Models of scientific software development

Over the past decade, I have performed several field studies with scientists developing software either on their own or together with software engineers. Based on these field study data, I identify a model of scientific software development as practiced in many scientific laboratories and communities. This model does not fit the standard software engineering models. For example, the tasks of requirement elicitation and software evaluation are not clearly delineated. Nevertheless, it appears to be successful within the context in which it is used. In the context in which scientists collaborate with software engineers, however, I describe problems which arose from the clash of this model with a traditional, phased software engineering model. Given these models, I discuss the issues which have to be addressed in order to determine the software techniques and tools which might best support scientific software development in different contexts.