Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/24/1996
Publication Date: N/A
Interpretive Summary: Farmers need crop varieties which grow well in their location. To provide these, seed companies and public Plant Breeders routinely grow new cultivars in many different locations to determine suitability. The study reported in this paper compared several different statistical methodologies which are routinely used to compare the suitability of different cultivars for different locations. The authors found that the different methodologies were based on different assumptions, and were not equivalent. They report that with the appropriate methodology, fewer testing locations are needed for ascertaining the best cultivar for different locations within a region, and that only a few cultivars are needed to provide the farmers with the most appropriate cultivar(s) throughout that region. These results should speed up as well as reduce the cost of crop breeding efforts.
Technical Abstract: To maximize yield throughout a crop's heterogeneous growing region, despite differences in cultivar rankings from place to place due to genotype environment interactions, it is necessary to subdivide an entire region into several relatively homogeneous megaenvironments and to breed and target adapted genotypes for each megaenvironment. This study compares sseveral statistical strategies for identifying megaenvironments. The four evaluative criteria were: flexibility in handling yield trials with various designs, focus on that fraction of the total variation which is relevant for identifying megaenvironments, duality in giving information on both genotypes and environments, and relevance for the primary objective of showing which genotypes win where. Several different megaenvironment analyses were compared. Linear regressions were often ineffective because they captured too little of the interaction. YS can be helpful if a single emegaenvironment suffices, but for large regional trials this outcome is rather rare. SHMM generates a large number of megaenvironments and fails to emphasize what wins where. Pattern analysis can be helpful, especially if its classification and ordination components are both chosen carefully. In general, AMMI was most helpful, using a modified biplot that represents genotypes by parameter regions in which they win rather than by the usual points. Preliminary results indicate that a small and workable number of megaenvironments often suffices to exploit interactions and to generate increased yields.