Fusing uncertain structured spatial information

Abstract

Spatial information associates properties to labeled areas. Space is partitioned into (elementary) parcels, and union of parcels constitute areas. Properties may have various level of generality, giving birth to a taxonomy of properties for a given universe of discourse. Thus, the set of properties pertaining to a conceptual taxonomy, as the set of areas and parcels, are structured by a natural partial order. We refer to such structures as ontologies. In fusion problems, information coming from distinct sources may be expressed in terms of different conceptual and/or spatial ontologies, and may be pervaded with uncertainty. Dealing with several conceptual (or spatial) ontologies in a fusion perspective presupposes that these ontologies be aligned. This paper introduces a basic representation format called attributive formula, which is a pair made of a property and a set of parcels (to which the property applies), possibly associated with a certainty level. Uncertain attributive formulas are processed in a possibilistic logic manner, augmented with a two-sorted characterization: the property may be true everywhere in an area, or at least true somewhere in the area. The fusion process combines the factual information encoded by the attributive formulas provided by the different sources together with the logical encoding of the conceptual and spatial ontologies (obtained after alignment). Then, inconsistency encountered in the fusion process may be handled by taking advantage of the existence of different fusion modes, or by relaxing when necessary a closed world-like assumption stating by default that what is true somewhere in an area may be also true everywhere in it (if nothing else is known). A landscape analysis toy example illustrates the approach. © 2008 Springer-Verlag.