Effect of orthogonal signal correction on the determination of compounds with very similar near infrared spectra

Abstract Orthogonal signal correction (OSC) is a spectral treatment that allows the portion of spectral information orthogonal to changes in analyte concentrations to be removed. In this work, the effect of using OSC on low-scatter near infrared (NIR) spectra was examined in order to determine analyte concentrations at different levels in samples containing additional components with very similar spectra. To this end, transmittance spectra for virgin olive oil were used to quantify oleic, linoleic and linolenic acids, which were found to account for 63–75, 8.9–13.4 and 0.2–0.9%, respectively, of total fatty acids. Partial least-squares regression (PLSR) was used for calibration and the results thus obtained were compared with those provided by the more usual NIR spectral pre-treatments. As shown in this work, OSC effectively removes information not correlated to the target parameter, which substantially decreases the number of PLS components required to construct calibration models. This, however, does not increase predictive ability because not all variability not correlated to the concentration of the target analyte is orthogonal to it.