How to enviromically predict breeding genotypes as if they were commercial cultivars?

Early trial rankings may not translate into cultivar‐stage performance, especially when genotype‐by‐environment interaction changes the target response across breeding stages. This study implemented and evaluated a bi‐trait genomic–enviromic reaction‐norm framework to predict breeding genotypes “as cultivars” by combining genomic relationships, environmental covariates, and covariance between trait‐stage expressions. Breeding trial yield (BTY) and cultivar target yield (CTY) were treated as distinct but genetically related responses, with variance–covariance components estimated through derivative‐free restricted maximum likelihood optimization. An illustrative simulation was used to develop the model, inspect its estimates, and visualize behavior. In this simulation, the single nucleotide polymorphism–based relationship matrix produced a more coherent BTY–CTY covariance structure than the identity matrix and enabled spatial recommendation maps. An empirical CassavaBase case study evaluated BTY‐to‐CTY prediction using records from multiple genotypes and locations, including bridge genotypes observed at both trait‐stages. In this setting, three models were compared: a genomic BTY‐only baseline, a bi‐trait genomic model, and a bi‐trait genomic–enviromic model using environmental principal components. The genomic–enviromic model showed its strongest advantage at the genotype level under environmentally supported leave‐one‐CTY‐location‐out (LOLO) validation, increasing prediction accuracy from 0.491 to 0.797 and reducing mean absolute error from 26.7% to 18.2% in LOLO‐ strict relative to the BTY‐only baseline. These results show that breeding‐stage genotypes can be projected onto the CTY scale across target environments by combining genomic, enviromic, and bi‐trait information.