1995 年 1 巻 1 号 p. 38-43
Multivariate analysis was applied to 35 volatile components and sensory scores of 16 Kenyan black teas made of tea leaves of two clones cultivated at eight areas. Although profiles of volatile components in black teas belonging to a clone deviated according to their harvesting areas, cluster analysis and factor score plots clearly showed differences in the two clones. Multivariate calibration methods provided equations predicting sensory scores using volatile components. Multiple linear regression analysis (MLR) selected 1-penten-3-ol, (E,E)-2,4-heptadienal and linalool oxide as effective components and the resulting multiple correlation coefficient (R) was 0.914. The optimum number of principal components indicated by cross-validation was three in the partial least squares (PLS) regression analysis with an R value of 0.946. A higher correlation of components eluted after linalool, i.e., α-cedrene, 3,7-dimethyl-1,5,7-octatrien-3-ol, cedrol and bovolide, to tea quality was suggested by factor loadings of PLS. Principal component regression (PCR) with four principal components showed the lowest R (0.811) among the three calibration methods.