What is a Geospatial Ontology? (2JO4)
The following summary is a non-technical start on this topic by Dr. Gary Berg-Cross. This discussion reflects some early work by SOCoP members and their demonstrations. It is intended to be a starting point for further input from the SOCoP and broader ontology community on this topic as an aid to the Interop effort. The discussion is at the conceptual level for domain specialists and does not try to discuss technical issues such as the use of ontology languages for representing ontological models. Within the Interop effort domain experts are not expected to have to deal with this issue (2JOF)
== Ontologies in General== (2JOQ)
In general' an ontology specifies a vocabulary of concepts together with some indication of their meanings. There is a range of levels of precision with which meaning is specified, but an overall goal is for the ontology to be in form (hence it is called a formal ontology) that express a common understanding of the structure of information suitable among people and-or software agents. Often there are twin targets to make the meaning clear to people while allowing a degree of "automated" processing. In this case a formal ontology can be used to make explicit the semantics, spatial cognition and knowledge contained within efforts such as software applications. (2JOA)
An ontological model is made up of classes/concepts (at least partial hierarchical) along with properties & attributes for these concepts (usually with descriptions to help humans)plus constraints on properties and attributes. The backbone on an Ontology is made up of Classes in a formal Hierarchy. Certain parts of an ontology are the primitives, foundational concepts from which other concepts are constructed. The concept of part and boundary are examples of primitives in some ontologlies. (2JOB)
A class is a concept in the domain so we may have a class of organizations (e.g. USGS) or a class of regions (e.g. Mid-Atlantic States). A class also has a population called instances which is the collection of the elements with similar properties defined by the class concept. Thus there are instances of states such as Maryland in the Mid-Atlantic class. (2JOC)
We can illustrate these general ideas using a very small, informal ontology made up of a few geospatial objects and some general relations. The Figure below illustrates an ontology using 3 general relations - part, sub-type and instance describing that portion of the USGS hydrology model for "hydrologic units". The ontological model has 4 sub-types showing a range of aggregation. Thus regions or sub-regions are a type of hydrologic unit and a specific instance of a region is the Mid-Atlantic, which includes as part of it Delaware. For accounting purposes Delaware may be dividing into parts - in this case and upper and lower part. These are asserted as both part of Delaware and as an instance of an "accounting Unit"
Ontology from Hydrologic Ontologies Framework (HOW) by Michael Piasecki, Bora Beran & Luis Bermudez (2JOD)
A final sub-type of hydrologic unit is a "cataloging unit" which has even finer part of accounting units and in this case the Schuylkill is an instance of this and part of the Lower Delaware accounting unit. A more complete ontology might include features of regions and units such as its size, borders etc. (2JOE)
As part of such an effort an ontology can be used to enable reuse of domain knowledge and to make domain assumptions explicit or to separate domain knowledge in a declarative form from the operational knowledge which can be implemented in software. (2JDH) (2JO9)
Geospatial Ontologies (2JOR)
Geospatial ontologies take as their domain a range of geospatial concepts such as geospatial objects (such as in the previous hydrologic example), relations (including special ones for spatial concepts) and features. (2JOG)
Relations (aka predicates as in the proposition "the door is in front of me) start with seemingly basic things such as topological and qualitative ideas of "near", "connected", "in front of" and "around" as well as other common spatial relationships in use, equals, disjoint, intersects, touches, crosses, within, contains, and overlaps. Spatial objects include abstract spatial notions such as "place" or "locations". Geospatial objects, such as a city or lake may be abstracted to geometrical concepts like a point or polygon area concepts in order to be understood at different levels of granularity. Thus an area like DC can be presented as a small oval on a national map or a complex area when zoomed up close. Features include "size" and "volume". These provide summative information about spatial things. Besides these simple ideas it also includes more macro, aggregated and complex concepts like "river", "estuary", "pond" and "lake". Knowledge about each of these would include relations to other things (e.g. each has boundaries with non-water objects, a river may connect to a lake etc.) Ontologies attempt to define and distinguish all of these concepts in explicit and precise ways to avoid confusions. It also my provide semantic glue by which diverse sources of information are brought together. At the 2009 SOCoP workshop John Bateman (University of Bremen) discussed ontolgoical issues in bringing together sources as diverse as: (2JOH)
- Location-based services (2JOI)
- Geographic Information Systems (2JOJ)
- Commonsense objects and activities (2JOK)
- Spatial awareness and understanding (2JOL)
- Natural language capabilities (2JOM)
- Robot perception (2JO7)
See John Bateman [slides ] (2JOO)
Getting such diverse areas of expertise to talk to each other is a challenge because the knowledge comes from different communities with different interests and it is often formalized using different (data) representations. For all of these reasons communities, including geo-science communities, wind up with divers pools of knowledge maintained in very different systems unable to share information. (2JOT)
An ontology helps by standardizing concepts expressed by a vocabulary. An example is the term and concept "near". One might want to distinguish something called DC that is near Baltimore and say: If DC is near Baltimore then Baltimore and DC are not connected, and also that if Baltimore is near DC then DC is also near Baltimore. (2JOP)
These can be stated in an ontology as axioms to rule out the possibility of interpreting “near” as “connected” (but a connected path with certain attributes such as above flood level between objects may be of interest). One would also need to distinguish near from far since neither is connected. (2JO6)
Even ontologies differ starting with different primitives and making different distinctions. But often simple alignments can be made to bring general and domain ontologies, such as geospatial ones together. An example of inserting some of the previous USGS region ontology into another ontology called SUMO (Suggested Upper Merged Ontology see http://www.ontologyportal.org/) is shown below.
(2JOS)