Skip to content
Open access

OMI-Based Estimation on Ground-Level O3 in U.S. Coastal Metropolitan Areas Using Multiple Observations, Model Simulation, and Machine Learning

2026 · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · Vol 19, pp. 32535-32550 · 0 citations · 82 references

Abstract

Ground-level ozone is a major air pollutant with strong spatial and temporal variability. Although the ozone monitoring instrument (OMI) provides long-term global ozone observations, its direct sensitivity to lower tropospheric O3 is limited. In this study, we developed city-specific light gradient boosting machine (LightGBM) models to improve OMI-based estimates of ground-level O3 over three U.S. coastal metropolitan regions: New York, Baltimore, and Houston. Our models integrate OMI ozone profile retrievals, Tropospheric NO2, HCHO, and UV information with NO3 and VOC emissions, and other data. Warm-season O3 estimates were generated at 1 km resolution for 2010–2019. The resulting products capture distinct regional O3 patterns. Tenfold cross-validation (CV) shows strong agreement with surface observations, with correlation coefficients of 0.906, 0.909, and 0.930 for New York, Baltimore, and Houston, respectively, and RMSE values of 5.276–6.569 ppb. Leave-one-monitor-out CV, in which entire monitoring sites are withheld from model training, yields correlation coefficients of 0.856–0.886 and RMSE values of 6.500–8.262 ppb. Comparison with community multiscale air quality simulations provides an external gridded benchmark and shows that the LightGBM products reproduce major tempo-spatial O3 variability, with correlations of 0.764–0.790 across the three domains. Feature-importance and SHAP analyses indicate that meteorological and temporal variables are the strongest predictors, with patterns consistent with temperature-driven photochemistry in New York and Baltimore and with wind-driven transport and coastal meteorology in Houston. These results demonstrate that multisource data fusion with explainable machine learning can enhance the utility of OMI observations for high-resolution urban ground-level O3 estimation.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.