Applying FAIR Principles to Numerical Modeling: Lessons from CSDMS

Abstract:

The motivation behind FAIR principles for data applies equally well to numerical models. Work with numerical models can include reproducing prior results, applying existing models to a new problem or location, linking models and/or data operations in a sequential workflow, coupling models to study feedbacks, adding new algorithms or capabilities, and building new models. Each of these can be made more efficient by applying FAIR principles. The Community Surface Dynamics Modeling System (CSDMS), an NSF-supported cyberinfrastructure facility that supports earth-surface geoscience research, uses a variety of approaches to promote FAIR development, application, and sharing of numerical models. A web-hosted Model Repository supports findability and accessibility by providing a catalog of models and tools with standardized metadata and bibliographies. CSDMS standards and tools include the Basic Model Interface (BMI), a language-agnostic interface standard that specifies a basic set of control and query functions that enable a model to act as a framework component, with the ability to advance execution iteratively and to exchange data interactively. The Babelizer addresses cross-language interoperability by wrapping a Python interface around models written in other languages. For model-based workflows, the Python Modeling Tool (pymt) provides a script-based framework that takes advantage of Python's rich collection of data analysis and plotting libraries. Model-data integration is facilitated by Data Components: small Python programs that use a BMI to access individual datasets, irrespective of the underlying format. The Landlab Toolkit, a Python library for building 2D numerical models, supports reusability by providing individual process components that can be assembled to create complete models. Lessons from CSDMS' experience include the importance of community building, the critical need for education in core computing and software skills, the value of documentation, and the importance of incentives to create robust and lasting digital artifacts. Perhaps the most important lesson is that a modest investment in the human and cyber infrastructure behind FAIR modeling can have a strong positive impact on research productivity.