An efficient strategy for genomic prediction in new locations via enviromic indexing

Abstract Genomic prediction (GP) is a core component of modern breeding programs, but its accuracy can decline significantly when predicting phenotypes in untested environments, largely due to genotype-by-environment interaction (G × E).eno To improve model generalization under these scenarios, we propose an alternative framework that integrates genomic information and environmental covariates (ECs) data through a novel Ridge-based environmental indexing approach. Unlike traditional methods such as partial least squares (PLS), our method instead uses ridge regression to derive environmental indices from stage-specific ECs, improving both model interpretability and stability. We evaluated the method on four real multi-environment datasets representing diverse breeding contexts. Benchmarking was performed against conventional GBLUP, PLS-index models, and synthetic factorial indices under a leave-one-environment-out (LOEO) cross-validation strategy. Across traits and datasets, our proposed model demonstrated robust and competitive predictive ability, frequently ranking among the top performers in terms of Pearson correlation (COR) and normalized root mean square error (RMSE). This study supports Ridge-based enviromic indexing as a scalable and practical tool for improving genomic predictions in new environments, especially when combined with other data modalities. The approach is readily extensible to untested genotypes and can serve as a foundation for future G × E-aware selection strategies.