There has been a significant trend in the use of different statistical tools to analyse genotype × environment (GE) interaction for grain yield in multi-environment trials. Several statistical models including 16 univariate stability methods and four multivariate models such as the additive main effects and multiplicative interaction (AMMI), GGE biplot (G+GE biplot), and factorial regression and partial least squares regression were applied to investigate the GE interaction for grain-yield data of 18 durum wheat genotypes grown in 14 environments (location-year combinations). The main objectives were to use the different statistical models to evaluate GE interaction for grain yield in durum wheat and to investigate the effect of some climatic variables on the interactions. The main effect of environment, genotype and GE interactions were significant (p < .01), and accounted for 85.1, 0.8 and 6.8% of total variation, respectively. Using the cluster and discriminant analyses, a pattern map developed simultaneously for clustering of stability methods and genotypes, which allowed identifying seven genotypic groups for genotypes and four groups for stability methods. The different stability groups explained genotypic performance differently, with or without respect to yield. The AMMI stability value, Wricke's ecovalence (Wi), Shukla's stability variance (σ 2 ), Perkins and Jinks's (β and Dj) indices, joint regression parameters (bi and S 2 di , R 2 ) and Tai's stability (σ and λ) methods did not correlate with genotypic mean yields, while the dynamic stability GGE distance and superiority measure (Pi) showed significant positive correlations with genotypic mean yields, showing selection based on these two methods would improve yield stability and performance. Using the applied methods, the breeding lines G12, G13 and G7 showed high mean yield and stability performance. The results also showed that the GE interactions were mostly influenced by the climatic data of rainfall, freezing days, minimum temperature and relative humidity. Although the multivariate methods provided valuable information on GE interaction, the univariate methods seem to be useful alternatives to complement improving screening efficiency.

No abstract

No abstract