Quantification of red wine phenolics using ultraviolet-visible, near and mid-infrared spectroscopy combined with chemometrics
Abstract
The use of multivariate statistics to correlate chemical data to spectral information seems as a valid alternative for the quantification of red wine phenolics. The advantages of these techniques include simplicity and cost effectiveness together with the limited time of analysis required. Although many publications on this subject are nowadays available in the literature most of them only reported feasibility studies. In this study 400 samples from thirteen fermentations including five different cultivars plus 150 wine samples from a varying number of vintages were submitted to spectrophotometric and chromatographic phenolic analysis. Anthocyanins, total phenolics, tannins, colour density and the most representative compounds within the main phenolic families (hydroxicinnamic acids, flavan-3-ols, flavonols and anthocyanins) were quantified. Spectra were recorded in different regions of the electromagnetic spectrum. Particularly the information contained in the ultraviolet-visible region as well as in the near and mid-infrared regions was collected. Regression models were built and validated. The interpretation of the loadings and coefficients of regression, the evaluation and analysis of the correlation among variables and the measured phenolic compounds as well as the chemistry basis behind each quantified compound was extensively investigated and reported. Spectral pre-processing techniques as well as variable selection tools were also investigated and selected based on model performance. Accurate models for most of the phenolic compounds and spectroscopies were obtained with residual predictive deviation (RPD) values higher than 2.5. The results obtained showed UV-visible and infrared spectroscopy as valid approaches for the quantification of the phenolic content throughout the winemaking process. Considerations such as easiness of use and the economical and human resources involved in the analysis will also be discussed.
Issue: Macrowine 2016
Type: Poster
Authors
*Stellenbosch University