To stop the spread of COVID-19 (2019 novel coronavirus), China placed lockdown on social activities across China since mid-January 2020. The government actions significantly affected emissions of atmospheric pollutants and unintentionally created a nationwide emission reduction scenario. In order to assess the impacts of COVID-19 on fine particular matter (PM2.5) levels, we developed a "conditional variational autoencoder" (CVAE) algorithm based on the deep learning to discern unsupervised PM2.5 anomalies in Chines cities during the COVID-19 epidemic. We show that the timeline of changes in number of cities with unsupervised PM2.5 anomalies is consistent with the timeline of WHO's response to COVID-19. Using unsupervised PM2.5 anomaly as a time node, we examine changes in PM2.5 before and after the time node to assess the response of PM2.5 to the COVID-19 lockdown. The rate of decrease of PM2.5 around the time node in northern China is 3.5 times faster than southern China, and decreasing PM2.5 levels in southern China is 3.5 times of that in northern China. Results were also compared with anomalous PM2.5 occurring in Chinese's Spring Festival from 2017 to 2019, PM2.5 anomalies during around Chinese New Year in 2020 differ significantly from 2017 to 2019. We demonstrate that this method could be used to detect the response of air quality to sudden changes in social activities.
Keywords: COVID-19; Conditional variational autoencoder; Emission reduction; PM(2.5).
Copyright © 2021 Elsevier B.V. All rights reserved.