A concept-drift-driven hybrid modeling framework for high-resolution ambient surface NO₂ retrieval in...

Long-term accurate exposure assessment of ambient air pollutants is critical for environmental management and public health protection, particularly in heavy-polluted and densely-populated regions such as Beijing-Tianjin-Hebei region (BTH). Despite severe nitrogen dioxide (NO₂) pollution in BTH, current high-resolution retrieval frameworks commonly overlooked concept drift (CD), i.e., temporal heterogeneity pervasively exists in data distributions and pollutant-covariate-associations, inducing substantial estimation bias, or arbitrarily partitioned datasets, still yielding inaccurate retrievals. In this study, we developed a concept-drift-driven hybrid modeling framework, CD-X, which delivered the temporally longest and most accurate daily 1-km seamless surface NO₂ retrievals for BTH during 2014-2022, reducing estimation bias by 35.06 % (to 6.95μg/m³) and increasing 10-fold cross-validation R² by 14.25 % (to 89.54 %) compared to the conventional static model. The retrieved dataset reproduced long-term trajectory of NO₂ levels, and revealed a persistent central-southern pollution hotspot and cleaner areas in the north and BTH-Shandong-Henan adjacent coast. These findings provide guidance for targeted "source-transmit-deposit" mitigation and inform dynamic emission scheduling, industrial timing and trans-boundary control. This study bridges gaps in long-term exposure assessment by explicitly accounting for concept drift, and CD-X offers a transferable methodological paradigm for exposure assessment in urban agglomerations facing similar temporal heterogeneity challenges, thereby contributing to more equitable air quality assessment and environmental justice.