This communication tends to elaborate a plea for the necessity of a specific modelling methodology which does not sacrifice two modelling principles: explanation Micro and correlation Macro. Actually, three goals are assigned to modelling strategies: describe, understand and predict. One tendency in historical and spatial modelling is to develop models at a micro level in order to describe and by that way, understand the connection between local ecological contexts, acquired through local ecological data, and local social practices, acquired through archaeology. However, such a method faces difficulties for expanding its validity: It is validated by its adequacy with local data but the prediction step is unreachable and quite nothing can be said for places out where. On the other hand, building models at a far larger scale, for instance at the continent and even the world level, enhances the connection between ecology and its temporal variability. Such connections are based on well-improved theories but lower the " small causes, big effects " emergence corresponding to agent-based approaches and the related inherent variability of socio-ecological dynamics that one can notice at a lower scale: for instance, the emergence of social innovations can be simulated only as an input parameter. We then propose a plea for combining both elements for building large-scale modelling tools, which aims are to describe and provide predictions on long-term past evolutions, that include the test of explaining socio-anthropological hypotheses, i.e. the emergence and the spread of local social innovations.