Style classification

The principal variables (volatiles) were calculated in LatentiX, based on the entire dataset as latent variables (Höskuldsson, 1994). The 24 volatiles that were most descriptive of the styles and explained more than 90 % of the total variation (Table 3.6) were used to calculate the PCA model. The data were subjected to auto-scaling transformation and validated by CV. CV is not as reliable as using a test set for validation, and can be too optimistic as the validation is based on the original model sample set; however the sample set was extremely small, and it was therefore decided to use CV to maximize the number of samples used for the classification model.

Table 3.6 Volatiles that described > 90 % of the total variance and were most descriptive of the three styles and the numbers associated with each volatile for the PCA loadings plot.

Compound class Number on PCA loadings plot Volatiles Explained variance (%) Alcohols 27 1-Octen-3-ol 90.0

53 2-Phenyl ethanol 98.5

Carbonyl compounds 19 Acetoin 97.9

6 Diacetyl 90.3

Wood derived 51 Whiskey Lactone 98.3

36 Guaiacol 91.0

30 Furfural 95.0

Chapter 3.    Research Results   

Figure 3.3 The PCA scores plot (left frame) shows clear differentiation between the wooded and non wooded Chenin blanc styles, where w (red, squares) is rich and ripe wooded, rr (blue, circles) is rich and ripe unwooded and ff (green, triangles) is fresh and fruity. The PCA loadings plot (right frame) indicated that higher content of hexyl acetate and ethyl hexanoate were associated with the rr style, and high content of isoamyl acetate and 2-phenylethyl acetate were correlated with the ff styles, whereas the w styles were associated with higher content of 2-phenyl ethanol and ethyl lactate (numbering for the volatiles is given in Table 3.6)

PCA was done using the volatile data, and some differentiation of the three styles was achieved. As illustrated by the PCA scores plot (Figure 3.3 (left frame), there was a distinction between the wooded and unwooded styles. Differentiation between the two unwooded styles (rr and ff in the PCA scores plot, Figure 3.3 (left frame), was also observed, but to a somewhat lesser extent, with the distinction not being as clear as between wooded and unwooded styles.

Volatiles associated with the fresh and fruity Chenin blanc style were in the top left quadrant of PC1 of the loadings plot (Figure 3.3, right frame, green, circles). These included acetate esters such as isoamyl acetate and 2-phenylethyl acetate. This result for the fresh and fruity style corresponds very well with literature regarding the sensory attributes associated with these compounds. Fruity aromas of banana and pear are associated with isoamyl acetate, while 2-phenylethyl acetate imparts rose, honey, fruity and flowery aromas to wine (Francis and Newton, 2005).

Ethyl butyrate, ethyl hexanoate and the terpenes geraniol and β-ionone, were positively correlated with the unwooded rich and ripe Chenin blanc style (rr), (blue, circles in the loadings plot Figure 3.2 (right frame, the numbers in the loadings plot are associated with the volatiles as shown in Table 3.6). Ethyl butyrate lends an apple aroma to wine, while ethyl hexanoate provides wine with fruity, floral, apple peel and strawberry notes (Francis and Newton, 2005). β-Ionone is known to lend aromas of violet, floral, raspberries and seaweed to wine (Francis and Newton, 2005) while geraniol is associated with aromas of rose and geranium above the odour detection value of 30-130 µg L^-1 (Swiegers et al., 2005).

The volatiles situated towards the positive side of PC1 of the loadings plot (circled in red, Figure 3.3 (right frame), with numbering associated with the volatiles given in Table 3.6), were associated with the wooded style Chenin blanc wines. Of these, ethyl lactate, acetic acid, diacetyl and acetoin are volatiles known to increase when wine undergoes malolactic fermentation (Bartowsky and Henschke, 1995; Gil et al., 2006). Although malolactic fermentation of Chenin blanc wine is usually not encouraged by winemakers, the wooded styles are the only South African Chenin blanc wines which are allowed to undergo between 30-70 % malolactic fermentation (O’Kennedy, 2009). Ethyl lactate contributes buttery, coconut and creamy aromas (Swiegers et al., 2005) while acetoin and diacetyl have been known to lend white wine nutty, toasty aromas and add complexity to the wine when below 30 mg L^-1 and 0.2-8 mg L^-1, respectively (Varnam and Sutherland, 1994; Martineau et al., 1995; Gil et al., 2006).

The rich and ripe wooded Chenin blanc wines were also positively correlated with fusel alcohols (e.g., 2-phenyl ethanol), furfural, guaiacol and furfuryl alcohol (circled in red, Figure 3.3, right frame). Furfural and guaiacol exhibit aromas of caramel, sweet, bread and almond, and smokey

and sweet, respectively (Prida and Chatonnet, 2010). 2-Phenyl ethanol confers attributes of honey, spice and floral (Francis and Newton, 2005).

Subsequently the chemical data was investigated as to whether or not the wines could be classified according to their vintages. The results were much less obvious than classification by styles (data not shown). However, some of the distinctive trends were that the younger Chenin blanc wines tended to exhibit higher content of isoamyl acetate and 2-phenylethyl acetate than the older wines. This corresponds with reported findings that as white wine ages, the concentration of isoamyl acetate and 2-phenylethyl acetate decrease significantly (Camara et al., 2006). This decrease in acetate esters as wines age could be a reason that white wines, during the first period of maturation, have a higher degree of fruitiness than in the following stages of maturation (Camara et al., 2006).

The more mature wines were correlated with higher content of acetic acid and furfural. The increase in acetic acid in ageing wine is due to ester hydrolysis, as initially noted by Simpson and Miller (1983) and more recently addressed by Sumby et al., (2010). Furthermore, furfural has been used successfully in the past to predict age of wine, as it has been reported to increase in a linear manner during maturation (Camara et al., 2006; Silva Ferreira et al., 2003).