Differential performance and changes in the rank of genotypes across environments make selecting the best genotypes challenging in both Multi-Environment Trials (MET) and Target Population of Environments (TPE).

Therefore, it is relevant and crucial to study the causality behind rank changes, known as genotype by environment interaction (GEI). With the integration of genomic selection in wheat breeding schemes, the need to develop genomic prediction models that incorporate GEI for better selection became apparent. Observed variance-covariance structures fitted in mixed model (GBLUP GxE) settings have been proposed.

However, such strategies are limited in their ability to predict beyond the tested environment boundaries. In contrast, the idea of using environmental covariates (EC) either in variance-covariance structure (GBLUP GxEC) or random regression model (RRM) has gained more attention due to its ability to better connect genotype to phenotype and its advantage in predicting genotype performance in untested environments.

Our main goal is to develop an improved genomic prediction model that integrates GEI and EC. We hypothesize that using EC will provide better characterization of environments and enhance prediction power for both tested and untested environments. In addition, using RRM will enable better characterization of genotypes. We are using 11 years grain yield dataset (2010-2020) from 4 locations in the National Wheat Breeding Program (WBP) of Uruguay.

We will fit different genomic prediction models, including GBLUP, GBLUP GxE, GBLUP GxEC, and RRM, and test their predictive ability using different cross-validation schemes (CV1, CV2, and CV0). The best selected model could be utilized in the breeding program for selecting candidates in wheat breeding pipeline considering GEI.

Along with that, RRM fitted with EC could help us for selection of adapted materials in a climate change context.