A Method for Classification and Evaluation of Testing Environments1

Abstract

A methodology is presented to improve the efficiency of cultivar testing first by clustering nursery environments based on selected environmental variables and second by identifying optimum selection environments within clusters by linear regression of the performance of genotypes within an environment on mean genotype performance over all environments. First, the site mean response for the trait is regressed on environmental variables at each site to identify the most predictive subset of variables. Next, the selected predictor variables are converted to standard units and then weighted by the sums of squares from the multiple regression. Finally, the weighted predictor variables are used in a cluster analysis to group sites. A genotypic index regression method is proposed to identify sites that are consistently able to discriminate genotypes. The model is a linear regression of the form yij. =aj+bjyi.+eijk, where yij. is the ith cultivar mean response to the jth environment, aj is the jth environment intercept, bj is the jth environment regression coefficient, yi. is the ith cultivar mean response, and eijk is the error term assumed to be from a normal distribution with a mean of zero and a variance of σ2. The mean cultivar response is an index of the genotypic value of the cultlvar. An environment that discriminates well among genotypes will have a high genotypic index regression coefficient (bj) and one that consistently predicts the performance of genotypes at other sites in the cluster will have a high coefficient of determination (r2). An optimum selection environment should have high values for both statistics. The methodology is applied to a set of data from the International Rice Cold Tolerance Nursery.