Ozone Variability and Precursor Sensitivity in Urban Taichung: Insights from Observation-Based Machine Learning and Chemical Mechanism Modeling

Ground-level ozone (O 3 ) remains a persistent air-quality challenge in rapidly urbanizing East Asian cities, driven by nonlinear interactions among nitrogen oxides (NO x ), volatile organic compounds (VOCs), and meteorology. In this study, we investigated O 3 ​ variability and precursor sensitivity in Taichung, Taiwan – a representative East-Asian urban area – using year-long hourly observations from 2023 combined with a Random Forest (RF) model and chemistry-based simulations with the Master Chemical Mechanism (MCM). The RF model reproduced observed O 3 ​ variability well (R 2 = 0.90, RMSE=5.53 ppbv), and model performance improved substantially when NO and NO 2 ​ were treated as separate predictors rather than as combined NO x ​, highlighting the importance of resolving the O 3 –NO–NO 2 ​ relationship in data-driven prediction. O 3 -NO x -VOCs sensitivity analysis based on the optimal RF model showed that O 3 formation in Taichung was predominantly VOC-limited, although some conditions approached a transitional regime. However, the response of O 3 ​ to changes in total VOC–OH reactivity (TVOCR) was relatively weak, implying that O 3 reductions from VOC controls may occur gradually. MCM simulations provided mechanistic support for these results by showing that daytime O 3 ​ production was dominated by the HO 2 ​ + NO pathway, while alkene ozonolysis contributed significantly to O 3 ​ loss and partly offset VOCs-driven O 3 production. These results indicate that urban ozone pollution in Taichung is governed by coupled effects of VOCs availability, O 3 ​-NO titration and radical termination, and demonstrate that integrating machine learning with explicit chemical mechanism modeling is a useful framework for interpreting urban ozone sensitivity and informing precursor-control strategies.