Does COVID-19 lockdown matter for air pollution in the short and long run in China? A machine learning approach to policy evaluation

Abstract

This paper leverages a data-driven two-step approach to effectively evaluate the effects of COVID-19 lockdown on air pollution in both the short and long-term in China. Using air pollution, meteorological conditions, and air mass clusters from 34 air quality monitoring stations in Beijing from 2015 to 2022, this study first employs a deweathering machine learning technique to decouple the confounding effects of meteorological on the air pollution. Furthermore, a detrending percentage change indictor is applied to remove the influence of seasonal variations on air pollution. The findings reveal that: (1) Human interventions are the primary drivers of changes in air pollution concentrations, whereas meteorological factors have a relatively minor impact. (2) During the COVID-19 lockdown, significant variations in air pollution levels are observed, with the effects of city lockdown ranging from a decrease of 40.11% ± 14.81% to an increase of 20.28% ± 14.36%. Notably, there is a decline in concentrations of NO2, PM2.5, CO, and PM10, while the levels of O3 and SO2 increase even during the strictest lockdown period. (3) In the year following the COVID-19 lockdown, there is a rebound in overall air pollution levels. However, by the second year, a general decline in air pollution is observed, except for O3. Therefore, it is imperative to integrate the confounding effects of meteorological factors into air quality management policies under various future scenarios: adopt high-intensity control measures for sudden air quality deteriorations, advance green recovery initiatives for long-term emission reductions, and coordinate efforts to reduce composite atmospheric pollution.