Fluorescence spectroscopy with xgboost discriminant analysis for intraregional wine authentication

Abstract

AIM: This study aimed to use simultaneous measurements of absorbance, transmittance, and fluorescence excitation-emission matrix (A-TEEM) combined with chemometrics as a rapid method to authenticate wines from three vintages within a single geographical indication (GI) according to their subregional variations.

METHODS: The A-TEEM technique (Gilmore, Akaji, & Csatorday, 2017) has been applied to analyse experimental Shiraz wines (n = 186) from six subregions of Barossa Valley, South Australia, from 2018, 2019 and 2020 vintages. Absorbance spectra and EEM fingerprints of the wines were recorded and the data were fused for multivariate statistical modelling with extreme gradient boost discriminant analysis (XGBDA) as reported by Ranaweera, Gilmore, Capone, Bastian, and Jeffery (2021) to classify wine according to their subregions. The cross-validated (k =10, Venetian blinds) confusion matrix score probabilities of classes were used to assess the accuracy of the classification models. A similar procedure was also carried out to discriminate subregions for a single vintage year. Basic chemical parameters (alcohol %v/v, pH, titratable acidity, and volatile acidity) were modelled with the partial least squares regression (PLSR) using A-TEEM data and reference chemical data.

RESULTS: Results have shown an unprecedented 100% correct classification of wines according to subregion across the three vintages and 98% accuracy for subregion in a single vintage year. Other model performance parameters of confusion matrix, including sensitivity, specificity, precision, and F1 score, were also showing the highest values (1.0) for each of the subregions. PLSR modelling revealed that A-TEEM data can also be used for a rapid assessment of basic wine chemical parameters. Notably, the results confirmed a distinct resolution among subregions despite their relatively close proximity within a single GI, indicating the effect of terroir on intraregional variation.

CONCLUSIONS

The sensitivity of A-TEEM allied with multivariate statistical analysis of fluorescence data facilitated the accurate classification of Shiraz wines according to the subregion of origin and production year. As a robust analytical method, A-TEEM can help identify the drivers of regional expression of wine and can potentially be developed for use within the supply chain to guarantee the provenance indicated on the label and to provide an assurance of quality. Overall, A-TEEM with XGBDA modelling continues to be shown as an accurate wine authentication tool that could even be applied at a subregional level.