Establishing the geographical origin of some dry white wines according to ICP spectrometric and chemometric analysis

Abstract

An important aspect of assessing authenticity of wines is determining their geographical origin. The aim of the present work was establishing geographical origin of wines produced from grape varieties of Chardonnay, Riesling and Muscat grown in different districts of Krasnodar Territory, from the results of ICP-spectrometry and chemometric methods. A significant difference in the concentration of Al, Ba, Ca and Rb in wines was observed depending on the grape variety, and a difference in Al, Ba, Rb, Fe, Li and Sr concentration depending on region of the grape origin. Concentrations of the elements in different groups of wines also had different deviation from the average values. A cluster structure of wine samples relative their origin districts, revealed using discriminant analysis, allowed to develop models with high prognostic properties for identifying geographical origin of wines. A quality criterion of the developed models was precision of classification, i.e. fraction of correctly identified wine samples. Neural networks demonstrated the maximum precision of classification (100 %) for all 153 wine samples used, followed by support vector machine (98.69 %) and general discriminant analysis (94.77 %). Among all metal concentrations, Sr, Li and Fe dominated in the importance of their contribution in the constructed statistical models for predicting the geographical origin of wines. The results of the studies showed that the machine learning methods oriented to high dimensional data together with ICP-spectrometry analysis can successfully solve problems of small dimension related to determining the geographical origin of wines on the basis of their component composition and name with the precision exceeding the traditional method of general discriminant analysis.