Exploring Applications of Machine Learning for Satellite Nitrogen Dioxide Slant Column Retrievals

Nitrogen dioxide (NO2) is an atmospheric trace gas harmful to human health at elevated levels and emitted naturally from soils, oceans, and volcanos. Combustion of fossil fuels is the primary anthropogenic source of NO2. NASAs Ozone Monitoring Instrument (OMI) NO2 standard product provides the slant column amount of NO2 retrieved using the Differential Optical Absorption Spectroscopy (DOAS) technique. However, the NO2 slant column retrievals accomplished with this method require complex fitting procedures, are computationally expensive, and are sensitive to noise. Following the application of a Principal Component Analysis (PCA), we investigate an alternative means of retrievals via neural network training on NASAs Version 4 NO2 product to exploit OMIs hyperspectral observations more rapidly in the visible blue-green range (~400-500nm). After training on a sample of standard NO2 retrievals, we apply the new method to independent data and evaluate the performance of the NN retrieval through comparisons with operational OMI NO2 slant columns.