This contribution depicts an approach enabling agricultural production systems to be represented at various spatial-temporal scales and organization levels: single plot or workshop, whole farm, group of farms within a territory, supply chain. This approach is based on an 'Action-Flow-Stock' ontology according to which each production unit is represented by a stock subject to filling-emptying actions controlled by conditions derived from states or events locally observed on some processes. Whereas stocks are computed as continuous variables, actions are represented by dynamical discrete functions. The models built with this ontology, what-ever the formalization framework used to implement them (systems dynamics, multi-agents systems, timed automata), aim at dynamically simulating both the material flows and the human activities within agricultural production systems. The intended use of such models is to help agricultural stakeholders design and assess actual or potential management strategies in the light of the new concerns brought by the sustainability issues.