|RAGHUVEER, SRIPATHI - University Of Wisconsin|
|CONAAGHAN, PATRICK - Teagasc (AGRICULTURE AND FOOD DEVELOPMENT AUTHORITY)|
|GROGAN, DERMOT - Department Of Agriculture, Food, And The Marine|
Submitted to: Crop Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 3/11/2018
Publication Date: 6/21/2018
Citation: Sripathi, R., Conaghan, P., Grogan, D., Casler, M.D. 2018. Modeling genotype x environment correlation structures in long-term multi-location forage yield trials. Crop Science. 58:1447-1457.
Interpretive Summary: Variety trials are the last stage of a plant breeding program designed to produce new varieties for producers. As such, both precision and accuracy are critical aspects of field-based variety trials. As part of a long-term project between USDA and its Irish counterpart, TEAGASC, this study aimed to create a more cost-efficient variety testing scheme as the last phase of the Irish perennial ryegrass breeding program. The study was based on long-term data collected over 11 years at five locations in Ireland. The study showed that the Athenry location was the least unique, always highly correlated with one or more of the other locations. Elimination of the Athenry location is expected to reduce trial costs by 20% with only an 8% reduction in trial efficiency, which could be made up by a slight increase in the number of field replicates at each of the other four locations. Using modern statistical computing methods, these results will allow the Irish to improve the prediction accuracy of forage yield estimates from field trials at a reduced cost.
Technical Abstract: Genotype x environment interactions (GEIs) are a critical aspect of field experiments to evaluate yield of forage cultivars. The objectives this study were to (i) model genotypic effects across years, locations, and harvest years of forage yield trials using variance-covariance structures, (ii) to predict genotypic performance across different environments, and (iii) to compare relative efficiency of different cost reduction scenarios based on locations, harvest years, and replicates per sowing year. This study is focused on long-term ryegrass yield trials conducted by the Department of Agriculture, Food and Marine (DAFM) in Ireland. Log likelihood responses of different models indicated that unstructured variance-covariance models (UN†) were superior to heterogeneous variance structure models (DIAG†) to model genetic effects across locations. The UN† models also improved predictability compared to DIAG† models to estimate genetic effects. Among locations, Athenry (representing west of Ireland) was consistently clustered with other locations in both first and second harvest years of late and intermediate maturity trials. For resource allocation, cost reduction via reducing the number of harvest years was not effective for the DAFM trials. One location or four replicates per sowing year can be excluded from the DAFM trials generating a 20% reduction in cost with only an 8% reduction in efficiency. Our results suggested that modeling genetic effects using UN† models not only allow us to group locations based on genetic correlations but also to improve prediction accuracy of genotypic effects across different locations and to reduce costs.