Four-Decade (1980-2023) Surface Ozone Concentrations across the Contiguous United States: Fine-Resolution Estimates...

Exposure to ambient ozone (O₃) poses significant acute and chronic health risks, but comprehensive evaluation of the O₃-related health burden is limited by the lack of high-resolution historical surface O₃ data. To address this gap, we developed a Light Gradient Boosting Machine modeling framework incorporating geophysical a priori information from numerical simulations with novel k-fold nearest neighbor distance matching cross-validation to account for geographical distribution patterns in training and prediction data. The models achieved strong predictive performance with cross-validation R² values of 0.64-0.80, 0.74-0.90, and 0.53-0.94 against observations on daily, monthly, and warm-season annual average scales, respectively. Using this framework, we generated daily maximum 8-h average O₃ concentrations at 0.01° (approximately 1 km) resolution across the contiguous United States from 1980 to 2023. Our analysis revealed distinct spatial patterns: rural areas have higher average exposure levels, while urban areas experience more days with high values (>70 ppb) and exhibit slower decreasing trends across most regions. Despite overall decreased exposure levels, attributable health burdens from both short- and long-term exposures increased by 25 and 18%, respectively, from 1990 to 2023, driven primarily by population aging and growth. The resulting 44-year data set represents the first comprehensive ambient O₃ reconstruction at this spatiotemporal resolution, enabling in-depth life-course epidemiological studies and supporting evidence-based air quality policy decisions.