|Yan, W - AGCANADA|
Submitted to: Euphytica
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: August 24, 2009
Publication Date: January 10, 2010
Citation: Yan, W., Holland, J.B. 2010. A HERITABILITY-ADJUSTED GGE BIPLOT FOR TEST ENVIRONMENT EVALUATION. Euphytica. 171(3):355-369. Interpretive Summary: Varieties that perform best in one environment are often not best in a different environment. Plant breeders refer to this as genotype-by-environment interaction. This is a major complication to plant breeding programs, as it reduces gain from selection and may require varietal selection for specific environments. Statistical analyses have been developed to graphically plot the adaptation of different varieties to a set of environments. In this study, we demonstrated that these analyses can be improved by the use of statistical models that assume a unique experimental error variation for each test environment. The results demonstrate that these methods can better identify which varieties should be retained in selection programs.
Technical Abstract: Multiple environment crop variety trials can be informatively analyzed as a multivariate data set in which genotype (G) main effects are not separated from genotype-by-environment (GE) interactions, referred to as GGE data. GGE biplot analysis of genotype-by-environment data includes three aspects: mega-environment analysis, test environment evaluation, and genotype evaluation. Previously, this analysis was used assuming all environments had similar levels of precision for variety comparisons. In this study, the standard unscaled data analysis was compared to three alternative methods that account for differences in precision among test environments: (1) a mixed model incorporating heterogeneous error variances, (2) data scaling based on the within-environment standard error (SE), and (3) data scaling based on the within-environment standard deviation among genotype means (SD). The analyses were demonstrated for grain yield of 37 oat (Avena sativa L.) genotypes evaluated across ten environments. Results indicate that the GGE biplot based on best linear unbiased predictors (BLUPs) from a mixed linear model assuming heterogeneous errors is the preferred GGE biplot for GED analysis as it appropriately accounts for the heterogeneity of the environments, is valid with unbalanced data, and displays the genetic variations in each environment. The SD-scaled GGE biplot does not reveal either the genotypic or phenotypic variation of the environments but it is effective at displaying the associations among test environments in relative genotype performances. After defining mega-environments based on GGE biplots, genotype comparisons with sound statistical inferences can be made with genotype BLUPs even with unbalanced data.