Abstract
The use of computers in geography and related subjects is not new. In its most trivial form geocomputation has been happening ever since quantitative geographers started doing their statistics and mathematical modelling using computers instead of calculators in the early 1960s. However, in the last five years there has been a very significant change in the size, the speed, and economics of high performance computing that its radically altering the opportunities for applying a computational approach. In an increasing number of sciences the leading edge research is becoming heavily biased towards a computational approach. The emergence of scaleable parallel computing hardware is beginning to dramatically increase the performance and memory sizes of high performance computing greatly increasing its usefulness and broadening the range of suitable applications. As a result massively compute intensive approaches to doing science have resulted in the emergence of high profile Grand Challenge research in physics, biology, chemistry, engineering, material science, geology, environmental and climatic modelling with a particular emphasis on: computational fluid dynamics, the modelling of macromolecules, simulation of materials, quantum chromodynamics, genome data management, combustion, aeronautical design, and the search for new chemical substances. Supercomputing is now an accepted technique that stands as an equal partner to observation, experimentation, and theory. Many sciences now have computational as an adjective. This is very sensible as the speed-up in computation looks set to continue more or less quadrupling ever two or three years with affordable machines capable of sustained terraflop speeds confidently expected before 2000.
So far the impact on geography and social sciences has been minimal. There have been no Reserarch Council initiatives in this area and no targetting of HPC as a core technology despite the recognition in the Foresight Exercise of its key importance to many areas of social science in public and commercial contexts. It is almost as if most of the researchers either have failed to note the tremendous changes in speed and size or that they have no idea of what to do with it due to lack of awareness. Yet it is quite likely that by 2000 the compute speeds available to geographers will have increased by about 109 times since the quantitative revolution of the mid 1960s, by about 108 times since the mathematical modelling revolution of the early 1970s and at least 106 times since the GIS revolution of the mid 1980s. It is surely time to re-assess what the implications are likely to be for what geographers and other social sciences do. Yet seemingly most geographers are still astonished by the appearance of PC systems with the performance of 1980’s mainframes. Indeed it is only recently that social science speed benchmarks have been produced so that it is now possible for the first time to convert gigaflops on a Cray T3D Massively Parallel Processor into a Pentium Pro PC equivalent valid for problems typical of those faced in geographical computation.
Currently, geographical perspectives on high performance computing are heavily biased towards GIS software which is notable for its lack of any need for high performance computing platforms. Despite a high level of current neglect there is no doubting the range of potential applications for doing or re-doing geography via a computationally intensive route. The applications are of various generic types: (1) legacy mathematical models whose resolution and precision can be improved by using more data points or small zones or finer interpolation grids or by removing the short-cuts and approximations that were once necessary because of restricted computer speeds and memory sizes; (2) legacy statistical methods that can be given an added value via computational add-ons; for instance, non-parametric estimation of variances and confidence intervals via computationally intensive statistical methods such as jack knifing and bootstrapping or the use of Monte Carlo significance tests in place of heavily assumption dependent classical alternatives; (3) legacy non-linear optimisation methods widely used in parameter estimation problems can be improved by the use of evolution strategies and genetic algorithms to remove the need for fundamental assumptions such global convexity and well behaved continuous functions; (4) likewise the substitution of classical compute adverse methods by more computationally intensive alternatives that promise better quality results; for example, the use of unsupervised neural network based classifiers with large spatial data sets or the use of simulated annealing optimisation to improve the quality of large spatial locational optimisation problems; (5) the replacement of conventional modelling tools by artificial intelligence based alternatives; for example, the use of supervised neural networks to model in an equation-free way non-linear relationships; (6) attempts to incorporate knowledge in existing techniques by developing new hybrid technologies; for example fuzzy modelling of spatial data can combine what qualitative knowledge exists with what can be discovered from a data base by a self-organising fuzzy adaptive trainer; (7) the tackling of old problems that were previously neglected or considered unsuitable or insoluble; for example, the use of artificial life concepts and distributed agents to hunt out patterns in data without being told in advance where to look, when to look, or what to look for; or the application of Genetic Programming to ‘breed’ or create new better performing mathematical models from data bases; and (8) the prospects of using robotic vision technologies to detect recurrent fuzzy patterns and relationships in GIS databases in a highly abstract and very general manner. In all these applications a common component is the desire to compute our way to better or new or novel solutions to a mix of old and new problems of various sorts. The increasingly spatial data rich world about us is giving a tremendous stimulus to these developments, some of which are pure research, and others which possess a high degree of applied relevancy.
The themes of GeoComputation ‘96 reflect the constituent technologies of high performance computing, artificial intelligence, and GIS. They offer a broad range of applications of geocomputational technologies being used. However, in many ways this is only the beginning of the extensive and broadly based development of a geocomputational paradigm that will over the next few decades grow in strength and expand in the breadth of applications as some exponential function of supercomputing speeds. The take-off has been delayed by an inadequacy of high performance computing hardware, the lack of appropriate tools and technologies relevant to geography that would need to be powered by supercomputers, and a deficiency in awareness and potential benefits to be gained. Suddenly the constraints have dissolved and we stand at the dawning of a new era of geocomputation.