New technologies based robotics and data analysis are starting to be used for the treatment of disease in horticulture field. Such autonomous systems evolving in dynamic environment need automatic operation and control based on dynamic scheduling. To do so, it is necessary to predict the behaviour of this environment to better control these autonomous systems. This paper aims to define the behaviour of plants infection by mildew disease in greenhouses in order to optimise the robotized treatment, which should be fully autonomous. We propose a Markovian approach to model individual plants behaviour and their interactions to predict the dynamics of mildew disease in greenhouses. A multi-agent based simulation is implemented to validate the model. Simulation results have shown the efficiency of the proposed approach to reproduce the behaviour.