In an industrial context of web-based decision support system for croppers, agronomists in our firm design crop models using a scientific programming language. Then in a second step this model is transposed in Java by computer engineers. Due to keeping both codes in evolution, this double implementation causes losses in cost production. To avoid such costs, this paper proposes to apply model driven engineering techniques to identify a metamodel for the domain of mechanistic modelling in agronomy. We started by a reverse engineering of three existing model implementations. The resulting metamodel is the basis for the definition of a mechanistic model factory dedicated to agronomists. This visual tool will offer automatic model linking and data conversion features: they rely on the definition of metadata associated with the inputs/outputs of the different models and on the use of adapters. These automation possibilities should grant easy design capabilities and optimal model reuse in the future. The designed models are linked to the metamodel proposed thus permitting the generation of a ready-to-use Java implementation of the mechanistic model.