Empirical Ozone Isopleths using a Machine Learning-based Computational Framework

Despite significant progress in air quality, surface ozone (O₃) pollution, which is a secondary pollutant, remains a challenge worldwide. The complexity of O₃ control can be attributed to the non-linear response of O₃ levels to alterations in the emission of its primary precursors, namely, nitrogen oxides (NOx) and volatile organic compounds (VOCs). This non-linearity is further compounded by significant variations in different meteorological conditions that regulate both the local formation of O₃ and its long-range transport. Therefore, accurate characterization of the non-linear relationships between O₃ and its precursors is a prerequisite for formulating effective O₃ control measures in a particular area.

In this study, we developed a Machine Learning (ML)-based computational framework to build O₃ isopleth diagrams from measured data on NOx and VOCs concentrations, specifically aromatic hydrocarbons and non-methane hydrocarbons (NMHCs). The framework, inspired by the Empirical Kinetic Modeling Approach (EKMA), integrated an XGBoost model with the Shapley additive explanation (SHAP) method and was implemented using hourly air pollutant and meteorological data, collected, from 2018 to 2022, near an oil pre-treatment plant in Southern Italy.

The results demonstrated the strong predictive performance of the XGBoost models (R² > 0.80), effectively capturing the nonlinear relationships between O₃ and its driving factors. The SHAP analysis revealed the dominant role of NOx and relative humidity in driving O₃ variability. The isopleths were derived by perturbing precursor concentrations while holding other covariates constant at their seasonal average. Different VOC species were analyzed to identify which compounds to target in order to optimize O₃ reduction.

Overall, the high computational efficiency of interpretable ML algorithms makes the developed framework an effective and flexible tool for developing empirical O₃ isopleths under different meteorological scenarios. ML-based isopleths can support the development of O₃ pollution mitigation strategies tailored to specific environmental and meteorological contexts.