Implementing reproducible research using the Madagascar open-source software package

Abstract:

Reproducible research is a concept pioneered by Jon Claerbout. It refers to the discipline of attaching software code and data to scientific publications, which enables the reader to reproduce, verify, and extend published computational experiments. The Madagascar open-source software package provides an example of implementing the reproducible research discipline in geophysical publications. The package contains not only software tools for making geophysical computations but also research papers complete with links to data and reproducible data-analysis workflows. When researchers discover a research paper published on the Madagascar website and install Madagascar, they are able to follow the links and replicate all computations to verify the published computational results. Of course, reproducibility is not the goal in itself. The goal is to be able to extend previously published research by, for example, trying new computations on previously used data or previously used computations on new data. The Madagscar collection currently contains about 150 research papers and book chapters and about 900 reproducible scripts. More than 80 people from different organizations around the world have contributed to the development. This experience shows that, instead of being the responsibility of an individual author, computational reproducibility can become the responsibility of open-source scientific-software communities. Our experience shows how a dedicated community effort can keep a body of computational research alive by actively maintaining its reproducibility.