Inconsistency detection and adjustment of spatial data using rule discovery

This dissertation presents a novel view of spatial data consistency, and introduces innovative techniques for the measurement and enhancement of the quality of spatial vector data. Consistency of spatial data is crucial for effective analysis and presentation of single or combined spatial databases. Spatial data stored in vector-based Geographic Information Systems (GIS) consist of explicitly stored objects representing real-world entities. Uncertainties are inherent in the processes of collection, capture, storage, analysis and display of this spatial data. Additional uncertainties become apparent when data from multiple sources are overlaid for analysis or presentation. Consistency is defined in the context of spatial data uncertainty; its causes and the need for its management are presented. Existing commercial GIS do not have decision support tools to handle spatial consistency. This thesis aims to fill this gap by proposing a framework for management of spatial data for consistency. The framework includes development of a data model that is suited for consistency analysis, discovery of rules that describe the structure of the data, and application of techniques for detection and adjustment of inconsistencies.; A data model is proposed that captures the implicit topological relations between neighboring objects. New geostatistical techniques are applied to discover patterns of object interactions. These patterns are formulated as rules that govern topological relations between pairs of objects as well as among multiple objects in a GIS, and take into account the semantic information of the objects. The rules are then used in a process of detecting inconsistencies, which are formulated as complex multi-object structures. Automated and semi-automated approaches for adjustment of inconsistencies among multiple objects are proposed based on localized and regionalized analysis of the inconsistencies, as well as on an innovative characterization of object's relative contributions to inconsistent behavior.; The proposed techniques were implemented, and tested on a number of commercial GIS databases. Based on these experiments, we find the techniques to be effective, with processing times that cannot be rivaled by human experts for both detection and adjustment of inconsistencies in spatial data.; These techniques improve the quality of spatial data. Consistent spatial databases can now be presented to users for effective spatial analysis. This results in increased levels of confidence in the use of spatial data. Finally, the ability to provide consistent representation of multiple combined data sets for analysis and presentation will result in many new application areas.