27 June 2003
| "For geographers, though, what matters is not
the behaviour of individual agents per se but what happens to the
spatial economy as the result of the actions of sets of agents."
(Macmillan 1989, page 103)
"But the greatest limitation is surely of our own making. It is a limitation of vision, purpose, and conviction. If we are to remodel geography we will have to recognize and take up the challenges posed by the great geographical issues of our day, we will have to elevate the demands of scholarship above all other considerations, and we will have to exhibit a greater collective faith in our scientific methods." (Macmillan 1989, page 313, emphasis added)
|
Complex geographic systems can be characterized by the degree to which a clear understanding of isolated system components is not sufficient to explain or predict the behavior of the system as a whole. Recent advances in computer science and computational power support the development of computational laboratories for specifying and evaluating models that begin to explore the cumulative effects of context-sensitive individual interactions. Define a computational laboratory as a well-specified cellular automata (CA) or spatial agent-based simulation (SABS) model coupled with careful experimental design. Spatial simulations are especially appropriate for modeling complex geographic systems because we can frame our hypotheses at the individual level, and explore the effects of individual interactions on the patterns and properties that emerge at macro levels. In addition, hypotheses can be related not merely to specific properties of individuals, but also to system dynamics and to the proportional and spatial distributions of agent characteristics and behavior which can vary across space in patterns that are relevant to the emergent behavior of the system as a whole.
These are quintessentially geographic issues, and models which allow us to explore their nature aim for the heart of geographical analysis. Yet computer science and computational power by themselves offer little guidance for appropriate scientific use of computational laboratories. Computational intensity alone does not guarantee good science, and the use of CA and SABS simulations presents modeling in geography with two potential dangers: First, the subtler opportunity-cost danger of building "models" that merely provide fancy computerized visualizations mimicking morphological effects, at the expense of more insightful exploration of the underlying causes for those effects (Macmillan 1989, page 90). Second, though not unrelated to the first, there is the danger that shortsighted, inappropriate, or sloppy employment of spatial simulation models may discredit the entire methodology before its more rigorous potential can be realized.
Yet what are the appropriate modeling and experimental skills for these new geographic laboratories? What are the appropriate domains for such models; which new questions can we ask, and which older questions can be addressed in new and more incisive ways? What constitutes good scientific practice with such models; under what conditions can a set of well-designed computational experiments provide trustworthy explanations or predictions of important geographic phenomena? This paper explores the standards for good scientific practice in working with computational laboratories in geography.
Keywords: computational geography, agent-based simulations, theory, complex systems, experimental design
References:
Macmillan, Bill. Quantitative Theory Construction in Human Geography. Macmillan, Bill, Editor. Remodelling Geography. Oxford: Basil Blackwell; 1989: 89-107.
Macmillan, Bill. Modelling Through: An Afterword to Remodelling
Geography. Macmillan, Bill, Editor. Remodelling Geography. Oxford:
Basil Blackwell; 1989: 291-313.
27 June 2003