Quantifying the drivers of ozone formation using localized photochemical reactivity and double machine learning.

Localized photochemically constrained double machine learning reveals a 4.04 μg/m³ increase in surface ozone per 1°C temperature rise in Yuncheng during 2020-2022.

• Why it matters: Understanding the nonlinear interactions among emissions, meteorology, chemistry, and transport is crucial for effective ozone pollution control, especially as precursor emissions decline.
• What they did: The study integrated hourly observations with a box model and developed a localized maximum incremental reactivity scale, then applied a random forest model with SHAP analysis and double machine learning to quantify ozone responses.
• The result: Findings show local precursors and photochemical drivers contribute 44.6% and 26.4% respectively to ozone formation, enabling targeted VOC-NOx control strategies tailored to Yuncheng’s chemical sensitivity.