Knowledge representation in geographic information systems.

Abstract

In order to satisfy increasing demand for better, smarter, more flexible land resource information an alternative form of representation is proposed. That representation is to be achieved through the coupling of Expert System methods and Geographic Information Systems. Instead of representing resource information using entities such as soil types, defined by rigid boundaries on a map, a more fluid presentation is proposed. Individual resource attributes will be represented by surfaces that describe their probability of occurrence, at a number of levels, across a landscape. Such flexible representations, which are designed to better capture the mental models behind their creation, are capable of being combined and synthesised to answer a wide range of resource queries.An investigation of methods of knowledge representation in a number of fields of research, led to the belief that a Bayesian Network provides a representational calculus that is appropriate to the "fuzzy" and imprecise conceptual models used in resource assessment. The fundamental mathematical principles of such networks have been tailored to provide a representation that is in tune with the intuitive processes of a surveyor's thinking.Software has been written to demonstrate the method and tested on a variety of data sets from Australia and overseas. These tests and demonstrations have used a range of densities of knowledge and range of acuity in evidential data. In general the results accord with the mental models used as drivers. A number of operational facets of the method have been highlighted during these demonstrations and attention has been given to a discussion of them.