摘要 : Coupled modelling is increasingly necessary to make progress in understanding the science of complex physical phenomena and a number of bespoke ('custom') coupled solutions to specific scientific challenges have emerged in recent ... 展开 Coupled modelling is increasingly necessary to make progress in understanding the science of complex physical phenomena and a number of bespoke ('custom') coupled solutions to specific scientific challenges have emerged in recent years. These coupled models generally consist of some framework code in which individual models are embedded. The framework code promotes the required interoperation of the models to solve the larger problem being addressed. Bespoke solutions limit the ability of scientists to share models and to couple them together flexibly to produce (efficient) implementations to address new problems. This paper presents an approach, GCF, which addresses several of these limitations. Individual model sharing and flexibility in composition and deployment is achieved by imposing some lightweight development rules for single models and capturing information relating to the models themselves, to their composition into coupled models and to their deployment onto computational resources as machine-readable metadata. These metadata can be processed to support the generation of an implementation of the coupled model required by the developer. For example, lean and efficient framework code for the specific coupled model and deployment described by the developer can be generated. Alternatively, GCF-compliant models can be automatically adapted for use within other, existing frameworks. This paper presents the design and implementation of a bespoke framework generator to achieve the former, and the flexibility in the composition of GCF-compliant models is demonstrated. 收起