Rapid Satellite Nitrogen Dioxide Data Processing Using Machine Learning

While highly spatially and temporally resolved satellite nitrogen dioxide (NO₂) observations from the current and upcoming missions continue to make far-reaching impacts on our understanding of emissions, atmospheric processes, and ecological exposures of nitrogen oxides, increasingly growing data volumes from these satellite sensors pose a challenge for effective and efficient data processing and analysis. Deriving tropospheric NO₂ column amounts from satellite-observed solar backscatter radiances involves a multi-step procedure consisting of 1) a spectral fitting process for NO₂ slant column retrievals (total amount sensed by a satellite instrument along the average photon path between the sun and sensor), 2) calculation of air mass factors for slant-to-vertical column conversion, and 3) a scheme to separate stratospheric and tropospheric NO₂ columns. The first step of retrieving NO₂ slant columns using traditional, state-of-the-art Differential Optical Absorption Spectroscopy (DOAS) method is highly accurate, but the algorithm is complicated, and the computation is slow and expensive. As an alternative to the existing approach, we develop a novel machine learning (ML) approach trained on data from traditional algorithms to rapidly process satellite NO₂ data with high accuracy. Using data from the Ozone Monitoring Instrument (OMI), a sensor making global observations from Low Earth Orbit (LEO), we will demonstrate how NO₂ derived from various ML architectures compares with NO₂ data from two independent retrievals from the OMI Standard Product (OMNO2) and Quality Assurance for Essential Climate Variables (QA4ECV). We will also discuss how the ML approach can help reduce noise in retrievals from traditional algorithms and offer an efficient alternative. The increased efficiency is important considering large data volumes from higher spatial resolution instruments in geostationary orbits such as the US Tropospheric Emissions: Monitoring of Pollution (TEMPO) and Korean Geostationary Environment Monitoring Spectrometer (GEMS) and advanced instruments in LEO such as the European TROPOspheric Monitoring Instrument (TROPOMI) and Sentinel-5.