TCTrack: Working with researchers to enable FAIR tropical cyclone tracking

Abstract

We will present our experiences working with researchers to produce TCTrack, an open source Python package for determining tropical cyclones tracks. We will discuss how we were able to both apply RSE expertise to improve the software design as well as building deep domain knowledge in the process, ultimately resulting in a better tool for research. The initial challenge was a fragmented software ecosystem in which different tropical cyclone tracking algorithms produced substantially different results while also differing in software interfaces with varying degrees of documentation (or lack thereof). Researchers needed a unified approach that reduced the barriers to running and comparing multiple approaches. As RSEs working on the project we identified further software engineering challenges. Rather than reimplementing tracking algorithms directly – which would increase maintenance burden and further fragment the ecosystem – we designed TCTrack to build upon existing tracking codes where possible, with a framework that is easily extensible to enable and encourage integration of other tracking approaches by the community. Furthermore, most tracking codes produce different, often customised, output formats with minimal metadata, falling far short of FAIR (Findable, Accessible, Interoperable, Reusable) data standards. To address this, TCTrack transforms outputs from all algorithms into a common domain-standard format, while preserving metadata from the input files, and storing all parameters for full reproducibility. Finally, we will present the dashboard provided by the library for interactive exploration of the results. This can visualise and compare the outputs for different tracking algorithms and data sources, and compute summary statistics.