Comparing different approaches for assessing the impact of COVID-19 lockdown on urban air quality in Reading, UK

Abstract

Many studies investigated the impact of COVID-19 lockdown on urban air quality, but their adopted approaches have varied and there is no consensus as to which approach should be used. In this paper we compare three of the main approaches and assess their performance using both estimated and measured data from several air quality monitoring stations (AQMS) in Reading, Berkshire UK. The approaches are: (1) Sequential approach – comparing pre-lockdown and lockdown periods 2020; (2) Parallel approach – comparing 2019 and 2020 for the equivalent time of the lockdown period; and (3) Machine learning modelling approach – predicting pollution levels for the lockdown period using business as usual (BAU) scenario and comparing with the observations. The parallel and machine learning approaches resulted in relative higher reductions and both showed strong correlation (0.97) and less error with each other. The sequential approach showed less reduction in NO and NOx, showed positive gain in PM10 and NO2 at most of the sites and demonstrated weak correlation with the other two approaches, and is not recommended for such analysis. Overall, the sequential approach showed −14, +4, −32, and + 56% change, the parallel approach showed −46, −43, −43 and + 7% change, and the machine learning approach showed −47, −44, −38 and + 5% change in NOx, NO2, NO and PM10 concentrations, respectively. The pollution roses demonstrated that the UK received easterly polluted winds from the central and eastern Europe, promoting secondary particulates and O3 formation during the lockdown. Changes in pollutant concentrations vary both in space and time according to the approach used, environment type of the monitoring site and the data type (e.g., deweathered vs. raw data). Therefore, the reported results (here or elsewhere) should be viewed in light of these factors before making any conclusion.