The use of ordination techniques in the preliminary evaluation of Stylosanthes accessions

Abstract

It is suggested that ordination is preferable to classification if two sets of data are to be rigorously compared. Ordination techniques have been applied to data from 154 Stylosanthes accessions grown at two contrasted sites. Eleven agronomic attributes were recorded for each site. Each set of data was submitted separately to a principal component analysis, and the first three components in each case are described and discussed. Techniques are explained as well as the way in which the components can be used for predicting the most promising accessions. The two sets of components were compared by canonical procedures, and although the between-attributes-between-sites correlation coefficients are everywhere low, the canonical coefficients are extremely high. There is, therefore, a robust attribute- pattern, common to the two sites, but exemplified by different species in each instance. If a desirable performance can be defined in terms of the canonical components, it follows that accessions can always be found that will meet it, but different species will be required at different sites. The results strongly suggest that, among the set of accessions tested, there may be forms potentially capable of out-yielding even the best of the existing cultivars.