Leveraging environmental data and dominance effects in genomic prediction may accelerate genetic gain in hybrid sunflower

Genomic prediction improves the efficiency of plant breeding by allowing the performance of a new, untested genotype to be predicted without conducting field trials. There is a dearth of knowledge on best practices for performing genomic prediction in sunflower ( Helianthus annuus L.), a hybrid crop. Our objective is to address this knowledge gap by performing genomic prediction for height, oil content, and yield using 115 hybrids derived from the Sunflower Association Mapping panel grown in ten environments in the northern Great Plains. We investigated three breeding scenarios and compared models using linear, Gaussian, and dominance genomic kernels, models incorporating an environmental relationship matrix (ERM), and five Bayesian models. The most difficult breeding scenario mimics the most common scenario in breeding programs, predicting untested genotypes in untested environments. The most complex predictive model uses an ERM to model covariance in the environment and genotype‐by‐environment terms and a genetic dominance kernel. The ERM was composed of the average precipitation and maximum and minimum temperature in four phenological stages defined by growing degree days. Overall, there was little to no improvement with the inclusion of an ERM in only the environment term. When the ERM was used to model genotype‐by‐environment (G E) covariance, there was an improvement in prediction accuracy for yield. The inclusion of dominance further improved yield prediction accuracy. Our results demonstrate the importance of incorporating environmental variables and dominance in genomic prediction models, and illustrate areas of improvement for sunflower genomic prediction.