This paper presents the use of a data mining algorithm to extract co-variation patterns from ecological data. The system of interest is a network of shallow fish-ponds in the region of the Dombes, in France, from which we gathered more than 4 years of measures. These data were analysed and the extracted patterns were structured in a hypergraph that allowed us to infer the most pertinent parameters from which construct a dynamical model of a major vegetation/plakton relationship.