INTERNATIONAL COMPUTER SCIENCE INSTITUTE1947 Center St. • Suite 600 • Berkeley, California 94704-1198 • (510) 643-9153 • FAX (510) 643-7684

MetaViz: Visual Interaction withGeospatial Digital Libraries

Volker Jung

Technical ReportTR-99-017

October 1999

Abstract

Recentinitiatives to geospatialdigital librariesprovide accessto a wealthof distributeddata,but offer only basiclevels of interactivity anduserassistance.Consequently, usersfind it difficult andtime-consumingto browsethroughdatacollectionsand locatethosedatasetsthat meettheir requirements.The MetaViz projectaddressestwo of the majorbarrierspreventingtheextensive useof digital libraries:lack of usabilityandinformationoverload.This researchfocuseson geospatialdata,makingit possibleto developeffectivevisualizationand interactionmethodsthat are basedon familiar spatialmetaphors.Thevisualizationmethodsdevelopedemploy three-dimensionaltechniques,combiningseveralcharacteristicsor dimensionsof metadatainto singlegraphicalviews. As thosevisualiza-tionsarebasedon mapandlandscapemetaphors,they areeasyto understandandprovideinstantoverviews of complex datacharacteristics.The visualizationmethodshave beenintegratedinto MetaViz, aninteractivesystemfor browsingandsearchinggeospatialdata.In MetaViz, graphicalviews of datacharacteristicscanbecreatedandcombineddynami-cally, levels of detail can be adjustedand the data sets found can be previewed andaccessed.MetaViz helpsusersto locateandselectappropriategeospatialdatafrom vari-ous sources and to combine and use them in an effective way.

1

1 Introduction

1.1 Geospatial Data

1.1.1 Types and Use

Geospatialdatacontaininformation aboutthe location,shape,and characteristicsof real-worldgeographicfeatures.Therearethreemain typesof geospatialdata,vector data, raster data (seeFig. 1) andgeoreferenced data. Vectordatastorethelocationandshapeof geographicfeaturesasgeographiccoordinates,linesandpolygons.Examplesof vectordataaredigitizedmapsandroaddatabases.Rasterdataarebasedon a cellular datastructurecomposedof rows andcolumnsforstoringimages.Groupsof cellswith thesamevaluerepresentfeatures.Examplesof rasterdataarescanned maps and digital satellite images.

Georeferenceddatadescribethecharacteristicsof geographicfeatures.They aretypically storedina tabular format andlinked to a vectorbasedataset.Examplesof georeferenceddataarecensusdata, linking statistical values to census blocks.

Geospatialdataareusedin a varietyof ways.Roaddatabasesarethedatasourcefor varioussim-ple mappingtools in the World-Wide Web that provide customizedmapsanddriving directions.RoaddatabasesalsosupportGPS-basedcarnavigationsystems.In thepublicandcommercialsec-tors,geospatialdatahave beenusedmuchlongerfor advancedmappingandplanningtasks,suchasregionalplanningandenvironmentalmanagement.Researchin earthsciencesandlife sciencesis another field that makes extensive use of geospatial data.

1.1.2 Characteristics and Metadata

Geospatialdatahave complex characteristicsof content,context, andaccess.To make valid andeffective use of the geospatial data, it is important to know and observe these characteristics.

Content characteristics specify the domain and themes covered by a data set.

Fig. 1: Vector data and raster data.

2

Context characteristics specify the spatial and temporal context of the data and may include

• the geographic area covered by the data set,

• the time period covered by the data set,

• data quality properties, e.g. location accuracy, and

• the scale of the data set.

Access characteristics specify the valid ways to access and use the data. They may include

• access paths to the data set,

• data set formats, e.g., raster and vector formats,

• the cost of the data set,

• the data set size, and many other characteristics.

Metadataarea way of formalizingandstoringthesecharacteristics.Usually, a geospatialdatasetwill be accompaniedby a metadatarecordin oneof a numberof standards.In the US, the bestknown of thesestandardsis the FederalStandardfor Digital GeospatialMetadata[14], in effectsince 1994. Other standards are used in Europe [9] or are proposals for ISO norms [21].

1.2 Searching Geospatial Data

1.2.1 Geospatial Digital Libraries

Geospatial digital libraries organizelarge repositoriesof geospatialdataandprovide supportforsearchingandretrieving geospatialdata.Digital librariesmaycontainseveralrepositories thatcanbephysically distributedandconnectedover a network. Theindividual repositoriestypically con-tain a numberof collections of thematicallyrelateddata.Individual collectionsconsistof spatialinformation objects (SIOs),which form thebasicinformationunit [3]. Someexamplesof SIOsareaerial photographs, vector data sets of national park boundaries, and seismic data sets.

Metadataarewidely usedto supportsearch,retrieval, transferandevaluationof geospatialdataindigital libraries.In addition to the metadatafor individual SIOs,geospatialdigital librariesalsocontain metadata for other levels, e.g. collection level metadata.

A well-known exampleof a geospatialdigital library is theAlexandriaSpatialDigital Library [7],locatedat the University of Californiaat SantaBarbara.ADL is a 4-yearresearchprojectwithintheNSFdigital library initiative.Currently, theADL repositoryincludesthefollowing collections[2]:

• theADL Catalog,acollectionof maps,aerialphotographs,satellitephotographsanddatabases(300,000+ SIOs),

• the ADL Gazetteer, a collection of geographic place-names (6,000,000 SIOs),

• the Earthquake Data Collection (300+ SIOs), and

• the Volcano Data Collection (1,500+ SIOs).

3

1.2.2 The Problem: Ineffective Search Process

An interestingfeatureof theADL is a Java-basedgraphicalsearchinterfacethatcanbeusedovertheWorld-Wide Web. Userscanselecta geographicregion of interestin a mapbrowser, specifyacollectionname,metadataattributevaluesor valuerangesandretrievea list of matchingSIOs(seeFig. 2).

ADL containsa detailedmetadataschemaof over 400attributes.Usersthatsearchfor geospatialdatawith veryspecificcriteriain mindcanusethisschemato formulateprecisequeries.Theresultset of these queries will generally be rather small and matching SIOs can be identified quickly.

However, usersoftendonothavedetailedandspecificcriteriain mindwhensearchingfor geospa-tial datasets.Instead,therearea numberof constraintsthatneedto besatisfied,suchasthegeo-graphicregion of interestand thematicconstraints.The userwould specify theseconstraintsassearchparameters,startthequery, studytheresultsetof thequeryto getanoverview of thecon-tentsof the collection,probablyrefine the searchcriteria or add new ones,start anotherquery,study the new result set, and so on.

In a nutshell, this search process will be rather repetitive and involve the following steps:

1. Specify search parameters and start query

2. Browse through result set (reading through a list of values)

3. For promising SIOs read and compare metadata (reading through a list of values)

4. Redefine query and start over

Fig. 2: Browsing search results with the Alexandria Spatial Digital Library

4

If theresultsetof thequeryis large,thissearchprocesscanbequiteineffective.In many cases,theresultsetwill be large (several hundredsof SIOs)or huge(over a thousandSIOs). In the ADLsearchinterface,theresultsetwouldbedisplayedasa largelist of metadataattributevaluesfor theindividual SIOs (Fig. 2, left window). To browse through the result set, the userwould scrollthroughthis list, looking for promisingSIOs, identifying them by specificattribute values.Forevery promisingSIO, theuserwould thenretrieve thecompletemetadataset(Fig. 2, middlewin-dow) and study the attribute values in detail.

1.2.3 Related Work: Visualizing the Result Set

This searchprocessis ineffective,becauseboth,steps2 and3 involve readingthroughpotentiallylargelists of values,comparingthesevaluesandidentifying specificoccurrencesof values.Thesetaskscould be mademuchmoreeffective, if the datawasnot shown in lists but presentedin agraphicalform. InformationVisualizationhasbeenusedfor sometime to transformdatasetsintoimagesthat areeasierto interpretfor users.Often, graphicalrepresentationsreveal featuresnotvisible in the original tabular displays [5, 31, 35].

Variousresearchersin InformationVisualizationhave shown that resultsetsof queriesshouldbepresentedvisually [1, 20,30]. With theTileBars system[19], Hearstshowedthatit is importanttoshow many, if possibleall, parametersof the resultsetsimultaneouslyandcompactly. Beard[4]introducedgraphicalviewsof geospatialmetadataandarguedthattheseviewsshouldfacilitatethequick comparisonof individual SIOsin theresultset.Variousexperimentshave shown thatcom-puter-generated3D scenescanbe very intuitive andeffective settingfor interactingwith digitallibraries [10, 12]. Thesefindings from relatedareasof researchform the basisfor the MetaVizproject.

1.3 The MetaViz Project

1.3.1 Motivation

Distributeddigital librariesoffer a degreeof informationaccessibilitynever reachedbefore.How-ever, having accessto large collectionsof datais of little useunlesspeopleareable to find theinformation they needand also know how to useit. Locating,selectingand using informationeffectively is especiallydemandingin thecaseof geospatialdata,with their manifoldcharacteris-tics of scale,quality, cost,format,andthematiccoverage.Recentinitiativesto geospatialdigitallibraries(GDLs) provide accessto a wealthof distributeddata,but offer only basiclevelsof inter-activity and user assistance.

Currentsearchinterfacesfor geospatialdataaremainly list-basedandusershave to go throughatediousprocessto find andcomparespatialinformationobjects(SIOs).Several stepsin this pro-cessareineffective becauseof the textual representationof metadatainvolved.Researchin infor-mation visualizationprovides several clues on how to improve this process.Someimportantrecommendations on presenting result sets are:

5

• always present the result sets visually

• present as many attributes of the result sets as possible

• facilitate quick comparison of individual SIOs

1.3.2 Goals

The MetaViz projectaddressestwo of the major barrierspreventing the extensive useof digitallibraries: lack of usability andinformationoverload.The researchis focusedon geospatialdata,makingit possibleto developeffective visualizationandinteractiontechniquesthat arebasedonfamiliar spatial metaphors like maps, landscapes and city models. The goals of this project are

• To find adequate visualization techniques for spatial metadata.

• To develop interaction metaphors for query result sets in GDLs.

• To create a GDL interface for visualizing, browsing, and interacting with query results.

1.3.3 Project Outline

The MetaViz project is centeredarounda setof visualizationmethodsfor geospatialmetadata.Geospatialmetadataare key to browsing and interaction in geospatialdata libraries, as theydescribecharacteristicsof datasetstypically usedto formulatequeries.The visualizationtech-niquesdevelopedtransformresultsetsof GDL queriesinto a visual form: so-calledresultscenes.In this transformationprocess,a numberof mappingsfrom metadataattributesto visualattributesof theresultsceneareperformedmakingsurethatmetadataattributescanbeprocessedeffectivelyand correctly by the human visual system.

The spatialmetadatavisualizationmethodsareaugmentedby a setof interactionmetaphorsandtechniquesfor GDL queryresultsets.Theseinteractiontechniquesoperateon the resultscenes.Specificinteractiontechniqueslet usersexplorebothdetailandcontext of a resultscene,compareindividual SIOs effectively, or use the result scene to refine the query.

Both, themetadatavisualizationandinteractionmethodsareintegratedinto aninteractive systemfor browsingandsearchinggeospatialdata.In this interactivesystemgraphicalviewsof datachar-acteristicscanbecreatedandcombineddynamicallyandthedatasetsfoundcanbepreviewedandaccessed.

2 Geospatial Metadata Visualization

2.1 Visualization Paradigm

Thegeospatialmetadatavisualizationtechniquesdevelopedtransformresultsetsof GDL queriesinto a visual form: so-calledresult scenes. In this transformationprocess,a numberof mappingsfrom metadataattributesto visual attributesof the result sceneare performed.Thesemappings

6

make surethatmetadataattributescanbeprocessedeffectively andcorrectlyby thehumanvisualsystem [5, 13, 22, 28].

Geographicspaceis mappedto thexy planeof the resultscene,sincethis is botha naturalandaveryeffectivemapping.It is natural,becauseit preservesthefamiliar landscapeor mapmetaphorsthatmake it easyfor usersto spatiallyorientthemselvesin theresultscene.Thepositionin thexyplaneis alsothemosteffective visualvariableto encodequantitative data[5]. Encodingthegeo-graphiccontext of geospatialdataeffectively is important,becauseit is themostimportantsearchcriterion for users [32].

Temporalattributesof themetadataarepreferablymappedto thez planeof theresultscene.This isalsoaneffectivemapping,reflectingthefactthatusersratethetemporalcontext of geospatialdataas the second most important search criterion.

Quantitative attributesof themetadataaremappedto color intensityand/orhue.Color intensityisthemosteffectivevisualvariablefor quantitativeattributessuchascostandsize.Colorhueis alsovery effective, if thesuccessionof huesis perceivedasnatural[8, 29,36]. The“hot iron” andthe“spectrum” scales used in the example visualization are effective for quantitative data.

Categoricalattributesof themetadata,e.g.datasetformats,arealwaysmappedto differentcolorhues.

2.2 Test Data

Themetadatausedin theexamplevisualizationshasbeenextractedfrom theCaliforniaEnviron-mentalInformationCatalog[10] andvariousothersources[15]. Metadataattributescoveredin thevisualizations are:

• Geographic Region - the boundingbox of the SpatialInformationObject(SIO) described.(Metadatarecordsthatdefineaboundingbox largerthanacertainthresholdor noboundingbox at all are not shown.)

• Time Period - the period covered by the SIO

• Publication Date - the date the SIO was originally published

• Update Frequency - continually, daily, weekly, monthly, annually, biannually, as needed, ir-regular, or, none planned

• Size - the size of the SIO in megabytes

• Scale - the map scale, e.g., 1:24,000

7

2.3 Surface Plots

Examiningtheresultsetof a search,oneof thebasictasksis to getanoverview of thespatialdis-tribution of SIOsmatchingthesearch.A surfaceplot givessuchanoverview, mappingtheabso-lutenumberof SIOscoveringapointon theplaneto boththeheightof thesurfaceabove thispointandto thesurfacecolor. In Fig. 3, two differentviewsof thesamesurfaceplot areshown, onefromabove, andonefrom the southsideof the planelooking north.The surfaceplot makesit easytoidentify several generalpropertiesof the resultset’s geographicdistribution: therearefour geo-graphicareaswith a high densityof SIOsin this search:GreaterSanDiego, the Monterey Bay,Lake TahoeandtheSanFranciscoBay. Otherareas,suchastheMojave Desertarenot coveredatall. Locatingtheseareasis madeeasyby renderingthesurfacetransparentlyonareferencemapasa backdrop.

Fig. 3

Technique:Surface Plot

Attributes:Number of SIOs:

mapped to height and color

8

2.4 Box Piles

Surfaceplots show the spatialdistribution of SIOs but give little information aboutthe spatialextentof individual SIOs.A techniquethatportraysthenumerousnessof SIOsat certainregionsbut alsoshows thespatialextentof eachsingleSIO is a box pile (seeFig. 4). In this visualizationmethod,eachSIO is representedby a flat box. The x andy sizeandpositionof the boxescorre-spondto the geographicregion attribute. Boxes that would overlap in the xy planeare simplystacked on top of each other.

In boxpilesthecolorof theindividualboxescanbeusedto mapanothermetadataattribute.Fig. 5mapsthe publicationdateto the box color; a logarithmic scalehasbeenused.Fig. 6 mapstheupdatefrequency, a categoricalattribute,to thebox color. Theplotsmake it easy, for example,toidentify areas with SIOs that are recent or that have a high update frequency.

Fig. 4

Technique:Box Pile

Attributes:Geographic Region

9

Fig. 5

Technique:Box Pile

Attributes:Geographic RegionPublication Date:

box color

Fig. 6

Technique:Box Pile

Attributes:Geographic RegionUpdate Frequency:

box color

10

2.5 Temporal Box Plots

Thebox pile techniquestacksboxeson top of eachotherto preventthemfrom overlapping.Thus,theverticalpositionsof theboxesareassignedrandomly. Temporalboxplotsusetheverticalposi-tion in a differentway: to expressoneof thetemporalattributesof theSIOs.In Fig. 7, theendofthe time period of the SIO is mapped to the vertical position.

A problemwith thetemporalbox plots is thatboxescanoverlap.Eventhoughthebox materialistransparent, some boxes may be obstructed or barely visible.

Again, thecolor of theboxescanbeusedto expressanadditionalattribute.In Fig. 8 thescaleofthe SIO is mapped to the color. Those SIOs that do not have the scale attribute are shown in gray.

2.6 Space Time Plots

In theTemporalBox Plotsaboveonly pointsin timecanbemapped.SpaceTimePlotsareagener-alizationof thattechniquealsopermittingtimeperiodsto bemapped.Heretheboxesareno longeruniform in height;instead,theheightcorrespondsto aduration.Fig. 9 showsaTemporalBox Plotof searchresultset.Examiningthescenein 3D makesthetemporalcharacteristicsof theSIOsvis-ible, e.g., those SIOs can be identified that cover a very long period in time.

Fig. 7

Technique:Temporal Box Plot

Attributes:Geographic RegionEnd of Time Period:

vertical box position andbox color

11

Fig. 8

Technique:Temporal Box Plot

Attributes:Geographic RegionEnd of Time Period:

vertical box positionScale:

box color

Fig. 9

Technique:Space Time Plot

Attributes:Geographic RegionTime Period:

vertical box position andsize

Start of Time Period:box color

12

As in theothervisualizationtechniques,thecolorof theboxescanbeusedto expressanadditionalattribute. In Fig. 10 thescaleof theSIO is mappedto thecolor. ThoseSIOsthatdo not have thescale attribute are shown in gray.

2.7 Glyph Plots

A shortcomingwith both,temporalboxplotsandspacetimeplotsis thatboxescanoverlap.Someboxesmay be obstructedor barelyvisible. It canbe especiallyhardto spotthe numerousnessofSIOs that have identical bounding boxes and time periods.

Glyph plotsavoid this shortcomingby mappinga compact“glyph” (simply a spherein theexam-ple visualizations)at or nearthecentersof theSIO boundingboxes.If two boundingbox centersare identicaloneof the glyphswill be slightly displacedin the scene.The areaof the boundingboxes can be mapped to the glyph size and/or color.

Fig. 11 shows a glyph plot of a resultsetwhereglyph sizeandcolor correspondto thegeographi-cal areacoveredby the SIOsandthe vertical glyph positioncorrespondsto the endof the timeperiod covered. In Fig.12, the glyph color shows an additional attribute, the size of the SIO.

Fig. 10

Technique:Space Time Plot

Attributes:Geographic RegionTime Period:

vertical box position andsize

Size (MB):box color

13

Fig. 11

Technique:Glyph Plot

Attributes:Center of Geographic Re-gion:

glyph positionArea of Geographic Region:

glyph size and colorEnd of Time Period:

vertical glyph position

Fig. 12

Technique:Glyph Plot

Attributes:Center of Geographic Re-gion:

glyph positionArea of Geographic Region:

glyph sizeEnd of Time Period:

vertical glyph positionSize (MB):

glyph color

14

2.8 Aggregation Plots

Glyph plotscanshow a largenumberof SIO attributessimultaneouslyandthereforegive a com-pact overview of the characteristicsof the result set.However, they can get crowded when theresultsetsarevery large.They alsodo not show theexactspatialextentof eachSIO but only mapthe geographical area covered to the glyph size.

A wayto reducethenumberof glyphsin aresultsceneis to show only asingleglyph for anumberof differentSIOsthathave very similar characteristics.Aggregationplots(seeFig. 15 on page16)aggregateall SIOsof a resultsetthatcover thesamegeographicareaandtime span(plus/minusasmalldelta)andthathavethesamevaluesfor all otherattributesshown in thescene(e.g.,thesamepublicationdatesin Fig. 15). Aggregationplotsusetwo differenttypesof glyphs:cubesfor indi-vidual SIOsandspheresfor aggregatedSIOs.Thesizeof thesphereis proportionalto thenumberof SIOsaggregatedin oneglyph. Aggregation plot glyphsalsousethin lines to show the exactwidth and height of the geographic area covered.

3 MetaViz

MetaViz is an experimental system for searchingand browsing geospatialdigital libraries.MetaViz is basedon thegeospatialmetadatavisualizationmethodspresentedin chapter2. It aug-mentsthesemethodswith dynamicinteractiontechniquesfor the resultsets.Specificinteractiontechniqueslet usersexplore both detail and context of a result scene,compareindividual SIOseffectively, or usethe resultsceneto refinethe query. MetaViz alsoprovidesa userinterfaceforspecifying GDL queries.

Fig. 13: MetaViz architecture.

GeodataCollection

QueryInterface

ResultSceneGUI

GeodataCollectionGeodataCollectionGeodataRepository

GeospatialMetadata

QueryEngine

MetadataVisualizer

Inter- QueryPresentation RefinementactionInitialQuery

Modified Query

Result Set

DataAccess

NETWORK

15

3.1 Architecture and Implementation Notes

Fig. 13 shows the architecture of MetaViz. At its core the system has a database of geospatialmetadata and a query engine on top of it. Users can specify queries in the graphical query interfacewhich is imbedded into the GUI. To specify the geospatial constraints of the query, MetaViz offersa map interface (Fig. 14), other constraint types offered are temporal constraints (time period, pub-lication date, metadata date), thematic constraints (collection name, themes, keywords), size,scale, and format. Once the query has been started, the query engine creates the result set metadataand hands it over to the metadata visualizer. The visualizer transforms the metadata into a three-dimensional graphic representation, the result scene. A 3D viewer window in the GUI renders theresult scene and lets users interact with it (see Fig. 15). Once the user decides to access one of theSIOs in the result scene, a network connection to the geospatial data repository is created and theSIO is retrieved and displayed.

MetaViz is a Java 2 application that makes use of the Java 3D extension. The geospatial metadatais stored in an mSQL database and accessed through the JDBC interface.

3.2 Interacting with the Result Set

3.2.1 Navigation and Selection

MetaViz offers several ways of interaction with the result set. The most basic interaction techniqueis navigation. Users can examine the scene by modifying the viewing parameters, rotating thescene, etc. In addition to this basic viewer navigation, users can also select one of the glyphs orboxes of a result scene to display a window listing all the attributes of the corresponding SIO (seeFig. 16). Both, the result scene window and the attributes window offer hyperlinks that can be fol-

Fig. 14: MetaViz: spatial query window.

16

lowed to modify parametersof the visualization.Following oneof the icons in the result scenewindow triggersthe generationof a new resultscenewith a differentvisualizationtechniqueorcolor mapbeingused.Following oneof theattributenamehyperlinkstriggersthegenerationof anew result scene that maps the attribute to a visual variable, e.g. color.

3.2.2 Keyword Searches

Following oneof thekeywordhyperlinksin theattributewindow triggersanew queryfor thiskey-word. For multiple keyword searches,MetaViz computesa scorefunction for eachSIO in theresultsetandmapsthis functionto a visualvariable,e.g.color. Fig. 17 shows anexampleof sucharesultsetwith ascorefunction.Thekeywordsearchthatleadto this resultsetwas“boundaryORwildlife OR refugeOR bay”. MatchingSIOswerefoundall over theUS but thescorewassignifi-cantly higher for SIOs in the New England states.

Fig. 15: MetaViz: result set.

17

Fig. 16: MetaViz: SIO attributes.

Fig. 17: MetaViz: Result Set for Keyword Search

18

3.2.3 Geographical Zoom

An importantinteractiontechniquefor result setsis the geographicalzoomfunction. Userscanselectapointon thebasemapandzoomin or zoomoutaroundthispoint. In generalthiswill trig-gera new querywith a different(smallerresp.larger)geographicregion of interest.For aggrega-tion plots this will generallyresultin fewer SIOsbeingaggregated(andconsequentlymoredetailbeingshown) for zoomin andin moreSIOsbeingaggregated(lessdetailbeingshown) for zoomout.

4 Discussion

4.1 Summary

Distributed digital libraries promisea level of information accessibilitynever reachedbefore.However, having accessto largecollectionsof datais of little useunlesspeoplearein apositiontofind theinformationthey needandalsoknow how to useit. Locating,selectingandusinginforma-tion effectively is especiallydemandingin thecaseof geospatialdata,with their manifoldcharac-teristicsof scale,quality, cost,format,andthematiccoverage.Recentinitiativesto spatialdigitallibrariesprovide accessto a wealthof distributeddata,but offer only crudelevelsof interactivityanduserassistance.Consequently, usersfind it difficult and time-consumingto browse throughdata collections and locate those data sets that meet their requirements.

This projectaddressestwo of the major barrierspreventingthe extensive useof digital libraries:lackof usabilityandinformationoverload.Theresearchwasfocusedongeospatialdata,makingitpossibleto develop effective visualizationand interactiontechniquesthat are basedon familiarspatial metaphors like maps, landscapes and city models.

In thefirst phaseof theprojectasetof visualizationmethodsfor spatialmetadatahavebeendevel-oped.Spatialmetadataarea key to browsing andinteractionin geospatialdatalibraries,astheydescribecharacteristicsof datasetstypically usedto formulatequeries.Employing three-dimen-sionalvisualizationtechniques,severalcharacteristicsor dimensionsof metadatacanbecombinedinto singlegraphicalviews. As thosevisualizationsarebasedon mapandlandscapemetaphors,they are easy to understand and provide instant overviews of complex data characteristics.

In thesecondprojectphasethesemetadatavisualizationmethodswereintegratedinto an interac-tive systemfor browsingandsearchinggeospatialdata.In this interactive system,graphicalviewsof datacharacteristicscanbe createdandcombineddynamicallyandthe datasetsfound canbepreviewed and accessed.

4.2 Conclusions and Further Work

MetaViz is an experimentalsystemthat demonstratesthe useof informationvisualizationtech-niquesto interactwith resultsetsof geospatialdatasearches.MetaViz offers usersof geospatialdigital librariesagraphicalview of searchresultsets.Comparedto traditional,list-basedpresenta-

19

tion of resultsets,graphicalviews make it mucheasierto gain a quick overview over theresults,takinginto accountmany differentattributesof theSIOs.Othertasksareeasierto performvisuallyaswell, suchasthecomparisonof attributesfor SIOs,detectionof patternsin theSIOdistribution,or the identification of promising candidates for more detailed inspection.

An interestingareafor furtherresearchwould bea detailedusabilityevaluationof graphicalinter-facesto result sets,suchasMetaViz, in comparisonto list-basedinterfaces,suchas the currentADL interface.A numberof psychologicaltestscouldbedevised,wheretheperformance(speed,accuracy) achieved with both interfaceswould be measuredandcomparedfor a numberof testusers.

Therearemany waysof extendingthesystem,bothby integratingnew visualizationmethodsandby developing other interaction techniques.In the current state, temporal attributes are onlymappedto two of thevisualvariables,theverticalpositionand/orthecolor of glyphs.Othermap-pingsmight involve animatingtheresultsceneor usingfurthervisualvariables[25, 27]. Non-spa-tial visualizations(wherethe geographicspaceis not mappedto the xy planeof the result set)should also be supported when the other attributes of the SIOs are more important to the user.

Navigation in the result scenecould be improved by providing a directed-flight option anddynamicglyphs.With the directed-flightoption,userscould click on oneof the glyphsto startasmoothflight that takesthemcloseto the glyph to show its neighborhoodin detail.Aggregationglyphs could also be madedynamic in a way that they automaticallyexpandto the individualglyphs (to a higher level of detail) as users move close to them.

Acknowledgements

Thiswork hasbeenmadepossibleby ascholarshipfrom theGermanAcademicExchangeService(DAAD).

References

[1] C. Ahlberg, B. Shneiderman. Visual information seeking: tight coupling of dynamic queryfilters with starfield displays,Proc. CHI’94., pp. 313-480, Boston, MA, April 1994.

[2] AlexandriaDigital Library. Adl ProjectNewsletter.Fall ’98.http://www.alexandria.ucsb.edu/docs/news_fall98.html

[3] K. Beard, T. Smith and L Hill. Meta-information models for georeferenced digital librarycollections. Proc. IEEE 2nd International Conference on Metadata. Silver Spring, MD,1997.

[4] K. Beard, V. Sharma. Multilevel and graphical views of metadata.Proc. IEEE Advances inDigital Libraries, Santa Barbara, CA, 1998.

[5] J.Bertin.Semiology of Graphics. The University of Wisconsin Press, Madison, 1983.Translated by William J. Berg.

[6] K. W. Brodlie et al.Scientific Visualization: Techniques and Applications. Springer, Berlin,1992.

20

[7] B. P. Buttenfield, M. P. Kumler. Tools for browsing environmental data: the Alexandriadigital library interface.Proc. 3rd Int. Conf./Workshop on Integrating GIS andEnvironmental Modeling. National Center for Geographic Information and Analysis, SantaBarbaraCA, 1996.

[8] H. W. Calkins,H. J.Dickson.Theeffectiveuseof color in cartographicdisplays.Proc.of theInt. GIS Symp., Vol. 3, ArlingtonVA, 1987, 189-199.

[9] CEN/TC 287, Geographic Information. European Prestandards. http://forum.afnor.fr/afnor/WORK/AFNOR/GPN2/Z13C

[10] CERES. California Environmental Information Catalog. http://www.ceres.ca.gov

[11] V. Coors, V. Jung. Using VRML as an interface to the 3D data warehouse.Proceedings ofthe VRML’98 Symposium (Monterey CA, Feb. 16-19, 1998), ACM Press, New York, 1998.

[12] P. Cubaud, C. Thiria, A. Topol. Experimenting a 3D interface for the access to a digitallibrary. Proc. ACM Digital Libraries, pp. 281-382, Pittsburgh, PA, 1998.

[13] B. D. Dent.Cartography:ThematicMapDesign. W. C. BrownPublishers,DubuqueIA, 4thedition, 1996.

[14] Federal Geographic Data Committee. Content standards for digital geospatial metadata.FGDC, U. S. Geological Survey, Washington, D. C., 1994.

[15] Federal Geographic Data Committee. National Geospatial Data Clearinghouse. http://130.11.52.178/clearinghouse_sites.html

[16] A. U. Frank. Geographic information systems: a challenge to computer graphics. State-of-the-Art Report, Eurographics ‘93, Barcelona, 93.

[17] C.G.Healey.Choosingeffectivecoloursfor datavisualization.Proc.IEEEVisualization‘96(held in San FranciscoCA, Oct. 27 - Nov. 1, 1996), IEEE Computer Society, 1996.

[18] Hearnshaw, H. M. and Unwin, D. J. (Eds.).Visualization and GIS. Wiley, London, 1994.

[19] M. Hearst.TileBars:Visualizationof TermDistributionInformationin Full Text InformationAccess.Proc. ACM SIGCHI’95, pp. 59-66, Denver, CO, May 1995.

[20] M. Hearst, C. Karadi. Cat-a-Cone: An Interactive Interface for Specifying Searches andViewing Retrieval Results using a Large Category Hierarchy.Proc. ACM SIGIR‘97.Philadelphia, PA, July 1997.

[21] ISO. Draft International Standard on Metadata for Geographic Information, ISO 15046-15.International Organisation for Standardization, Geneva, 1998.

[22] V. Jung.Integrierte Benutzerunterstützung für die Visualisierung in Geo-Informationssystemen. Ph.D. Diss, Techniche Universität Darmstadt, FachbereichInformatik, 1998. Fraunhofer IRB Verlag, Stuttgart, 1998.

[23] V. Jung.Knowledge-basedvisualizationdesignfor geographicinformationsystems.In Proc.of the3rd ACM Int. Workshop on Advances in Geographic Information Systems(BaltimoreMD, 1995), ACM Press, New York, 1995, 101-108.

[24] P. R. Keller, M. M. Keller.Visual Cues: Practical Data Visualization. IEEE ComputerSociety Press, Los Alamitos CA, 1992.

21

[25] M.-J. Kraak, A. M. MacEachren. Visualization of the temporal component of spatial data.Proc. of the 6th Int. Symp. on Spatial Data Handling, Edinburg, 1994, 391-409.

[26] G. E. Langram.Time in Geographic Information Systems. Taylor & Francis, London, 1992.

[27] A. M. MacEachren.Timeasacartographicvariable.In H. M. Hearnshaw,D. J.Unwin (Eds.)Visualization and GIS, Wiley, London, 1994, 115-130.

[28] J. Mackinlay.Automatic Design of Graphical Presentations.PhD thesis, StanfordUniversity, Computer Science Dept, Stanford CA, 1986.

[29] J.Olson,C. Brewer.An evaluationof colorselectionsto accommodatemapuserswith color-vision impairments,Annals of the Association of American Geographers, Vol. 87, No. 1,1997.

[30] R. Rao,J.O. Pedersen,M. Hearst,etal. RichInteractionin theDigital Library,Comm.ACM,Vol. 38, No. 4, 29-39, April 1995.

[31] G. G. Robertson,S.K. Card,J.D. Mackinlay.Informationvisualizationusing3D interactiveanimation.Comm. ACM, Vol. 36, No.4, 1993.

[32] P. Sallis. L. Hill, M. Larsgaard, K. Lovette, C. Masi, M.-A. Rae. Alexandria digital libraryuser and use evaluation: experiments with a neural network method for log data analysis.Presentation at the ASIS 1999 Mid-Year Meeting, Pasadena, CA, May 24-26, 1999.

[33] A. Turk. Towards an understanding of human-computer interaction aspects of geographicinformation systems.Cartography, Vol. 19, No.1 (June), 1990, 31-60.

[34] A. Turk. Cogent GIS Visualizations. In H. M. Hearnshaw and D. J. Unwin (Eds.),Visualization and GIS, Wiley, London, 1994, 26-33.

[35] E. R. Tufte.The Visual Display of Quantitative Information. Graphics Press, Chesire CT,1983.

[36] C. Ware. Color sequences for univariate maps: theory, experiments, and principles.IEEEComputer Graphics & Applications, Vol. 9, September 1988, 41-49.

[37] R.Weibel.Geographicinformationsystems(GIS)andvisualization.State-of-the-ArtReport,Eurographics ‘94, Oslo, 1994.

[38] J. D. Wood, K. W. Brodlie, H. Wrigh. Visualization over the World Wide Web and itsapplication to environmental data.Proc. IEEE Visualization ‘96(held in San Francisco CA,Oct. 27 - Nov. 1, 1996), IEEE Computer Society, 1996.