The way marine ecosystems evolve is the result of numerous interactions between species. These interactions directly or indirectly determine services such as climate regulation, supply of food products or regulation of coastal water quality. Understanding these interactions is therefore key in determining the trajectory of such ecosystems under the influence of global change. Phytoplankton (microscopic algae) is a compartment at the base of trophic webs (“food chains”). It is extremely sensitive to environmental variations and therefore serves as an indicator. Long-term observation networks with high sampling frequencies have been set up since the 1990s.

The creation of models to represent the mutual influences between the different species (such as competition) is therefore of great interest. Using existing sampling data, this thesis aims to propose a method to automatically build models of the interactions between species. The preferred approach will be Learning From Interpretation Transition (LFIT), a collection of logic learning algorithms that produce logical programs from observations, while ensuring their explicability, contrary to common statistical approaches. From these logic programs, a reaction network will be extracted to either temporally simulate the model, or to study the annual fluctuations of the populations of phytoplankton species using abstract interpretation methods.

Detailed subject (in French) here.