Flexible Support For Spatial Decision-Making

ProblemIdentificationProblemIdentificationLayering of spatialaspectsInstantiation of spatialmodels in a single layerwith appropriate data/layer/ themeInstantiation of non-spatialmodels for a single aspectwith appropriate dataNeedTransformation?NeedTransformation?3. The FSDSS FrameworkExecution ofspatial modelswith appropriatesolverIntegration ofnon-spatial aspectsIntegration ofspatial layersIntegration results of spatial and nonspatial aspects(Modelling of wledgeScenarioExecutionSolverSpatial and Non-Spatial Object RepositoryChoiceScenario evaluationand sensitivityanalysisApply DecisionMakersPreferenceorBuild MCDMModelDecision ScenarioMaking EvaluationExecutionof scenariosKernelModelNoYesUser InterfaceBase ObjectsScenarioInstantiationInstantiation of scenarios(multiple instances)We propose a flexible spatial decision supportsystem framework (Figure 2) to support the decisionmaking process and overcome the problems identifiedearlier.DesignExecution ofnon-spatial modelswith appropriatesolverModelExecutionYesYesthe decision maker to further develop a more desirablesolution to a particular problem. Scenario evaluationranks the many alternative scenarios based on decisionmakers’ preferences. Sensitivity analysis are employedas a means for achieving a deeper understanding of thestructure of the problem by changing the inputs e.g.data, solver or evaluation model. This helps to learnhow the various decision elements interact and allowsthe decision makers to determine the best solution. Incompleting of the above processes, the best-evaluatedscenario is selected. As there is no restriction on howthe user chooses to solve a problem, decision makerscan select the phases to follow based on the nature ofthe specific problem and their specific purposes.NoModel IntegrationScenario ModellingNoModelInstantiationGrouping nonspatial aspectsProblemModellingCreation of Modelfor spatial and non-spatial aspectsIntelligenceProceedings of the 37th Hawaii International Conference on System Sciences - 2004Figure 1. Spatial Decision-Making ProcessThe process is iterative in nature so that multiplescenarios instances can be generated using the samescenario structures. The scenario integration processenables the decision maker to combine both spatial andnon-spatial scenarios to create a complex multi-criteriaspatial scenario that addresses all the requirements of acomplex spatial problem. When required, theinstantiated scenarios are called for execution usingdifferent solvers. The execution of the scenario allowsObject, Layer&Theme olKnowledgePoolFigure 2. The FSDSS FrameworkThe FSDSS framework is comprised of six majorDSS objects or components namely, data, models,solvers, visualisations, scenario and knowledge. Theseobjects are stored in the object repositoryindependently, and they communicate through thekernel, which is the programmatic engine that makesthe system run. The framework accommodates spatialdata (spatial objects, layers and themes) and nonspatial data. It contains both spatial and non-spatialmodels, solvers, scenarios and visualisations. The0-7695-2056-1/04 17.00 (C) 2004 IEEE3

Proceedings of the 37th Hawaii International Conference on System Sciences - 2004knowledge is the output of the decision-makingprocess and can be stored in the system for futurereference. The decision maker interacts with thesystem through the user interface. Different data,model and solver can be selected from the objectrepository and mapped together to generate a scenario,or a specific decision support system that is tailored fora particular problem domain. This framework allowsgenerating multiple scenarios at one time and storesthem in the scenario pool. The framework supports theintegration of several simple scenarios into a complexmulti-attribute scenario that contains both spatial andnon-spatial aspects through scenario integrationprocess. Similarly, the knowledge can be stored in andretrieved from the knowledge pool.4. The FSDSS ArchitectureWe propose the FSDSS architecture thatimplements the framework and supports the proposeddecision-making process, as shown in Figure 3. TheFSDSS architectural components are organised intofive distinct layers, these are: persistence layer, objectservices layer, DSS objects layer, integration layer,and presentation layer. These layers and theircomponents are briefly described as follows.User Interface and Presentation LayerKernelIntegration LayerValidationMappingBase ObjectsComponent ControlObject Services LayerDSS Objects LayerScenario ManagerKnowledgeScenarioData/Layer/ThemeKnowledge StorageModelEvaluation(MCDM)SolverKnowledge RetrievalVisualisationSpatial ObjectsSpatial Data/Layer/ThemeSpatial VisualisationPersistence LayerObject LibrarySpatial ModelSpatial SolverScenarioSpatial and Non-SpatialKnowledgeSpatial and Non-SpatialNon-Spatial ObjectsNon-Spatial DataNon-Spatial SolverNon-Spatial ModelNon-Spatial VisualisationFigure 3. The FSDSS ArchitecturePersistence layer contains the object library used tostore the system objects. This includes the storage ofnon-spatial data and the variety of spatial data (objects,layers, themes and map). It is also responsible for thestorage of models, solvers, visualisations, scenariosand knowledge, either spatial or non-spatial in nature,using the object-oriented database managementsystem.Object services layer manages the system objectsthrough the component control that contains severalparameters and methods to coordinate the componentpool and the application. It exports objects from theobject library to the DSS objects layer, as well asimporting the resulting scenarios and knowledge fromthe DSS objects layer back to the object library. It alsofacilitates dynamic creation, updating and deleting ofthe objects.DSS objects layer supports independentdevelopment and use of the decision-supportcomponents including both spatial and non-spatialdata, models, solvers and visualisations, for generatingsimple spatial and non-spatial scenarios. It isresponsible for integrating scenarios to developcomplex spatial scenarios. It supports the evaluationand ranking of multiple scenario instances using theevaluation model. This layer also facilitates the storageand reuse of the result from the decision-makingprocess (the knowledge). It also provides graphical andmap-based presentation of data, scenarios orknowledge. The data component includes both nonspatial and spatial data i.e. spatial objects, layers,themes and maps. The model can be of the primitivetype or the compound type [2]. Primitive type modelparameters are directly derived using base data typevariables or executed model values of the base models.The compound type parameters inherit and/oraggregate the base models as well as adding someother user-defined parameters. The non-spatial model0-7695-2056-1/04 17.00 (C) 2004 IEEE4

Proceedings of the 37th Hawaii International Conference on System Sciences - 2004handles non-spatial problems or non-spatial aspects ofa spatial problem. Spatial models cater for spatialproblems. The evaluation model is made of differentparameters as well as the weights for each of thesemodel parameters. The FSDSS architecture containsthe spatial-oriented solvers (contain a parameter oflocation) and generalised solvers that can be used forboth spatial and non-spatial models. The scenariocombines data, model, solver and other relevantinformation. The scenario structure and its multipleinstances can be stored in the database. The FSDSSsupport three types of visualisation i.e. spatial, nonspatial and map-based visualisations. Spatialvisualisation is used to represent spatial data, scenariosand knowledge. Non-spatial visualisation e.g. 3Dgraphs are used to present the output of analyticalresults. In addition to the general graphical reportfunctions, the FSDSS visualisation is particularlyimportant when used with maps. Different spatialobjects, layers or themes are overlaid to generate a newmap. The Knowledge component contains the finalresults of the decision-making process, includinginformation about the decision maker, the rules thatwere applied, alternative scenarios, the final decisionas well as the system components used in reach theparticular decision.Integration layer contains the communicationcomponents i.e. kernel, mapping and validationcomponents. In addition to activating and using thecomponent functions, the kernel works as a userinterface and is responsible for the communicating andintegrating of system components. Mapping enablesthe model component to communicate with data andsolver components properly through model-data andmodel-solver mapping processes. The model parameteror attributes are fixed; the user selects the dataattributes for model-data mapping and selects thesolver name and solver attributes for model-solvermapping. Validation enables proper communicationbetween system components. It is responsible forchecking the input data type to the model and to thesolver during the mapping process. The model-datavalidation tests whether the data type of the modelattributes is similar or convertible to the data attributes,while model-solver validation checks whether the datatypes of the attributes of the model instance are similaror convertible to data type of the solver attributes.Presentation layer or user interface provides all theinteractions between users and the system. It isdesigned to be technology independent so that thisarchitecture can be implemented using other platforms.It provides a flexible environment where spatial andnon-spatial components are used together to create thecomplex spatial results.A simple decision-making flow in Figure 3illustrates how the FSDSS architecture supports thedecision-making process. The decision maker initiatesthe decision-making process at the interface layer andinteracts with the system through the kernel. Thecomponent control picks up the relevant componentsfrom the persistence layer. The selected data, modelsand solvers are combined in the integration layer todevelop scenarios using the mapping component. Thescenario manager manages these scenarios and theevaluated scenarios can be presented using theappropriate visualisation component. The output of thedecision-making process can be saved in thepersistence layer as knowledge. The interactionbetween the DSS objects layer and the persistencelayer are bi-directional. On the one hand, thearchitecture allows flexible selection of objects fromthe object library. On the other hand, the executedresult (e.g. scenarios generated) can be stored back tothe object library.5. The FSDSS ImplementationA prototypical FSDSS is implemented to prove thevalidity of the spatial decision-making processes aswell as the FSDSS framework and architecture.Object-oriented concepts, object-oriented databasemanagement system and the object-orientedprogramming language are the tools and technologiesused to develop the FSDSS prototype. Jade(www.jadeworld.com), a fully integrated developmentenvironment [6] with its own object-oriented databaseand program language was selected for implementationplatform. The complete prototype was developedwithin Jade without having to take recourse to anyother tool. The proposed spatial decision-makingprocess and the implemented FSDSS are evaluatedthrough five scenarios across spatial decision problemdomains including location, allocation, routing and/orlayout. Table 1 gives details of the type of spatialproblems and the specific domains where we tested theprototype.Table 1. Spatial Problems andImplementation DomainsSpatialProblemApplicationDomainExample Spatial ProblemsAllocationGeoMarketingFind geographical distributionsLayoutRunningRoutingDeliveryDesign and select best runningpathIdentify the fast routeLocationHousingSearch the most suitable houseSpatioTemporalHealthTrace the spread of a disease overspace and time0-7695-2056-1/04 17.00 (C) 2004 IEEE5

Proceedings of the 37th Hawaii International Conference on System Sciences - 2004The same environment was used in the testing, butwith different data, model and solver sets. In thefollowing section, we explore the interaction with theFSDSS in the context of the house location problem.In the next step, the decision maker models thisproblem using the proposed modelling approach byseparating the spatial and non-spatial aspects of acomplex spatial problem.6. Sample Session with FSDSS6.2This section illustrates the implemented FSDSS tosolving a location problem using the proposed spatialdecision-making process. Each process step, as shownin Figure 1, is described in detail.The problem modelling involves both spatial andnon-spatial aspects. Quality and economic are nonspatial in nature whereas accessibility criteria are of aspatial nature. On the non-spatial side, cost and qualityof the property can be analysed using non-spatialmodels and solvers. The spatial aspect of the problemfocuses on the location of the property, as it is animportant criterion when people rent or buy a house.Location is a complex criterion that has multiplespatial dimensions e.g. environment and distance tomain facilities. These spatial dimensions need to beanalysed one by one in order to find a best location.In this illustration, the decision maker first broadlyselects a target area then carries out accessibilityanalysis. The analysis involves both the non-spatialmodel and spatial model and it uses both non-spatialsolvers and spatial solvers. The problem is solvediteratively by firstly, considering spatial and nonspatial data, models, solvers and scenarios; secondly,applying spatial and non-spatial criteria and finally,using the goal- seeking and sensitivity analysis.6.1Step 1: Problem IdentificationThe problem presented in this session is to identifythe optimal location of a property that maximises“return” i.e. the satisfaction level that is measured onthe basis of the following three criteria. The value treeof the problem analysis is presented in Figure 4.Number of RoomsNon-SpatialFunctionsHas GaragePrice .House LocationProblemSchool ZoneEnvironment Spatial CriteriaDistance to MainFacilities Distance toSchoolDistance toHospitalDistance to PublicTransportation ProblemIdentificationProblemIdentificationSimple Spatial andNon-spatialScenarioDevelopmentComplex Spatialand Non-spatialScenarioDevelopmentFigure 4. Value Tree of Location Problem Quality criteria e.g. construction material, builtyear, size, number of rooms and functions.Economic criteria such as market price or rentalcost; andLocation e.g. property accessibility, vicinity, andenvironmental conditionsSome of these factors are difficult to evaluate orpredict, as relative impacts for some of these factors onreturn remain unknown. It is hard to structure theproblem in its entirety at one time i.e. precisely defineand measure the objective for every possible solution.6.3Step 2: Problem ModellingStep 3 and 4: Scenario DevelopmentThe decision maker now needs to load relevantdecision-making components. These include theproperty table and relevant map in which the propertiesare located, the various models, solvers andvisualisations to be used for building the differentscenarios. A simple non-spatial scenario and a simplespatial scenario are developed separately at first; theyare then integrated into a combined scenario. Thesescenarios are then transformed into a complex multicriteria scenario through a structural integrationprocess. The scenario development process isillustrated as follows:Simple Non-Spatial ScenariosThe non-spatial scenario is created using the nonspatial Filtering model and the Range solver. In thisexample, we have selected the 3-bedroom flat with aprice range between 300,000 and 400,000. Severalproperties are identified through this filtering processas shown in Figure 5. These stored in the database asScenario 1 (4 instances).0-7695-2056-1/04 17.00 (C) 2004 IEEE6

Proceedings of the 37th Hawaii International Conference on System Sciences - 2004non-spatial filteringcriteria:Price: 300k to 400kHouse Style: flatRoom Numbers: 3Non-SpatialScenarioScenario 1SpatialModelResultingScenarioScenario 3During this integration process, the four non-spatialFiltered results:4 propertiesSaved asScenario 1Figure 5. Simple Non-Spatial Scenario CreationSimple Spatial ScenarioThe decision maker has selected a buffer zone (a500-meter radius circle) around a particular location(e.g. x, y coordinates: 200,200). The filtering model isinstantiated with the property data and executed usingthe Distance solver to find the properties within thedefined circle. This process develops many scenarioinstances (as shown in Figure 6). These scenarioinstances are then stored in the database as Scenario 2(14 instances).Figure 7. Integration of Non-spatial withSpatial Scenariosfiltered scenario instances of scenario 1 as describedearlier are supplied as input to the spatial Filteringmodel. The resulting scenario instances are stored asScenario 3 (3 instances).The second way for integration of Scenario 1 andScenario 2 is to supply spatial Scenario 2 as input intothe non-spatial filtering model and then apply the nonspatial Range solver for execution, as illustrated inFigure 8. The process develops three instances that arestored in the database as Scenario 3.SpatialScenarioScenario 2Non-SpatialModelResultingScenarioScenario 3non-spatial filteringcriteria:Price: 300k to 400kHouse Style: flatNumber of Rooms: 3load Scenario2Figure 6. Simple Spatial Scenario CreationCombined Scenario (Pipelining Integration)Pipelining integration of spatial and non-spatialscenarios can be done in two ways. The first way is tocreate non-spatial Scenario 1 and then execute thegeographical filtering model using spatial solvers e.g.Distance or Point-in-Polygon solver (Figure 7).filtered usingnon-spatialcriteria3 Instancesoutput saved asScenario 3Figure 8. Integration of Spatial withNon-spatial Scenarios0-7695-2056-1/04 17.00 (C) 2004 IEEE7

Proceedings of the 37th Hawaii International Conference on System Sciences - 2004The scenario pipelining integration process can takeplace bi-directionally, either from non-spatial to spatialor from spatial to non-spatial. The flexible use andintegration of spatial and non-spatial models, solversand scenarios is one of the most important features ofFSDSS. The above process helps the decision maker tochoose the properties that satisfy the non-spatialcriteria e.g. quality, cost and the basic locationrequirement such as area. The following sectionillustrates another aspect of the location problemnamely, accessibility analysis.Complex Spatial Scenario (Structural Integration)The complex spatial scenario is generated using theProperty data, Distance model and Distance solver asshown in Figure 9.The previously created Scenario 3 and its threeinstances are loaded from the scenario pool. Now, thedecision maker focuses on distance to major facilitiesfor accessibility analysis.The distance has multiple dimensions. It includesthe distance from a particular spatial object (e.g.property 0014) to another object (e.g. hospital 2). Thedistance from one object to a spatial layer (e.g. schoollayer) returns multiple values, in this case the systemreturns the shortest distance from the target object to asingle object (e.g. school 1) in that layer.6.4Step 5 and 6: Scenario Integration andInstantiationThe decision maker integrates the simple combinedscenario (Scenario 3) structure with these newlydeveloped distance parameters to develop a morecomplex scenario that contains all the criteria for theproblem.Figure 9. Multi-Attribute Spatial ScenarioCreationThe structural or permanent scenario integrationtakes place in two steps. First, a bare scenario templateis created as shown in Figure 10. Then multiplescenario instances are created (Figure 11).The decision-making selects a scenario instancefrom Scenario 3, and calculates each of the distanceparameters in the scenario template. Once all therelevant distance values have been calculated, ascenario is then instantiated with these values.original parametersnew parametersFigure 10. Scenario Template for Integration of Spatial and Non-Spatial ScenariosFigure 11. Multi-Criteria Spatial Scenarios0-7695-2056-1/04 17.00 (C) 2004 IEEE8

Proceedings of the 37th Hawaii International Conference on System Sciences - 2004The scenario integration process is

a spatial decision-making process (Figure 1) by synthesising ideas of the decision-making processes [7] and the multi-criteria decision-making process [3]. It also integrates concepts from spatial modelling, model, scenario and knowledge management as well as MCDM methodology. The process contains nine