Our objective was to develop a model (SUNFLO) which can help to improve genotypic assessment in the sunflower crop. The present work aimed at identifying, quantifying and modelling the phenotypic variability of crop performance in response to the main abiotic stresses occurring in the field (light, temperature, water, nitrogen) but also in the expression of genotypic variability.
We therefore include just enough genetic information to enable the model to be used with new genotypes. Each genotype was thus defined by chosen phenotypic traits which were transcribed into a set of 12 genotype-specific parameters.
The model's performance was evaluated in both specific field experiments and generic multi-environment trials (MET). The first evaluation assessed model robustness: no variables had a large prediction error, indicating that the final output error results more from poor prediction for all variables than from error compensation. An ANOVA on the simulated MET dataset showed that although the model simulates less variability than in reality (60 % ), there was genotype–environment interaction and the ranking of the ANOVA factors was identical in both observed and simulated networks. The model's accuracy was sufficient to discriminate between genotypes from different breeding periods, but was similar to the difference in performance between actual genotypes (0.2 t ha−1).
To improve the understanding of crop physiology and crop–environment interactions, this kind of model shows weaknesses, especially when dealing with environmental stress integration or biomass allocation. On the other hand, SUNFLO seems sufficiently robust to estimate the influence on yield of breeding traits or to explore new management practices.