Scientific workflows have emerged as a new paradigm to facilitate and automate scientific processes. During workflow execution, the need often arises to capture and store the data derivation history, known as provenance, which describes the steps that yielded workflow output results. Workflow tools available today address this need by employing systems that capture provenance, store it in a database, and provide an interface for scientists to explore the stored data. However, these systems differ both in their functional and non-functional characteristics. This fact indicates that there is no agreement yet in the community about the essential capabilities that a provenance system has to provide. It is therefore important to develop a set of requirements for a scientific workflow provenance system. Furthermore, due to the lack of understanding of what these standard requirements should be, it is only natural that a standard system architecture for provenance management is missing. To address these shortcomings, in this paper, we 1) identify a set of functional and non-functional requirements for scientific workflow provenance management systems that cover key aspects of provenance storage, exploration, and reasoning, and 2) propose a reference system architecture for scientific workflow provenance management.