Coupling Numerical Integration Models: Granularity and Computational Sequence

This paper aims at bringing some solutions to the issue: how to allow the coupling of models ? More precisely: like there are ways to design extendable classes, are there means to design "couplable" models ? As classes are designed only for some uses, models can be designed only for some coupling. Obviously, all the situations can not be considered, but efforts can be done to anticipate the coupling of models, as generically as possible.

This questioning on models coupling is highly motivated by a need to easily change the granularity, i.e. the level of details, of a submodel inside a whole model. This study is focused on numerical integration models, i.e. models mainly ruled by ordinary differential equations. First, the modeling of the interface between two models is studied, which introduces prerequisite notions and modeling methodology. Then, a major issue for the coupling of numerical integration models is tackled: the reordering of the integration variables of the coupled models for the derivation process.

An extension of UNIF (Unified Numerical Integration Framework), a generic simulation framework for numerical integration models, is proposed and experimented on a simplified version of the coupled model GEMINI (Grassland Ecosystem Model with INdividual-centered Interactions).