Introduction

The first international conference on 'GeoComputation', hosted by the School of Geography at the University of Leeds in 1996, heralded the launch of a new research agenda in geographical analysis and modelling (Openshaw and Abrahart, 1996). A measure of the interest generated in this field is that the conference has become established as a yearly event (Otago, New Zealand, 1997; Bristol, UK, 1998; Virginia, USA, 1999; Greenwich, UK, 2000) that will next be held at the University of Queensland (Australia) in 2001. Although there is certainly a reliance on computer science as an enabling technology, the term 'GeoComputation' is not intended as a synonym for GIS or spatial information theory — these being somewhat at odds with the original and ongoing vision. This brief note attempts to explain that vision, to describe the enabling technology in terms of what it offers geographical analysis and to outline some of the research questions that GeoComputation faces.

The Murky History

The late 1960's and early 1970's gave rise to a quantitative revolution in geography, as labelled by Haggett and Chorley (1967). Computing became widely accessible as a geographic tool, enabling a range of previously intractable problems to be analysed. An excellent account of the somewhat turbulent history of that time is given by Macmillan (1997). One interpretation of this period is that the computers became the slaves of the quantitative geographers. Lots of (very messy) Fortran code was written, lots of punched card stacks were dropped (or thrown) and some respectable geographers joined the ranks of the bearded, nocturnal hackers that haunted the 'machine room'. Geographers began to look for more data (because now they could actually analyse it) and to build more complex models.

Disabling Technology

The late 1970's and early 1980's saw the rise of databases; large monolithic systems that standardised on interfaces, file structures and query languages. Like all generic solutions there was a cost — flexibility and performance were sacrificed for robustness and ease of application construction. Early GIS show evidence of these same engineering compromises. GIS saw to it that geographers became the slaves of the computer, having to adopt the impoverished representational and analysis capabilities that GIS provided, in exchange for ditching the Fortran, getting some sleep and producing much prettier output.

GIS was, for some, a backwards step because the data models and analysis methods provided were simply not rich enough in geographical concepts and understanding to meet their needs. (It is entirely possible that computer scientists invented GIS out of spite, being fed up with all those quantitative geographers hogging their CPU cycles and clogging up the disks with kilobytes(!) of data.) Consequently, many of the geographical analysis problems that gave rise to the quantitative revolution in the first place could not be addressed in these systems. Quantitative geographers switched over to GIS or they went to the back of the research funding line.

In the intervening time, GIS have improved somewhat and geography has become very much richer in digital information. The requirement to build complex applications and simulations has not receded, if anything it has become more urgent, with the need to plan for a changing climate, to feed an increasing population and to provide pinpoint marketing analysis for digital encyclopaedia salespeople.

The Focus of GeoComputation

GeoComputation represents a conscious attempt to move the research agenda back to geographical analysis and modelling, with or without GIS in tow. Its concern is to enrich geography with a toolbox of methods to model and analyse a range of highly complex, often non-deterministic problems. It is about not compromising the geography, nor enforcing the use of unhelpful or simplistic representations. It is a conscious effort to explore the middle ground from the doubly-informed perspective of geography and computer science. A true enabling technology for the quantitative geographer, a rich source of computational and representational challenges for the computer scientist.

Enabling Technology

GeoComputation is enabled in part by the continuing development of analytical statistics that can be applied to clustering, search and measures of association over space and through time. However, the successful scaling of these analysis methods to larger problems relies on a number of significant advances in computer science:

Computer architecture and design:

Improvements in computing performance and the development of massively parallel architectures have enabled previously intractable analysis problems to be addressed via deterministic means. Hence the interest in both coarse and fine-grained parallelism (e.g. groups at Department of Geography, University of Edinburgh and Centre for Computational Geography, University of Leeds.

Search, classification, prediction and modelling:

Progress in pattern recognition, classification and function approximation tools, originating from the artificial intelligence community (such as decision trees, neural networks and genetic algorithms), now provide sophisticated capabilities for tackling a range of non-deterministic problems. These techniques make some significant advances over traditional techniques, such as k-means clustering measures, principal component analysis and maximum likelihood classification in that they are (by comparison at least): robust in the presence of noise, flexible in the statistical types that can be combined, able to work with attribute spaces of very high dimensionality, require less training data and make fewer prior assumptions about data distributions and model parameters. (e.g. Donnet neural network landcover classifier from Curtin University, Australia; recreation behaviour simulator from the University of Arizona, USA; and agent-based computational economics tools from Iowa State University, USA)

Knowledge discovery:

Advances in the mechanisms by which knowledge can be distilled from large datasets as a result of data mining and knowledge discovery tools. The sheer volumes of geographically-oriented data now available in some spatial databases defy conventional approaches to reporting and analysis. (e.g. the recent NCGIA Varenius meeting on knowledge discovery and data mining, knowledge discovery in spatial databases, Computer Science, Munich, Germany, analysis and exploration machines for geography, Centre for Computational Geography, Leeds University, UK).

Visualisation:

Advances in visualisation as a means of data exploration are providing new tools and approaches to help gain insight into complex and multi-dimensional datasets. Visual portrayal can illicit understanding by adopting a representational form that is much more readily assimilated by people than statistical summaries of correlation or clustering. (e.g. GeoVista Center, Penn State Geography or Geography, Leicester).

Some problems for GeoComputation

Problems remain before this new technology is effectively harnessed. Many of these are methodological; sophisticated tools require sophisticated set-up and operation. Although there are many reported examples of the use of these tools in the earth sciences literature, their use often incurs a considerable investment in terms of customisation, set-up, experimentation and testing before useful results are obtained. So, GeoComputation must overcome some significant challenges if the techniques are to become established in the toolbox of the geographer. These challenges include:

1. the inclusion of geographical 'domain knowledge' within the tools to improve performance and reliability
2. the design of suitable geographic operators for data mining and knowledge discovery
3. the development of robust clustering algorithms able to operate across a range of spatio-temporal scales
4. obtaining computability on geographical analysis problems too complex for current hardware and software
5. visualisation and virtual reality paradigms that support a visual approach to exploring, understanding and communicating geographical phenomena.

In short, there is a gap in knowledge between the abstract functioning of these tools (which is usually well understood in the computer science community) and their successful deployment to the complex applications and datasets that are commonplace in geography. It is precisely this gap in knowledge that GeoComputation aims to address.

Mark Gahegan,
Department of Geography, The Pennsylvania State University, USA. Email: mark@geog.psu.edu.

On behalf of the GeoComputation International Steering Group.

References

Haggett and Chorley (1967) (Eds.), Models in Geography, Methuen, UK.

Macmillan, W. D. (1997), Computing and the science of geography: the postmodern turn and the geocomputational twist. Proc. 2nd International Conference on GeoComputation (Ed. Pascoe, R. T.), University of Otago, New Zealand, pp. 1-11.

Openshaw, S. and Abrahart, R. J. (1996), Geocomputation. Proc. 1st International Conference on GeoComputation (Ed. Abrahart, R. J.), University of Leeds, UK, pp. 665-666.