Machine Learning-Based Analysis of Driving Factors for PM_2.5 and O₃ Pollution in Urban Beijing

To address the limitations of conventional models constrained by emission inventory uncertainties and statistical methods in disentangling multi-factor interactions, this study developed a random forest-meteorological normalization model based on hourly observations of PM_2.5, O₃, and meteorological factors in urban Beijing (2015–2023). The model quantifies the nonlinear impacts of emission controls and meteorological variations on these pollutants. Results show that the declining trend in annual PM_2.5 concentrations was driven by emission source restructuring and favorable meteorological conditions, while O₃ levels remained around 185 μg·m^-3 due to combined effects of anthropogenic emission reductions, regional transport, and adverse meteorological conditions (e.g., high temperature and low humidity). Feature importance analysis identified relative humidity (41.6%) and temperature (60.2%) as the dominant factors influencing PM_2.5 and O₃, respectively. High PM_2.5 pollution typically occurred under conditions of 80%–100% relative humidity and temperatures between ?15°C and 0°C, whereas O₃ pollution intensified at 30°C–40°C with 0%–20% humidity. PM_2.5 formation and regional transport were enhanced under northeast-southwest winds (5–7.5 m·s^-1), while O₃ pollution worsened under southwest-southeast winds (0–5 m·s^-1) due to enhanced precursor mixing and transport. We recommend sustained anthropogenic emission reductions in Beijing and strengthened regional joint prevention and control mechanisms, particularly targeting pollution transport along the northeast-southwest urban corridor.