Statistical evaluation of the group structures of five venetian wines from chemical measurements

Exploratory statistical methods are used to elucidate the group structures pertaining to a chemical data set obtaned from 184 samples of five typical white wines from the Venetian Region: Soave Classico, Presecco di Conegliano-Valdobbiadene, Verduzzo del Piave, Tocai di Lison, Tocai delle Grave del Friuli. Analytical data included 19 classical parameters and 9 aroma components. Correlation matrices showed that the same three sets of intercorrelated variables were present in all the groups, and that aroma components were nearly uncorrelated with classical parameters. Principal components analysis allowed a subspace of reduced dimensionality to be derived for each group. However, because of similar pattern of the correlation matrices, a high degree of similarity was also observed among the subspaces. An efficient differentiation of the groups was obtained by canonical variates analysis. The most important chemical parameters were potassium, ash content, total nitrogen, cis-3-hexen-1-o1 and 1-hexanol. Classification of samples by the Euclidean distances from the centroids in the canonical space gave on average 94% correct results.