The functioning of marine ecosystems depends on key processes such as climate regulation and water quality. Phytoplankton, unicellular microalgae at the base of marine food webs, play a central role in these dynamics that are threatened by global change. Although abiotic factors (e.g., temperature) are well studied, biotic interactions (i.e., between species) are poorly understood. Classical machine learning models, while effective at prediction, often operate as black boxes and provide limited biological interpretability. Meanwhile, symbolic and qualitative modeling approaches, which could offer greater explanatory power, remain largely underused in marine ecology. To address this gap, we investigate phytoplankton interactions using an explainable machine learning method, LFIT (Learning From Interpretation Transition), which infers logical rules from observational time series data. Adapting this framework required methodological contributions, notably a species-specific discretization strategy informed by ecological theory. The extracted rules enable the construction of an interaction graph, where edges represent probable interspecies interactions. This graph offers an interpretable representation of the dynamics of the community and helps identify key drivers of the development of phytoplankton.