Port wines attract a price premium dependent on their age and variety. In this study (open access) the authors built an FT-IR model to differentiate Colheita, Tawny, Vintage, and Late Bottled Vintage ports by variety and by age.
The authors relate the differences in spectral signatures back to different oxidation conditions and temperatures during the port’s production and maturation, and discuss the chemical changes such as Strecker degradation, Maillard-type pathways, and the formation of furanic compounds derived from sugar degradation, and show that the consequential spectral changes can be predictable with age.
The authors analysed a data set of 7281 spectra from the four commercial categories using classical chemometric methods and machine learning models.
They found that for category classification, a support vector machine with a radial basis function kernel gave the best performance. For age prediction, a deep ensemble model achieved an R2 value of 0.9803 and a mean absolute error of 1.25 years. A hybrid two-stage random forest model, designed to account for category-specific aging regimes, achieved comparable performance, with an R2 of 0.9630 and a mean absolute error of 1.28 years, while requiring lower computational resources. Leverage-based prediction intervals provided per-sample uncertainty estimates with a 97.7% empirical coverage.
Feature importance analysis identified three main spectral regions, around 1740, 1050, and 1600–1450 cm–1. These features could be related back to known chemical pathways occurring during Port wine aging.
They conclude that the results demonstrate that FT-IR spectroscopy, combined with appropriate modeling strategies, can support rapid, nondestructive, and scalable estimation of Port wine age, offering a promising complementary tool for certification and authenticity control.
Photo by Irene Kredenets on Unsplash
Comments