Abstract
Agroforestry is a promising agricultural system to face climate change. It combines trees and crops (field crops, meadows, etc.). However, designing an agroforestry plot is a difficult task, since it requires consideration of both spatial and temporal dynamics. We conducted surveys of agroforestry advisors on design approaches, which revealed a diversity of methods and tools used. But we identified a lack of tools for visualizing the future appearance of plots and the ecosystem services provided.In this thesis, we propose a computer processing chain to assist advisors in agroforestry plot design workshops, and propose a new approach to agroforestry system representation.In our chain, a first module captures information from a physical model of an agroforestry system, using a deep learning system to identify the model's components and their positioning. A second module builds an abstract representation of the system using combinatorial maps. This modeling, based on the system's adjacencies and structural hierarchies, proposes a new formal graph representation of the agroforestry system, and we show that the dual graph can be used to represent and estimate the system's ecosystem services. Finally, a third module, exploiting Unity's real-time potential, enables interactive visualization in augmented reality of the variety and complexity of plantations, including the representation of plant growth and the production of ecosystem services.Two prototypes have been developed for embedding into mobile devices. One application superimposes virtual images on physical model views during the design workshop, and the other, for outdoor use, integrates virtual images directly on the plot views. A panel of users is currently testing their use.This interdisciplinary work (agronomy, computer science, human-machine interfaces) will be useful to advisors and farmers in designing agroforestry plots, taking into account the relationships between their structure and functions and, ultimately, the ecosystem services produced. The systems designed and their future evolution will thus be better adapted to farmers' constraints and objectives.The proposed structure-function representation of systems also opens up new research prospects as a support for participatory design approaches, and provides a framework for mechanistic modelling of the functioning of and interactions within agroforestry systems