Effects of resolution, spectral window, and background on multivariate calibrations used for open-path Fourier-transform infrared spectrometry

The performance of CLS, PLS-2, and PLS-1 on calibration/validation sets of mixtures of five short-chain alcohols was examined as a function of resolution, spectral window, and type of background single beam to determine the feasibility of achieving automated calibrations for open-path Fourier-transform infrared (OP-FTIR) spectrometry. The backgrounds used in the study were created by ratioing actual single-beam OP-FTIR spectra. Some backgrounds were created by ratioing spectra measured at the same path length, and others were created by ratioing a long-path sample spectrum against a short-path reference. After conversion to absorbance, these background spectra provided a noise term that is representative of what is seen in actual field deployment. The absorbance spectra of mixtures of the spectra of up to five short-chain alcohols (chosen because of their spectral similarity) were added to these spectral baselines. CLS was shown to perform well only when using equidistant backgrounds and TO-16 spectral window selection, with average errors of 2–20%. CLS calibrations under other conditions tested typically produced average errors of greater than 100%. Under most conditions the measurement of OP-FTIR spectra at low (8 cm⁻¹) resolution produced an increase in the accuracy of the determinations over higher-resolution spectra. Both PLS-1 and PLS-2 were able to predict the concentrations of the alcohols under any conditions, including short-path backgrounds, with an error of less than 5%; errors of less than 2% were achieved in most cases. The results indicate that an automated calibration system is feasible for OP-FTIR spectra measured at low resolution, even when a short-path background is used. © 1999 John Wiley & Sons, Inc. Field Analyt Chem Technol 3: 117–130, 1999