This paper presents the 'Environment and landscape geo-knowledge' (GEMINAT) project which aims to build an infrastructure favouring the cross-analysis of spatio-temporal heterogeneous data sources recorded at the Chizé environmental observatory since 1994. From a case study, we summarise the difficulties encountered by biologists and ecologists when maintaining and analysing collected environmental data, essentially the spatial organisation of the landscape, crop rotation, and wildlife data. We show how a framework which uses a spatio-temporal ontology as a semantic mediator can solve challenges related to the analysis and maintenance of these heterogeneous data.