Addressing the long term preservation issues associated with scientific data is a complex challenge compounded by: the scale and multidisciplinary nature of the problem the wide range of formats and data types involved and the often proprietary and transitory nature of the hardware and software used to generate the data. In this paper, we present the PANIC system - an integrated, extensible architecture based on preservation metadata, automatic notification services, software and format registries and semantic grid services - that we believe, offers a sustainable, dynamic approach to the long term preservation of large collections of heterogeneous scientific data.