Improved fine-scale change detection of ambient NO₂ during COVID-19 lockdowns : satellite-based estimation approach

Abstract

Environmental quality is of paramount importance in contemporary society. The deteriorating air quality during periods of urbanization and industrialization has prompted collective management efforts to pursue clean air for the sake of public health. The global crisis of COVID-19, often seen as a double-edged sword for global society, unexpectedly resulted in improvements in air quality. While an abundance of literature has documented changes in various air pollutants during lockdowns, the available results are limited to short-term observations, and the quality of air quality information is questionable. The main argument is that neither the sparse ground-level air monitoring network nor the satellite observations from space can provide accurate estimations of air pollution concentration that covers an entire city. These data flaws can lead to inaccurate evaluations of the effectiveness of interventions on air quality, thereby affecting the value of the estimated results for the public and policymakers. Furthermore, the lack of comprehensive data has limited the scope and depth of research. The data deficiency can be addressed through coordinated cross-disciplinary collaborations. The remarkable advancements in artificial intelligence and remote sensing applications in environmental science are noteworthy. Machine learning and deep learning approaches, with their unparalleled capacity for feature extraction and representation, have expanded our ability to model ground-level air pollution over long periods and at fine scales. This thesis conducts air pollution modeling and empirical analysis in the Beijing-Tianjin-Hebei (BTH) region, China, and England, the United Kingdom (UK), to explore the dynamics of NO2 during the COVID-19 pandemic. Fine-scale ground-level NO2 data derived from machine learning techniques enable us to quantify the impact of lockdown measures at the district level in the BTH regions. Furthermore, the effectiveness of deep learning methods in estimating ground-level NO2 concentrations was examined in England, particularly in the face of significant data gaps in satellite imagery. Additionally, the disparities in the changes in NO2 exposure among different ethnic groups were further examined. The findings suggest that modeling ground-level NO2 concentrations can bridge the gap between surface monitoring data and satellite column density observations, providing spatially continuous and accurate information regarding NO2 dynamics at a fine scale. Specifically, districts in the BTH region with different characteristics experienced varying magnitudes of reduction in NO2 during the lockdown period. The presence of spatial spillover effects highlights the importance of collaborative efforts in NO2 management. Furthermore, the lockdown measures led to a reduction in ethnic disparities in NO2 exposure in England. Ethnic minorities experienced a greater decrease in NO2 concentrations compared to the white population, even after accounting for mobility and other socioeconomic factors. The findings presented in this thesis contribute significantly to our understanding of the impact of lockdown measures on local-scale NO2 air quality. They emphasize the importance of having spatially resolved air quality information for informing policy decisions and conducting health-related research. Additionally, these findings will assist regulators and the public in identifying opportunities for individual behavioral changes that can lead to improved air quality and environmental sustainability.