PRINCIPAL COMPONENT AND CLUSTER ANALYSES OF COTTON CULTIVAR VARIABILITY ACROSS THE UNITED-STATES COTTON BELT

Though few examples of multivariate analyses of agronomic trials exist in the literature, such analyses can provide useful additional information as a supplement to normal univariate analyses. In this study, data from the Regional Cotton Variety Tests were used to study the interrelationships of cotton (Gossypium hirsutum L.) cultivars based on agronomic and fiber trait measurements. The study of agronomic variability among cultivars was reflective, in part, of genetic variability, and gave graphical, nonnumerical assessments of genetic variability. Principal component, hierarchical (Ward's minimum variance) cluster, and nonhierarchical (k-means) cluster analyses were computed on data from seven of the nine regions of the tests. Check cultivars were common across regions, enabling comparisons of results within and across regions. Three-dimensional plots displayed results of the principal component and k-means cluster analyses, while results from Ward's minimum variance clustering were presented as dendrograms. Cultivars in the Mississippi Delta, Central, and Texas High Plains regions did not cluster tightly, whereas cultivars in the Eastern, New Mexico, and San Joaquin Valley regions clustered comparatively tightly. This implies greater intercultivar differences in the first three regions than in the latter three, which shows the genetic bases of the first three to be broader. New Mexico Acala cultivars clustered less tightly than Acala cultivars in the San Joaquin Valley, likewise implying a broader genetic base in the New Mexico cultivars. Check cultivars tended to cluster as outliers outside their region of adaptation.