In many domains, low turnaround time is highly desirable between obtaining new geographical data of an area and having the information suitable for simulation systems and training. In modern warfare for example, proper tactical planning and training needed to prepare effectively for a certain mission, mandate familiarity with details of the area of operation. Most existing techniques would not achieve a high level of fidelity when rendering the area in 3D unless the GIS data is further augmented and refined by humans. For instance, given initial geographical source data layers consisting of elevations, road surface features and imagery, many techniques would only render road texture over steep terrain. Whereas, a human would immediately distinguish this as improbable by collectively looking at the data layers and note a missing element, an overpass or tunnel.

This paper describes a system which uses deductive reasoning in conjunction with specialized per-element spatial tests and applies it to the GIS data to extract, identify and classify individual spatial elements along with values for their properties. An expert cartographer's knowledge is formalized by means of an ontology. Description Logic reasoners are then used to infer information about instances revealing their true identities and to provide associated property values. In our specific example above, from the analysis of the road feature, the elevation pattern under this road, and image analysis of the specific sub-region in the imagery, the reasoner draws the correct conclusion of the existence of an overpass or tunnel and provides quantitative information needed for 3D rendering, such as location and other parameters for a procedural model. Previous semantics based research in this domain has concentrated more on improving the fidelity through the addition of artifacts like lights, signage, crosswalks, etc. Our work differs in that separating formal knowledge from data processing allows fusion of different data sources which share the same context.