Connexion Agenda

Valorisation > Conférences et soutenances

Séminaire Clément Aldebert, MIO (Mediterranean Institue of Oceanography à Marseille)

27 juin 2018, Salle -115L

Facing current socio-environmental issues, such as species extinctions and loss of ecosystem services, requires to make ecological predictions with a level of accuracy that is not yet achieved. Uncertainty arises in predictions made by mathematical models, which are perceived as objective tools but remain simplified representations built on somewhat arbitrary choices. Indeed, when several scales of organization, space and/or time are entangled in a complex process like predation, its description by equations is necessarily uncomplete. This approximation implies that the same process can be described by different equations depending on the modeller assumptions. Whereas different assumptions about the emerging process shape (e.g. density-dependence, group behaviour) have been widely considered in the literature, almost no attention was given to the choice between similar equations (i.e mechanisms) to model the same process shape (e.g. process rate increases with population abundance). However, recent studies showed that this choice can deeply affect model predictions, a problem coined structural sensitivity.

This seminar will be a journey through these recent developments about structural sensitivity in ecology. After explaining this problem from a biological perspective, I will show how this "grain of sand" can affect the predicted dynamics and resilience of a predator-prey system. Then, some of these results will be upscaled to realistic food webs with tens of interacting species. Finally, I will present a possible solution to deal with structural sensitivity by taking into account some basic (but too often neglected) features : mass-balance (with an explicit resource dynamics), and individual metabolism (maintenance, energy reserves). The seminar will end with some perspectives toward a probabilistic use of predictions based on deterministic models in order to present predictions together with their associated uncertainty.

publié le