Parametric model based assessment of genotype×environment interactions for grain yield in durum wheat under irrigation

The objectives of this study were to assess the genotype (G) by environment (E) interaction of grain yield of durum wheat(Triticum durum Desf.) based on parametric models, additive main effects and multiplicative interaction (AMMI) and joint linear regression models; and compare the relative efficiency of the two models in explaining the GE effects. Twenty-three genotypes were evaluated across 12 environments (location-year combinations) in 2003 and 2004. Combined analysis of variance showed that the environment (E) accounted for a high percentage of sums of squares (remaining after removing the sums of squares due to error and replications). The genotypic variability of grain yield among genotypes was small. The best genotype 6 (DBSP02/8) out yielded the check by 0.24 ton/ha. Based on full AMMI model analysis, AMMI-1 was found to explain up to 94% of the main and interaction effects, and AMMI-2 was found to fully capture target percentage sums of squares in the GE interaction pattern. The biplot based on the first bilinear AMMI model terms indicated that genotype 7 (DBSP02/9) and genotype 19 (DBSP03/16) could be suited for cultivation across the test environments.  However, no genotype had superior performance in all environments. Reitrivier normal planting date in 2003 (E403) was the most favorable environment for yield, whereas, Upington in 2003 (E603) was the least favorable one. Model comparison criteria showed that AMMI model was superior to joint regression model in terms of its predictive accuracy and efficiency of explaining the pattern of GE sum of squares. It was concluded that AMMI biplot clearly facilitate identification of mega-environments and cultivars for specific recommendations. The differential response of genotypes observed in this study reaffirms the necessity of multi location evaluations to identify superior and stable genotypes. However, trends in specific adaptation could be detected using the which-won-where pattern of the AMMI analysis, and site-specific breeding may be exploited when feasible.