Abstract
Ecosystems unfold over time in response to successive events that modify their components. These events, whether random or not, are the phenomena by which ecosystems assemble and disassemble. Each sequence of states - or trajectory - is a possible history of the ecosystem. Yet, if these events, or their order, had been different, the future state could have been different. However, the order of events, due to the limits of our knowledge, is often uncertain. When this uncertainty is important, it is therefore reasonable to conceive the dynamics of the ecosystem as a set of possible trajectories. If the set of possible trajectories were known, it could be useful to know if a desirable state is attainable, and if so, by which path(s).In this thesis, we propose a modelling approach named EDEN (for Ecological Discrete-Event Networks) to answer this question. It is presented in the form of a formalism whose variables are qualitative and whose values are modified by "if-then" rules representing the events of the ecosystem. These rules are executed one by one, in a non-deterministic way and without any probability. This non-deterministic and non-probabilistic approach has therefore been called possibilistic. The dynamics are then presented in the form of a state-transition graph whose dynamic properties (such as the fact that a species can persist) can be verified.First, a brief history of dynamical modelling in ecosystem ecology is presented, with a particular focus on qualitative modelling approaches, to which EDEN belongs to. Based on the limitations of existing approaches to answer certain questions, we propose the EDEN approach and justify the relevance of its properties with simple examples.Then, the EDEN approach is illustrated on semi-arid ecosystems in East and West Africa in order to know if certain states of interest are reachable (such as the maintenance of vegetation in savanna, or soil fertility in an agroecosystem). To do this, we present and use tools that have not been used much in ecology until now, such as summary graphs and temporal logics. For a given scenario, these tools enable to define (1) the possible transitions between the states of interest, (2) the conditions under which they can occur and (3) the events responsible for these transitions. In East Africa, the EDEN approach was able to determine the long-term effects on vegetation and human activities of reduced/increased surface water availability, indicating a strong indirect influence of the latter on woody vegetation. In West Africa, using EDEN, it was possible to determine the conditions that allowed small-scale producers in southwestern Burkina Faso to develop persistent agropastoralism, as well as the events necessary to achieve such a state.Finally, we discuss the contributions of the EDEN set of models to various fields of ecology, as well as possible improvements and bridges to be built between ecology, systems biology and theoretical computer science.