In recent years, multiple geospatial ontologies have been developed for a
wide range of di erent spatial databases. In addition, the development of
volunteered geographic information both challenges and provides opportunities to
the traditional authenticated geospatial information. Though volunteered
geographic information is typically not as reliable and structured as the
authenticated geospatial information, it often re ects changes in the real world more
quickly and contains richer information related to human activity [
We propose a new semi-automatic method to align geospatial ontologies,
based on coherence and consistency checking in description logic, as well as
domain experts' knowledge. We evaluate it on real world data and compare it to
two state of the art ontology mapping systems, CODI [
Step 1 : Retractable assumptions include disjointness axioms and mapping axioms. For TBoxes, disjointness axioms are generated for sibling classes. Initial mapping axioms between TBoxes are generated by stating equivalence of atomic concepts with identical names. Initial mapping axioms between ABoxes are generated based on three criteria: location, lexical labelling, and cardinality of mapping (one-to-one or one-to-many). We ensure that the geospatial instances from di erent sources are rst represented at the same scale and using the same coordinate reference system scaling and transforming the input data as necessary. Given two instances, if their geometries are not spatially disjoint, we rst generate a candidate `sameAs' axiom for them. (When dealing with polygon geometries, the geometry checking is based on spatial disjointness, rather than shapes or sizes of geometries or their percentages of overlapping, because two corresponding geospatial individuals may be represented di erently in di erent datasets, and the representations may be of di erent geometry accuracy levels.) Then, each correspondence will be checked lexically. If the labels of the instances cannot be matched, we remove the correspondence. After that, the mapping will go through cardinality checking. In the case that several instances are mapped to the same instance, we change `sameAs' relation to `partOf' relation in the corresponding axioms. The geometry, lexical and cardinality checking are all necessary, since di erent geospatial individuals may share the same label or the same location in an ontology, and a same geospatial individual may be represented as a whole in one ontology, whilst as several parts of it in the other.
Step 2 : Two ontologies are aligned by calculating a CAS with respect to them.
We use Pellet [
The method is implemented as a system called GeoMap. We evaluate it using
the Ordnance Survey of Great Britain (OSGB) Buildings and Places ontology [