University of Houston

Advances in database and data acquisition technologies have resulted in an immense amount of spatial data, much of which cannot be readily explored using traditional data analysis techniques. The goal of spatial data mining is to automate the extraction of interesting and useful patterns that are not explicitly represented in spatial databases. The motivation for regional knowledge discovery is driven by the facts that global statistics seldom provide useful insight and that most relationships in spatial datasets are geographically regional, rather than global. This dissertation focuses on designing and implementing an integrated framework to systematically discover regional knowledge from spatial datasets. Challenges of the project include finding regions of arbitrary shape at different levels of resolution, providing suitable plug-in measures of interestingness to guide discovery algorithms what to seek, ranking identified regions by relevance, and providing pruning and other sophisticated search strategies to seek highly ranked regions efficiently.

This dissertation work consists of three parts: discovering scientifically interesting places, identifying regional associations and determining the scopes of these associations, and mining controlling factors of geospatial variables. The proposed framework has been evaluated in three real-world applications: (1) identifying spatial risk patterns of arsenic in the Texas water supply; (2) analyzing feature-based hot spots of ground ice on Mars; and (3) studying controlling factors for high vegetation cover in the United States. The experimental results not only rediscover regional knowledge that has been reported in the scientific literature, but also suggest promising new hypotheses for future exploration.