Chemometric application with Fourier-transform near infrared spectroscopy in the quantification of greenhouse gases with high humidity levels

Greenhouse gases (GHG) emission has been increasing due to human activity. The continuous increase in GHG concentrations in the atmosphere may be linked to climate changes and global warming. Thus, there is an interest in monitoring the sources of GHG emissions, with special attention to the CO2, CH4 and N2O. For this reason, it has been observed an increase of different approaches in the development of analytical techniques for quantification and in situ observation. The analytical techniques, which are often used for trace gas analyses, are gas chromatography (GC) and non-dispersive infrared spectroscopy (NDIR). One of the difficulties encountered by spectroscopic techniques is the interference generated by ambient water, due to spectral overlapping. However, the use of near infrared (NIR) spectroscopy when combined with chemometrics is able to overcome this limitation. It enabled the development of a method with a greater selectivity in determining these gases even at high humidity levels. Therefore, in this study it was developed a multivariate model to predict the concentration of CO2, CH4 and N2O in an atmosphere with high humidity, using NIR spectroscopy and chemometrics.
In this way, analytical curves were prepared employing a Shimadzu GC-MS QP 5050A for each GHG. Spectra of mixtures containing the three GHG were obtained in a Fourier-transform NIR ABB Bomem 160D (4000 – 6600 cm-1) coupled with a multipass cell with optical path of 105 m. Pure gases were used in the preparation of GHG mixture (99.95 % for each GHG). The analytical curves for CH4 and N2O (0.5 to 32 ppmv) and CO2 (100 to 1000 ppmv) were used to evaluate the concentration of GHG in the mixture. The mixtures were prepared in the multipass cell with the addition of the GHG in wet synthetic air, with different relative humidity values (40 to 90 %). A total of 55 samples of gas mixtures prepared were analyzed by FT-NIR and GC-MS. The GHG mixtures were divided into two sets, using the Kennard-Stone algorithm: 30 samples to the calibration set and 25 to the external validation set. All spectral information was mean-centered and treated with second derivative, using the Savitzky–Golay algorithm and employing 21 points window and 2nd order polynomial to correct variations in baseline. PLS models using NIPALS algorithm and one-leave-out cross validation were obtained for the target gases, employing the full range of the spectral. The Jackknife algorithm was used to select the variables with the greatest contribution for the construction of robust PLS models. All chemometric treatments were made with software The Unscrambler X 10.3.
The chemometric method developed presented RMSEP values of 0.71 ppmv (N2O), 0.88 ppmv (CH4) and 38.98 ppmv (CO2) in the presence of water vapor ranging to 40 up 90% in relative humidity. When compared to other techniques found in the literature, the proposed method showed a similar ability to quantify the GHG, with advantages such as the determination of such gases in situ and without being necessary the removal of atmospheric water.