Abstract : The mutual influences between human activities and natural resources are at the heart of societal debates. The challenges of water management depend on these interdependences differently according to the local situation. In this context, the hydrology research community is investigating the relevance of its models to represent human activities interacting with water resources. The literature counts many studies in which the complexity of modelled processes concerns hydrology. But human activities such as irrigation are represented in a more or less aggregated way and in a trend or even stationary way over time. As shown by the reflections in progress within "socio-hydrology", it is necessary to better document the various interactions and feedbacks due to irrigation within agro-hydrosystems. Short-term interactions are particularly under-explored. However, in the short term, irrigation management is based on operational constraints, such as those inherent to water distribution, which can significantly impact the future state of crops and harvests. Conversely, the state of crops influences the frequency and spatial distribution of irrigation operations that locally modify the state of water resources. Moreover, most approaches to representing human actions within these systems conceptualize action with its decision phase, often ignoring the operational level. Thus, the question explored in the thesis is the following:

How can we represent at the operational level the actions of irrigators in space and time to take into account in a dynamic and situated way their interactions with the agrohydrological components of the system?

And what could this representation bring to the water management discussions of a specific case study?

We first propose to mobilize the Affordance concept to build an agent-based model (WatASit) explicitly representing the possibilities of action of irrigators in situations of tension for water sharing. Applied to a typical gravity-fed network of the Buëch catchment in the Durance region (France), we show that the trajectories of the agents depend on the evolution of their possibilities during the irrigation campaign and that the analysis of these possibilities helps in the interpretation of individual and collective behaviours. In particular, the consequences of the abandon of the coordination by daily slots of the network, observed during the field surveys, does not seem to impact all the irrigators in the same way by reinforcing the spatial inequalities between the upstream and downstream parts of the network.

We then propose the COPAT (COupling Plant and Agent Trajectories) framework to couple a crop model at the plot scale (Optirrig) and the WatASit model at the irrigation network scale. The temporal consistency of the coupling is based on the derivation of the crop model as a daily function. At each daily time step, the irrigation received by each plot is determined by the operations of the agents constrained by the sharing of water within the collective network. An
earlier water stress is observed compared to irrigation that would not depend on such sharing, but the coordination of the network by daily slots tends to limit this stress. Finally, we propose COWAT (COupling Water and Agent Trajectories) as a coupling methodology with the spatialized hydrological model J2000 and highlight some points of vigilance to ensure spatial coherence at a fine scale. In the end, the Affordance concept connects with a wider interdisciplinary debate on the representation of human actions and hydrology is thought in interaction with these actions and the operational issues of water management.

Keywords : collective irrigation; operational management; agent-based model; crop model; distributed hydrological model; Buëch.