On the calculation of line-by-line absorption coefficients for gas mixtures using machine learning method

High resolution spectral gas radiative properties are essential for atmospheric radiation research and applications. This study aims to evaluate the applicability of using a single neural network structure to train models to calculate high resolution spectral absorption coefficient for various gases within the same wavenumber range. The developed model is trained separately for ozone, carbon dioxide, and water vapor in the 2550–2650 cm⁻¹ range. The results show that the trained model is highly accurate for each gas, with root mean square errors smaller than 10⁻⁶. The mixture absorption coefficients, obtained by adding the contributions of the three gases, also agree very well with the reference method. This preliminary work demonstrates the applicability of a neural network-based line-by-line model for gas mixtures and suggests the possibility of developing a surrogate model that includes more gases and larger wavenumber ranges for atmospheric radiation.