A data integration approach to dynamically fusing geospatial sources

Accurate and efficient integration of geospatial data is an important problem with implications in critical areas such as emergency response and urban planning. Some of the key challenges in supporting large-scale geospatial data integration are: (1) automatically representing a large number of geospatial source available on the web by utilizing various geospatial data access standards, (2) handling different geospatial data formats and access patterns, (3) assessing the quality of the data provided by a large number of geospatial sources, and (4) automatically providing high quality answers to the user queries based on a quality criteria supplied by the user. In this thesis I describe my research on efficient and accurate integration of geospatial data from a large number of sources. In particular, I describe a representation methodology for declaratively describing the content and the quality of data provided by sources in a data integration system. I discuss methods to automatically generate the descriptions of both the content and the quality of data provided by geospatial sources. I describe a quality-driven query answering algorithm that exploits the descriptions of the content provided by the geospatial sources to generate an initial data integration plan that answers a given user query and optimizes the generated plan by utilizing the description of the quality of data provided by the sources and the quality criteria specified by the user. I also present a mapping of the generated integration plan into a program that can be efficiently executed by a streaming, dataflow-style execution engine. I implement my techniques in a framework called Quality-driven Geospatial Mediator (QGM). My experimental evaluation in automatically representing over 1200 real-world geospatial sources shows that QGM accurately generates the descriptions of the content and the quality of geospatial sources. The empirical evaluation of QGM's query answering techniques using over 1200 real-world sources shows that QGM provides better quality data in response to the user queries compared to the traditional data integration systems and does so with lower response time.