Data fusion in air pollution exposure assessment: Methods, applications, and future directions

Over the past two decades, major advances in observational techniques and modeling capabilities have enabled the development of air pollution data fusion methods that integrate information from regulatory monitoring networks, low-cost sensors, satellite remote sensing, and chemical transport models (CTMs). These approaches combine measurements from regulatory monitoring networks with complementary data sources, including satellite remote sensing, chemical transport models, low-cost sensor networks, and ancillary geospatial predictors, to overcome the spatial limitations of individual datasets. This review synthesizes the evolution of air pollution data fusion methodologies and their applications in exposure assessment and environmental health research. After reviewing the fusion framework and commonly used data sources, we describe five major methodological families that have emerged in the literature: data assimilation, deterministic regression methods, Bayesian statistical models, machine learning techniques, and hybrid or ensemble frameworks that combine multiple modeling strategies. Subsequent sections discuss how these methodological developments have been implemented in practice using an in-depth discussion of representative studies in each of the methodological families and important downstream applications. While fused products should be interpreted with appropriate attention to uncertainty, they provide valuable tools for exposure assessment, air quality analysis, and regulatory planning, particularly in regions where monitoring infrastructure remains sparse. Continued methodological innovation and integration of emerging observation platforms are expected to further enhance the accuracy and applicability of fused air pollution datasets.Implications: Data fusion is now an integral component of air pollution exposure assessment, enabling spatially and temporally resolved concentration estimates that enhance epidemiologic research, policy evaluation, and environmental justice assessments. By integrating ground observations, satellite data, and model simulations, these approaches improve exposure characterization beyond the limitations of individual data sources. Ongoing advances in model transparency, interpretability, and uncertainty characterization will be essential to increase the reliability and applicability of fused products. When applied appropriately, data fusion can support more robust, evidence-based decision-making and contribute to improved public health protection.