Extending genomic prediction to future climates through crop modelling. A case study on heading time in barley

Integrating genomic prediction (GP) and biophysical crop models has the potential to support plant breeding in defining adapting strategies to climate change. However, whether this integrated approach can actually broaden the prediction domain to unexplored environments is still unclear. We showed how crop models can extend GP to new environments and capture genotype-specific response to future climate conditions.

Days to heading (HD) in spring barley was used as a case study, given the primary role of phenology for crop adaptation. Phenotypic and genomic (50 K Illumina SNP array) data for 151 two-row genotypes were used, with HD determined on 17 site × season combinations spread from the Mediterranean basin to Northern Europe. A dedicated modelling solution was developed by integrating approaches from the WARM and WOFOST crop models, with optimization algorithms used to derive accession-specific values of most relevant model parameters. GP was carried out on the model parameters with the R package rr-BLUP, and the resulting markers-derived parameters were used to simulate heading date with the crop model explicitly reproducing G × E × M interactions. Prediction accuracy for unknown genotypes was evaluated through ten-fold cross validation, whereas the capability of crop models to extend GP to unexplored environments was evaluated with a leave-one site-out cross validation.

Results were encouraging, with average R², RMSE and Nash-Sutcliffe efficiency for unknown genotypes in unexplored environments equal to 0.97, 9.27, and 0.94, respectively. Accession-specific HD under future climates (20 seasons centred on 2060 generated for the IPSL-CM6A-LR realization of SSP3–7.0) highlighted large G × E × M interactions, confirming the potential of integrating crop modelling and GP to supporting breeding programs targeting adaptation to climate change.