Retrieving CO Concentrations from FT-IR Spectra with Nonmodeled Interferences and Fluctuating Baselines Using PCR Model Parameters

Abstract

It is demonstrated that good predictions of gas concentrations based on measured spectra can be made even if these spectra contain totally overlapping spectral features from nonidentified and nonmodeled interfering compounds and fluctuating baselines. The prediction program (CONTOUR) is based solely on principal component regression (PCR) model parameters. CONTOUR consists of two smaller algorithms. The first of these is used to calculate pure component spectra based on the PCR model parameters at different concentrations. In the second algorithm, the calculated pure component spectra are subtracted one by one from the contaminated spectrum, and the length of the spectral contour within specified wavenumbers is then calculated. When the length of the contour is at a minimum, a condition is reached where the pure component part of the measured spectrum is absent and only the background signal remains. The assumptions are that the background and analytical signals must be additive and that no accidental match between these signals takes place. The best results are obtained with the use of spectra with a high selectivity. The use of the program is demonstrated by applying simple single-factor PCR models based on pure gaseous 1 and 4 cm−1 CO Fourier transform infrared (FT-IR) spectra (50–400 ppm) measured at ambient temperatures. The program is validated with measured CO spectra containing interferents such as N2O, CO2, and added Hitran-simulated H2O, CO2, and COS spectra, representing strong features in the CO spectral region.