Hybrid Physical–Statistical Reanalysis of Urban PM and NO₂ for High-Resolution Exposure Assessment in Epidemiological Studies

High-resolution exposure data for particulate matter (PM) are a critical determinant of accuracy in environmental epidemiology and urban health impact assessments. However, conventional chemical transport models (CTMs) are limited in representing fine-scale spatial variability of PM at the urban scale, while purely statistical approaches often struggle to maintain physical consistency and interpretability over long-term time series.

In this study, we developed a hybrid physical–statistical reanalysis framework to construct high-resolution exposure datasets for urban PM (PM₂.₅ and PM₁₀), with complementary treatment of traffic-related NO₂, suitable for national-scale health impact studies. The proposed framework consists of three main components. First, a regional CTM (CMAQ, 9 km resolution) was used to generate national-scale background concentrations of PM and gaseous pollutants, along with meteorological reanalysis data, for the period 2013–2024. Second, physically based dispersion patterns derived from the CALPUFF model were applied to redistribute primary PM concentrations to a 100 m grid through a hybrid downscaling approach, enhancing the representation of intra-urban spatial gradients. Third, to improve the temporal accuracy of traffic-sensitive NO₂, an auxiliary XGBoost-based error correction layer was implemented to reduce model uncertainty while preserving the underlying physical structure.

The framework was applied to seven major metropolitan areas and key industrial and traffic-influenced regions in South Korea. The results demonstrate that the hybrid reanalysis effectively captures urban PM concentration gradients and roadside pollution hotspots, yielding substantial improvements over conventional coarse-resolution CTM outputs. The final exposure datasets were integrated with national health cohort data, providing multiple exposure metrics including short-term lagged PM exposures and medium- to long-term moving-average indicators.

By combining high predictive performance with physical consistency, this hybrid approach offers a robust alternative to purely data-driven downscaling methods for PM exposure assessment. The resulting high-resolution PM exposure datasets enable precision environmental health studies at both community and individual levels and provide a scientific basis for evidence-based urban and national environmental health policy development.

 

Acknowledgement

This research was supported by the Korea National Institute of Health (KNIH) research project (Project No. 2024-ER0606-01) and the Particulate Matter Management Specialized Graduate Program through the Korea Environmental Industry & Technology Institute (KEITI), funded by the Ministry of Environment (MOE).