A further numerical contribution to the classification of the Poaceae

Abstract

A sample of 92 widely representative grass genera scored for 50 attributes was subjected to the classificatory programmes MULTBET and MULTCLAS. Of these two the former was accepted as the more useful in that it was the more successful in uniting genera known to hybridize. It organized the genera into five groups here designated as bambusoid, pooid, panicoid-phragmitoid, eragrostoid, and andropogonoid. On comparing these in pairs and further combinations by means of a GROUPER analysis it was found that leaf and fruit attributes were principally responsible and that spikelet attributes contributed only slightly to the definition of groups. A principal coordinate analysis of the same data separated the bambusoid genera from the remainder, which were arranged in an elongated ellipsoid when considered in terms of the three largest axes. Within the ellipsoid the pooid, panicoid-phragmitoid, eragrostoid, and andropogonoid genera occupied discrete but contiguous regions. A GOWECOR analysis of the data showed that the attributes most closely correlated with the three longest ordination axes were those of the leaf and fruit. As with the GROUPER analysis, spikelet attributes were of little importance.