Modern breeding programs are accumulating large historical datasets as annual genotyping and field trials are routinely stored in databases. These investments in phenotyping and genotyping can be leveraged far after the year they are evaluated. Field locations year to year can have very different environmental factors that have a significant impact on performance.

Breeders try to control extreme environments by performing multi-environment trials and evaluating lines across many different environmental conditions and across years to best understand performance and stability. Multi-environment genomic prediction models can be a great tool for predicting genotype performance in specific historical environments, or to leverage sparse testing designs to optimize field resources. The difficulty lies in the unreliability of local climate year to year. Our research suggests that a sparse testing design is the best way to leverage shared genomic information to get the best prediction accuracies.

However, publicly available climate information can be used to inform new or unknown environments. Using a gaussian distance kernel to model environmental relatedness generally outperformed using a kinship kernel to inform predictions.

Evidence suggests that the climate data up until mid-season can be a useful tool to provide breeders with in-season predictions without a significant drop in overall PA. Climate and soil characteristics are only part of what make up the total environmental effects, and prediction accuracies can be influenced by other factors such as unusual biotic stresses.

Comparisons of advancement and discard decisions between breeder phenotypic selection and model predicted genetic merit were summarized.