The tragedy of the common is a dilemma often used to describe a scenario in which a group of users share a common limited pool of a certain resource. One of the most vital examples of these common pools is groundwater. In the last decades aquifers are over exploited, and about 70\% of groundwater withdrawals is used for agriculture globally. Water demand management often addresses this issue with a centralized, top-down approach. Irrigation networks however, are often organized as decentralized systems, in which small farmers have access to their own well and their abstractions are difficult to monitor. One of the main problems of these kind of systems, is the achievement of sustainable use of resources with pro-active participation of all the stakeholders. A deep and comprehensive understanding of the system and the behaviour of its stakeholders is necessary for developing effective tools for sustainable water management. The objective of this study was to develop an Agent Based Modelling (ABM) tool able to simulate not only the hydrological aspect, but also the social and behavioural ones. This was done by designing an hybrid ABM model, which considers in the decision process of the agents both behavioural strategies and feedback from the system based on their satisfaction. Participants were introduced to a simulation game developed to obtain a better insight on the real behaviour of individuals. This study showed that a hybrid ABM model is a promising tool to analyze and manage decentralized irrigation systems. The present approach showed the capability to generate time-efficient simulations of possible agricultural scenarios and demonstrate that cooperation is vital to maintaining the sustainable levels of water resources and that communication and incentives are important factors that enhance the control and coordination of the resources. The proposed methodology matches the paradigms of the current IAHS scientific decade ‘Panta Rhei’. Since the model is highly sensitive to the behaviors of the agents, data acquisition is a necessary element for improving model performance.