Code Design and Quality Control

ESM codes have mostly been written directly by scientists, as experts in the various geoscientific domains relevant for thesemodel. This contrastswith commercial forms of software development, where there is usually a clear division of labour between systems analysts, who determine what the system should do, and programmers, who are responsible for writing the code. Having scientists write their own code avoids many of the communication errors and misunderstandings common in commercial software practices, and fits well with exploratory and incremental nature of model development (Easterbrook and Johns 2009).Decisions aboutwhat to add to themodel are tightly bound with exploration of scientific research questions, as it is hard to know what will be needed ahead of time. Furthermore, model development requires deep knowledge of the physical processes and the impact of different implementation choices, so that it is usually easier to write the code directly, than to explain to a programmer what is needed. While this approach has been remarkably successful in building the first generations of general circulation models, it does not scale well. Understandably, scientists tend to focus on producing working code, and testing it for scientific validity, while postponing consideration of code quality issues, such as structure and readability of the code, portability, flexibility, modifiability and reusability. As ESM have become more complex, with more component models, more complex interactions between components, and more diverse user communities, these code quality issues become vital. The current generation of ESM are significantly larger and more complex than their predecessors. For example the National Center for Atmospheric Research (NCAR) Community Earth SystemModel (CESM) and the UKMet Office’s Unified