Provenance in support of workflows for science

In experimental science, claims must be supported not only by a guarantee of reproducibility of the results, but also with transparency of the underlying assets, either physical or digital. In this setting, scientists should be able to ask "why and how" a result has been generated, and there is an expectation that the response will include the experiment’s input data (or physical assets), through a sequence of transformations, generally encoded by algorithms. In addition, while designing and building experiments, as well as reproducing others, transparency for debugging is essential. The provenance of data and processes contribute to answering these requirements, in turn helping engender trust in the results. In this chapter, we provide an overview of provenance solutions in the specific context of computational models and systems that support experimental science by facilitating the production, automation, replication, and development of scientific code. Under the general category of "workflow support tools", these systems range from formal workflow systems, data analytics environments, and more general-purpose script-based frameworks. These systems entail different challenges and requirements for managing the provenance of the underlying data, requiring provenance architects to explore the associated trade-offs. We begin by summarising conceptual models, namely PROV and PROVONE, for representing provenance in the specific context of data-intensive processing. We then discuss the properties of provenance management systems in terms of provenance capture, storage, and query capabilities, with reference to different subtypes of workflow support tools, discussing a few examples in detail from a large space of existing systems. We conclude by offering future perspectives and presenting open challenges for provenance, in line with the evolution of the computational environments and with the increasing needs for data and process transparency and explainability.