Varietal classification of white wines by fluorescence spectroscopy

Abstract

The Slovak Tokaj region is one of the producers of high-quality white wine having protected designations of origin. The main grape varieties of this region are Furmint, Lipovina and Muscat blanc, which have specific sensory characteristics. This research work presents a strategy for the classification of three mentioned varieties of white wines using fluorescence spectroscopy with chemometrics. Emission and synchronous fluorescence spectra were obtained for bulk as well as diluted samples, principal component analysis (PCA) was applied for exploratory analysis and the scores of the selected PCs were used in linear discriminant analysis (LDA). For undiluted samples, the best PCA-LDA models based on either emission spectra excited at 370 nm or synchronous fluorescence spectra obtained at wavelength difference of 40 and 100 nm provided total correct classifications of 100, 100 and 93% for the calibration, validation and prediction steps, respectively. For diluted samples, the best PCA-LDA models (excitation at 280 nm; wavelength difference of 40 nm) again provided total correct classifications as mentioned above.

Keywords: Fluorescence; Linear discriminant analysis; Principal component analysis; Varietal classification; White wines.

Fig. 1

Total luminescence a–f and total synchronous fluorescence, g–l spectra of undiluted (a–c, g–i) or diluted (d–f, j–l) white wine samples. (a, d, g, j, Furmint; b, e, h, k, Lipovina; c, f, i, l, Muscat blanc)

Fig. 2

Score plot (a, c, e, g) and loading plot (b, d, f, h) for (a, b, e, f) and diluted (c, d, g, h) white wines produced from Furmint, Lipovina and Muscat blanc grape varieties. Spectral data: emission excited at 370 nm (a, b) and 280 nm (c, d); synchronous fluorescence at Δλ = 40 nm (e–h)

Fig. 3

Percentage of correct classification of PCA-LDA models corresponding to different excitation wavelengths λ_ex (a, b) and Δλ (c, d). (Orange, f, Furmint; pink, l, Lipovina; purple, m, Muscat blanc; green, t, total. Grid, calibration; parallel lines, cross-validation; no pattern, prediction) (color figure online)