Estimating surface sulfur dioxide concentrations from satellite data over eastern China: Using chemical transport models vs. machine learning

Abstract

Sulfur dioxide (SO2) is an important air pollutant that contributes to negative health effects, acid rain, and aerosol formation and growth. SO2 has been measured using ground-based air quality monitoring networks, but routine monitoring sites are predominantly in urban areas, leaving large gaps in the network in less populated locations. Previous studies have used chemical transport models (CTMs) or machine learning (ML) techniques to estimate surface SO2 concentrations from satellite vertical column densities, but their performance has never been directly compared. In this study, we estimated surface SO2 concentrations using Ozone Monitoring Instrument (OMI) retrievals over eastern China from 2015–2018 utilizing GEOS-Chem CTM simulations and an extreme gradient boosting ML model. For the first time, we quantified methodological uncertainties for both methods and directly compared their performance on the same truth dataset. The surface concentrations estimated from the CTM-based method had similar spatial distributions (r = 0.58) and temporal variations compared to the in situ measurements but were underestimated (slope = 0.24; RPE = 75 %) and had worsening performance over time. The ML-based method produced more accurate spatial distributions (r = 0.77) and temporal variations with a smaller discrepancy (slope = 0.69; RPE = 30 %) and stable performance over time. Despite the higher accuracy of the ML-based method at the monitoring sites, the CTM-based method produced more reasonable gridded spatial distributions over areas without monitoring data. These results suggest that satellite data could be a reliable way to estimate global SO2 concentrations to parameterize other chemical processes in the atmosphere.