GEOGRAPHIC KNOWLEDGE BASE DESIGN AND IMPLEMENTATION (ARTIFICIAL INTELLIGENCE, GEOGRAPHIC INFORMATION SYSTEMS, SPATIAL DATABASES, REPRESENTATION, MACHINE LEARNING)

The absence of intelligent search and learning capabilities is a prime weakness in current Geographic Information Systems (GIS). The resulting rigidity of their operation severely limits their use. In response to this problem, research and development of a KBGIS (Knowledge Based Geographic Information System) based on Artificial Intelligence (AI) techniques was carried out by a team of researchers at The University of California at Santa Barbara. The objective of the KBGIS system was to respond intelligently to user queries on large spatial databases. The spatial database included topographic data, remote sensing data and thematic maps which were implemented using the Quad(,-)tree data structure.;The design and implementation of most of KBGIS's components is described. Dual spatial and propositional data structures were used for representing the spatial knowledge. The principle of representational duality runs as a major design thread at the macro (KB-GIS), meso (KB) and micro (spatial object) structural levels.;The design was guided by the hypothesis that machine spatial cognition entails a shift in importance from low-level spatial representation to high-level symbolic representation. Implemented data structures included the following knowledge representations: logic (predicate calculus), AND/OR trees, frames and discrimination nets. Procedures included: initialization, user interface, query parser, rote learning and parts of location learning and example learning.;The most apparent success in terms of performance turned out to be the location learning process and its spatial inheritance mechanism. It was found that the dynamic location learning capability, in conjunction with a likely user querying behavior, results in progressively diminishing search times for related queries. The extremely important example learning process was found to be too crude and very slow. The failure to provide adequate handling of spatial relations within and between objects is viewed as the main shortcoming. It was concluded that the basic KBGIS architecture can serve as a prototype for future generations of knowledge based geographic querying systems. The theoretical ramifications of several general principles involving spatial knowledge which have emerged from this work, transcend beyond the relative narrow scope of the system described.