Mapping Air Pollution Concentrations in Greater Boston Using Hyperlocal Mobile Sensing and Machine Learning

ABIR SAHA, Mars Keesey, Yang Zhang, Shang Liu, Northeastern University

     Abstract Number: 301
     Working Group: Urban Aerosols

Abstract
Although many regions in the United States have met the National Ambient Air Quality Standards (NAAQS), the assessment of attainment is based on sparsely located air quality networks. Localized pollution still exists and varies spatially, especially experienced by disadvantaged communities. The growing body of low-cost sensor deployments has greatly improved our understanding of the spatial patterns of air pollution, but they suffer from measurement uncertainties and still lack coverage at the neighborhood level. To overcome this limitation, we have built a mobile laboratory equipped with advanced real-time instruments that are capable of accurately measuring gas- and particulate-phase air pollutants (i.e., O₃, NO, NO₂, NOx, SO₂, CO, benzene, toluene, ethylbenzene, xylene, PM1, PM2.5, PM10, PM composition, and size distribution) and greenhouse gases (CO₂, CH₄). We will conduct an intensive, 30-day mobile measurement field campaign in Greater Boston, Massachusetts, in the summer of 2025, focusing on two neighborhoods: Brookline, an affluent neighborhood, and Chelsea, a low-income neighborhood. Geospatial analysis will be performed to identify air pollution hotspots, and comparable analysis will be conducted between the two communities. Positive Matrix Factorization (PMF) will be applied to the mass spectrum measured by the High-Resolution Time-of-Flight Aerosol Mass Spectrometer (HR-ToF-AMS) for spatially resolved source apportionment. Expected PMF components include hydrocarbon-like organic aerosol, secondary organic aerosol, cooking organic aerosol, and others. By combining hyperlocal chemically resolved concentration measurements and source apportionment with real-time street view imagery, traffic density, and Google point of interest data, we will develop machine learning-based high-resolution land use regression models. These models will aid in predicting air pollution concentrations in these neighborhoods when measurement data is not available. The integration of chemical speciation, mobile sensing, and machine learning represents an emerging framework to tackle air pollution at street scales.