ASPRS 2007 Annual Conference Tampa, Florida ♦ May 7-11, 2007 ONTOLOGY-SUPPORTED AUTOMATIC SERVICE CHAINING FOR GEOSPATIAL KNOWLEDGE DISCOVERY Liping Di, Professor and Director Peng Yue, PhD student Wenli Yang, Principle Scientist and Associate Director Genong Yu, Post-doctoral Research Associate Peisheng Zhao, Research Assistant Professor Yaxing Wei, PhD student Center for Spatial Information Science and Systems (CSISS) George Mason University 6301 Ivy Lane, Suite 620, Greenbelt, MD 20770 ldi@gmu.edu pyue@gmu.edu wyang1@gmu.edu gyu@gmu.edu pzhao@gmu.edu ywei@gmu.edu ABSTRACT With the advances in sensor and platform technologies, the capability for collecting geospatial data has significantly increased. Large volumes of data have been collected using remote sensing. While those data are potentially valuable for the benefit of society, they must be converted to geospatial knowledge before they are useful. The traditional methods — only geospatial experts analyze data — fall far short of today’s increased demands for geospatial knowledge. As a result, significant amounts of data have not even once been analyzed after collection. Recent progress in the geospatial semantic Web has shown promise for developing automatic geospatial knowledge discovery methods for solving application problems, which otherwise require considerable resources. This paper presents an approach for automatically solving geospatial problems in the geospatial semantic Web environment. The approach simulates the process used by geospatial experts who first use backward reasoning from the required knowledge to the available raw data to select a set of available geo-processing functions, and then execute the functions sequentially, starting from raw data, to derive the desired knowledge. This backward reasoning effectively creates a path from raw geospatial data to the desired geospatial knowledge. With rich semantic descriptions of services and the support of ontology, the path can be formed automatically through backward reasoning from the desired result to raw geospatial data using semantic Web services. Such a path can be instantiated to become an executable workflow to generate the result automatically. A prototypical system is implemented to demonstrate the above concept and approach. INTRODUCTION With recent advances in sensor and platform technology, the capability for collecting geospatial data has significantly increased. Large amounts of geospatial data have accumulated over the years, through established satellite observations, emerging sensor networks, and the spread of location-sensitive data collectors. While those data are potentially valuable for societal benefits, they must be converted to geospatial knowledge before they become useful. A gap needs to be filled with proper geospatial processing before the end users can extract what they really want in terms of knowledge and information. For example, a fire chief for fire may be interested in information about where fires occur and how large an area is affected. A risk manager may be interested in knowing the potential risks of natural hazards in his territory. Such information does not pop out from the remotely sensed data directly, but can be achieved through a series of geospatial processing steps, i.e. image pre-processing, classifier training, classification, statistical summarizing, and result presentation. A series of processes may need to be invoked during the processing. This can become problematic and laborious when the data are fragmented across different sectors of agencies(Brodaric and Gahegan, 2006). Expected increases in Web-based services for both information and processing capabilities will exacerbate difficulties in finding, integrating, and using such services to