Meteorology-normalized variations of air quality during the COVID-19 lockdown in three Chinese megacities

Abstract

To avoid the spread of COVID-19, China implemented strict prevention and control measures, resulting in dramatic variations in air quality. Here, we applied a machine learning algorithm (random forest model) to eliminate meteorological effects and characterize the high-resolution variation characteristics of air quality induced by COVID-19 in Beijing, Wuhan, and Urumqi. Our RF model estimates showed that the highest decrease in deweathered PM2.5 in Wuhan (−43.6%) and Beijing (−14.0%) was at traffic stations during lockdown period (February 1- March 15, 2020), while it was at industry stations in Urumqi (−54.2%). Deweathered NO2 decreased significantly in each city (∼30%–50%), whereas accompanied by a notable increase in O3. The diurnal patterns show that the morning peaks of traffic-related NO2 and CO almost disappeared. Additionally, our results suggested that meteorological effects offset some of the reduction in pollutant concentrations. Adverse meteorological conditions played a leading role in the variation in PM2.5 concentration in Beijing, which contributed to +33.5%. The true effect of lockdown reduced the PM2.5 concentrations in Wuhan, Beijing, and Urumqi by approximately 14.6%, 17.0%, and 34.0%, respectively. In summary, lockdown is the most important driver of the decline in pollutant concentrations, but the reduction of SO2 and CO is limited and they are mainly influenced by changing trends. This study provides insights into quantifying variations in air quality due to the lockdown by considering meteorological variability, which varies greatly from city to city, and provides a reference for changes in city scale pollutant concentrations during the lockdown.