GeoDynamics. Atkinson, P.M., Foody, G.M., Darby, S.E., Wu, F. (eds), 2005, CRC Press.

Brian G Lees, Professor, School of Physical, Environmental & Mathematical Sciences, The University of New South Wales, at the Australian Defence Force Academy, Canberra ACT 2600, Australia

This text is not so much the compilation of a set of conference papers, but a well-selected group of papers from keynote speakers and special sessions at GeComputation 2003. As such, it rises well above the usual run of post-conference publications and will earn its shelf space many times over.

To have a preface, a foreword and an introduction seems a little like overkill. But when these are by such luminaries as Peter Burrough, Paul Longley and the editors themselves, they are interesting to read as each sees something slightly different in the compilation. This does tell us something about the state of the field. All note the attempts move away from aggregation and generalization, approaches to dealing with complexity which both simplify and generate error. Quite where these moves will lead is not yet clear, but this volume contains many pointers.
GeoComputation is a name given to a series of conferences started by Stan Openshaw as a reaction to the limits imposed on geographical analysis by GIS software. It aimed to showcase methods to model and analyse a range of highly complex, often non-deterministic geographical problems. Those who saw GeoComputation as a meeting place tended to be unhappy about the compromises over-simplistic representations had forced on those geographers who used GIS. Computer scientists who became involved were attracted by GeoComputation as a rich source computational and representational challenges. Interestingly, in those early meetings there was an emphasis on classification and aggregation which later meetings, like this one, have begun to move strongly away from. In one sense these meetings seem to have a more geographical feel than the similar, but perhaps more computer science dominated, GISc conferences. Perhaps this derives from GeoComputation's emphasis on dealing with real-world applications whilst GISc tries to eschew these.
Giles Foody et al's introduction to the first part of the book (Yes, another introduction!) is a very useful, concise and clear summary of the issues facing the remote sensing of land cover and land cover change as it tries to move away from simple classification. Arora, Shalan and Elgy's chapter on classification accuracy measures is a nice review of this fast moving field. It has been overtaken by recent publication of some of the measures it foreshadows, but is nevertheless a clear and informative contribution. Carmel's following chapter continues the theme of reducing error by decomposing it into location error and classification error and argues that aggregation is a means of increasing thematic map accuracy. One is tempted to read that title as 'Decreasing Precision can increase Thematic Map Accuracy', but it too is a useful contribution. The rest of this section of the volume looks at land cover change detection using a variety of strategies.
The second section of the volume deals with spatially distributed modeling. This is a field which was hardly visible ten years ago and now is possibly the fastest growing area of research activity in our sub-discipline. The papers emphasise the shift from mapping to process modeling, but not process models of the traditional, deterministic kind but rather those based on cellular automata, genetic algorithms, fuzzy theory and stepwise models which use high levels of granularity. Given the number of papers on landslide prediction using GIS currently appearing before editors, it was nice to see a very clear example from the Bonn area. Although little was made of this, most of the models reflected a move from polygon-based spatial data to cell-based data. This is a significant trend and distinguishes these very diverse examples from, say, those of five to ten years ago.
Part three of the volume deals with human processes and represents about half of the book. Again, this reflects a change in the balance of activity from a few years ago when natural system models and processes dominated. There is a heavy emphasis on cellular automata.
The emphasis on cellular automata does not mean that this section is overly repetitious, rather the strategies of using cellular automata to deal with human environments are explored. Issues of ethics, scale, complexity and the limits to sensible use are each the topics of separate chapters.
Very importantly, a number of the chapters in this section argue strongly for more testing and validation of models. The issues which dominate section one of the volume, accuracy and error tracking, are not yet seen, it is argued, as important and necessary steps in a project. Keith Clarke, in his chapter 'The Limits of Simplicity', points out that very few models in the urban growth modeling literature are calibrated, far less validated or sensitivity tested.
One important paper in this section, by Anthony Gar-On Yeh and Xia Li, does focus on error propagation and model uncertainties in urban simulation with cellular automata in a GIS environment. There is much in this chapter reminiscent of the earlier chapter in section one by Arora et al., dealing with error propagation and model uncertainties in land cover and land cover change assessment. I suspect that the close similarity in title is coincidence, but the emphasis reflects a broader concern in the GeoComputation community.
The final chapter, like the early chapters, provides an excellent overview of the field. This chapter, and indeed the whole volume, would make perfect reading for a post-graduate starting to build a proposal as it cover the field well and is fairly up to date. It also makes useful reading for researchers specializing in other areas of the literature to keep up with developments in this fast moving sub-discipline.

e-mail: b.lees@adfa.edu.au

imageThis site has been developed by Reed Business Information for ISPRS