Urban air quality monitoring using ground-based hyperspectral imaging of vegetation
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
University of Delaware
Abstract
Urban ambient air pollution is not uniformly distributed due to multiple factors that include the range of built and natural surfaces, the diversity and distribution of built structures, the complex interactions between pollutants, and the presence of numerous pollution sources. With the majority (55%) of the global population currently residing in urban areas, and on track to reach 68% by 2050, there is a need to characterize urban air quality at high spatio-temporal granularity accurately. This can assist in identifying the true impact of air pollution at the community level, locating the sources of pollution, and informing decisions about improving air quality in areas in which it is needed the most. Traditional air quality monitoring networks are generally too sparse and thus incapable of capturing the true distribution of ambient air pollutants in urban environments. Therefore, this dissertation presents a new method for the hyperlocal monitoring of ambient air quality that exploits the abundant presence of urban vegetation, the impact of air quality on their health, and the ability to identify changes in their health using remote, ground-based, hyperspectral imaging. ☐ First, an investigation into the processes and variables that may have influenced the largest cities in the US to adopt and maintain neighborhood-level hyperlocal air quality monitoring programs (HAMPs) over the past decade is presented using policy innovation and diffusion models. This study, together with an evaluation of the longest running HAMP with the most geographically comprehensive network of monitors, provides the motivations and need for the development of a new solution for the hyperlocal monitoring of urban air quality. Machine learning models capable of classifying the materials observed on a pixel-level in ground-based hyperspectral images of urban scenes are developed and tested for their ability and limits to automatically identify vegetation pixels. Frameworks for the atmospheric correction of radiance spectra are presented to allow for the accurate extraction of surface reflectance profiles of vegetation. Furthermore, an investigation of the impact of changes in air quality on the health of urban vegetation, and the ability to identify such changes in the observed vegetation spectra, is provided. Finally, an investigation into the capability of a deep machine learning model at predicting air quality parameters from the reflectance spectra of vegetation, identified and atmospherically corrected using the AI-driven solutions explored in the previous chapters, is presented. Combining these methods and frameworks illustrates the feasibility of developing an automated solution to extract air quality parameters from their impact on the reflectance spectra of vegetation across large swaths of urban areas with high spatiotemporal granularity using a single instrument.
Description
Keywords
Urban air quality, Hyperlocal air quality monitoring programs, Hyperspectral imaging, Vegetation