Estimation of Ground Ozone Concentration by Using the Synthetic Data of Geostationary Satellite and Machine Learning Techniques

Ground ozone is the most concerning air pollutant because of its direct threats to public health, agricultural production, etc. It is also the most challenging question for satellite remote sensing community because of the naturally uneven vertical distribution of ozone in the atmosphere. The next generation geostationary satellite, Tropospheric Emission Pollution (TEMPO), is a revolutionary instrument that aims to provide fine spatial resolution air pollutant observations over North America from southern Canada to Mexico every hour at daytime. These fine temporal and spatial resolution observations will improve our understanding of the air quality studies and our air quality prediction capability significantly. In addition to the best in class observations on NOx, SO2, etc., TEMPO will also provide some sensitivities on ground ozone. Before the launch of TEMPO in 2020 approximately, TEMPO science team has provided TEMPO synthetic data for the user community to test TEMPO data in their applications and studies.

Ground ozone concentration is controlled by complex processes including chemical productions, horizontal and vertical transports, emissions and depositions, atmospheric boundary layer processes, etc. Many studies have revealed the relationship between ground ozone concentration and its precursors, meteorological fields and land cover/land use, etc. However, these estimated ground ozone concentration contains large uncertainties. Machine learning techniques have been proved to be capable of finding the unknown relationships in complex systems. With the help of big data and machine learning techniques, ground ozone concentration can be estimated by using TEMPO’s multiple ozone precursor observations, ground ozone sensitivities, and meteorological fields.

We use a new method to estimate ground ozone concentration over North America by using TEMPO satellite synthetic data and machine learning techniques. Machine learning techniques can find the nexus among ground ozone concentrations, TEMPO’s sensitivities on the ground ozone and high accurate measurements of ozone precursors. Our ground ozone concentration can be estimated based on these findings by machine learning.