The term geocomputation, as defined by geocomputation.org, refers to “the art and science of solving complex spatial problems with computers”. This definition captures the essence of the term geocomputation, but is too broad for the purposes of this Guide. It potentially embraces all manner of concepts, tools and techniques that form part of mainstream GIS and spatial analysis toolsets. Indeed, as individual geocomputational methods become accepted as providing effective solutions to specific spatial analysis problems, so they start to appear in more generic software with widespread usage. This is particularly apparent in some areas of remote sensing data analysis, visualization tools and in association with a number of statistical and optimization problems.
Micro-scale simulation of spatial phenomena is regarded by many as perhaps the archetypal geocomputational tool. Geosimulation (a term introduced by Benenson and Torrens, 2004) is discussed in Section 8.2, Geosimulation, where cellular automata, agent-based models and the ever-expanding set of software tools to support such modeling are described. This is a very active and sometimes controversial area of geospatial modeling, which attracts much discussion amongst researchers. An important aspect of geosimulation has been the growing appreciation of its role in researching processes, particularly those involving emergent and unpredictable behaviors and outcomes, rather than always (or ever) having a strong predictive function. Section 8.2 also includes a wide-ranging discussion of the broader issues relating to spatio-temporal modeling.
Section 8.3, Artificial Neural Networks (ANN), addresses the use of so-called neural networks to solve particular spatial problems, with applications ranging from image analysis to spatial interaction modeling and spatial optimization. An important area of geocomputational research relates to the problems of inferring land cover and land-use from remotely-sensed data, as well as the quest to better understand the dynamics of land-use change. A relatively recent development in this field has been the use of computational neural network (CNN) methods for both supervised and unsupervised multi-band image analysis. At one level the use of neural networks can be seen as yet another technique in the ever-expanding collection available in GIS packages such as TNTMips and Idrisi and mathematical packages such as MATLab. However, with neural network methods selection of the type of model to be used, its configuration and training, coupled with the very specific and computationally intensive procedure then adopted, separate this approach from many others available.
The final area of geocomputation we address is a class of procedures inspired by biological reproduction and ‘survival of the fittest’ (Section 8.4, Genetic Algorithms and Evolutionary Computing). The principal topic we cover is that of genetic algorithms (GAs) and their application to a wide range of problems in spatial analysis. Again, this is a developing field, which (for many applications) has yet to prove its superiority to alternative procedures, and is thus rarely implemented in GIS or other commercial spatial software applications. However, it does offer a novel family of methods for solving problems that have a complexity not readily addressed by other approaches. And, rather like geosimulation, the dynamics of the process can be visualized and examined, in order to obtain a clearer understanding of the underlying problem and data.
Readers of earlier chapters will have observed that many other geocomputational methods have already been described in connection with a variety of different spatial analysis problems. Examples include: the use of computationally intensive procedures to achieve desired zoning patterns (see Section 4.2.11, Districting and re-districting); cluster-hunting software such as GAM/K (see Section 5.2.6, Cluster hunting and scan statistics); the widespread use of Monte Carlo and random permutation techniques to generate pseudo-probability distributions and associated confidence intervals; the development of interactive data mining tools, such as the visualization techniques offered by brushing and linking (see Section 5.2, Exploratory Spatial Data Analysis); conditional choropleth mapping (see 5.2.3, Cross tabulations and conditional choropleth plots); the processing of raster datasets using techniques such as distance transforms (see Section 4.4.2, Length and area for raster datasets); surface analysis (see Section 6.2, Surface Geometry) and visibility analysis (see Section 6.3, Visibility); and the application of heuristics and metaheuristics to combinatorial optimization problems (see Section 7.2.2, Heuristic and meta-heuristic algorithms). These procedures are not discussed further in this chapter and readers are referred to the individual sections of this Guide and the associated reference materials for more details in each case. The ground-breaking book “GeoComputation” edited by Openshaw and Abrahart (2000) is also a useful source discussing many of the topics covered in the subsections below.