Huge amounts of data available in a variety of digital forms has been collected and stored in thousands of repositories. However, the information relevant to a user or application need may be stored in multiple forms in different repositories. Answering a user query may require correlation of information at a semantic level across multiple forms and representations. We present a three-level architecture comprising of the ontology, metadata and data levels for enabling this correlation. Components of this architecture are explained by using an example from a GIS application.

Metadata is the most critical level in our architecture. Various types of metadata developed by researchers for different media are reviewed and classified wrt the extent they model data or information content. The reference terms and the ontology of the metadata are classified wrt their dependence on the application domain. We identify the type of metadata suitable for enabling correlation at a semantic level. Issues of metadata extraction, storage and association with data are also discussed.