Multivariate data analysis in classification of musts and wines of the same variety according to vintage year

Abstract

Multivariate analysis has been used to discriminate among different vintage years (1986, 1987,1988)of Chardonnay musts and wines from 31 growing areas in a relatively small geographical region ofTrentino (Italy). Musts have been analyzed as to their content of amino acids and wines with respectto amino acids, volatile compounds (i.e., alcohols, esters, amides, acids, and others), and metal ions.Principal component analysis has been used as a display technique. To identify the most significantparameters for discrimination, an appropriate stepwise procedure for feature selection was carried outbefore linear discriminant analysis was applied, which led to quite satisfactory classification and predictionrates. Biochemical and oenological arguments have been proposed to account for the significance ofthe selected sets of variables. On the other hand, unsatisfactory results were obtained by univariateapproach based upon calculation of Fisher weights.