Machine Learning-Based Bias-Corrected Future Projections of Ozone Concentrations from a Chemistry-Climate Model

Reliable projections of future surface ozone are crucial for air quality management and health risk assessment. However, potential biases in spatial distribution, magnitude, and trends in ozone simulated by global chemistry-climate models limit their applicability in regional evaluations. In this study, LightGBM, a machine learning (ML) algorithm, is applied to correct biases in CESM2-simulated ozone over China, the United States, and Europe and to calibrate future projections under two Shared Socioeconomic Pathways (SSP1-2.6 and SSP5-8.5) from 2020 to 2060. The ML-based correction significantly improves spatial distribution and reduces bias by 40 to 60%, also reversing the potentially incorrect trend under SSP1-2.6 in eastern China. When the ML-based correction is applied to CESM2 projections, the warm-season mean ozone shows substantial changes from 2020 to 2060. Under SSP1-2.6, corrected ozone decreases by 13.5, 17.9, and 13.7 μg/m³ in China, the United States, and Europe, respectively. In contrast, under SSP5-8.5, ozone increases over the same period by 9.4, 2.0, and 5.2 μg/m³ in these regions. The decomposition analysis shows that anthropogenic emission changes dominate future ozone trends, while a strong climate penalty occurs in polluted eastern China and climate benefits are found in western China, the United States, and Europe under SSP5-8.5. These findings demonstrate the value of combining ML with chemistry-climate models to produce more accurate air quality projections, indicating more effective and region-specific environmental protection strategies.