How Not to Drown in Code: Strategies for Managing Research Software

Modern research crucially depends on research software; yet, software development is rarely a core part of formal scientific training. Although software is a tool for most research, and most time should be spent on collecting, analyzing, and discussing results, a significant amount of time is still spent on implementing, debugging, and maintaining code.

Over the last years, our group has had to implement a variety of research software to autonomously collect sensor data, process this data, and combine it with atmospheric transport simulation runs in an optimization problem. Throughout these projects, we have established techniques and strategies to reduce the time spent on software development while improving the quality of our code.

In this presentation, we aim to share a few of these methods that are applicable to most research software efforts. This includes how to organize a long-term software project, manage a large number of configuration options, ensure reproducibility of results, reduce the effort of documenting code, avoid bugs in production runs, improve code reusability, and deal with legacy code.

Finally, we will discuss when it is worth investing time and effort into using these methods, emphasizing that not all research code needs to meet production-level quality standards.