We then describe the coverage we believe a widely-used geospatial ontology would have to have. It should include the possibility of rich descriptions of the topology of complex regions and three-dimensional structures. It should include a way of talking about direction, multiple frames of reference, shape, and size. It should have access to large compendia of natural, man-made, and geopolitical entities. One should be able to view geospatial fields and objects at different granularities. It should be possible to combine it with ontologies of time and of events and processes to produce ontologies of motion and change of varying complexity.
We examine the state of the art in spatial representation and reasoning in artificial intelligence, particularly with respect to qualitative topological information, hybrid qualitative-quantitative representations, directions, and combining space and time.
We then survey representative examples of several categories of geospatial resource, looking at their implicit or explicit ontology. The types of resource include geospatial datasets, such as the Getty Thesaurus of Geographic Names; geographic information systems, such as ArcGIS; geospatial ontology standards, such as the OpenGIS Feature Geometry; and large-scale research efforts on geospatial ontology, such as ResearchCyc and SUMO.
We discuss the problem of uncertainty in geospatial data, and list several different kinds of uncertainty. We briefly survey work on reasoning about and visualizing uncertainty in information.
We close with three principal recommendations:
The full report is available in PDF (104pp).