JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 101, NO. B4, PAGES 8741-8743, APRIL 10, 1996
A global,self-consistent, hierarchical,high-resolutionshoreline database Pill Wessel Schoolof OceanandEarthScienceandTechnology,Universityof Hawaii at Manoa,Honolulu
Walter H. F. Smith NOAA Geosciences Laboratory,NationalOceanService,SilverSpring,Maryland
Abstract. We presenta high-resolution shorelinedatasetamalgamated from two databases in the publicdomain. The datahaveundergone extensiveprocessing andarefreeof internalinconsistenciessuchaserraticpointsandcrossingsegments. The shorelines areconstructed entirelyfrom hierarchicallyarrangedclosedpolygons.The datacanbe usedto simplifydatasearches anddata selections or to studythe statisticalcharacteristics of shorelinesandlandmasses.The datasetcan be accessed bothelectronically overInternetandfromtheNationalGeophysical DataCenter, Boulder, Colorado; it comes with accesssoftware and routines to facilitate decimation based on a
standardline-reductionalgorithm.
Introduction
With ever-increasing amountsof remotelysenseddata,it often is necessaryto perform intricate data searchesand selections basedon multiplecriteria. A particularcriterionis whetheror not the data representvaluesover land or water. In many studies, shorelinedata are usedto constructlogicaldatamasksin orderto manipulateand "mask out" parts of a large data set. In other cases,the intersectionbetweendataprofilesand shorelinesmust be determined. Finally, in somestudiesthe shorelinedata themselvesare the objectof study. A systematicapproachto these typesof data retrieval and processingrequiresa self-consistent, hierarchical,high-resolutionshorelinedatabase.By self-consistent, we mean that all shorelinesare representedas continuous closedpolygonsandthe dataare free of shorelineintersections or other artifactscausedby data inaccuracies.By hierarchical,we meanthat the shorelinesare orderedso that the polygonsrepresentingocean-landboundariesmay be distinguished from those outliningland-lakeboundariesand alsothat eachpolygoncan be rankedaccordingto how muchareait encloses. We presenta digitaldatasetthat fulfills theserequirements.It was constructedfrom two well-known,publicdomaindata sets. The World Data Bank II (WDB; also known as CIA Data Bank)
containscoastlines,lakes,politicalboundaries,andrivers. These datahavean approximateworkingscaleof 1:3 million, meaning the featuresare considered to be accuratelylocatedon mapsusing that scale or smaller.
The other data set is the World
Vector
Shoreline (WVS), which only contains shorelinesalong the ocean/land interface (i.e., no land-locked bodies of water). The
WVS data set is superiorto the WDB data set in quality and resolution(its working scaleis approximately1:100,000),but it lackslakes. Althoughnot explicitly given, the precisionof the WDB dataappearto be in the 500-5000 m range,while the preci-
Copyright1996 by the AmericanGeophysical Union. Papernumber96JB00104.
sionof WVS is an order of magnitudebetter. We producedour datasetusingthe WVS datawhenpossibleandsupplementing it with WDB
data.
We obtained these data sets over the Internet.
They arealsoavailableon CD-ROM from the NationalGeophysical Data Center (NGDC) [1994].
Processing To facilitateland/waterdeterminations, it is necessarythat the shorelinedata be organizedin closedpolygons. Both the WVS andWDB dataconsistof unsortedline segments;no information is providedwith them to indicatewhich line segmentsbelongto the samepolygon. In addition,polygonsenclosingland mustbe differentiatedfrom polygonsenclosingwater (e.g., land-locked lakes)sincetheymay be usedin differentcontexts. The WVS and WDB togetherrepresentmore than 100 Mb of binarydataandcloseto 15 million datapoints. The largeamount of datanecessitated automaticprocedures for datamanipulation. Our first processing stepwasto removepointduplicates(repeated values)andoutliers(identifiedassinglepointsalongthe shoreline whosetwo immediateneighborswere identical.) In nature, no shorelinecan crossanothershoreline,but the digitizedrepresentations of shorelinesoften do cross;correctingsuch artifactsbecomesa complicatedprocessingstep. Crossingsegmentswere automaticallyedited,providedthatonly a few pointshad to be deleted. We determinedcrossoverlocationsusing the crossover routinesof Wessel[1989]. Crossoversmostlikely arosebecause manualdigitizationoftenproduces slightoverlapsinsteadof exact closure. We foundthat the majorityof segmentcrossoverswere near the segment'sendpoints. Hence endpointswere automatically removeduntil no crossingsremained. In a few hundred casesthe crossoverandeditingalgorithmswouldhaveeliminated morethan5% of the pointsin a segment.In thesecaseswe visually examinedthedatato determine(subjectively) whichpointsto manuallyedit in order(1) to avoidcrossingsand (2) to keep the segmentascloseto its originalshapeaspossible. Next, we examinedall loose segmentsto determinewhich segments shouldbe joinedto produceclosedpolygons.Because
0148-0227/96/96JB-00104505.00 8741
8742
WESSEL AND SMITH: A SELF-CONSISTENTSHORELINE DATABASE
mostof thesegments didnotjoinexactly(i.e.,therewerenonzero Oncethehierarchical levelsof thepolygonsweredetermined, gapsbetween somesegments), wehadto findall possible combi- we enforceda commonhandedness for all polygons,i.e., we arnations of groupings andchoose thesimplest combinations (i.e., rangedthemsothatwhenonemovesalonga polygon's perimeter thatgavethesmallest segment separation). TheWVSsegmentsfrombeginningto end,theareaimmediatelyto one'sleft is land.
joinedtoproduce morethan180,000 polygons, thelargest being Thuslevel 1 and3 polygonsgo counterclockwise, while level 2 the completeAfrica-Eurasia polygonwhichhasmorethan1.4 and4 polygons go clockwise.At thisstepwe alsocomputed the millionpoints.The WDB dataresulted in a smallerdatabase, areaof all polygons.
about 20% the size of WVS.
The next stepwas to combinethe WVS and WDB databases.
Themaindifficultyin thisstepwasthepresence of duplicate Examples polygons:obviously,most of the featuresin WVS are also in
Shorelines areusedin a varietyof situations.For instance,an applicationin satelliteradaraltimetrymightrequirea land/water maskmadefromall polygons enclosing an arealargerthanthealwhichonesto ignore.We usedtwotechniques to address this timeter's"footprint,"that is, the regionwhich backscatters radar
WDB. However,because theresolution of thedatadiffers,it is nontrivialto determine whichpolygons in WDB to includeand
problem.First,we lookedfor crossovers between all possible energy,or anareaperhaps 50 km: in area. The hierarchical and pairsof polygons. Because of thecrossover processing discussedorientedpolygonsmakepossiblegraphicsfill operationsof feaabove,we knewthattherewerenoremaining crossovers inter- tureswith holesin them,sothatlandareasmaybe paintedwith a nallywithinWVSandWDB;thuscrossovers couldoccuronly maskwhichallowswater-covered areasto showthrough,regardbetween WVS andWDB polygons.If crossovers weredetected, lessof whethertheyaremarineor lacustrineareas.This featureis
theycould indicate oneoftwoscenarios: (1)A slightly misplaced usefulfor plottingmapsof griddeddatawhichareonly validover WDBpolygon crosses a moreaccurate WVSpolygon, bothrepre- wet areas[e.g., Sandwellet al., 1995]. senting thesame geographic feature, or(2)a smallWDBpolygon Full resolutionof the datasetis vital whenworkingin rela-
representinga coastallake crossesthe more accurateWVS shore-
tivelysmallareasbutbecomes impractical for regionalandglobal
line.Wedistinguished between these cases bycomparing thearea applications.It thereforeis desirableto obtainreducedversionsof andcentroidof thetwopolygons.In almostall casesit wasobvi-
the completedatabase,corresponding to different resolutions.
ouswhenwehadduplicates; a fewcases hadtobeinspected visu- Suchdecimationcanbe carriedout usingthe Douglas-Peucker ally. Unfortunately, onmanyoccasions theWDBduplicate poly- line-reductionalgorithm[Douglasand Peucker,1973]. The rougondidnotcross itsWVScounterpart butwaseitherentirely in- tineworksto reducetherichnessof texturealonglinesby removsideor outsidetheWVS polygon.In thosecaseswe reliedonthe
ingpointsfromtheshoreline segment, thusgivinga straighter line
area-centroid tests.
segment. This processdependson a tolerancevalue: if the re-
We nexthadto assigna hierarchical levelto eachpolygon. movalof a pointcausestheresultingstraight-linesegmentto deHere,level1 polygons represent oceanboundaries, level2 poly- part morethan A km from the actualdata pointsoriginallybegonsrepresent lakeboundaries, level3 polygons represent island- tweenthesegment's newendpoints, thenthepointis kept. Figure in-lakeboundaries, andlevel4 polygons represent pond-in-island- 1 illustratesthe effect of the line-reductionalgorithmon the in-lakeboundaries. Level4 wasthehighestlevelencountered in shorelineof the islandof Sardinia,Italy. Here, we haveusedvalthe data. To automaticallydeterminethe hierarchicallevels,we uesof A = 0.2 km, 1 km, 5 km, and25 km whichtypicallyleadto compared all possible pairsof polygons to find howmanypoly- -20% reductionin data size for each stepin resolution. These gonsa givenpolygonwasinside.
8'
five data sets, derived from the data set discussedin this note,
9'
10'
8'
N= 1161 41'
•'
9'
10'
• '•
41'
40'
40'
39'
"'•F! A= 0.2
8'
9'
10'
39'
A=Skm
I 8'
9'
R 10'
A
2
05% 8'
9'
10'
Figure1. Example ofhowtheverydetailed polygon representing theisland ofSardinia (8816points) maybereduced bychoosing various tolerances. N indicates thenumber ofpoints inthepolygon. Rrepresents thepercentage of pointsin relation totheoriginalfullresolution polygon (ontheleft).
WESSELAND SMITH: A SELF-CONSISTENTSHORELINE DATABASE
8743
of approximately 175m:. Forthemostpart,thedistribution of
10 6
land areas(level I WVS polygons)followsa powerlaw (Figure 2), asonemightexpectof the contoursof a fractalsurface.The
departure fromthispower lawatareas lessthan0.1km:probably
10 5 -
reflectsundersampling of suchsmallfeatures. 0 l
•., 104 -
Appendix The shorelinedata file is a single 89 Mb binary file using a
simple,straightforward integerformatthatis described in theaccompanying documentation; sampleprograms distributed withthe .Q
data showusershow to accessthe file as well asdecimateit using theline-reduction routineabove. Extractingdatafrom the file can
10 2-
bedoneona personal computer withverymodest memory.How101 -
ever,theline-reduction routine(asimplemented) requires-36 Mb
of memoryto process the largestpolygon;we thereforeprovide 10 0
10 '4
10 '2
10 ø
10 2
10 4
10 6
10 8
polygonarea (km2) Figure 2. Land polygonsfollow a powerlawdistributionover a largerangeof polygonareas. Departurefrom the lineartrendat smallareasmostlikely represents undersampling.
the four lower resolutions usedin Figure I in additionto the full resolutionset. All files and programsmay be obtainedfrom the WorldWideWebat http://www.soest.hawaii.edu/wessel/wessel. html or from the National GeophysicalData Center,Boulder, Colorado.
Acknowledgments. R.E. Arvidson,D. Merritts,andJ. H. Willeminprovidedthorough reviews.Thisworkwassupported by grantEAR-9302272 from the National Science Foundation. School of Ocean and Earth Sci-
enceandTechnologycontribution 4048.
make up the binnedshorelinedata distributedwith the Generic MappingTools (GMT) [ Wesseland Smith, 1995]. (The GMT softwarepackagealsocontainstoolsto createdatamasksbasedon thesefive resolutions,with the optionsto ignoresmall featuresas discussedabove. However, GMT is not used to access the data discussedin this note.)
References
Douglas,D. H., andT. K. Peucker,Algorithmsfor thereduction of the numberof pointsrequired to represent a digitizedlineof itscaricature, Can.Cartogr.,10, 112-122, 1973. NationalGeophysicalData Center(NGDC), Global Relief Data CDROM, Boulder,Color., 1994.
Sandwell,D. T., M. M. Yale, and W. H. F. Smith,Gravity anomalyprofiles from ERS-I, Topex, and Geosataltimetry,Eos Trans.AGU, 76(17),SpringMeet.Suppl.,S89, 1995. Wessel,P., XOVER' A cross-overerror detectorfor track data, Cornput.
The user may createreduceddata setsof arbitrary resolution usingsoftwarearchivedwith thedatadescribedhere. The line-reductionalgorithmmay producesegmentsthat crossone another, so that if it is applied without further processingas outlined Geosci., 15,333-346, 1989. above,self-consistency of the resultscannotbe guaranteed. Wessel,P., and W. H. F. Smith, A new versionof the GenericMapping Tools(GMT), EosTrans.AGU, 76(33), 329, 1995.
Statistics
The completedatabasecontains188,628polygonsrepresenting 10,222,509datapoints.The meanpointseparation is 178 m, with valuesrangingfrom a few metersto an extremeof 24 km in Antarctica. The largestpolygonrepresentsthe combinedEurasiaAfrica continents;it contains 1,435,084 points. The smallest polygonis a smallarcticislandnearQueenElizabethIslandsoff northernCanada;it is madeup of only 4 datapointsandhasa size
W. H. F. Smith, NOAA Geosciences Laboratory,N/OES12, NationalOceanService,SilverSpring,MD 20910-3281.(e-mail: walter@amos. grdl.noaa.gov) P. Wessel,Department of GeologyandGeophysics, SOEST, Universityof Hawaiiat Manoa,2525CorreaRoad,Honolulu,HI 96822. (e-mail:
[email protected])
(ReceivedJuly 11, 1995;revisedDecember6, 1995' acceptedDecember28, 1995.)
