Analysis of ageing and typification of vintage ports by partial least squares and soft independent modelling class analogy

Abstract

Vintage port is a wine produced in very limited amounts in the Portuguese region of the Douro Valley. It ages very slowly in bottles and few analytical data are available. Thus, the composition of 24 wines from the cellars of Dow's and Graham's, corresponding to 12 different years between 1963 and 1990, was analysed. Forty-one analytical parameters determined on each sample were used to predict the year using a partial least squares (PLS) regression and to identify the cellar using a soft independent modelling class analogy (SIMCA) model. Both problems fall naturally within the field of food chemometrics and present the statistical difficulty of having fewer objects than variables. It is, therefore, impossible to apply normal multiple regression and discriminant analysis techniques. The PLS model requires two latent variables and explains 97.4% of the variance of the age variable and 90.2% of the cross-validated variance. As for the characterization of both brands, a SIMCA model with two components for each model correctly classified 22 of the 24 samples; it gives a sensitivity of 83.3% and a specificity of 91.7% for Dow's vintage ports, whereas for Graham's vintage ports the sensitivity and specificity are 91.7 and 100%, respectively.