Lessons in FAIR software from the Community Surface Dynamics Modeling System
Journal Article
Overview
abstract
The Community Surface Dynamics Modeling System (CSDMS) is a US-based science facility that supports computational modeling of diverse Earth and planetary surface processes, ranging from natural hazards and contemporary environmental change to geologic applications. The facility promotes open, interoperable, and shared software. Here we review approaches and lessons learned in advancing FAIR principles for geoscience modeling. To promote sharing and accessibility, CSDMS maintains an online Model Repository that catalogs over 400 shared codes, ranging from individual subroutines to large and sophisticated integrated models. Thanks to semi-automated search tools, the Repository now includes ~20,000 references to literature describing these models and their applications, giving prospective model users efficient access to information about how various codes have been developed and used. To promote interoperability, CSDMS develops and promotes the Basic Model Interface (BMI): a lightweight, language-agnostic API standard that provides control, query, and data-modification functions. BMI has been adopted by a number of academic, government, and quasi-private institutions for coupled-modeling applications. BMI specifications are provided for common scientific languages, including as Python, C, C++, Fortran, and Java. One challenge lies in broader awareness and adoption; for example, self-taught code developers may be unaware of the concept of an API standard, or may not perceive value in designing around such a standard. One way to address this challenge is to provide open-source programming libraries. One such library that CSDMS curates is Landlab Toolkit: a Python package that includes building blocks for model development (such as grid data structures and I/O functions) while also providing a framework for assembling integrated models from component parts. We find that Landlab can greatly speed model development, while giving user-developers an incentive to follow common patters and contribute new components to the library. However, libraries by themselves do not solve the reproducibility challenge. Rather than reinventing the wheel, the CSDMS facility has approached reproducibility by partnering with the Whole Tale initiative, which provides tools and protocols to create reproducible archives of computational research. Finally, we have found that a central challenge to FAIR modeling lies in the level of community knowledge. FAIR is a two-way street that depends in part on the technical skills of the user. Are they fluent in a particular programming language? How familiar are they with the numerical methods used by a given model? How familiar are they with underlying scientific concepts and simplifying assumptions? Are they conversant with modern version control and collaborative-development technology and practices? Although scientists should not need to become software engineers, in our experience there is a basic level of knowledge that can substantially raise the quality and sustainability of research software. To address this, CSDMS offers training programs, self-paced learning materials, and online help resources for community members. The vision is to foster a thriving community of practice in computational geoscience research, equipped with ever-improving modeling tools written by and for the community as a whole.
