Use of machine learning and principal component analysis to retrieve nitrogen dioxide (NO₂) with hyperspectral imagers and reduce noise in spectral fitting

Abstract. Nitrogen dioxide (NO2) is an important trace-gas pollutant and climate agent whose presence also leads to spectral interference in ocean color retrievals. NO2 column densities have been retrieved with satellite UV–Vis spectrometers such as the Ozone Monitoring Instrument (OMI) Tropospheric (TROPOMI) that typically resolutions of order 0.5 nm or better spatial footprints small 3.6 km × 5.6 km. These observations are used estimate emissions, monitor pollution trends, study effects on human health. Here, we investigate whether it possible retrieve amounts lower-spectral-resolution hyperspectral imagers Ocean Color (OCI) will fly Plankton, Aerosol, Cloud, Ecosystem (PACE) set for launch early 2024. OCI a resolution 5 ∼ 1 global coverage 1–2 d. At this resolution, small-scale structure from absorption still present. We use real spectra OMI simulate turn slant (SCDs) artificial neural network (NN) trained target While obtain good results no noise added simulated spectra, find expected instrumental substantially degrades Nevertheless, information may be value retrievals can temporally averaged over timescales months reduce provide higher-spatial-resolution maps useful downscaling lower-spatial-resolution data provided by instruments TROPOMI; could potentially enable higher-resolution emissions estimates other applications. In addition, show NNs coefficients leading modes principal component analysis radiance inputs appear reduction Once trained, speed up fitting algorithms applied OMI, TROPOMI, similar flying soon geostationary orbit.