As example, we studied the peach-brown rot system and used the ¡®Virtual Fruit¡¯, a process-based model that has been extensively tested, to perform virtual experiments. The challenge was to optimize the trade-off between antagonistic criteria of major importance for both fruit quality (increasing fruit mass and sweetness) and sensitivity to brown rot (decreasing skin density of cracks) in four different cultural scenarios. A multiobjective evolutionary algorithm, namely NSGA-II, was applied to solve this multiobjective optimization problem based on the ¡®Virtual Fruit¡¯. The optimized variables were six parameters of the ¡®Virtual Fruit¡¯, selected on the basis of a sensitivity analysis.
This optimization method provided a large diversity of solutions among which the decision-maker can choose the best suited trade-off between criteria according to a particular objective. Most of the optimized solutions were distributed along Pareto fronts suggesting a good convergence of the algorithm. Moreover, it also provided some solutions located in non-crowded zones which constitute some original alternatives for the final decision-maker.
The results confirmed the strong antagonism between the criteria considered. Large fruits had a weak sweetness and high crack density and for a given mass, those with improved sweetness had higher crack density. In a current breeding scheme, fruit mass would be the only criteria considered but alternative schemes could be considered for future, favouring organoleptic quality or environment friendly practices. In those cases, some interesting optimized solutions were identified.
The work described in this paper supports that multiobjective evolutionary algorithms should be used to optimize parameters of process-based models and help identifying trade-off in complex systems. The use of the ¡®Virtual Fruit¡¯ to design sustainable production systems combining genotypes and innovative practices is further discussed.