The current research is undertaken in the high-risk environment of poor farmers in Mali (West Africa) where agricultural biodiversity is crucial and people are always aiming to be more efficient in their use of the natural resource base. It introduces participatory modelling as a way to strengthen human and social capital in ways that support the management of the natural capital, including plant genetic resources. During a first stage of the research project, data was gathered and processed in a suitable spatial framework. This paper gives an insight into the methodology we use during the second stage of the project, which introduces Agent-Based Models (ABM). We present a first prototype that formalize existing assumptions and data within a dynamic framework. The development of this prototype has resulted in the identification of different decision contexts around farmers' sorghum seed management. It has led us to the next stage of our research that is to gain more knowledge about these decision contexts by the mean of fieldwork and role-playing game sessions. These information will be used to produce a second ABM resulting from a common understanding of farmers' seed management.