Atmospheric CO2 retrieval from satellite spectral measurements by a two-step machine learning approach

Abstract In the present study, a two-step machine learning approach is proposed and tested based on the Greenhouse gases Observing SATllite (GOSAT) spectral measurements to efficiently retrieve the atmospheric CO 2 column density. A simple one-dimensional line-by-line forward radiation model is developed to simulate the GOSAT observed spectra, which is used to generate the training data for the two-step machine learning model. By using the two-step machine learning model, the atmospheric spectral optical thickness can be retrieved first, and then the CO 2 column density is retrieved from the optical thickness spectrum later. As a proof-of-concept study, the model has been applied to retrieve CO 2 column density in Australia for clear sky conditions from the GOSAT observed 1.6  μ m spectra. Results have shown that the accuracy of the proposed method for CO 2 column density retrieval is about 3 ppm but the efficiency is found to be excellent. Further improvements for improving the retrieval accuracy are discussed.