2 Ontology for Spatio-temporal DatabasesIntroductionOntology to Drive Information System DesignOntological Problems of Geographic Information Systems and Other Spatio-temporal Information SystemsStructure of the ChapterThe Notion of OntologyClassical ViewSocial RealityApplication DomainsTable-Top SituationCityscapeGeographic LandscapeModel of Information SystemsInformation Systems as Vehicles of Exchange between Multiple AgentsCorrectness of Information System Related to ObservationsSemantics for Terms in Information SystemsGrounding of Semantics in Physical OperationsThe Five Tiers of the OntologyPhysical Reality Seen as an Ontology of a Four-Dimensional FieldObservation of Physical RealityOperations and Ontology of IndividualsSocial OntologyOntology of Cognitive AgentsThe Language to Describe the OntologyTools to Implement OntologiesMulti-agent Systems and Formalization of Database OntologiesOntological Tier 0: Ontology of the Physical RealityPropertiesPhysical Space-Time FieldOntological Tier 1: Our Limited Knowledge of the World through Observations of RealityObservationsMeasurement UnitsClassification of ValuesSpecial Observations: Points in Space and TimeApproximate LocationDiscretization and SamplingVirtual Datasets: Validity of ValuesOntological Tier 2: Representation -- World of Individual ObjectsObjects Are Defined by Uniform PropertiesGeometry of ObjectsProperties of ObjectsGeographic Objects Are not Solid BodiesObjects Endure in TimeTemporal, but A-Spatial ObjectsOntological Tier 3: Socially Constructed RealitySocial Reality Is Real within a ContextNamesInstitutional RealityOntological Tier 4: Modeling Cognitive AgentsLogical DeductionTwo Time PerspectivesSources of KnowledgeOntological Commitments Necessary for a Spatio-temporal DatabaseExistence of a Single RealityValues for Properties Can Be ObservedAssume Space and TimeObservations Are Necessarily LimitedProcesses Determine ObjectsNames of ObjectsSocial, Especially Institutionally Constructed RealityKnowledge of an Agent Is Changing in TimeConclusionsReferences2 Ontology for Spatio-temporal DatabasesAndrew U. FrankTechnical University of Vienna, Austria2.1 IntroductionOntology and the related term “semantics” have recently found increased atten-tion in database discussions. Early discussions of ontology issues important fordatabases [126,78] were lost in a sea of papers on technical, mostly performanceissues, despite the fact that textbooks as early as [134] discussed briefly the re-lationship between information system and real world. This is different today;interest in semantics has increased, and this will be more so in the future giventhe current interest in the Semantic Web [11,10] (see Chapter 9 of the book forrelated discussion).Information systems and their implementation as databases rest on onto-logical commitments. Decisions about the type system used, how identifiers aremanaged, and so on, are derived from a specific view of the world to whichthe database relates, in other words from a specific ontology. The ontologiesof standard database models make very limited assumptions and therefore thedata model is widely applicable. Spatio-temporal databases must make strongercommitments to capture the meaning of space and time. Such an ontology isnecessarily more involved and the connection to the application area stronger.The designer of a database application has to reconcile the ontological conceptsfrom the application area with the ontology built into the database. Optimally,a spatio-temporal database involves in its built-in ontology a minimal commit-ment on how space and time is structured and is thus most open for applicationspecific refinements. Exploring the minimal set of ontological commitment is thegoal of this chapter.The ontology built into a DBMS can be insufficient or it can be too re-straining. It is insufficient if the ontological categories necessary for numerousapplications are not available and must be reconstructed for each applicationanew; the resulting incompatibilities will be very costly to correct later [81]. It istoo restraining if the ontology commits those who apply it to assumptions whichdo not hold for novel applications. Spatio-temporal databases are typically con-structed to integrate the knowledge of many agents and face the problem ofheterogeneous environments, a point already raised by Wiederhold et al. [190,Chapter 22]. Current databases do not allow us to model joint beliefs of groupsof agents which do not correspond to similar beliefs of other groups of agents;for example, Reuter works with groups of scientists, who manage terabytes ofreports of results from experiments in cellular biology, where the validity of theresults and their interpretation are debated among the groups. Current onto-logical investigations related to databases and information systems have beenT. Sellis et al. (Eds.): Spatio-temporal Databases, LNCS 2520, pp. 9–77, 2003.c Springer-Verlag Berlin Hei delberg 200310 Andrew U. Frankextended into the spatial domain [36,37,62,63,65,160,188], but their extensioninto the spatio-temporal domain [44,97,98,120,139] has proved more difficultthan expected [86,182]. An overview of Time Ontology for computer science waspublished by [170]; Montanari and Pernici discuss the different proposals fortemporal reasoning [190, Chapter 21].I will investigate the questions which arise when information systems are builtfor purposes involving the representation of real space and time. Examples fromthe domain of Geographic Information Systems demonstrate the issues. Geo-graphic Information Systems are especially suited for our purposes because theymodel real-world situations including their spatial and temporal aspects. Theirapplication area is very broad and extends from the administrative and legalrules governing land ownership and registration [54] to systems built for envi-ronmental purposes [111] and for research into global change [145]. The situationis not substantially different for other spatio-temporal systems, like systems formotor traffic monitoring or tracking airplanes. Spatio-temporal databases areoften built from data from many different sources, which is notoriously difficult[103,199]. Data to be integrated differ in their semantics and representation, anda meaningful combination requires bridging the gap created by ontological as-sumptions as well as translations between the representations once their meaningis in the same context. But