Acceptance of domain-specific visual languages for environmental scientific computing

Abstract

Software is developed and used for many different scientific purposes. However, the development of scientific software has traditionally involved little application of software engineering. This thesis describes an investigation of the use of software engineering in scientific software development, culminating in a study of the application of an existing software engineering methodology, namely domain-specific visual languages, to the area of scientific software. This research focuses on the acceptability of the domain-specific visual language approach by environmental scientists, resulting in a new manner of recording context and provenance information using the domain-specific visual language paradigm, which integrates such information alongside experimental specifications and designs. The application of domain-specific visual languages to the scientific software field was justified based on the results of an initial survey, which investigated the current state of scientific software development. In particular, the use of software engineering techniques, the lack of which was thought to be the cause of many issues in scientific software development, was investigated. The results identified that although the situation had improved from that reported in previous literature, there remained areas for potential improvement. The proven benefits of domain-specific visual languages, demonstrated in other areas of software use, matched up well with the areas of potential improvement in scientific software. Although the benefits of domain-specific visual languages has been studied in detail, there has been little attention paid to the acceptance of the approach by users, particularly in the scientific software area. Therefore this research focused on the acceptability of the domain-specific visual language approach by users in the scientific software area. A study was performed comparing the acceptance of a domain-specific visual language based tool and an existing form based tool, both of which are used to specify an environmental sciences experiment as a software simulation. The two tools were found to not differ significantly in acceptance, although the domain-specific visual language tool was perceived to be significantly better for understanding, exploring, and communicating the experiment itself. Additional information collected in the study revealed that environmental scientists focused on context information, which explains what is being tested and why it is of interest, when explaining an experimental scenario. Apart from its commonplace inclusion in scientific papers, there has been little research into how context information is recorded by scientists, although there exist some tools for the related area of provenance information recording. A second study was performed to measure the acceptance of an extended domain-specific visual language tool, which allowed scientists to record context and provenance information in an integrated environment alongside the experimental setup. This new method of recording context and provenance information, integrated with experimental setup within a domain-specific visual language, received positive feedback from potential users. Potential benefits of this approach are that it forces experiment developers to explicitly consider and capture context and provenance information while creating their experiments, and allows them to directly link such information with components of the experimental design.