Air pollution sensing to urban air quality insight: review of spatially interpolated data products from sensor networks

Air pollution adversely impacts health globally, disproportionately affecting urban populations. Regulatory monitoring is often too spatially coarse to capture true intra-urban variability, needed to prioritize and guide localized interventions and policies. High-density lower-cost sensor (LCS) networks, mobile monitoring, satellite-derived products, and advanced interpolation techniques are emerging to support higher spatial resolution characterization. However, results from related studies are varied, disconnected across different fields, and as yet with no defined standards or guidance. This review examines the literature from 2005-2025 to systematically compare data collection and modeling pipelines for accuracy and application to the intra-urban challenge. 166 studies from regional to neighborhood scale were analyzed quantitatively on monitoring network characteristics (geographic location, spatial density, target pollutants, study duration) and interpolation modeling approaches (algorithms, predictive covariates, performance metrics). Studies from Europe, Asia, and North America dominated the literature, with particulate matter being the most studied pollutant. While 95% of studies had measurement density below 20,000 km2/device (metro scale), none had sufficient resolution for capturing intra-urban heterogeneity. Modeling approaches shifted over time from primarily geostatistical methods toward increasingly sophisticated machine learning and deep learning techniques incorporating diverse covariates, including meteorology, land use, traffic patterns, and satellite data products. More than measurement density or model complexity, strategic use of both static and dynamic covariates in data driven models was most consistently associated with performance improvement for interpolation of both PM2.5 and NOx. However, major gaps remain including (1) imbalance in data availability geographically, temporally, and across pollutant species, leading to major challenges in assessing and achieving transferability of models globally, (2) limited data availability at appropriate intra-urban scale for most metro areas, (3) lack of evaluation and reporting standards necessary to truly compare performance of different data-model pipelines, and (4) general lack of generating or reporting uncertainty quantification on model outputs, critically needed for interpretability and sensitivity analyses. These gaps point to an urgent need for not only interdisciplinary collaboration and innovation but also discipline-wide guidance that will enable objective assessment of model performance and state-of-the-art, including shared benchmark datasets, standardized evaluation metrics, mandatory unit reporting, and guidelines for complete model specifications.