CLUSTERING ENVIRONMENTS BASED ON CROSSOVER INTERACTIONS AND USING GRAPHICAL APPROACHES TO VISUALIZE CLUSTERS

Abstract

Crossover interactions occur in evaluation trails when ranks of cultivars change across environments. Determining groups of environments within which crossover interactions are minimized may facilitate making cultivar recommendations. Model-based approaches to finding such clusters have been previously described. Our goal was to describe a new, non-model based approach of defining these clusters and then apply this method to a 59 environment x eight maize (Zea mays L.) cultivar data set. Hierarchical clustering of a 59 x 59 distance matrix defined two environmental clusters within which the total crossover interaction was reduced by approximately one-third and four clusters within which the crossover interaction was reduced by one-half. Four graphical approaches to visualizing the environmental clusters in this data set also were considered. Multi-dimensional scaling (MDS) allowed visualization of clusters when the dimensionality of the crossover space was reduced by considering only some of the crossover interactions between pairs of cultivars. Another benefit of MDS may be identification of specific environmental variables associated with crossover interactions.