Abstract
Decision support systems, statistics and expert systems were some of the mainstay techniques used for modelling environmental phenomena. Now, many modelling techniques employ artificial intelligence (AI) techniques for the extra computational analysis they provide. This paper illustrates which new techniques are available, what these techniques are currently used for and which techniques best suit which applications (for example, in the spatial environment, artificial neural network techniques are usually used to aid automatic classification through data recognition and training; expert systems are usually used for simulation and prediction).
Whilst operating in a toolbox environment and by adopting AI techniques, the geographic information system (GIS) modellers therefore, have many options available for tackling problems. With the advent of these new computational techniques, can knowledge be saved in this toolbox environment? This paper will address the issues concerning, transferring, sharing, integrating and modelling spatial data which have been used in different knowledge bases by different analytical techniques. For example, with data restructuring, can the knowledge developed for an expert system be used as a knowledge-base for another reasoning system? How much restructuring needs to be done? How do these new techniques differ and what are the advantages and disadvantages of each technique? In order to address some of the above issues one technique is used as a reference point to indicate how data can be restructured (reused from data collected and structured for an old technique to be used by another technique).
The way techniques store and use their working units (rules, cases) determines what tasks they can perform, therefore in order to solve a problem it needs to be determined how the problem data can be represented, and which of these working units does the problem domain allow? If many working units are allowed, which technique should be adopted?
This paper will examine and summarise the available artificial intelligence techniques and the tasks adopted by geographic information systems. What tasks AI techniques perform and which tasks can be best applied to spatial phenomena. By identifying this new technique the example highlights the under utilisation of artificial intelligence techniques in the spatial modelling community. Some of the techniques offered by AI techniques may not be required by the spatial modelling community, these issues will be addressed. In particular, this paper outlines a new approach in applying artificial intelligence techniques to solve spatial problems. The approach consists of combining case-based reasoning (CBR) with geographic information systems. This combination allows both case-based reasoning and geographic information system technologies and techniques to be applied to solve spatial problems.
This paper focuses on CBR as one of these new tasks, and documents the saving data concept in going from decision trees to CBR.
More specifically this paper applies this combination of techniques to the problem of soil classification. Spatial cases are defined and analysed using the case-based reasoning techniques of retrieve, reuse, revise and retain. After the structure of cases are defined it is possible to compile a case base. Once the case-base is of sufficient size, the problem of soil classification is tested using this new approach. Essentially the problem is solved by searching the case base for another spatial case similar to the problem case. Then the knowledge from the searched case is used to formulate an answer to the problem case. The paper then compares the results from this new approach with a traditional method of soil classification. The application of case-based reasoning techniques to solving environmental dilemmas has associated problems and this paper highlights the benefits and limitations of this technique. The logistics of the problems that are characteristic of case-based reasoning systems are discussed. For example, putting the spatial domain of an environmental phenomena into a case base. What are the constraints of CBR, what data is lost, and what functions are gained (analysis)? Finally, the following question is posed "to what real world level can the environment be modelled using GIS and case-based reasoning techniques"?