A machine learning model to estimate ground ozone concentration in California, using TROPOMI Satellite Data

Abstract:

Exposure to the ground-level ozone can trigger a variety of health problems as well as ecological impacts. To date, few studies have used satellite-based statistical models to estimate ground-level ozone concentration as current satellite ozone column products lack sensitivity in the boundary layer. The Troposphere Monitoring Instrument (TROPOMI) aboard the Sentinel 5 Precursor satellite can provide high quality and relatively high-resolution trace gas retrievals worldwide. To evaluate TROPOMI’s feasibility to estimate ground-level ozone, we developed a machine learning model to estimate the daily maximum 8-hour average ground-level ozone concentration at 10 km spatial resolution in California from May 2018 to April 2019. In addition to satellite parameter, we also included meteorological fields from the High-Resolution Rapid Refresh (HRRR) system and land-use information as predictors. To estimate the boundary layer ozone abundance, we applied the Ozone Monitoring Instrument (OMI) ozone profile to derive the boundary fraction of the TROPOMI total ozone column. Our model achieved an overall 10-fold cross-validation (CV) R² of 0.84 with root mean square error (RMSE) of 5.91 ppb, indicating a good fit between model predictions and observations. Our model indicated high ozone pollution in South California, and the eastern side of the Central Valley. The suburban areas of the Los Angeles Metropolitan areas see the highest ozone levels, while the Bay Area has the lowest. We also observed low ozone concentration by the coast of the Pacific Ocean. The introduction of the TROPOMI data product in the model improves the estimate of the extreme value of ground ozone concentration compared to the model without TROPOMI. Our model accomplished a good prediction on the ground-level ozone concentration in California, supporting the feasibility and advantage of application TROPOMI satellite product and machine learning method in the prediction of ground-level ozone concentration.