Dynamic patterns of components of genotype × environment interaction for pod yield of peanut over multiple years: A simulation approach

The relative importance of the genotype × year (G × Y), genotype × location (G × L) and genotype × location × year (G × L × Y) interactions has significant implications on the testing strategy of crop breeding lines. The goal of this study was to examine the dynamic patterns of these three interactions for pod yield of peanut using a crop simulation model. Pod yields of 17 peanut lines in the early-rainy, mid-rainy and dry seasons at 112 locations covering all peanut production areas in Thailand were simulated for 30 years (1972–2002) with the Cropping System Model (CSM)-CROPGRO-Peanut. Combined analyses of variance were preformed for individual seasons and for overall seasons, with the number of year incrementally increasing from 2 to 30, and the relative contributions of the individual sources of variation were determined. This procedure was repeated four times with different starting years. The results showed that the environmental effects accounted for the major proportion of the total yield variation, followed by the genotype effects, while the genotype × environment (G × E) effects were rather small. The contributions of the individual sources changed as the number of years in the analysis changed. Increasing number of years in the analyses resulted in an increase in the magnitude of the G × Y and G × L × Y interactions, but a decline in the G × L contribution. The contributions of the G × Y and G × L interactions were greater and more fluctuated in the dry season, while those of the G × L × Y interactions were greater in the mid-rainy season. Notable increases in the G × Y interaction in the dry season were observed in certain years. The decline in the G × L interaction with increasing number of years was closely associated with the increase in the G × L × Y interaction, and both became stable when 6 or more years were included. Several cross-over in the ranks of peanut lines for mean pod yield in two contrasting years were also observed for the mid-rainy season. These results raise a question on the effectiveness of the strategy for using locations to replace years in varietal testing that is normally employed by breeders. The practical limit of multi-year evaluation of crop breeding lines could be overcome by the use of a crop simulation model.