Abstract :

Genomic prediction is effective in simulating the average performance of varieties from genotyping information, but does not predict well the variation in yield for each variety according to local environmental conditions (genotype x environment interactions). This situation could change thanks to two sets of recent advances. (i) The rapid development of sensor networks and environmental grids makes it possible to characterise the environment in any agricultural plot, and thus to use yield simulation models. (ii) Phenomics allows the characterisation of hundreds of varieties, in particular the response of key physiological traits to environmental conditions, which allows simulation models to be fed with genotype-dependent information.

The aim of the thesis is to predict the behaviour and simulate the yields of maize varieties, each characterised by allelic combinations; in hundreds of locations in France and Europe, characterised by environmental conditions measured or deduced from environmental grids. To do this, it will combine (i) the measurement of genotype-dependent physiological traits by phenomena, (ii) the genomic prediction of these traits, (iii) the environmental characterisation of plots, and (iv) a crop model parameterised on the basis of the phenomena data The main areas of application for ARVALIS are the enrichment of the annual characterisation of maize varieties in post-registration, thanks to an in silico evaluation of their performance and stability in current or future scenarios. In addition, this approach would make it possible to identify varieties with complementary interactions that would be recommended in clusters for better stability.