The global wheat industry is threatened by variability in the rate of productivity growth due to factors such as climate change, biotic and abiotic stress. Designing new wheat varieties with genetics tailored to target environment and management will provide flexibility for growers aiming to maximise productivity and contribute to the resilience and profitability of wheat crops.

This can be achieved by leveraging both genomic and interaction effects, including epistasis (G#G), environment (GxE) and management (GxExM) in evaluating genomic merits for breeding selection. In practice, understanding the correlation from these interactions, particularly the impact weight of each environment covariate is difficult and has been less considered. We have developed a new statistical and machine learning framework to address this challenge.

Specifically, we extended the traditional linear mixture model, GBLUP, by leveraging the flexibility and scalability of Bayesian stochastic Gaussian Processes (GP) to integrate genomic, environmental, and non-additive genomic interaction effects from data collected at farm and landscape scales to robustly evaluate genetic merits in different environments.

The model also provides biological insights into how the genome together with environment (weather, soil and disease) interact to drive wheat yield efficiently. We demonstrate the approach using a large-scaled data collected across an Australian wheat breeders’ trial network (AGT) consisting of ca. 20K genotypes with matched genomic SNP data and two-stage yield estimates (ca. 75K), together with environmental, soil and biotic covariates spanning max. 79 environmental trials (ET: site + year).

Model performance was evaluated against four breeder scenarios, predicting observed genotypes into observed environments; unobserved genotypes into observed environments; observed genotypes into unobserved environments and unobserved genotypes into unobserved environments.

By directly incorporating environmental covariates, this framework enables breeders to predict genotype merit into new ET, helping to extend evaluations beyond their current sites and make effective decisions to support continued genetic gain under environmental uncertainty.