Conceptual Modelling and Spatiotemporal

Information Systems: How to Model the Real

World�Agnar Renolen

Department of Surveying and MappingNorwegian University of Science and Technology, Trondheim

e-mail: agnar@iko.unit.no

ScanGIS’97, June 1 - 3

Abstract

Throughout the relatively young history of research on spatiotem-poral modelling, a substantial number of models have been presen-ted. However, since a spatiotemporal model represent a closer ap-proximation to the real world than is the case for traditional models,such models must be based on a thorough understanding of how ob-jects ’behave’ in reality. One way to acquire such knowledge is byusing conceptual modelling methodologies. In this paper, an overviewof different modelling principles are reviewed. Then a selection ofconceptual modelling languages are presented together with examplesrelated to a spatiotemporal problem. Finally, we conclude that spatio-temporal information seems to have a more complex structure thanpreviously assumed. As a step towards models that to a larger de-gree depicts temporal behaviour in the real world, we suggest ameta-model which is based on a simple behavioural model.

1 Introduction

Most current geographical information systems (GIS)manages spatial data-bases that to a large extent is based on the paper map model. Newer re-search however, are focusing on models that more closely ‘depicts’ con-�This work is funded by the Norwegian Research Council (project 31387)

1

cepts in real world. This includes semantical modelling, fuzzy modelling,and spatiotemporal modelling.In the field of information systems engineering, an abundance of mod-

elling methodologies have existed for a long time. But, “Geographical in-formation systems are often build without due considerations to this dis-cipline, which results in poor portability, maintainability and scalability atthe best” [HT96]. “GIS technology, quite naturally, concerned itself withproblems of capture, storage, manipulation and presentation of informa-tion that is referenced to space. This history of neglect is probably partlydue to the inadequacy of the technical support, both in terms of hardwareand software” [Wor95].“We view informations systems as models of reality. Thus, modelling is

the essence of systems analysis and design. Accordingly, we propose thata theoretical foundation for information systems development theory canbe based on ontology, which is the branch of philosophy that deals withmodelling the real world” [WR89]. That provides, we must emphasizethe difference between information systems analysiswhich aims to acquirea better understanding of the real world, and design which aims to cre-ate a model of the implementation. Because a spatiotemporal informationsystem (STIS) indeed models the real world more closely than traditionalGIS (since such models not handle changes over time), a STIS data modelmore heavily relies on a thorough understanding of phenomena in the realworld.Conceptualmodelling techniques, such as the entity/relationship model

[Che76], are often viewed as tools for describing and developing computerprograms and data structures. However, the view that the same meth-odologies can be used to model real world phenomena seem to have lesssupport amongst software engineers. According to [SK96], “a conceptualmodel is the phenomenon of a domain at some level of approximation ex-ternalized in a semi-formal or formal language”. This definition clearlysupports both views. That is, conceptual modelling techniques can be usedas a tool both in the analysis phase (i.e. real world modelling) as well as inthe design phase (i.e. computer model).In the case of STIS, most research on spatiotemporalmodels has focused

on the development of computer models that are based on simplified con-cepts such as models that view changes as sudden events, i.e. a transaction-oriented view. To create more realistic models, a thorough understanding ofconcepts in the real world is required. The aim of this paper is thus to focuson the analysis phase and present a selection of modelling methodologies

2

and models that are useful in acquiring such knowledge.

2 An Introduction to Conceptual Modelling

Computers and computer languages generally have an abstract nature. Lar-ger systems are compiled from thousands of statements in written sourcecode, and it is virtually impossible for even a skilled programmer to get ageneral view of a large program without any visual support in the form offigures and diagrams. Moreover, in the development of computer systemsit is important that all participants understand the domain to be implemen-ted. In general, software designers and software users represents differ-ent level of knowledge about programming, and communication betweenthem may easily be distorted due to misunderstandings and lack of in-sight. The key to achieve a successful communication among participantsis therefore to make them share relevant conceptual knowledge about thedomain of discourse. This is usually achieved by developing so-called con-ceptual models [SK93].A conceptual model usually have a diagrammatic form (‘boxology’)

with a grammar which usually consists of boxes and links between them.In information systems engineering, a conceptual model serves as a tool forsense-making, as a vehicle for communication, and as documentation andbasis for design and implementation. Conceptual modelling techniques arenot only useful in the design and development of computer data structures,they have also proven to be a valuable methodology in the acquisition ofknowledge of real world phenomena.An interesting side kick here, would be to compare conceptual models

and cartographic maps. Amap can be defined as a selective, symbolic, gen-eralized image of an object presented in a given scale [Bjø]. Comparing thisdefinition of a map with the definition of a conceptual model given above,one may indeed conclude that a map is a conceptual model. A map dorepresent a phenomenon of a domain (a part of the real world seen fromabove), it is representing some level of approximation (it is selective andgeneralized) , and it is presented in a formal language (a symbolic imagewhere the symbols are defined through some legend). This may explainwhy the paper map model has been such a popular basis for spatial com-puter models in GIS, but why this detour? In newer research (such as[HT96]), models are based directly on perceptions of the real world, andnot via the map model.

3

process

P1

store

S1

flow

external

entity

Figure 1: The functional perspective: Symbols in the DFD language.

In the remainder of this section we will give a brief review of the differ-ent modelling principles and then discuss the difference between domainmodels and meta-models.

2.1 A Brief Review of Conceptual Modelling Perspectives

Throughout the history of computers and computer system development,an abundance of different conceptualmodelling languages have been presen-ted. The entity-relationship diagrams and data-flowdiagrams for example,are twowell known modelling languages. In general, modelling languagescan be divided into classes according to the structural principle or the per-spective of the language. In [SK96], the following seven perspectives aredescribed:

The structural perspective: The focus of the structural perspective is ondata and data modelling. The main components of structural modelsare entities, relationships, attributes, and constraints on relationships(i.e. cardinalities). It was the development of the Entity-Relationshiplanguage of Chen [Che76] that represented the breakthrough of thismodelling perspective.

The functional perspective: The functional perspective focuses on processesrather than the data. The best known conceptual modelling languagewith a process perspective is the the data flow diagrams (DFD) whichdescribes the universe of discourse (UoD) in terms of external entit-ies, processes, data stores, and (data-) flows between these. Figure 1shows the Gane Sarson notation of the DFD language [GS79].

The behavioural perspective: The basic concepts of the behavioural per-spective are states and transitions that transform the system from one

4

state to another. One of the problems with this perspective is thatlarger and complex systems quickly becomes unmanageable with analmost infinitely number of possible states. To overcome this, somelanguages such as Statecharts add hierarchical abstraction mechan-isms in the form of AND and XOR decompositions to simplify theview [Har87].

The rule perspective: The main application of the rule perspective is inknowledge systems and artificial intelligence (AI). In general, a rulehas the form: if hconditionithenhconsequencei:The main application of rules is to express constraints on other mod-els such as structural or functional models. One drawback with rulesis that conditions are expected to be either true or false, while in car-tography as in many other applications natural conditions seem tohave a rather fuzzy nature [Bjø95].

The object-oriented perspective: The object perspective have basically emergedas a result of the need to support object-oriented programming lan-guages like SmallTalk, C++ and Eiffel. However, object-oriented ana-lysis and design have truly become a branch on its own, applying thesame concepts that were introduced in the object-oriented program-ming languages (i.e. the object-oriented paradigm).

The communication perspective: The communication perspective are basedon the assumption of language/action theory developed by Austinand Searle called the speech act theory [Aus62, Sea69, Sea79]. A fewmodelling languages exist such as theActionWorkflowdiagrams [M+92].

The actor role perspective: The actor and role perspective is based on ideasdeveloped during work on object-oriented programming languagesand intelligent agents in AI. Basic constructs of this perspective areactors, roles and agents.

At first glance, it is quite evident that several of these perspectives areuseful in the development of a STIS. Behavioural models may help us un-derstand how objects changes over time and functional models may pos-sibly support us in that process. Structural models on the other hand nor-mally represent the real world in a static fashion, although extensions tothe ER-language that supports changes over time— such as the ERT-model

5

[MSW92] — have been developed. The object-oriented approach is closelyrelated to the structural approach since it describes relationships betweenobjects, but as the approach incorporates the functionality of the objects,relations to the functional approaches can be drawn as well. Moreover,object-oriented models may further be supported by Objectcharts [CHB92]which is an object-oriented adaption of Statecharts.Also the rule perspective is of great value in the development of tem-

poral information systems in general. Rules that are coupled with businesspolicy may change over time, and historical data should be viewed in con-text with the current policy at the time in question. The ability to expresspolicy in the form of explicit rules is therefore critical.

2.2 Domain Models and Meta-models

Normally, conceptualmodelling dealswith developing domain modelswhichattempts to describe some domain in the real world or in a computer sys-tem, as for example a cadastre. When implementing these models we needan underlying data structure — a meta-model — upon which the higherlevel domain model can be implemented. For example, we can use a poly-gon mesh to implement the spatial description while ER-models can bedeployed for expressing the topological and non-spatial characteristics ofour model. In general, the main uses of meta-models are [met]:� As conceptual schemas for repositories that hold software engineer-

ing and related data.� As conceptual schemas for modelling tools such as CASE1 tools.� Todefine modelling languages, such as for object analysis and design.� As part of technology (together with a transfer mechanism) that al-lows interoperability of modelling tools.

Much research in spatiotemporal issues is devoted to rather low levelmodels. Examples of models that have been presented are the space-timecomposite model [Lan88] and the Multimodels/Metamap model [BS93,Mis93]. Consider the case of a cadastral database. At the lower level, wemay implement the space-time composite model for describing the spatialconcepts as the parcels, and the boundaries between them. At the higher1CASE: Computer Aided Software Engineering

6

43/162

43/161

43/66,70

43/14843/157

43/158

43/159

43/163

43/156,160

Figure 2: A cadastral map showing parcels and a street.

level, we need to describe relationships with objects, their spatial represent-ation, and their relationship with other objects (such as a persons owner-ship in a parcel, and the boundaries relationship with boundary markers).Apparently, there seem to be little interest in such higher level modelling(i.e. domain modelling) in GIS today, and the meta-models does often notallow higher level relationships to be established.On the brighter side, in [LT93] good examples of domain models can

be found, and in the case of the Metamap model higher level relationshipsare embedded as part of the model. Also, many commercial GIS uses thegeo-referenced data model which separates the spatial description of fea-tures from the aspatial description which may be handled in a standardrelational database system. These systems allows for higher level relation-ships of the domain model to be implemented through the this databasesystem.

3 Structural Modelling and the Time Dimension

3.1 Entity RelationshipModels

The Entity Relationship model, or in short the ER-model, developed byChen [Che76] was not the first language for semantical data modelling,but it certainly became the most popular one. The main reasons for its pop-ularity is the simple diagrammatic representation and the easy transition to

7

Parcel

shipOwner-

Owner

N

1

BoundaryBoundarySegment

Boundary

Street

N

1

Markings BoundaryMarker

N2 1 N

Figure 3: A sample ER model describing entities in a cadastre

tables in relational databases. Although, the intension of the ER-modellinglanguage was to describe the structure of (relational) databases, the ER-language is highly appropriate for modelling real world domains as well(i.e. not only the part to be stored in a database).There are two basic constructs of the ER-model: entity, relationship, and

attributes. Chen defined an entity as “a thing that can be distinctly identi-fied” and a relationship as “an association among entities”. This definitionof an entity is clearly a wide one, which is really a strength of the ER-model.Each entity is characterized by a set of attributes which is common to allentities of that type. Figure 3 shows an ER-diagram modelling a cadastre,describing the relationship between parcels, owners and streets. The modelis based on the concepts outlined in figure 2 where streets and parcels areconsidered to be two different types of entities.However, for modelling concepts in the real world the ER-language

have several shortcomings. The EER language is an extension of the ER-model which also included concepts of sub-typing and association [T+86],concepts that today are part of the object-oriented approach.

3.2 The ERT language

With the recent research in temporal databases, the need for an accompa-nying conceptual modelling language has become evident. Since much ofthe research on temporal databases have been focusing on extending therelational model, it is not surprising that most conceptual modelling lan-guages supporting time are extensions to the ER or the EER-model. Onesuch language is the ERT modelling language (Entity-Relationship with

8

Street

T

TOwner

TParcel T TBoundary

TBoundaryMarker

Company

Person

PrivatParcel

StateProperty

T T

ID

NameAddress

T

T

(1:N)

(1:2)

separates

marked by(0:N)

marks (1:1)

(1:1)

Owns (1:N)

(1:N)

(1:1)

of

has

(1:1) of

(1:1) has

has

owned by

has

(1:1)of

(1:1)

Figure 4: An example of an ERT model showing a cadastre

Time) [MSW92]. In this language, which is based on the EER-model, sup-port for temporal concepts have been applied by the use of time-stampingof entities and relationships.The basic constructs of this language is the entity class (denoted by a

rectangle) which denotes a set of objects which share the same set of at-tributes, the value class (denoted by a rectangle with a black corner) whichis used to describe an entity’s attributes, and the relationship (denoted bya line with small black square) which describes the associations betweenan entity and a value class or another entity. Furthermore, the inheritanceextension is represented by a relationship with a circular join. If the circle issolid then the sub-classes are disjoint or total, if open then the sub-classesare overlapping or partial.To implement the temporal dimension to the ERT diagrams, entity classes

9

and relationships may be either T- orH-marked. If an entity is T-marked itmeans that the entity only exist at certain times (or ticks) in our UoDmean-ing that the entity is undergoing temporal variation. If a relationship is T-marked, it means that the relationship between the two entities it involvesexists for only a subset of the time (number of ticks) for which both the en-tities it associates exist. If a relationship between two entities of which atleast one is T-marked is not T-marked, it means that the relationship existas long as both entities co-exist in our UoD.The H-mark is used to indicate that a relationship has a historical per-

spective. This means that a relationship involvement may exist betweentwo entities that do not co-exist in time. For example, we may say that oneperson has a grandparent that is another person, but the two persons didnot co-exist in our UoD if the grand parent died before the grand child wasborn. However, in the grand parent example, wemight want to say that thegrandparent is related to its grandchild from the time that the grandchildbegins to exist. For this purpose, we will use the TH-mark.Figure 4 shows an improvedmodel of a cadastre using the ERT-language.

In this model, we have distinguished between streets, state properties andprivate parcels. The latter type can be owned by either a person or a com-pany. Because the ERT model both supports temporal aspects and sub-typing it is particularly interesting in the design of an object-oriented sys-tem. Furthermore, the notation of the ERT-language allows us to read outthe relationships, as for example the relationship between private parcelsand owners: A parcel is owned by exactly one owner, while an owner ownsfrom 1 toN parcels.However, the ERT language does not distinguish between the differ-

ent relation types used in object oriented analysis such as association andaggregation. For this reason the ERT model can be augmented for use inobject-oriented analysis.

4 Modelling Temporal Behaviour

Understanding temporal behaviour is one of the most fundamental issuesof STIS engineering. A simplistic view that many researchers of temporaldatabases as well as of temporal GIS seem to adopt, is to represent objectsonly in terms of static representations, viewing changes as sudden events.This view most likely stem from the fact that changes (or transactions) indatabases always are of a sudden nature. However, we know that many

10

stepwise constant

t

acontinuous

a

t

uniformlysmoothlyiregularly

discretea

tt

agradually changing

Figure 5: Temporal behaviour of an attribute

Staticstate

Changingstate

Event C

Ceased

Temporal ObjectCessation

Cessation

ReincarnationCreation

Figure 6: Generic behaviour of temporal objects using the Statechart nota-tion

11

changes in the real world have duration. Actually, features in the real worldexhibit a wide range of temporal behaviour. Some features such as admin-istrative units are indeed subject to instantaneous changes. Other featuressuch as glaciers are changing continuously, while yet another group of fea-tures may sometimes be in a changing state while other times in a staticstate. An example of the latter type of behaviour is roads. While the road isunder construction or reconstruction, it is in a changing state, but when theconstruction is done the roadmay be considered to be in a static state. Sum-ming up what can be found in [SS93, MP93, Ren97] a total of four differentbehaviour types can be found. These are as shown in figure 5:

Stepwise constant: A feature of this type is considered to persist until someinformation is introduced about suddenly changing its state.

Gradually changing: A feature of this type is considered to persist in someperiods, while changing in other periods. Changes of this type arenormally considered as events with duration.

Continuously changing: Features of this type, are always considered to bein a changing state. Whether we may interpolate according to a givenfunction for values not explicitly stored in a dataset is determined byhow smoothly or irregularly the value changes.

Discrete values: Features of this type are considered to apply only on thetimes for which they are specified; the number of cars passing a cer-tain point on a road is an example of such values.

Tomodel temporal behaviour, it seems natural to use a language such asStatecharts [Har87] or Objectcharts [CHB92]. However, these languages aresimilar to finite state machines which is based on instantaneous transitionsbetween states. Thus at first glance, they seem of less value for modellinggeographical objects as we know that changes may have duration. On theother hand, if we introduce the state of change as a distinct state, we mayobtain a generic Statechart model (i.e. a meta-model) for spatiotemporalobjects as shown in figure 6. In this model an object may either be in a staticstate, a changing state or it may be ceased (We prefer to use the term cease orcessation here as we do not physically destroy the database representationof the object). Whether the creation leads to a changing or to a static statedepends on the type of object, hence the Conditional (C) transition. Themodel also allows an object that is ceased, to be reincarnated some time later[Lan92].

12

C

B

DE

F

A

AB

C C

E

D

F

D

T1 T2 T3 T4T1

T2

T3

T4

Time cross section

B M3

C

M4 F

D

EM1 M2A

T1 T3 T4 tT2

Figure 7: The story of a land area (above) shown in the history graph nota-tion (below).

It was the idea of this model that lead to the definition of the historygraph notation [Ren96]. In this language, an object’s history may be de-scribed through a series of consecutive versions (i.e. static states) and trans-itions (i.e. changing states). The versions are denoted by a square rectangle,while the transitions are denoted with boxes with circular ends. Both theversions and the transition can be characterized by a time interval. Ob-jects that changes suddenly would then be described by transitions withzero duration (i.e. events), while objects that changes continuously woldbe described by version with zero duration (i.e. snapshots) describing in-termediate states. Figure 7 shows a sample story where parcel are split andmerged. Since the changes in this story are considered as sudden events,the transitions are shown as circles.Studying figure 7, onemay identify a number of distinct transition types,

such as splitting and merging of objects. In total, six different transitiontypes were identified. These are:

Creation: An object is created.

Alteration: An object is changed or modified.

Cessation: An object is destroyed or removed.

13

reincarnation

alterationcreation

cessation

merging / annexation splitting/deduction

Figure 8: Six basic types of changes (Shown in history graph notation)

Reincarnation: An object that previously has been destroyed or removedis reintroduced, possibly with a new state and location.

Split/Deduction: An object is subdivided in two or more new objects orone ore more objects is deducted form an existing object.

Merge/Annexation: Two ormore objects are joined together to form a newobject or one ore more objects are ’swallowed’ into another object.

One could also add ‘reorganization’ to the six alreadymentioned (whichwould be a many-to-many relationship between object versions), but wethink that such transitions are sufficiently rare not to justify its inclusion.Figure 8 shows the six different change types in the history graph notation.Albeit the history graph notation helps us understand temporal beha-

viour in particular cases, it does not allow us to describe general behaviourof certain feature types. For example in a cadastre we may identify a finitenumber of change-classes all of which are events with zero duration:� A parcel may be created by deducting a part form another parcel. This

also means that we have to associate the new parcel with an owner.� A parcel may be ceasedwhen another parcel annexes it into its extent.� A parcel may change owner.� A boundary between two parcels may be adjusted after resurveying,after resolving a disagreement, or by deed of conveyance.

14

� A corner between three or more parcels may be adjusted after resur-veying or after resolving a disagreement amongst the owners.

Looking at figure 6, it is obvious that the change state is non-existentin a cadastre since all changes are events. More interesting though is tomake transition specifications of the changes outlined above. Formally ina Statechart, a transition specification comprises the initial and final stateof the transition and the service name for the transition, together with aprecondition and a postcondition.A problem with the Statechart is that they badly express the interaction

of one object with other objects. For example, how canwe comprehensivelydescribe a deduction. A deduction generally means the creation of oneobject and the alteration of another. Obviously, such issues have not beenaddressed, although the history graph notation to a large extent can assistin such specifications.

5 Object-OrientedMeta-models

A popular approach in GIS modelling in general and spatiotemporal mod-elling in particular is the object-oriented approach. Severalmodels have beenpresented [RMD94, Ham95, Wor94] although most of them are rather lowlevel-models. Montgomery lists fourmain advantages of an object-orientedmodel in a temporal GIS [Mon95]:

1. The complete history of an entity can be represented as one singleobject.

2. Since the complete history of an entity can be represented as a singleobject, queries become less complicated, because they do not considerthe dispersion of the entity over many tuples.

3. Since complex object queries are executed efficiently, the correspond-ing temporal data should be handled efficiently as well.

4. Handling of temporal and non-temporal data can be done in a uni-form way.

The object-oriented approach provides concepts as object class, associ-ation, aggregation, and generalization. A popular modelling language is

15

Class Exactly one

Class Many (zero or more

Class Optional (zero or one)

Class One or more

Class Numerically specified

Multiplicity of Associations (cardinalities):

1+

1,2-4

Class name

Class-2Class-1Association Name

role-1 role-2

Association:

Assembly Class

Part-1 Class Part-2 Class

Aggregation:

Superclass

Subclass-1 Subclass-2

Generalisation (Inheritance):

Class Name

list of attributes

list of operations

Class:

Figure 9: The most common symbols of the OMT-language

the one of the object modelling technique (OMT) [R+91]. The most com-mon symbols of this language is shown in figure 9.So far, we have been treating a spatiotemporal feature as a single unit

that is subject to temporal variation. However, a temporal object may becharacterized by a number of different attributes and descriptions, all ofwhich may change independently in time. In the case of a land parcel,the spatial description of a parcel may remain constant while the parcelschange owner several times. As will be shown in the sequel, object-orientedmethods have proven to be a magnificent tool for creating meta-modelsthat deals well with this problem.The Temporal BaseModel (TBM) is one example of such amodel [Ren97].

The model is based on the Multimodel concept [BS93, Mis93], extendedwith the concept outlined in figure 6. In this model which is shown in fig-ure 10, each object can be characterized by a number of temporal descriptorswhich may vary independently in time. Spatiotemporal objects is thus re-quired to have at least one temporal descriptor that provides a spatial de-scription. Another descriptor may describe an object’s relationship withother objects while a third descriptor may provide a set of attributes for the

16

Version

TransitionVersionLevel

TransitionObjectLevel

TemporalObject Descriptor

Temporal

Transition

Touch

1+1+ 1+

Figure 10: The Temporal BaseModel (TBM) shown using the OMTnotation

object.Each descriptor comprises a set of descriptor versionswhich represent the

static state of a descriptor. A transitionmay associate objects and descriptorversions in predecessor-successor relationships representing the changingstate of a descriptor. Having a closer look at TBM, one may realize thatthe model solves the problem of interaction between different objects. Thishad lead to three types of transitions: object-level transitions that providea temporal topolgy or historical links between objects, version-level trans-itions that provides a history graph-like structure for descriptor versionsand touches that represent some kind of action on objects that generally donot represent any change. For example, an object may have been inspectedand verified to be in correspondence with its database representation, or itmay be subjected to maintenance activities.At a higher level, there are currently no object-oriented languages that

support temporal concepts. However, the ERT-language supports the mostimportant concepts and as stated, slight modifications to the ERT can beapplied to support the whole range of object-oriented concepts.

6 Rules and Business Policy

In cartography, rules have mostly been applied to issues related to carto-graphic generalization. In conceptual modelling on the othee hand, rules

17

can be used to express constraints on conceptual schemas (such as cardin-ality constraints).Most often, rules are implemented directly in source code. In inform-

ation systems this represent potential problems since business policy maychange over time, making current information systems obsolete [M+91].Historical data should always be viewed in context with the rules that werecurrent at the time in question. For example in a cadastre, one may at onetime state that a parcel cannot be owned by more than one person (or insti-tution), while at an earlier or later time a group of people would be allowedto share the ownership in the same property.A promising solution is to explicitly express rules in a External Rule Lan-

guage (ERL) and store them together with the system. There are two ap-proaches to this problem. One is to store the rules in a temporal database,such that rules easily can be obtained for specific points or periods of time.Another approach is to let the time validity be an inherent part of the ruleif the rule is expressed aswhenhtimeiifhconditionithenhconsequenceiThe rule can then be viewed as describing some logical constraint on themodel, which must hold at every moment (or tick) whilst the informationsystem is active [M+91].7 Discussion

In this paper we have presented a selection of conceptual modelling ap-proaches and demonstrated some of their applications in the developmentof a STIS. Through this, we have shown that the structure of temporalinformation can be more complex than current research seem to acknow-ledge. This is illustrated by the temporal base model which is inspired bythe Multimodel and Metamap concept [BS93, Mis93], and extended withideas incorporated in the history graph model. Behind all these concepts isthe behavioural model that is presented in figure 6 which states that objectsare either in a static state, in a changing state, or in a ceased state.On top of this underlying meta-model, we need to create application-

specific domain models which precisely describes a domain in the realworld. An example of such a domain is a cadastre, which may describethe relationship between parcels, streets, boundaries and land owners.

18

A major challenge though, is to deal with the fact that constraints ondomain models may change over time as a result of changing (business)policy. A promising solution to this is to express such constraints in theform of explicit rules. Rules can be stored in their own temporal databaseand made available to the system to ensure that the data comply with theserules at all times.Finally, we suggest that the design of a STIS should be based on the

meta-models presented in this paper. Saying this, we emphasize the differ-ence between the analysis and the design phase, and the importance of notrestricting oneself by implementation constraints in the analysis phase.

References

[Aus62] J.L. Austin. How to do things with words. Harward UniversityPress, Cambrigde MA, 1962.

[Bjø] Jan T. Bjørke. Digital kartografi, del 1. Compendium in courseDigital Cartography.

[Bjø95] Jan T. Bjørke. Fuzzy set theoretic approach to the definition oftopological spatial relations. In Proceedings: ScanGIS’95, the 5thScandinavian Research Conference on Georgraphical Information Sys-tems, Trondheim, Norway, 1995.

[BS93] Olav Mork Bjørnas and David Skogan. Multimodels and spatio-temporal modeling in object-oriented GIS. Master’s thesis, Nor-wegian Institute of Technology, 1993.

[CHB92] Derek Coleman, Fiona Hayes, and Stephen Bear. Introducingobjectcharts or how to use statecharts in object-oriented design.IEEE Transactions on software engineering, 18(1), january 1992.

[Che76] P. P. Chen. The entity-relationship model: Towards a unifiedview of data. ACM Transactions on Database Systems, 1(1), 1976.

[GS79] C. Gane and T. Sarson. Structured System Analysis: Tools and Tech-niques. Prentice-Hall, 1979.

[Ham95] Torill Hamre. Development of Semantic Spatio-temporal Data Mod-els for Integration of Remote Sensing and in situ data in a Marine

19

Information System. PhD thesis, University of Bergen, Norway,1995.

[Har87] David Harel. Statechart: A visual formalism for complex sys-tems. Science of computer programming, (8):231–274, 1987.

[HT96] Thanasis Hadzilacos and Nectaria Tryfona. Logical data model-ling for geographical applications. Int. J. of Geographical Informa-tion Systems, 10(2), 1996.

[Lan88] Gail Langran. A framework for temporal geographic informa-tion systems. Cartographica, 25(3), 1988.

[Lan92] Gail Langran. Time in Geographic Information Systems. Tech-nical Issues in Geographic Information Systems. Taylor & Fran-cis, 1992.

[LT93] Robert Laurini and Derek Thompson. Fundamentals of Spatial In-formation Systems. Academic Press, 1993.

[M+91] Peter McBrien et al. A rule language to capture and model busi-ness policy specifications. In Advanced Information Systems En-gineering, CAiSE’91, Trondheim, Norway, May 1991.

[M+92] R. Medina Mora et al. The action workflow approach to work-flow management technology. In Proceedings of CSCW’92, 1992.

[met] metamodel.com. Whatis metamodelling, and what is a metamodel good for? WWWsource, url=”http://www.metamodel.com/”.

[Mis93] Gunnar Misund. Multimodels and metamap— towards an aug-mented map concept. Master’s thesis, University of Oslo, 1993.

[Mon95] Leslie David Montgomery. Temporal geographic informationssystems technology and requirements: Where we are today.Master’s thesis, The Ohio State Univeristy, 1995.

[MP93] Angelo Montari and Barbara Pernici. Temporal reasoning. InTemporal Databases. Theory, design, and implementation. The Ben-jamin/Cummings publishing company inc., 1993.

20

[MSW92] Peter Mc.Brien, Anne Helga Seltveit, and Benkt Wangler. Anentity-relationship model extended to describe historical inform-ation. In Proceedings: CISMOD ’92, 1992. Bangalore, India.

[R+91] James Rumbaugh et al. Object-Oriented Modelling and Design.Prentice-Hall International, Inc., 1991.

[Ren96] Agnar Renolen. History graphs: Conceptual modelling ofspatiotemporal data. In GIS Frontiers in Business and Science,volume 2. International Cartographic Association, 1996. Brno,Czech Rebublic.

[Ren97] Agnar Renolen. The spatiotemporal object model: A conceptualmodel based on ontology. Article submitted to Geoinformatica,1997.

[RMD94] Bhaskar Ramachandran, Fraser MacLoad, and Steve Dowers.Modelling temporal changes in a GIS using an object-orientedapproach. In Advances in GIS research, 1994. Sixth internationalSymposium on Spatial Data Handling.

[Sea69] J.R. Searle. Speach acts. Cambrigde University Press, Cambrigde,1969.

[Sea79] J.R. Searle. Expression and Meaning. Cambrigde University Press,Cambrigde, 1979.

[SK93] A. Sølvberg and D.C. Kung. Information System Engineering, AnIntroduction. Springer-Verlag, 1993.

[SK96] Arne Sølvberg and John Krogstie. Information Systems Engineer-ing – Advanced Conceptual Modeling. draft, 1996.

[SS93] Arie Segev and Arie Shoshani. A temporal data model based ontime sequences. In Temporal Databases. Theory, design, and imple-mentation. The Benjamin/Cummings publishing company inc.,1993.

[T+86] Toby J. Teorey et al. A logical design methodology for relationaldatabases using the extended entity-relationship model. ACMComputing Surveys, 18(2), June 1986.

21

[Wor94] Michael F. Worboys. Unifying the spatial and temporal compon-ents of geographical information. In Advances in GIS research,1994. Sixth international Symposium on Spatial Data Handling.

[Wor95] Michael F. Worboys. GIS: a Computing perspective. Taylor & Fran-cis, London, 1995.

[WR89] Y. Wand and R.Weber. An ontological evaluation of system ana-lysis and design methods. In E.D. Falkenberg and P. Lindgren,editors, Information Systems Concepts: An in-depth Analysis. El-sevier Science Publishers, New York, 1989.

22