JointEUROGRAPHICS- IEEETCVG SymposiumonVisualization(2004)O. Deussen,C. Hansen,D.A. Keim,D. Saupe(Editors)

Visualization For Public-Resource Climate Modeling

J.P.R.B.Walton�, D. Frame

�andD.A. Stainforth

�

�TheNumericalAlgorithmsGroupLtd, WilkinsonHouse,JordanHill Road,OxfordOX2 8DR,UnitedKingdom�

Atmospheric,OceanicandPlanetaryPhysics,Universityof Oxford,ClarendonLaboratory, ParksRoad,OxfordOX1 3PU,UnitedKingdom

AbstractClimateprediction.netaimsto harnessthespare CPU cyclesof a million individual users’ PCsto run a massiveensembleof climatesimulationsusingan up-to-date, full-scale3D atmosphere-oceanclimatemodel.Althoughit hasmanysimilarities with other public-resource computingprojects,it is distinguishedby the complexity ofits computationaltask, its systemdemandsand the level of participant interaction, data volumeand analysisprocedures.For simulationsrunningon individual PCs,there is a requirementfor compellingvisualizationsthatare readilygrasped,sincemostusers will beinterestedin theoutputfromthemodel,but will havea limited levelof scientificexperience. Thispaperdescribesthedesignandimplementationof thesevisualizations.

CategoriesandSubjectDescriptors(accordingto ACM CCS): I.3.8 [ComputerGraphics]:ApplicationsI.6.3 [Sim-ulationAnd Modeling]: ApplicationsC.1.4[ParallelArchitectures]:DistributedArchitecturesJ.2[PhysicalSci-encesAnd Engineering]:EarthandAtmosphericSciences

1. Introduction

The increasein the speedandcapacityof widely-availablenetworkedPCsis providing anopportunityto usedistributedcomputingto tackle modelingtasksthat would previouslyhave requiredaccessto a supercomputer. ProjectssuchasSETI@home[KWA � 01] have demonstratedthe principlesof public-resource computingwhich allows researcherstotap into the idle processingcapacity of millions of PCsvolunteeredby businessesandthe generalpublic. Previousprojectsof this typehaverequiredtheinstallationof ananal-ysis applicationon the participant’s machine,followed bythedownloadfrom acentralserverof asmallamountof datathatis testedby theapplication.Thetestresult,whichis usu-ally of the form “match” or “no match”, is thenreturnedtotheserver, andthenext pieceof datais downloaded.As eachtest takesonly a few hoursto process,the participantneednothavea long-termcommitmentto theproject.

This paradigm has been taken a step further by cli-mateprediction.net [SKA � 02], which differs from previousprojectsin thevolumeandmanagementof distributeddata,the complexity, duration and granularity of the modelingtask,andthe needfor tools which will maintainlong-termparticipantinterest.More specifically, the applicationthatparticipantsareinvitedto install is afull-scaleclimatemodelwhich is run locally to simulatethe Earth’s climate up to

2050.Eachrun is characterizedby a uniqueset of valuesfor certainparametersusedby themodel,leadingto differ-ent representationsof the physics in eachsimulation.Thecombinationof resultsfrom all runscanbeusedto make aprobabilisticclimateforecastwhichaccountsfor uncertaintyin themodelformulationandtheprocessesit represents.

The high resourcerequirementsfor eachrun (each100yearsimulationtakesaroundthreemonthsona1.4GHzma-chine),thelargenumberof runsneededfor anaccuratefore-castandthefactthatclimatechangeis of interestto thegen-eralpublicall make thisprojectagoodcandidatefor public-resourcecomputingusingvolunteeredresources.Successfulprojectsof thisnaturehavebeencharacterizedby thestimu-lationof thepublic’s imaginationandafeelingof beingableto contribute to the solutionof a problemwhich is—in thepresentcase,literally—globalin its effect.

If the project is to have a good level of uptake, the ap-plicationmustbesimpleenoughto berun by inexperiencedPCuserswith slow Internetconnections.At thesametime,theoutputfrom thepackageshouldbecompellingenoughtostimulatecontinuedparticipation.Visualizationplaysanim-portantrole in meetingthis demand,sinceit allows theuserto view theprogressof their run, whilst alsoraisingaware-nessof theissuesbeingstudiedby theproject.

c�

TheEurographicsAssociation2004.

Stainforthetal / VisualizationFor ClimateModeling

This paperdescribesthe designand implementationofreal-time visualizationfor the climateprediction.net desk-top package.We introducetheprojectin the following sec-tion,anddiscussits visualizationcomponentin §3, includingsomeexampledisplays.A final sectiondiscussesthepresentstatusof theprojectanddescribesthefutureevolution of itsvisualizationrequirements.

2. Global Climate Prediction

2.1. Project Rationale

Recentsimulationsusingclimatesystemmodelshaverepro-ducedmany large-scalefeaturesof recent(circa 150years)climate changewith a good degreeof accuracy (see,e.g.[STJ� 00]). Thesemodelscanbe usedto make projectionsof the effect of variouslevels of greenhousegasemissionson future climatechangefor the Earth.However, therearenoobjectiveuncertaintylimits oncurrentpredictions,whichmakesthemlessusefulto policy makers.Recognizingthis,the Inter-GovernmentalPanel on Climate Change(IPCC)hasrecentlycalledfor theimprovementof methodsto quan-tify theseuncertainties[WG1], andthis is themainscientificgoalof climateprediction.net.

The modelscontaina numberof adjustableparameterswhosevaluesare poorly constrainedby available dataonthe processesthey aresupposedto represent.For example,cloudsplay a key role in the climatesystemthroughtheirstrongfeedbackon radiative forcing—they caneithercoolor warm the surfacevia their Albedo or greenhouseeffect.Someof the model parameterswhich affect cloud forma-tion includethecritical relative humidity at eachheight,thethresholdfor liquid watercontentin theatmosphereoverseaand land, and the rate of conversionof cloud liquid waterto precipitation.Theeffect of perturbingparameterssuchasthesewithin their rangeof uncertaintyis non-additive,sotheonly way to investigatetheir impactis performa simulationfor eachpoint of interestin anN-dimensionalspace,whereN is the numberof parameters.Whilst intelligent samplingof this spaceshouldhelpreducethenumberof simulations,it hasbeenestimated[SKA � 02] that betweenoneandtwomillion independentsimulationswill berequired.

2.2. Model Details

We useversionsof the ClimateModel of the Hadley Cen-tre for ClimatePredictionandResearchat the UK Meteo-rological Office [Cul93]. Onepart of the model representstheocean,andtheothertheatmosphere.The latter is mod-elledin 3D (19 levels in thevertical,with a horizontalreso-lution of 2.5� in latitudeby 3.75� in longitude)andinteractswith thelandsurfaceandcryosphere.This is thesameasthemodelusedby theMet Office in numericalweatherpredic-tion, but we areusinga coarserresolutionin orderto sim-ulate the long timescalesassociatedwith climate research.Our resolutionis almostpreciselythesameasthatof all the

modelsusedin themostrecentIPCCclimatechangeassess-ment [WG1], and it is believed that many of thesemodelswill beusedat this resolutionin thenext assessmentin 2007[All03].

Visualizationcomponent

Modelcomponent

Experimentalcontrol

SM1

SM2

uses / produces

downloads / uploads

manages

monitorsmonitors

monitors

readscreates / updates

executes

Input / outputfiles

Client component

Figure 1: Architecture of theparticipantpackage. All com-ponentsareinstalledontheparticipant’smachine. Detailsofparameterperturbationsare downloadedfrom climatepre-diction.netservers and the simulationresultsare uploadedto theseservers at the end of the run. SM1 and SM2 aretwo sharedmemoryarenaswhich are usedfor communica-tion betweentheclient,modelandvisualizationcomponents(SM1)andfor transferof datafromthemodelto thevisual-ization(SM2).

2.3. Implementation

We choseWindows as the initial target platform for cli-mateprediction.netbecauseof theexperienceof [KWA � 01]who found the contribution of Windows participantstobe an order of magnitudegreaterthan that of usersrun-ning on otherplatforms.Accordingly, the modelof §2.2—consisting of around 0.5M FORTRAN lines—hasbeenported[SKM � 02] to Windows.(Otherenvironmentsarealsounderconsideration—see§4, below.)

The first part of the main experimentconsistsof simu-lations of the 1950-2000period. During eachrun, resultsarecomparedto historicaldatafor variablessuchassurfacetemperature,anda quality-of-fit value is calculated.Thoseparametersetsleadingto modelswith theclosestfit to pastobservationswill then,in thesecondpartof theexperiment,beusedin simulationsof the2000-2050period.Theresultswill helpwith anobjectiveassessmentof theuncertaintyforclimateprojectionsto 2050underarangeof futureemissionscenarios.

To participatein climateprediction.net, a user initially

c�

TheEurographicsAssociation2004.

Stainforthetal / VisualizationFor ClimateModeling

downloadsthepackagefrom adistributionserver. Theirfirstsetof parametervaluesareobtainedfrom a client interac-tion server and the simulationis startedasan ‘idle’ prior-ity process.No further network communicationis requireduntil the run is completed—usuallyafter several weeks—whenthepackageuploadsa subsetof thesimulationresults(whosesizeis about10 Mb) andcontactstheclient interac-tion server to getdetailsof thenext run to beperformed.

Figure 1 shows the structureof the participantpackagein moredetail. The modelandvisualizationarekept sepa-ratebecauseit is envisagedthat the modelexecutablewillbecomparatively stableover thelifetime of theexperiment,whilst thevisualizationandtheclient will probablyrequireperiodicupgrades.In addition,sincethe modelcomponenttakesup a substantialfractionof thepackage,it is desirableto limit the numberof timesit hasto be downloadedfromthe distribution server. More detailson the overall systemdesigncanbefoundin [SKM � 02].

3. Real-Time Visualization

3.1. Motivation

As notedabove, an essentialrequirementof the designisthatparticipantsshouldbeableto seewhat their simulationis doingin realtime.Oneof thereasonsfor this is simplysotheparticipantcan“seesomethinghappening”,whetherforreassurance,or curiosity, or scientificinterest.Anotherrea-sonfor compellingvisualizationis to raiseawarenessof theissuesunderstudy. Climatechangeandtheeffect of green-housegasemissionson global warmingaretopicsof wideconcern.The scaleof the problem,coupledwith the de-greeof commitmentwhich is requiredto addressit, canbedaunting.Watchinga desktopsimulationof changesin theearth’s climatecould stimulatefurther considerationof theissues,along with a feeling of personalsatisfaction at be-ing involved—athoweversmalla level—in amajorresearchexperimentin this field.

3.2. Previous Work

The GLOBE program [GLO] is a popular public-accessprojectaimedat schoolswhich allows participantsto sharetheir measurementsof varioustypesof environmentaldataover the internet,to plot them as mapsand graphsand tocollaborateonspecificenvironmentalprojects.Visualizationin GLOBE is doneusing lightweight clients within a webbrowser, andhasnot to dateinvolved3D display.

Middleton andhis colleaguesat NCAR have createdvi-sualizationsof climatechange[VETS] calcluatedby large-scaleclimatemodelsandhaveperformedcollaborativeanal-ysisusinglargedisplaysandtheAccessGrid.Otherworkershave useda visualizationtoolkit for climatemodelingdis-play. Treinish hasproducedcompellingdisplaysof ozonedepletion[Tre93] usingOpenDX.SheldonandVahlenkamp

[GFDL] have usedIRIS Explorerto createvisualizationsofglobalwarmingandhurricaneformation.Finally, VisAD hasa datamodelwhich mapswell to thevisualizationof mete-orologicaldata,andhasbeenusedasthe basisof NCAR’sVMET tool for the displayandanalysisof meteorologicaldata[VMET].

Althoughmuchwork hasbeenperformedon thedisplayof resultsfrom live feedsor datawarehouses,noneappearsto have beendonewith real-timedisplayof simulationout-put in a public-resourceenvironment,but it shouldbe pos-sible in this work to make useof someof the techniquespreviouslyappliedto larger-scaledata.

3.3. Visualization Implementation

At eachtime step,the simulationoutputsa seriesof two-dimensionalarraysto the visualizationvia the SM2 sharedmemoryarena(seeFigure1). The model interval betweentimestepsis 30minutesandtheresolutionof eacharrayis 96by 73 (see§2.2). Amongstothermodelvariables,thearrayscontainvaluesfor:

� cloudcoverat threeheightsabove theearth,� snow depthandseaicecoverage,� surfacetemperature,� precipitation.

Two of thethreeheightsat which cloudcover is selectedare the lowest and highestof the atmosphericmodel’s 19verticallayers;in addition,weselectthe10thlevel asrepre-sentativeof themiddleof theatmosphere.Thesnow andicedatasetsare complementary(sincethe former is only cal-culatedover land) and so can be combinedinto a singlesnow/icecoveragearray.

Initial visualizationprototypeswerebuilt usingIRIS Ex-plorer, but its useasa basisfor thevisualizationcomponentof the participantpackagewas not deemedacceptablebe-causethe sizeof the component—andhence,its contribu-tion to the packagedownloadtime—wastoo large. Subse-quentwork was thereforedoneusing a lower-level graph-ics library, in order to reducethe size of the component’sfootprint.SinceOpenInventor[Wer94] wasusedastheba-sisfor thegraphicsmodulesin theIRIS Explorerprototype,this wasthelibrary of choice.UsingCoin’s implementation[COIN] of the OpenInventor API, the size of the visual-izationcomponentwas3.1Mb. This includesCoin librarieswhich will only needdownloadinga few timesduring theproject;thevisualizationexecutablealoneis 0.1Mb, whichis asmallproportionof 7.1Mb, thecurrentsizeof thewholeparticipantpackage.

In choosingtechniquesfor thedisplayof simulationdata,therewasa requirementfor imageswhich would be easilyunderstoodby theparticipant,even in theabsenceof scien-tific knowledge.Considerationssuchasthis(andsomeof thework in §3.2) suggestedmappingthesimulationoutputonto

c�

TheEurographicsAssociation2004.

Stainforthetal / VisualizationFor ClimateModeling

Figure 2: Continuouscloudcover of entire globe(top) anda close-up(bottom)usingcell-baseddisplay.

a globewhich couldbemanipulatedin 3D. Two variantsonthis themeweredeveloped,discussedbelow.

3.3.1. Cloud Display

Here,theglobeis formedby texture-mappinga spherewitha satelliteimageof the earth’s surface.Next, the snow/icedatais mappedonto a texturedsphereconcentricwith theglobe(thetexturedsphere’s radiuswasslightly greaterthanthe globe’s to ensurethat the snow/ice layer was alwaysrenderedon top of the earth’s surface). The texture is atwo-component(intensity, alpha)image,wherethevalueforeachcomponentis given by the snow/ice coverageat thatpoint. In asimilarway, thecloudcoveragefor eachheightismappedontoa texturedspherecentredon theglobe,with aradiusscaledapproximatelyby its heightbut representingagreatlyexaggeratedatmosphericdepththatclarifiesthesep-arationof differentlevels.Thesuperpositionof texturedlay-ersaroundtheglobegivesa senseof depthin a pseudo-3Deffect(seeFigure2).Wefoundthatthis typeof visualizationof the cloud datawas quicker to generateand manipulatethanfuller alternativessuchascalculatingisosurfacesof thecompleteclouddataset,whilst still producingimageshavingagooddegreeof realism.

Texturing thespheregeneratesa smoothlyvarying inten-

sity andtransparency. An alternative representation,whichreflectstheresolutionof thesimulation,is to divide thesur-faceof the sphereup using the underlyingcomputationalmesh,and to assignan intensity/alphavalue to eachcellbasedondatavalue.Theresultis shown in Figure2.

Oneof theobjectivesof thistypeof displayis to produceacaricatureof theappearanceof theearthfrom space,includ-ing its illumination by the sun; incorporatingthe directionof sunlight in the visualizationalso highlights the diurnalvariationin modelresults.To simulatesunlight,weaddadi-rectionallight which rotatesaroundtheglobeonceevery 48time steps.The sub-solarlatitudeand longitudearecalcu-latedusing

φ � ζcos nd 201� (1)

θ � 180� 360nh � nm 60

24(2)

whereζ is the earthtilt angle( � 23� 5 � ), andnd, nh andnm arethe day, GMT hour andminutenumber. The modelmakesthestandardassumptionthateachmonthhas30daysandeachyearis 360dayslong; thentheshift of 201 in (1)givesφ � 0 at themodelequinoxes,namelyApril 21 (nd �111)andSeptember21 (nd � 291).Similarly, the180offsetin (2) makes midday (nh � 12� nm � 0) correspondto theprimemeridian(θ � 360 � 0).

3.3.2. Field Display

This typeof sceneis intendedfor thedisplayof theremain-ing modelvariablessuchassurfacetemperatureandprecip-itation. The first visualization(shown at the top of Figure3) is similar to thecell-basedclouddisplay(§3.3.1)—here,eachcell is coloredaccordingto field value,while contextis providedby a coastlineoverlay. An alternative methodofdisplayingfield datais to usesolid contouring(bottom ofFigure 3). This can be done[HS93] in OpenGL(or OpenInventor)by first creatinga texture with a one-dimensionalcolormapandthe field valuesastexture coordinates.Then,if thetexturequality is setto a low value,thetexturewill berenderedwith no interpolationbetweendatavalues,result-ing in solid regionsof color for datawithin somerange.Thetexturecanbeappliedto a simplequadrilateralmeshthat isthenrenderedoffscreento produceanimagewhichis in turntexture-mappedarounda sphere.Scalingthesizeof theoff-screenviewport providesa tradeoff betweenimagequalityandrenderingspeed.

3.4. Interface And Performance

Theupdateof thevisualizationis controlledby a timer sen-sor [Wer94] that periodically interrupts the display eventloopto checkwhetherthetimestephaschangedandrebuilds

c�

TheEurographicsAssociation2004.

Stainforthetal / VisualizationFor ClimateModeling

Figure 3: Two options for visualizing field data—here,temperature—using(top) cell-baseddisplay and (bottom)solidcontours.

thesceneusingthe latestdatasetif it has.Someconsidera-tion of theperformanceof thesimulationis requiredin orderto determinea sensiblevaluefor thetimer interval. Besidesbeingmachine-dependent,performancedetailsaresensitiveto othermachineactivity sincethe model is runningasanidle process,but a representative resultmay be quoted:ona 1.8 GHz Athlon XP, it takesabout195 secondsto simu-lateaday. It shouldbenotedthateverysixth timesteptakesabouttwelve timeslongerowing to an extra calculationofthe radiationbalancein the atmosphere;thus, the shorterstepstake aboutone second,and this is the initial defaultvaluefor thetimer interval. As thevisualizationproceeds,itstoresthemoving averagesof theshortertime perstep,andusesthisto adjustthetimerinterval to becommensuratewithit. In practice,we have not observedany effect of the timerinterrupton userinteractionwith the visualization.Onceanew datasethasarrived,thetime takento rebuild thesceneandto renderit hasbeenmeasuredat around0.1 secondonan SGI 540 with a Cobalt graphicscard having 48 Mb ofmemory. Thecomparisonof this timewith thetimestepdu-rationabove is reinforcedby moreextensive betatestingofthepackage(§4) on a varietyof hardwareplatforms,whichshows that the visualizationtime is alwaysmuchlessthanthecalculationtime.

The basic user interface for 3D display and manipula-tion is that of the genericOpenInventorExaminerviewer[Wer94], allowing rotation, zoomingand panningthroughclick-and-drag mouse actions. We have customizedtheviewer in two ways. Firstly, screen-basedobjectssuchascaptionsand legends(seeFigure 3) have beenintroducedinto thedisplayby overridingtheExaminerviewer’s redrawmethod.Secondly, apopupmenuin ourcomponentgivesac-cessto application-specificoptionslike switchingbetweendisplayoptionsfor differentfieldsandpausingthesceneup-date.Our menualsocontainssomehelpful genericoptionsselectedfrom the original Examinerviewer menusuchasheadlightcontrol, stereodisplayandthe storageandrecallof a homeviewing position(otheroptionsthatmight proveconfusingto usersarenotexposed).

4. Current Status. Future Work

Following beta testing, the participantpackage(includingthe visualization component)was releasedat the formallaunchof theprojectin September2003.In orderto encour-agetheparticipationof thegeneralpublic in theexperiment(the aim is to have the packagerunning on arounda mil-lion PCsworldwide), theorganiserslooked for—andgot—extensive coverageof the launchevent from both the printand broadcastmedia.The visualizationsprovided imagesdescribingtheexperimentwhichwereverypopularwith themedia(see,e.g.[BBC]).

Sincetheprojectlaunch,participantshave suggesteden-hancementsvia the project website.Somewhich are cur-rently under considerationinclude making the navigationthroughthe 3D sceneeasierthroughthe useof pre-loadedcamerasto specify views of interestand the addition of amodewhichonly allowsrotationabouttheglobe’sN-Saxis.Thedisplayof othermodelfieldssuchassurfacepressure—possiblyin compositedisplayswith thecloudcoverage—hasalsobeensuggested.

The emphasisof projectdevelopmenthaschangedsincethe releaseof the package.More work is now being doneon “sense-making”toolsto helpparticipantsunderstandtheclimatemodelresults,andplanneddevelopmentsinclude:

� Customizable“userprofile” pagesatthewebsiteonwhichparticipantscanview statisticsandother informationontheir simulationruns.� A semanticweb browsing capabilityto help participantsbrowsethesiteandthewider webby providing guidanceandlinking to their results.� A secondary, downloadable visualization applicationwhich providesaccessto thediagnosticmaterialheldontheparticipant’s disk.

This final point deserves amplification.Whilst we haveconcentratedin thispaperontheimplementationof thereal-timevisualizationcomponentin theparticipantpackage,this

c�

TheEurographicsAssociation2004.

Stainforthetal / VisualizationFor ClimateModeling

projecthasmany furthervisualizationrequirements,partic-ularly now that the post-processinganalysisis underway.In this secondphaseof theproject,the techniquesfrom thereal-timecomponentwill be re-usedalong with other ap-plications.For example,the displayof resultsfrom severalsourceswill requiretheuseof distributedvisualizationsys-tems,probablywith collaborative extensionsfor multi-userdisplay[BDG � 03, ESG]. Making useof theGrid asa plat-form for analyzinglargedistributeddatasets,theprojectwillbeableto build on otherwork linking visualizationsystemswith Grid middleware[Gal02].

In parallelwith thesedevelopments,theclient will beex-tendedto allow theembeddingof regionalmodelswithin theexperimentalarchitecturealongwith the inclusionof otherclimatemodels.It will alsobemadeavailableon otherplat-forms suchas Linux and Macintosh,probablyby makinguseof theBOINC platform[BOI] which is aimedat public-resourcedistributedcomputingprojectsof this type.

5. Acknowledgements

Early work on the cloud display was carriedout by AndyHeaps. We thank Richard Gillis, Carl ChristensenandRichardGaultfor helpwith systemintegration,MylesAllenfor helpful discussions,RebeccaWaltonandananonymousreviewer of an earlier versionfor useful commentson thetext, andDavid Carlislefor helpin preparingthispaper.

References

[All03] ALLEN M. R.: Personal communication(2003). 2

[BBC] http://news.bbc.co.uk/1/hi/sci/tech/3100024.stm. 5

[BDG � 03] BRODLIE K. W., DUCE D. A., GALLOP J. R.,WALTON J. P. R. B., WOOD J. D.: DistributedAnd CollaborativeVisualization.In Eurograph-ics2003StateOf TheArt Reports(2003),Mon-taniC.,Pueyo X., (Eds.),TheEurographicsAs-sociation,pp.139–166. 6

[BOI] http://boinc.berkeley.edu/index.html. 6

[COIN] http://www.coin3d.org/. 3

[Cul93] CULLEN M. J. P.: TheUnified Forecast/ Cli-mate Model. Meteorological Magazine122,1449(1993),81–94. 2

[ESG] http://www.earthsystemgrid.org/.6

[Gal02] GALLOP J.: gViz: VisualisationMiddlewareFor E-Science:A New Project In The UK E-ScienceProgramme.ERCIMNews21 (October2002). 6

[GFDL] http://www.gfdl.gov/~jps/GFDL_VG_Gallery.html. 3

[GLO] http://www.globe.gov/. 3

[HS93] HAEBERLI P., SEGAL M.: TextureMappingAsA FundamentalDrawing Primitive. In FourthEurographicsWorkshopon Rendering(1993),CohenM., PuechC.,Sillion F., (Eds.),pp.259–266. 4

[KWA � 01] KORPELA E., WERTHIMER D., ANDERSON

D., COBB J., LEBOFSKY M.: SETI@home—Massively Distributed Computing For SETI.Computing in Scienceand Engineering3, 1(Jan–Feb2001),78–83. 1, 2

[SKA � 02] STAINFORTH D. A., KETTLEBOROUGH J.,ALLEN M. R., COLLINS M., HEAPS A.,MURPHY J.: DistributedComputingFor Pub-lic InterestClimateModelling Research.Com-puting in ScienceandEngineering4, 3 (2002),82–89. 1, 2

[SKM � 02] STAINFORTH D. A., KETTLEBOROUGH J.,MARTIN A., SIMPSON A., GILL IS R., AKKAS

A., GAULT R., COLLINS M., GAVAGHAN

D., ALLEN M. R.: Climateprediction.net:Design Principles For Public-ResourceMod-eling Research. In Proceedings of the14th IASTED International Conference: Par-allel And Distributed ComputingAnd Systems(2002),pp.32–38. 2, 3

[STJ� 00] STOTT P., TETT S., JONES G., ALLEN M.,M ITCHELL J., JENKINS G.: ExternalControlOf Twentieth Century TemperatureVariationsBy NaturalAnd AnthropogenicForcings. Sci-ence290(2000),2133–2137.2

[Tre93] TREINISH L.: VisualizationTechniquesForCorrelative Data Analysis In The Earth AndSpaceSciences. In Animation And ScientificVisualization: Tools And Applications (1993),Earnshaw R. A., WatsonD., (Eds.),AcademicPress,pp.193–204. 3

[VETS] http://www.vets.ucar.edu/vg/CSM/index.shtml. 3

[VMET] http://www.4dwx.org/vmet. 3

[Wer94] WERNECKE J.: TheInventorMentor:Program-ming Object-OrientedGraphicswith OpenIn-ventor, Release2. Addison-Wesley, 1994. 3, 4,5

[WG1] WORKING GROUP 1: ClimateChange 2001:The ScientificBasis.Third AssessmentReportof the Intergovernmental Panel on ClimateChange. CambridgeUniversityPress,2001. 2,2

c�

TheEurographicsAssociation2004.