Biological reality and parsimony in crop models—why we need both in crop improvement!

The potential to add significant value to the rapid advances in plant breeding technologies associated with statistical whole-genome prediction methods is a new frontier for crop physiology and modelling. Yield advance by genetic improvement continues to require prediction of phenotype based on genotype, and this remains challenging for complex traits despite recent advances in genotyping and phenotyping. Crop models that capture physiological knowledge and can robustly predict phenotypic consequences of genotype-by-environment-by-management (G×E×M) interactions have demonstrated potential as an integrating tool. But does this biological reality come with a degree of complexity that restricts applicability in crop improvement? Simple, high-speed, parsimonious models are required for dealing with the thousands of genotypes and environment combinations in modern breeding programs utilizing genomic prediction technologies. In contrast, it is often considered that greater model complexity is needed to evaluate potential of putative variation in specific traits in target environments as knowledge on their underpinning biology advances. Is this a contradiction leading to divergent futures? Here it is argued that biological reality and parsimony do not need to be independent and perhaps should not be. Models structured to readily allow variation in the biological level of process algorithms, while using coding and computational advances to facilitate high-speed simulation, could well provide the structure needed for the next generation of crop models needed to support and enhance advances in crop improvement technologies. Beyond that, the trans-scale and transdisciplinary dialogue among scientists that will be required to construct such models effectively is considered to be at least as important as the models.