Static Analysis for Climate Code

Climate modelling encompasses a wide range of sub-domains and analytical techniques incorporating both numerical and data-driven methods. A typical general circulation model (GCM) includes a number of inter-connected earth, ocean, and atmosphere components with advanced numerical simulations and subgrid models ("parameterisations") - all within a single codebase. Such models can be large, complex, and often developed over many years, making them vulnerable to issues of correctness, maintainability, transparency, and reproducibility. These issues can be further exacerbated by the context in which some climate models are developed; and, depending on the relevant institution or research group, robust software development practises may not always be in place.

Static analysis and automated refactoring tools have the potential to significantly improve the quality of source code and identify various kinds of problems. Highly developed modern static analysis tools, such as SonarQube, Coverity and Infer, provide support for general-purpose code review; however, they do not generally address the scientific or numerical correctness of model implementations nor the particular programming patterns of climate models. A small number of domain-specific tools, such as the CamFort project developed by the Universities of Cambridge and Kent, address the numerical consistency of computational model implementations focussing on issues like units-of-measure correctness and array access shape; however, these platforms do not incorporate the same code quality features available elsewhere and have not been proven to scale to large climate models. In order to safeguard their future development, climate models require both good quality code and scientific correctness in their implementations.

In addition to a substantial design and implementation phase, this project will include a study of the problem domain and existing solutions. An initial study will evaluate the prevalence...