Sensor web applications such as real-time environmental decision support systems require the use of sensors from multiple heterogeneous sources for purposes beyond the scope of the original sensor design and deployment. In such cyberenvironments, provenance plays a critical role as it enables users to understand, verify, reproduce, and ascertain the quality of derived data products. Such capabilities are yet to be developed in many sensor web enablement (SWE) applications. This paper develops a provenance-aware "Virtual Sensor" system, where a new persistent live "virtual" sensor is re-published in realtime after some model-based computational transformations of the raw sensor data streams. We describe the underlying OPM (Open Provenance Model) API's (Application Programming Interfaces), architecture for provenance capture, creation of the provenance graph and publishing of the provenance-aware virtual sensor where the new virtual sensor time-series data is augmented with OPM-compliant provenance information. A case study on creating real-time provenance-aware virtual rainfall sensors is illustrated. Such a provenance-aware virtual sensor system allows digital preservation and verification of the new virtual sensors.