Cell Mean Versus Best Linear Unbiased Predictors in Biplot Analysis of Genotype × Environment Interaction in Barley.

In multi-environment trials, accurate estimation of yields in individual environments and astute choice of models to extract and display agronomically relevant signals enhance genotype evaluation and accelerate breeding progress. The objective of this study is to (i) compare patterns of genotype × environment interaction (GE) using additive main effect and multiplicative interaction (AMMI) biplots arising from cell means versus best linear unbiased predictors (BLUPs), and (ii) examine some features of the genotype main effect plus GE interaction (GGE) in relation to AMMI in comprehending the GE patterns. A data set generated from 39 barley genotypes grown in 18 environments (three sowing dates and two crop protection treatments over three years) in the central highlands of Ethiopia was used. AMMI analysis of variance based on cell means depicted the first five principal components (PCs) to be significant. However, only the first two PCs were significant when BLUPs were used. Partitioning of the original GE sum of squares into signal and noise confirmed that only the first two AMMI PCs contained signals required to explain the real GE pattern. AMMI PC1 contained 76.5% and AMMI PC2 15.9% of the total GE variance. AMMI biplot based on BLUPs depicted patterns that were more in tandem with agronomic interpretations than biplot based on cell mean data. PC1 of GGE contained 66.9%, PC2 11.2% and PC3 14.5% of the total GE variance. AMMI2 explained as much GE variance as PC1, PC2 and PC3 of GGE put together. AMMI2 biplot depicted a GE pattern that was not obvious from GGE2. AMMI2 biplot was more similar to GGE PC1 versus PC3 biplot than GGE2 biplot. AMMI2 was more efficient than GGE2 for displaying patterns of GE interaction in this data set. However, GGE2 was quite elegant and simple for presenting G and GE combined in a biplot graph including the which-won-where pattern. BLUPs might improve yield estimation and pattern recognition, and that attempting both AMMI and GGE analysis might provide important insights on genotype performance and GE.