En savoir plus

Abstract :

Progresses of high-throughput phenotyping, genomic prediction and modelling may jointly provide novel tools for breeding schemes and variety recommendation to farmers, in a context of climate change and water scarcity. Integration of these approaches requires new methods that tackle genotype x environment interactions and evaluate the comparative advantages of varieties in contrasting environmental scenarios. In this thesis, we developed and evaluated a new approach for predicting maize leaf area and grain number across multiple environments, which combined genomic prediction models, novel phenomics methods and a crop model (Sirius Maize). The latter can simulate, based on explicit physiological processes, yield and other traits for multiple genotypes in a large range of environmental conditions, provided that genotype-specific parameters are estimated for many hybrids and fed to the model. We tested our approach by using three panels of maize hybrids: a diversity panel, a panel that captures the genetic progress and a panel of recent hybrids, with 246, 56 and 86 hybrids, respectively. Genotype-specific traits were measured in indoor or field experiments, related to plant phenology, architecture, leaf growth, responses to soil or air water status, and maximum grain number. We first showed that traits measured indoor can translate to the field, either directly or via the use of a model. Then, we showed that they can be successfully estimated, for a larger number of hybrids, via genomic prediction. Finally, we converted these traits into genotype-specific parameters, either via explicit equations or via scaling. Appreciable prediction accuracies were achieved by the crop model for leaf area index and grain number of studied hybrids, simulated in 9 and 21 experiments, respectively, with contrasting environmental conditions. In the thesis, we discuss the relevance of each of these steps, needed for integrating the knowledge from genetics, ecophysiological models and phenomics. We also identify areas to improve the approach and its prediction accuracy and for further applications in a plant breeding or variety recommendation context.