GIS, Expert Systems and Interoperability

Linda Lilburne1, George Benwell2 and Roz Buick1
1Landcare Research, P. O. Box 69, Lincoln, New Zealand
2Department of Information Science, Otago University, P. O. Box 56, Dunedin, New Zealand

A GIS can not do everything, despite the vendors' presumed ambition. The increasingly functional GIS software never seems to catch up with the needs of users, who are trying to solve more complex problems. Techniques from a variety of disciplines can potentially be utilised. While the expansion of GIS functionality, especially of analytical operations, is welcomed, it is also believed that moving towards a truly inter-operable GIS is essential in the new distributed and inter-linked world. An inter-operable GIS offers users the opportunity to effectively use each technique or system for its designed purpose. This promotes efficiency via the artificial intelligence principle of using suitable representation and search methods.

Integration offers the potential of harnessing the spatial power of a GIS with strengths found in other systems. In reality however, integration is at best a compromise. The sequential nature of many integrations and the impedance mismatch between system models tend to result in inflexible, application specific systems. This is demonstrated by reference to a 3-D conceptual model of integration, which incorporates the functional perspective as well as the data and interface perspectives.

Incorporation of knowledge is important to add meaning to the variety of data and analyses a GIS can typically manage and perform. Expert systems are designed to represent and reason with knowledge. They can combine experience, intuition and judgement to solve ill-structured problems where knowledge and data may be incomplete and/or imprecise. Expert systems are thus well suited to decision support environments. But expert systems are not designed to efficiently manipulate spatial data. Real world problems happen in a spatial context, so the integration of expert systems and GIS is desirable. There has been much research in this area already using well known integration techniques.

Computing advances are creating new opportunities for closer interaction between software packages. These opportunities open the way for progress to be made in combining knowledge with spatial concepts. Two advances in particular, the client/server paradigm and the object-oriented paradigm are considered in this paper.

The client/server paradigm is perceived to offer considerable potential in enabling a powerful, full-featured and flexible combination of systems, whose strengths are not compromised, without the need for extensive programming resources. It enables two independent processes to communicate with each other.

The object-oriented (OO) paradigm provides the ability to create a system model that relates closely to the real world, and is consequently an appropriate environment in which to model real world knowledge. A key component of the OO model, which is essential to the integration of an expert system and a GIS, as described in this paper, is inheritance. Objects representing real world phenomena are linked to spatial classes from which they can inherit appropriate spatial behaviour, in addition to domain related behaviour. The other key components of abstraction and encapsulation are also very important in enabling the domain relevant data only, to be modelled in a knowledge-base. This makes the complexities of GIS data representation more transparent.

The client/server and OO paradigms are employed to investigate their potential to easily create a flexible linkage between an expert system and a GIS. The strengths and weaknesses of these two systems are described to demonstrate how, conceptually, an expert system and a GIS can complement each other. A GIS can take the server role and provide spatial information to an expert system, in which tool and/or domain knowledge is represented. This knowledge enables the intelligent classification, prediction, interpretation and explanation of spatial data, as well as expert guidance, simulation and control of spatial processes. In other words, the GIS is a spatial repository, capable of providing spatial information to an intelligent front-end. With the GIS, taking a client role, the expert system can be used to provide knowledge of semantic associations between data layers, classification heuristics, meta-data (knowledge about data) and intelligent consultation sessions.

Two powerful tools (ARC/INFO and Smart Elements, previously known as Nexpert) are integrated in a client/server relationship using an object-oriented approach, to explore the limitations and advantages of these technologies in linking a GIS and an expert system. The culmination of this research is SES, a Spatial Expert Shell, which is a domain independent tool box capable of combining expert knowledge of some activity, process or phenomenon with a spatial nature or character. SES enables the computational, data storage and presentational strengths of a GIS to co-operate with the knowledge related strengths of an object-oriented expert system. The SES toolbox demonstrates the power of the client/server and object-oriented paradigms.

This paper presents the design of SES, with examples of its use. The limitations of the approach are described, including the implications of the synchronism of the client/server link. SES is contrasted with other documented expert system-GIS linkages, in order to highlight the benefits of using a client/server approach. Finally SES is represented on the 3D conceptual model of integration