A Machine Learning--Based Surrogate EKMA Framework for Diagnosing Urban Ozone Formation Regimes: Evidence from Los Angeles

TL;DR

This study develops a machine learning–based surrogate EKMA framework to diagnose urban ozone formation regimes using Los Angeles observations from 2024–2025. A random forest surrogate, trained on hourly O$_3$ and precursors with cyclic temporal encodings, achieves $R^2=0.857$ and $RMSE=0.006$, and is used to perform EKMA-like sensitivity analyses by scaling NO$_2$ and CO. Results indicate a predominantly VOC-limited regime under observed conditions, suggesting VOC emissions controls are more effective for reducing peak ozone than NO$_x$ reductions alone. The framework offers an efficient, data-driven alternative to chemical transport models for regime diagnosis and policy support, with potential for ongoing updating as new data become available.

Abstract

Surface ozone pollution remains a persistent challenge in many metropolitan regions worldwide, as the nonlinear dependence of ozone formation on nitrogen oxides and volatile organic compounds (VOCs) complicates the design of effective emission control strategies. While chemical transport models provide mechanistic insights, they rely on detailed emission inventories and are computationally expensive. This study develops a machine learning--based surrogate framework inspired by the Empirical Kinetic Modeling Approach (EKMA). Using hourly air quality observations from Los Angeles during 2024--2025, a random forest model is trained to predict surface ozone concentrations based on precursor measurements and spatiotemporal features, including site location and cyclic time encodings. The model achieves strong predictive performance, with permutation importance highlighting the dominant roles of diurnal temporal features and nitrogen dioxide, along with additional contributions from carbon monoxide. Building on the trained surrogate, EKMA-style sensitivity experiments are conducted by perturbing precursor concentrations while holding other covariates fixed. The results indicate that ozone formation in Los Angeles during the study period is predominantly VOC-limited. Overall, the proposed framework offers an efficient and interpretable approach for ozone regime diagnosis in data-rich urban environments.

Figures (8)

• Figure 1: Monthly mean O$_3$ in Los Angeles during 2024--2025. Ozone concentrations exhibit pronounced seasonality, with elevated levels in late spring and summer and substantially lower concentrations during winter months.
• Figure 2: Monthly climatology of O$_3$ in Los Angeles aggregated across 2024--2025. The solid line indicates the mean concentration, while the shaded band represents the interquartile range, highlighting increased variability during the warm season.
• Figure 3: Mean diurnal cycle of O$_3$ by season (DJF, MAM, JJA, SON) in Los Angeles during 2024--2025. Ozone concentrations increase rapidly after sunrise, peak in the early afternoon, and decline during the evening, with the largest diurnal amplitude observed in summer (JJA).
• Figure 4: Diurnal ozone cycles averaged over weekdays and weekends in Los Angeles during 2024--2025. Daytime ozone concentrations are systematically higher on weekends, consistent with the ozone weekend effect and indicative of nonlinear ozone--precursor interactions.
• Figure 5: Observed versus predicted surface ozone (O$_3$) for the random forest surrogate on the 2025 test set. The red line indicates the 1:1 reference. Reported metrics are $R^2=0.857$ and $\mathrm{RMSE}=0.006$.