Application of Principal Component Analysis in Processing of Time-Resolved Infrared Spectra of Greenhouse Gases
Yu-Hao Wang,Jian-Guo Liu,Liang Xu,Ya-Song Deng,Xian-Chun Shen,Yong-Feng Sun,Han -Yang Xu
DOI: https://doi.org/10.3964/j.issn.1000-0593(2023)07-2313-06
2023-01-01
Spectroscopy and spectral analysis
Abstract:Fourier Transform Infrared (FTIR) spectroscopy enables simultaneous measurement of the concentrations of various greenhouse gas components. However, instrument noise and spectral line overlap can affect the quality of spectroscopic data, thereby influencing the accuracy of component concentration inversion. In this study, we address this issue by reconstructing the time-resolved measurement spectroscopic matrix using different numbers of principal components. The Euclidean and cosine distances between the reconstructed and original spectroscopic matrices are employed as criteria for dynamically selecting the number of principal components, thus enhancing the quality of time-resolved spectroscopic data. The proposed method is applied to numerical simulation spectra, standard gas measurement spectra, and field experiment spectra. The results show that the structure characteristics of the simulated spectra, with an additional 0. 001 RMS noise, remain essentially unchanged after spectral reconstruction. The residual standard deviation between the reconstructed and original spectra is 4. 191 x 10(-4), effectively reducing the impact of noise in the measurement spectra. The method is further utilized to reconstruct the average measurement spectra of standard gases, and the accuracy of component concentration inversion between the reconstructed and average spectra is compared. The precision of component concentration inversion for 1-minute average measurement spectra is as follows: CO2 : 0. 24 mu mol center dot mol(-1), CH4; 5. 24 nmol center dot mol(-1), N2O: 2. 92 nmol center dot mol(-1), and CO: 4. 72 nmol center dot mol(-1). Similarly, for 5-minute average measurement spectra, the precision is: CO2 : 0. 24 mu mol center dot mol(-1), CH4 : 5. 24 nmol center dot mol(-1), N2O: 2. 92 nmol center dot mol(-1), and CO: 4. 72 nmol center dot mol(-1). The precision of component concentration inversion for 1-minute reconstructed spectra is: CO2 : 0. 17 mu mol center dot mol(-1), CH4 : 2. 97 nmol center dot mol(-1), N2O: 0. 72 nmol center dot mol(-1), and CO: 1. 40 mu mol center dot mol(-1). For 5-minute reconstructed spectra, the precision is: CO2 : 0. 15 mu mol center dot mol(-1), CH4 : 1. 74 nmol center dot mol(-1), N2O: 0. 29 nmol center dot mol(-1), and CO: 0. 97 nmol center dot mol(-1). Utilizing reconstructed spectra improves the accuracy of gas concentration inversion, with the precision of component concentration inversion for 5-minute reconstructed spectra meeting the requirements of the World Meteorological Organization/Global Atmosphere Watch (WMO/GAW) for extended measurement accuracy. In field experiments, the correlation coefficient between the concentration of CO2 obtained from 1-minute reconstructed spectra and that from 1-minute average spectra reaches 89. 40%. The comprehensive analysis demonstrates that the dynamic selection of principal components for FTIR time-resolved spectroscopic data reduces the influence of noise and effectively preserves the characteristic variation information of time-resolved measurement spectra.