Neural network predictive model for quantitative analysis of atmospheric gas absorption IR spectroscopy data

In this paper, predictive models are proposed for estimating the concentration of small atmospheric gas components based on a training sample of model IR absorption spectroscopy data. The spectra of the targeted gases were taken from the HITRAN database. The following machine learning methods were studied: support vector regression, random forest, LASSO, and artificial neural network of forward propagation. A new machine learning pipeline based on an artificial neural network has been proposed, surpassing the accuracy of other methods. The obtained models can be used to estimate the concentration of target gases in the atmosphere based on IR absorption spectroscopy data.

Keywords

Authors

References