Wood anatomical classification using iterative character weighing

In this paper we investigate the pattern of wood anatomical variation in some groups of Rubiaceae (i.e. Cinchoneae, Rondeletieae and Condamineae) by using a numerical pattern detection method which involves character weighing (Hogeweg 1975). In this method character weights are obtained iteratively on the basis of the distribution of character values in previously generated classes; these classes being generated by agglomerative cluster analysis, initially with characters weighed equally and subsequently with characters weighed differentially as indicated above. The result of such a study consists of a sequence of dendrograms together with the character weights by which these are produced. Our biological conclusions include: The so obtained results confirm the conclusions drawn by Koek-Noorman & Hogeweg (1974) that the pattern of variation in the wood anatomical structure of these taxa is consistent with the existing classifications at the genus level but does not warrant the higher level classification in Cinchoneae, Rondeletieae and Condamineae as each of these groups show two majorly different woodtypes (differing most conspicuously with respect to libriform fibres and fibre tracheids) while these wood types constitute the major pattern of variation in the sample. However, during the iteration, the Rondeletieae and Cinchoneae tend to be distinguished as separate groups on a level below this main bipartition. The Condamineae remain scattered. The genus Cinchona becomes sharply separated from the other groups during the iteration. Its intermediate position with respect to fibre tracheids and libriform fibres was noted prior to this analysis by Koek-Noorman (Koek-Noorman & Hogeweg 1974). Rather surprisingly, ambiguities (in the sense of being differently classified by different authors) in the existing classifications based mainly on flower morphology reoccur in our iteration based on wood anatomical data in the sense that these taxa change position in the later steps of the iteration. Finally we note that character weights obtained by our method do not spoil the grouping in genera as did character weights obtained by considering the literature on evolutionary trends in wood anatomy (Koek-Noorman & Hogeweg 1974), and that the character weights obtained by our method are not counter to intuition.