Convergence in integrated modeling frameworks

Abstract

There are three important reasons for the use of modeling frameworks in environmental science: dealing with complexity, re-using modules for different models, and providing support for commonly needed services. Literally dozens of modeling frameworks are being used by environmental scientists, several of which are under active development. It was our objective to determine just how much common ground there is between these frameworks. A review of how the various frameworks support decomposition of the modeling problem, specification of the model (including sub-models and compositions of sub-models), event-handling, numerical integration and run-time execution of models, revealed that it is helpful to make a distinction between implementation-level and modeling-level frameworks. An implementation-level framework's primary purpose is to link existing model implementations. With an implementation-level framework, the user is responsible for fully specifying links between modules and ensuring their correctness with respect to intention and logic. Differences between the various implementation-level frameworks are relatively unimportant. However, representing even a moderately complex system with an implementation-level framework requires the user to spend much effort to make sure that the necessary modules are properly connected. Modeling-level frameworks attempt to unburden the user by allowing domain-specific terms to be used to specify which models or modules should be used and how they should be linked. There is a direct, if usually implicit, link between both kinds of frameworks: before a model specified with a modeling-level framework can be executed, it must be translated into a general-purpose programming language, which will be done using an implementation-level framework. It is argued that users and builders of frameworks in environmental science will benefit from making the distinction between implementation-level and modeling-level frameworks and that they should be explicit about the implementation-level frameworks targeted by the code generators of modeling-level frameworks. If at all possible, a widely accepted target implementation-level framework should be chosen. Then, at some future time, we may see re-use and extension of an existing framework instead of the creation of YAMF (Yet Another Modeling Framework).