Qualitative georeferencing and proximity modeling for geospatial information systems.

摘要

Due to the growing popularity of information technology, more people, especially in the general public, have access to computerized geospatial information systems. However, the general public's geographic view is often qualitative, while the geospatial information systems are mainly quantitative. This dissertation research aims to bridge the gap between the study of qualitative spatial reasoning, which deals with qualitative spatial relations, and the metric spatial information systems, which handle quantitative spatial relations. The concept of Qualitative Georeferencing is introduced for this purpose. As part of the effort toward the realization of qualitative georeferencing, the research explores methodologies and empirically conducts the process of constructing a context-contingent model for proximity spatial relation.; Qualitative Georeferencing is defined as “a mechanism to locate places or spaces in a metric geospatial information system according to their qualitative spatial relations to some place(s) or space(s) that is (are) already referenced in the metric system.” To enable current geospatial information systems with qualitative georeferencing capability, the dissertation proposes a conceptual framework to couple qualitative spatial reasoning models with the metric geospatial systems such as a geographical information system.; Among the components in the conceptual framework, the proximity relation is the least studied of all spatial relations. This research explores methodologies to construct context-contingent proximity models. The modeling purpose is to set up a translation mechanism between linguistic distance measures and metric distance measures according to context. The research uses two modeling methods and compares their results. The first method is Ordered Logit Regression, a statistical method for ordinal dependent variable. The second method is Neurofuzzy Inferencing, a member in the Fuzzy Neural Network family, which models fuzzy relationships with the help of neural networks. Although the models constructed in this research are specific to the sampled population, the modeling methodology can be easily applied to other proximity data.; The research results can be used to extend the capabilities of current geospatial information systems. Secondly, it contributes to the study of proximity spatial relations. And thirdly, it contributes to the research agenda of Naïve or Common-Sense Geography.