Integration of Environmental Covariates in Trial Analysis Using a Crop Model and Genomic Selection

Use of environment covariates in trial data analysis has been a subject of interest for a long time but is limited by the same problems as those addressed by GS: a large number of correlated predictors each explaining a small amount of the total variance, in addition to a non-linear response of genotypes to stresses. One strategy is to consider crop model parameters as traits with higher heritability and a less complex genetic architecture to better understand and predict economic trait behavior. Those approaches are potentially powerful but require specific phenotypic data. Here, we take a different approach that merges crop modeling and GS, starting from the trait of interest rather than from a crop model. This approach uses a crop model to perform a physiological integration of the environment data to derive stress covariates by development stages, leveraging agronomic knowledge for better trial analysis. In large multi-environment trials, the analysis is usually complicated by variation in phenology that could be confounded with genotype performances. Those covariates can be further used to explicitly model G*E. Using the sparse group lasso, we show how a better environment mean modeling can be achieved with inclusion of environment covariates, providing a first step toward merging GS and crop modeling approaches. The usefulness of the new proposed methods is demonstrated using a large winter wheat dataset, representative of the type of data available in a large-scale commercial breeding program.