Assessment and spatial characterization of ground-level ozone exposure using a low-cost sensor network

Ground-level ozone (O₃) remains a pervasive secondary air pollutant of public health concern, particularly in regions with complex emission profiles and limited monitoring infrastructure. This study investigated the hyperlocal spatial variability of ambient O₃ in the Greater Springfield area, Massachusetts using a calibrated network of 13 metal oxide semiconductor-based low-cost sensors (LCS). A machine learning-based calibration model was developed using random forest regression trained on a one-year dataset, which demonstrated robust predictive performance for O₃ estimation (R² = 0.82) relying solely on sensor-derived variables. The model was influenced by the local chemical regime, reflecting the VOC-limited environment in the study region. Analysis of the calibrated sensor measurements revealed that 37.7% of the observed O₃ variance was attributable to regional-scale photochemical processes, while localized sources, such as industrial activities and urban traffic, accounted for an additional 22.3%. The spatial patterns exhibited coherence and alignment with known emission sources and land use features. This study demonstrates the potential of leveraging LCS networks to capture spatially resolved dynamics of O₃. Operation of these networks in resource-limited communities will facilitate equitable access to air quality information by populations with enhanced vulnerability to the associated health impacts.