Genotype by environment interactions affecting grain sorghum. I. Characteristics that confound interpretation of hybrid yield

Past sorghum hybrid trials in north-eastern Australia have detected substantial genotype by environment (GxE) interactions for yield in sampling a variable target population of environments (TPE) that is affected by spatial and seasonal differences in crop water supply. Three datasets, comprising yields of commercial and final stage experimental hybrids and covering 9-17 years (Y) and up to 30 locations (L), were analysed to quantify variance components for trial error, genotypic (sigma(g)(2)), and GxE (sigma(gl)(2), sigma(gy)(2), and sigma(gly)(2)) interaction effects. Whereas trial means varied 2-3-fold across seasons, a greater range was estimated for variance components of trial error (range of 0.05-0.5), G (0->0.3), and GxL interaction (0.05->1.0). There was substantial seasonal variation in the ratio of sigma(g)(2) to (sigma(g)(2)+sigma(gl)(2)), and in two datasets, 73% of the seasonal sigma(gl)(2) was due to poor genetic correlations among locations. This implies that any given set of hybrids in a random set of locations would be ranked differently from season to season. Analysis of locations over years detected 90% of the total GxE interaction as GxLx Y, rather than GxL or GxY, although this was reduced by accounting for genotype maturity. To achieve repeatabilities of >80%, trials would need to be conducted over at least 5 years and 20 locations per year. The variable and unpredictable nature of much of the GxE interaction in the region implies that broad adaptation to different water regimes is required, unless prior knowledge of the seasonal weather can be used to choose 'narrowly adapted' cultivars. With current approaches, a large sample of environments is needed to identify such hybrids, and testing across locations and years is equally important. Alternative breeding strategies based on classifying environment types are discussed.