Genotype by environment interactions affecting grain sorghum. II. Frequencies of different seasonal patterns of drought stress are related to location effects on hybrid yields

Abstract

Genotype × environment (G×E) interactions due to variation in soil moisture and rainfall complicate the interpretation of sorghum hybrid performance trials over locations (L) and years (Y). This paper aims to use pattern analysis to explain measures of the G×L interaction for yield, and whether these can, in turn, be explained using simulation models to determine the occurrence of environment types (within-season patterns of drought). The aim of this work is to simplify the analysis of G×E by explaining it in terms of interactions of genotypes with environment types (ET) that are not ‘fixed’ to locations and years. In a sequential analysis of 17 seasons, 18 locations were separated into groups that tended to represent either the northern (i.e. central Queensland, CQ) or southern Queensland (SQ) regions. For a subset of 6 locations, ordination partially explained differences among locations as being related to latitude ( r = 0.88) and rainfall ( r = -0.46), but they were better related ( r > 0.9) to the frequencies of 3 stress ETs as determined by long-term crop simulations. These 3 environment types were: ( 1 ) low stress (occurring in 33% of seasons); ( 2 ) severe terminal stress with an early-season (9%) or midseason time (29%) of onset; and ( 3 ) intermediate terminal stress with a midseason (9%) or late-season (20%) time of onset. Low stress ETs were more common in two SQ locations than in CQ. Stress ETs as defined by simulation models and pattern analysis had more consistent relationships with simulated yields than did the fixed descriptors of locations and years. Sorghum hybrid trials for broad adaptation in Queensland should include locations at least from each of the 2 regions and the results should be interpreted in the context of the season in which they are conducted. To match the long-term patterns in the 6 locations of the analysis, trial yields would need to sample from at least 3 yield ranges: <1 t/ha, 1–3.5 t/ha, and >3.5 t/ha. Additional seasons of testing are likely to be required when the locations used during a season do not adequately represent the target population of environments over all locations and years.