A machine learning based line-by-line absorption coefficient model for the application of atmospheric carbon dioxide remote sensing. (February 2023)

• Record Type: Journal Article
• Authors: Xie, Fengxin; Ren, Tao; Zhao, Ziqing; Zhao, Changying
• Is Part Of: Journal of quantitative spectroscopy & radiative transfer. Volume 296(2023)
• Publication Date: 2023-02
• Subjects: Line-by-line -- Absorption coefficient -- Machine learning -- OCO-2 -- GOSAT; Spectrum analysis -- Periodicals; Radiation -- Periodicals; Analyse spectrale -- Périodiques; Rayonnement -- Périodiques; Radiation; Spectrum analysis; Periodicals; 543.0858
• DOI: 10.1016/j.jqsrt.2022.108441
• Languages: English
• ISSNs: 0022-4073

Abstract

Highlights: A machine learning based LBL absorption coefficient model is proposed. The LBL absorption coefficient model is efficient and accurate. The model has been tested by simulating the GOSAT observed spectra. Abstract: Space-based remote sensing, which is used to infer CO 2 concentration from the satellite-measured atmospheric spectral absorption signals, is an effective way to obtain CO 2 concentration data for greenhouse gas monitoring. The next-generation greenhouse gases monitoring satellites mainly address the challenge of improving the spatial and temporal resolutions of observations, which will dramatically increase the computational power required for the CO 2 retrievals. One of the bottlenecks is the high computational cost for the radiative heat transfer calculations, which involve inefficient high-resolution (usually requires a line-by-line spectral resolution) spectral modeling. Therefore, developing a fast and accurate spectral modeling method becomes necessary to tackle this problem. In the present study, we presented a machine learning based line-by-line absorption coefficient calculation (prediction) method for CO 2 in the applications of atmospheric remote sensing. By training an artificial neural network with data randomly generated from a line-by-line CO 2 absorption coefficient look-up table, a compact, accurate and efficient absorption coefficient prediction model can be developed, which only takes the CO 2 thermodynamic states as input. The proposed method has been tested by developing an absorption coefficient prediction model for the CO 2 1.6 μ m spectral band, which was later used to simulate the measured spectra for clear-sky conditions from the Greenhouse gases Observing SATellite (GOSAT) for several different locations around the world. Results have shown that the model is both accurate and efficient. In addition, the same approach has been applied to fit the absorption coefficient tables provided by the Orbiting Carbon Observatory (OCO)-2 mission and an accurate prediction model was also presented.