Researchers are increasingly using complex models to understand socio ecosystems (SES) with biophysical and social dynamics, and their interactions. The initialization and parameterization of these models require a lot of data, and their semantics and media are heterogeneous. Despite various efforts, the proposed approaches are fairly ad hoc and no general framework has been proposed to initialize complex SES models. This paper aims at providing a generic framework for the initialization problem, bearing in mind that model initialization is not only building the initial state of the model, but also specifying time series (for example, climate series) and parameterizing processes. We propose to reformulate the model initialization issue as a transformation of data (collected by thematicians) into data structures as used by computer programs. This enables the use of Model Driven Engineering (MDE) concepts and the implementation of domain specific languages (DSL) needed to initialize the models.