Distributed frameworks and parallel algorithms for processing large-scale geographic data

The number of applications that require parallel and high-performance computing techniques has diminished in recent years due to to the continuing increase in power of PC, workstation and mono-processor systems. However, Geographic information systems (GIS) still provide a resource-hungry application domain that can make good use of parallel techniques. We describe our work with geographical systems for environmental and defence applications and some of the algorithms and techniques we have deployed to deliver high-performance prototype systems that can deal with large data sets. GIS applications are often run operationally as part of decision support systems with both a human interactive component as well as large scale batch or server-based components. Parallel computing technology embedded in a distributed system therefore provides an ideal and practical solution for multi-site organisations and especially government agencies who need to extract the best value from bulk geographic data. We describe the distributed computing approaches we have used to integrate bulk data and metadata sources and the grid computing techniques we have used to embed parallel services in an operational infrastructure. We describe some of the parallel techniques we have used: for data assimilation; for image and map data processing; for data cluster analysis; and for data mining. We also discuss issues related to emerging standards for data exchange and design issues for integrating together data in a distributed ownership system. We include a historical review of our work in this area over the last decade which leads us to believe parallel computing will continue to play an important role in GIS. We speculate on algorithmic and systems issues for the future. (C) 2003 Published by Elsevier B.V.