary PM_2.5 production. Our results revealed that NO₂ in Wuhu decreased significantly due to short-term epidemic lockdown, but PM_2.5 and O₃ pollution were not effectively reduced. To continuously improve future urban air quality, joint reductions in emissions from multiple anthropogenic sources and multiple pollutants are required.