varieties were defined as grouping variables. The recognition abil- ity was 100% for all the considered classes. The prediction abilit which was evaluated by cross-validation (Table 2 was 100% for Starkrimson, Qinguan, la, Golden Delicious, and Fuji juices but only 88.9% for Jonagold juices. LDA produced some misclas fication of the samples. One sample from Jonagold was assigned as Qinguan SLDA. Although all classes appeared separated from the others prediction ability showed that LDA based on the scores matrix of the 1st 17 principal components did not enable completely sorting of the samples from the 6 varieties. Therefore, we tried to perform SLDA directly on the data matrix to select the most useful variables for the differentiation of the 6 varieties and to achieve maximum separation among the given classes. Only 10 variables were retained and used as input in LDA classification LDA
The result obtained after performing LDA on the reduced data set showed unequivocal discrimination between the 6 varieties. Hundred percent recognition ability and 100% prediction ability were obtained. The prediction ability of the SLDA model was evaluated by leave-one-out method. The eigenvalues, explained variances, and canonical correlations for the 5 discriminant func tions which are statistically extremely significant (Wilks' Lambda values) are shown in Table 3 The coefficients of the variables in the 1st 5 discriminant functions (Table 3) reveal which variables laave a greater influence on those. The graphical representation of the juices in the plan defined by the 1st 2 discriminant functions is presented in Figure 2. This figure revealed a clear separation between the 6 varieties.
Identification of discriminant markers for varietal classification.
A step based on the identification of the volatile compounds selected by the SLDA was carried out to gain further knowledge about the results obtained from a chemical standpoint The volatile markers identified in accordance with the 10 reten tion times selected by SLDA for the discrimination of apple juices from different varieties are summarized in Table 4 The contribution of the volatile compounds to the flavor has been investigated in previous studies in order to establish relation- ships between the volatile compound and sensory attributes. All the ester markers identified are described to have apple-like i odorants, with fruity, sweet, and ripe descriptors (Young and oth- ers 1996; Dixon and Hewett 2000). Ethyl 2-methylbutanoate is correlated with strawberry attributes (Mehinagic and others 2006) and 2-methylbutyl acetate contributes to banana n (Williams and Knee 1977). These 2 compounds have been recognized to contribute significantly to Fuji and Golden Delicious aroma (Mehinagic and others 2006). Moreover, hexyl acetate which smells strong sweet-fruity odor (Rizzolo and others 1989) has been reported as the main odor contributor in Fuji, Golden Delicious and Jonagold apples (Echeverria and others 2004; Mehinagic and others 2006: Kom thong and others 2007). 1-Hexanol which has earthy and unpleasant odor is indicated as the negative odor con- tributor (Rizzolo and others 1989). This compound suppresses the apple-like odor in the apple model mixture (Bult and others 2002)
Classification of apple juices according to geographical origin
LDA. Twelve apple juice samples from Liquan, 12 from Sanyuan, 12 from Xunyi, and 13 from Yongshou were subjected to LDA. The sample from Chunhua was excluded from further analysis due to the very small number of samples from this county. 17 principal components of the samples The scores of the 1st were set as independent variables and geographical origins were set as grouping variables. The total recognition ability was 91.8%, while the prediction ability evaluated by cross-validation was only 63.3% (Table 5). The juices from Sanyuan were clearly separated from those of Liquan, Xunyi, and Yongshou. However, the juices from the latter 3 counties were not sufficiently separated. The prediction ability for juices from Xunyi (41.7%) and Yongshou (38.5%) was not acceptable due to a serious overlapping between them