Application of Principal Component Analysis in Processing of Time-Resolved Infrared Spectra of Greenhouse Gases
WANG Yu-hao1, 2, LIU Jian-guo1, 2, XU Liang2*, DENG Ya-song2, SHEN Xian-chun2, SUN Yong-feng2, XU Han-yang2
1. School of Environmental Science and Optoelectronic Technology, University of Science and Technology of China, Hefei 230026, China
2. Key Laboratory of Environmental Optics and Technology, Anhui Institute of Optics and Fine Mechanics, Hefei Institutes of Physical Science, Chinese Academy of Sciences,Hefei 230031, China
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 4.191×10-4, 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 μmol·mol-1, CH4: 5.24 nmol·mol-1, N2O: 2.92 nmol·mol-1, and CO: 4.72 nmol·mol-1. Similarly, for 5-minute average measurement spectra, the precision is: CO2: 0.24 μmol·mol-1, CH4: 5.24 nmol·mol-1, N2O: 2.92 nmol·mol-1, and CO: 4.72 nmol·mol-1. The precision of component concentration inversion for 1-minute reconstructed spectra is: CO2: 0.17 μmol·mol-1, CH4: 2.97 nmol·mol-1, N2O: 0.72 nmol·mol-1, and CO: 1.40 μmol·mol-1. For 5-minute reconstructed spectra, the precision is: CO2: 0.15 μmol·mol-1, CH4: 1.74 nmol·mol-1, N2O: 0.29 nmol·mol-1, and CO: 0.97 nmol·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.
Key words:FTIR;Principal component analysis;Spectral noise reduction;Greenhouse gas
王钰豪,刘建国,徐 亮,邓亚颂,沈先春,孙永丰,徐寒杨. 主成分分析在温室气体时序红外光谱处理中的应用研究[J]. 光谱学与光谱分析, 2023, 43(07): 2313-2318.
WANG Yu-hao, LIU Jian-guo, XU Liang, DENG Ya-song, SHEN Xian-chun, SUN Yong-feng, XU Han-yang. Application of Principal Component Analysis in Processing of Time-Resolved Infrared Spectra of Greenhouse Gases. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(07): 2313-2318.
[1] Lan X, Thoning K, Dlugokencky E. Laboratory Measurements: Version, 2023, 6: 1.
[2] Hall B D, Crotwell A M, Kitzis D R, et al. Atmospheric Measurement Techniques, 2021, 14(4): 3015.
[3] Lenssen N J, Schmidt G A, Hansen J E, et al. Journal of Geophysical Research: Atmospheres, 2019, 124(12): 6307.
[4] Tan X, Liu Y, Dong H, et al. Environmental Geochemistry and Health, 2022, 44: 2955.
[5] Esler M B, Griffith D W T, Wilson S R, et al. Analytical Chemistry, 2000, 72(1): 206.
[6] Smale D, Sherlock V, Griffith D T, et al. Atmospheric Measurement Techniques, 2019, 12(1): 637.
[7] Liu H, Li Y, Zhang Z, et al. Optics Express, 2018, 26(18): 22837.
[8] Statheropoulos M, Pappa A, Karamertzanis P, et al. Analytica Chimica Acta, 1999, 401(1-2): 35.
[9] Liu T, Liu H, Chen Z, et al. IEEE Transactions on Industrial Informatics, 2018, 14(12): 5268.
[10] Kistenev Y V, Skiba V, Prischepa V, et al. Journal of Quantitative Spectroscopy and Radiative Transfer, 2022, 289: 108278.
[11] Griffith D W T. Applied Spectroscopy, 1996, 50(1): 59.
[12] Jolliffe I T, Cadima J. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 2016, 374(2065): 20150202.
[13] Deborah H, Richard N, Hardeberg J Y. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2015, 8(6): 3224.
[14] Haney R, Siddiqui N, Andress J, et al. 41st International Conference on Environmental Systems, 2011: 5091.