A Data-Based Approach for Selecting Locations for Regional Trials

Two approaches to selecting locations for a regional trial based on the genotype × environment (GE) interaction are investigated. The purpose is to find the subset of locations which is essential for generating the structure of GE interactions found in the sampled data. Approach 1 is to calculate the largest GE interaction mean square (MS) for each number of locations. The algorithm does not require calculation of all MS. The largest MS can be synthesized in a stepwise fashion in a backward direction from all locations to two locations using the pairwise GE interaction MS. These largest MS are then plotted against the number of locations to determine the optimum number required. Approach 2 is to group the locations by the similarity of their GE interaction using cluster analysis. A location is then selected from each of the resulting groups. The GE interaction MS generated from the group structured combinations (Approach 2) are very stable. For the present set of data, the ratio between the largest and the smallest sets (in terms of MS) is about 1.3 times, while for the unstructured combinations, the ratio is about 31 times. Since Approach 1 gives the best mathematical solution (the largest MS), but provides no information about other subsets which are near optimum in terms of MS, while Approach 2, does the opposite, simultaneous use of both approaches is recommended.