Abstract:

The mathematic classification of 46 species in Rhododendron was studied with 44 morphologic characters, including 18 dualistic characters, 13 multi-characters and 13 quantitative characters, by using SPSS( 10.0. The squared Euclidean distance coefficient was used in case clustering and the Pearson correlation was used in variable clustering by within-groups linkage. The 46 species of Rhododendron were divided into two caboodles by case clustering: one was lepidote and the other was evergreen. The different groups were formed under two caboodles and the groups were similar to the traditional classification except R.calophytum and R.asterochnoum. The results of variable clustering showed that the various characters were decided in different groups. Some characters had strong correlativity within a group, such as the position of squama, flower formation, the size of external organs. In principal component analyses (PCA), the accumulative contribution of the first three principal components was up to 60.9%, which showed there were some representative characters which could be used in classification of Rhododendron, such as squama, gland, hair and the size of leaves, flowers and fruits. However, the number of stamen, hair on sepal, size of sepal, color of crown, hair on lower silk and hair on crown etc were not so reliable for the classification. The results of PCA were consistent with that in case clustering, which suggested that we should pay more attention to choose characters in classifying.

Key words: Rhododendron, mathematic classification, principal component analyses (PCA), cluster analysis