The effects of resolution, spectral window, and background type on the predictive capability of partial least squares regression (PLS) on spectra measured by an open-path Fourier transform (OP/FT-IR) spectrometer were tested with spectra of mixtures of alkanes and chlorinated hydrocarbons. The results were compared with the results obtained with the identical data sets using classical least squares regression (CLS). It is shown that the most accurate predictions are obtained using the same conditions that were optimal for CLS, namely spectra measured at low resolution and ratioed to background spectra over the same path length, with the calculations made over limited spectral windows. However, good predictions could be achieved with background spectra measured over a very short: path. Even in the worst cases, the relative error of predictions made by PLS was usually less than 5%. On average, the predicted concentrations of the components of mixtures containing up to five chemically similar analytes made using the PLS algorithm are 120 times more accurate than the predicted concentrations of the components of the identical data sets made using CLS.
