"Our work is motivated by the application of the so-called Model-based Predictive Control (MPC) in the context of irrigation networks. The main obstacles for applying classical centralized methods to such complex systems are the huge computational effort required to solve the corresponding optimization problems and the lack of flexibility in the problem structure's description. In this work we propose the application of the multiagent paradigm to a distributed scheme of MPC. The global optimization problem is firstly decomposed into sub-problems which are then solved in parallel by agents. The communication between agents is then used to guarantee the convergence of this distributed scheme to the global optimum. In addition, agents are given the ability to rearrange and reconfigure themselves in order to improve the fault tolerance of the control scheme. Finally, the proposed approach is validated in a simulated irrigation system made of interconnected canal reaches interconnected through actuated gates."