Dec 10, 2000 - solutions slightly and consistently differ from NNR-A. The systematic disagreement between the two geode- tic solutions (ENS 97 and ITRF 97) ...
JOURNALOF GEOPHYSICAL RESEARCH,VOL. 105,NO. B12,PAGES28,279-28,293, DECEMBER10,2000
New constraints on Antarctic plate motion and deformation from GPS data Marie-No•lleBouin Laboratoire deRecherche en G•od•sie,EcoleNationaledesSciences G•ographiques Marnela Vall•e, France
Christophe Vigny Centre Nationaldela •echercheScientifique, UMPt8538,EcoleNormaleSup•rieure, Paris,France
Abstract. Fouryears(1995-1998)of continuous GlobalPositioning System (GPS)datarecorded by 11 permanent International GPSService(IGS) stations inAntarcticaand surroundingregionshavebeenprocessed usingan optimized methodfor regional geodeticnetworks. The 4-year time serieshas allowedus to extract nonlinear variations and constant horizontal velocities for all the stations.
In December1997 we have set up a permanent GPS station at Dumont d'Urville inTerreAd•lie. The resultingdata, alongwith thoseof the ScientificCommittee
onAntarcticResearch(SCAR) campaigns(1995and 1996)are includedin our processing. A coseismic displacement (1 to 2 cm) is detectedin the horizontal components of this station. We relate this to the March 25, 1998,M•,=8.1 Balleny Islands earthquakefor whicha dislocationmodelyieldspredictionin the right direction.For the six Antarctic stationsthe horizontalrates are very consistent withthe rigid plate rotation, all the residualvelocitiesare negligible(lessthan
2 mm/yr), exceptat O'Higginsin the headof the peninsula(7 mm/yr toward thecontinent).The positionof the rotationpole (62.0øN,146.7øW)and its rate (0.264ø/Ma)are significantlydifferentfrom the NNR-Nuvel-iA predictionsfor the Antarctictectonic plate but are consistentwith the Australian relative motion. Thehighhorizontalresidualin the peninsulaarea, whereicebergcalvingrates and iceshelfdisintegrationhave recently increased,can be interpretedas the horizontal component of the elastic crustal responseto the load variations.
1. Introduction
Antarctic coasts.The expectedassociatedviscoushorizontal motions consistin a slight divergenceof 1 to Present-daycrustal motions occurring over the Antarctic continent canbeproduced bya varietyofphe- 2 mm/yr overthe wholecontinent,which couldhardly be detected by current geodetictechniques. Moreover, nomena.First, variations of the mass of the Antarctic uncertaintiesin the predicted rates of both vertical and icesheetsincethe Last GlacialMaximum(LGM) inhorizontal componentsare important. The crustal beduce a viscoelastic response involvingcrustalandmanhavior is influencedby many parameters such as the tlemechanisms [Farrell,1972]. Variationsof the ice lithospheric thickness,mantle viscosityand deglaciation thickness in East Antarctica as deduced from relative history that remain largely unknown. sealevelrecordscoveringlong periods[Nakadaand In addition to these long-term ice mass changes, Lainbeck, 1989;Peltier,1994;Tushingham andPeltier, the Earth crust reacts more locally, elastically, to 1991] couldreachvaluessuperiorthan 1000m since the present-day evolution of the Antarctic ice sheet. theLGM(18,000years).It'hasbeenshown by James This behavior is controlled by accumulation, dependandIvins[1998]usingthe deglaciation modelof Dening mainly on climatic parameters like air temperatonetal. [1991]thanthe resulting present-day vertical ture and humidity and by the ablation rate of the ice
motions wouldreach15 to 20 mm/yr nearareaswith shelves. thelargest deglaciation rates,astheRossEmbayment
The recent evolution of the West Antarctic
Ice
Sheet is still poorly constrained:the air temperatures
butwould remain inferior than10mm/yrontheEast are increasingon the Antarctic peninsula,resultingin Copyright 2000bytheganerican Geophysical Union. Paper number 2000JB900285. 0148-0227/00/2000 JB900285$09.00
higheraccumulationand ablationrates. Differentrealistic ice sheet modelspredict either rapid melting and collapseof the ice shelvesor increasingaccumulation rates without any changein the ablation rates [Bind28,279
28,280
BOUIN AND VIGNY' ANTARCTIC PLATE MOTION FROM GPS DATA 0•
schadler,1997;Huybr'echts, 1994]. The predictedelastic uplifts around the Ross Ice Shelf and Filclmer-Ronne Ice
Shelf are slightly more than 10 mm/yr but, reachonly
6-7 mm/yr on the peninsula[,Jamesand I'vins,1998]. This scenm'iodoes not include the recently observed
celerationof melting on the major West Antarctic Ice Shelves. As far as the horizontal motions are concerned,
the present-day ice mass unloading could result in a slight convergencearound the main dischargecenters. This is in contrast with the divergenceobtainedin the viscouscase,an elastic trend involvingboth lithosphere and umlerlying mantle. At, a larger scaleit is necessaryto take into account dist)lacementsof [he tectonic plates. Estimating current values of relative plate motions and comparing them
witt• thoseover severalmillions years [DeMct• et al., 1990]hasbeenan importantgoalof GlobalPositioning System(GPS) analysisover the past 10 years and Hcfiiu, 1995]. The GPS networkincludesseveral stai ions on the tectonic plates surroundingAntarctica[, that is, Santiago(South America), Perth, Hobart and Auckland(Australi•t),and Chatlmm(Pacific),thusallowing•s to deducethe relative motionof the Antarctic plate. All the plate boundariesaround Antarctica as chm'actcrizedby seismicor magneticprofilesare ridges. A less-knownregion is located betweenthe southeast end of South America and the Antarctic peninsula,
with the Scotiamicroplate[Pelayoand W'iens,1989]. The preciselocationsand mechanisms of the different }•oundaricsof this plate are not, completelyknown yet. The soutl• boundary along the Antarctic plate is usu-
Perth
180'
Figure 1. GPS stations used in this study. The tectonic plate boundaries•tre from the N•vel-1 model
[DeMet•'et al., 1990]andthe seismicity aroundtheplate is after t he centroi{1moment t{•sor catalog, earthquakes witIx .h[• >_4.5since 1977. 2. Data
and
Measurements
Our analysisincludesall the permanentGPS stations availableon the Antarctic plate since 1995. Most
of these stations belong to the IGS global network
ally definedasa setof ridgesand transformfaultscalled and were installed in 1995, providing a quasi continand South Scotia Ridge. It, extends fi'om the Shackleton uousdata flow sincethen: Casey,Davis, lXlcMurdo, O'Higginson the AntarcticcontinentitselfandKergueFr•cture Zone and Bransfield Strait at its west end to len on the corresponding islands.Exceptfor O'Higgins. the South Sandwich Trench at the east. From the inall the stations are far away from the plate edgeand traplate tectonicpoint of view, the Antarcticplate is should document the rigid plate rotation.In December characterizedby a stableand quiet behavior.The seis1997 a new IGS-like permanent stationwasinstalled mic activity remains located on the plate boundaries at the French base of Dumont d'Urville, thanksto the [Okal,1981](Figure1). Thereforeweexpectverylitfinancial and technical help of the Institut Fran•ais tle intraplate deformationfrom tectonicorigin. pour la Recherche et la Technologie Polaire (IFRTP). Becauseof the expectedlow amplitude (lessthan Because 10mm/yr) andtimedependence ofthemotionswewant This GPSstationis nowrunningcontinuously. of technical problems in data transmission by Internato determine,we have chosento analyzethe Antarctic Organization, it isnotpossible International GPS Service(IGS) data with a specific tionalMaritimeSatellite to make them available within a reasonable time period. processing strategy. This strategyis adaptedto the Nevertheless, this station should soon be included inthe tectonicsand postglacialrebound of Antarctica, as opIGS network. The data. from late 1997 till the end of posedto theprocessing strategyusedbytheglobalIGS analysis centers.
1998havebeenprocessed in ouranalysis. Thegeodetic
To sumup, the aim of this studyis to assess thelevel pillar chosenfor this permanentstation has beenpre-
usedduringthe1995and1996Scientific Comof rigidityof the Antarcticplate,to compare its overall viously
onAntarctic Research (SCAR)campaigns. Data motion to the Nuvel-lA model, and to detect horizon- mittee havealsobeenincludedin our tal crustal deformations due to past and present-day from thesecampaigns
sitesarepictured Figure1, while deglaciation. First, we shallgivea detailedpresenta- study.All analyzed tion and discussionof our data analysis. Second,the
Figure2 shows thedataprocessed from1995to 1998.
There are four IGS stations on the Antarctic conquantitativeresultswill be described.Readersinterdatamoreorlesscontinuously estedmainlyin this secondpart canproceedandread tinentitself,providing it directly.
since1995. All of them are locatedalongthe coasts.
BOUIN AND VIGNY- ANTARCTIC PLATE MOTION FROM GPSDATA
nationalTerrestrial Reference FrameITRF 97). This
* AMUN * DUM1 * MAW1 * GOUG * VESL * PALM SANT PERT * OHIG * MCM4 MAC1 * KERG HOB2 * DAVl CHAT * CAS1 AUCK
28,281
transformation will occura posteriorirelativeto our GPSdata analysis.Thusthe sitesusedasreference stationsmustexhibita stablebehaviorfor the periodof theanalysis (1995-1998). Moreprecisely, theirITRF 97
I!nnnun!lmn
•,,,,,
,,,,,,,
•ql•
,,•,,i
,,,,
......
•
.....
,
,•,,,
•
• ....
1995.0
•.... ,,,•,,,
•,,,
.,.• ,,,,,,
iIia•Iili ,,,, .....
constantvelocitymust be representativeof their motion during this period.
,•,,.
••!•••I•
I•li
,,, ,,
,,,,,,,
................
i ailaii•l •
,,
, ,,
,
,,
,,,,,,•
Amongall theIGS sitessurrounding Antarctica,we choose to usethesixstationsshownon Figure1. Severalstationswereelimina.ted for differentrea. sons,such
I Iliiiil!li
......
•i• ,,,•,•
•,,,
I I
I
i .... i .... i .... •.... l .... i .... i .... i .... i .... l .... i .... i .... i.... f.... 1995 5
1996 0
1996 5
1997.0
1997 5
1998 0
1998 5
1999 0
Figure2. AvailableGPS data for all the sitesusedin thisnetwork,sincethe beginning of 1995, tbr the sta-
tions ofAmundsen (AMUN), Auckland(AUCK), Casey (CAS1),ChathamIsland (CHAT), Davis (DAV1), Dumont d'Urville (DUM1), GoughIsland (GOUG), Hobart(HOB2), Kerguelen (KERG), Macquarie Island(I•IAC1),Mawson(MAW1), McMurdo(MCM4), 0'Higgins (OHIG), Pahner (PALM), Perth (PERT), Santiago (SANT), and Sanae(VESL) (seeFigure 1 fbr stationidentification).The stationswithin the Antarcticplateare marked by an asterisk.
as poor geodeticquality of the data, like in La Plata
(Argentina),or stationswhichare too remotefromthe centerof our network(SouthAfrican stations),or installedtoo recently(EasterIsland). Finally,somestations would be redundant with others in the same area
(Australian stations).All thedataincluded in ourprocessingare presentedin Figure 2.
3. Data Analysis GPS data processing in Antarcticasuffersfrom several specificitiesof this area of the xvorld and benefits
from a few of them. On the one hand, the size of the continent,the sparsenetwork,and ionospheric activity
considerably increase the difficultyof,the task. On the In Antarctica,the main difficulty in installing permanentGPS stationsfor geodeticor geophysicalpurposes is to find ice-free areas, with bedrock outcrops. Directaccessto the solid earth is prevented by the coveringof the whole continent by an ice sheet up to 4500 m, excepton a few areas, all located near the edgeof thecontinent. This distribution is not ideally suited for the detectionof vertical glacial rebound, as the mostimportant postglacialmovementsare expectedin-
otherhand, the extremedrynessof the atmosphereand the unusuallylong satellite orbital arcs create more favorableconditions. All the data have been processed with the GPS scientificsoftwareGAMIT 9.1, developedat Massachusetts Institute of Technology (MIT). The current version of this software is an evolution of
the first versiondescribed by Kin# and Beck[1993]. GAMIT usesboth carrierphaseand pseudorange observables. The most precise information can be ob-
land.Nevertheless, it providesa networkgeometrywell tained by the carrierphase,but with ambiguitieson adapted to measure platetectonics velocities (withex- the integernumberof wavelengths.An unambiguous cellent spatialcoverage)and to detectpostglacialhori- signalcan be obtainedby fixing thesereal numbersto zontaldivergence.Data from four additionalstationsin themostprobableintegervalue. The success of thisstep Sanae (Veslekarvet), Amundsen, Mawson,andPalmer dependsessentiallyon the quality of the signaland the areavailable since1998(seeFigure2). The Amundsen lengthof the baselines.It is usuallyconsideredthat this permanent station is located at the South Pole base and
procedure is valid for baselinesunder 500 to 700 kin. In
builtonthe ice. Its motionis representative of the ice Antarcticathe ionospheric activity,whichis especially displacement only,but we havechosento includeit in high, introducesan additional noisewhich, under some ournetwork,in order to reinibrce the density of sites circumstances,can produce cycle slips or prevent the withinthe Antarctic continent. signalrecordingitself. Using the ionosphere-free linear To thosestations located on the Antarctic continent combinationof two frequenciescan help to removethe itself, wecanaddtheKerguelen station,withinthetec- first-orderof this ionosphericnoisebut resultsin an actonicplate,far enoughfrom the Australian-Antarcticcentuationof the nondispersive noise. The lengthsof ridge.As alreadymentioned, the only stationin the the baselinesof our netxvork, between 1000 and more vicinity of a plateboundary is O'Higgins, at thehead than 10,000kin, make ambiguityfixing impossible.All ofthepeninsula, nearthe Bransfield Straitrift. For the solutionspresentedin this studyleavethe ambigutworeasonswe have chosen to include some IGS sta- ity nonfixed. tions outside the Antarcticplatein ournetwork.The The considerablelength of the baselinesin the first reason islinked withtheaccuracy ofouranalysis.Antarctic network also reducesthe quantity of double
The addition ofexternal stations increases thestrengthdifference data (twosatellites,two stations)whichcan ofthenetwork.Second, wecanusethoseadditionalbe usedfor the GPS inversion.In this respectthe instations to tie ournetwork to a reference frame(Inter- troductionof Amundsenstationincreasedthe average
28,282
BOUIN
AND VIGNY:
ANTARCTIC
PLATE
numberof doubledifferences from 120to 136 and significantly contributed to the improvementof the repeatability after August 1998. The dry componentof the zenith troposphericdelay is implemented in the software by the $aastaraoi•e• [1971]model. The wet componentis estimatedduring the inversion, with zenith-delay adjustmentsevery 3 hours,which apparently correspondto the besttradeoff betweenformal uncertainty and repeatabilityfor two or moresuccessive sessions [Walpersdorf,1997].However, the troposphericactivity aboveAntarctica is quite simpleto model, the atmospherebeingone of the driest in the world. Di•culties in evaluatingthe tropospheric zenith-delayparametersare generallydue to large and rapid variations in the atmospheric rate of humidity, but such a rate southernmost
remains
more or less constant
in the
latitudes.
We have chosenthe elevationcutoff angle as a tradeoff between the amount of data finally used in the processingand the relative accuracy.Tile geometryof the apparentorbits of the satellitesoverthe SouthPole,for positionsat the Earth's surface below 60øS, is rather exceptional. Satellites avoid the zenith, remain for a long time at very low elevation,and make up for a bad zenithal distribution with a very goodazimuthal layout. The elevation cutoff angle we used is 15ø. This allows us to keep as much data as possibleand to free ourselvesfrom ionosphericand troposphericnoise coming from the lowest layers of the atmosphere. Becauseof the length of the baselinesin our network, the accuracyof our positioningis greatly influencedby the orbit accuracy, accordingto a direct proportional law. We have used the IGS precise combined orbits, whichwereavailablesincethe beginningof our analysis. A comparative study demonstratedthat their precision is weaker over the Southern Hemisphere,becauseof the lack of IGS tracking stations. Nevertheless,the precision of these IGS
orbits over the southernmost
areas
MOTION
FROM
GPS DATA
onlyconstraintto a well-determined reference frame is implicitlyprovidedby the fixedIGS orbits.This reference frameis absolute but arbitrarilydefined and changesfrom day to day. In order to usethe coor-
dinatesderivedfromour dailysolutions for geophysical interpretation,we need to transformall ourfree solutionsinto a consistent and well-identified reference
frame(ITRF 97)validforthetimespanofouranalysis. 4.1.
Reference
Stations
An optimal approachto expressour free network in the ITRF is to selecta subsetof referencestations.Our
dailypositionson thesestationsandthe ITRF 97positions are constrainedto be the same. For thesestations at least, this impliesthat we assumea more or lesslinear behavior, as describedby the referenceframe solution
itself (ITRF 97 is limited to linear displacements for all sites). Our referencestations,outsidethe Antarctic continent,have beenchosenaccordingto geographical or geodeticcriteria. The first set of referencestations consistsof three stationsamong the classA or B IGS stations, as described in the ITRF 96 and ITRF 97
quality assessment [AItamimi,1998; Altamimiet al., 1999]:Hobart (Australianplate), Kerguelen(Antarctic plate), and Santiago(SouthAmerica).In Santiago we have to deal with a specificproblem: althoughthe IGS station belongsto the classA stations for position and velocity both in the ITP•F 96 and ITRF 97 classifications,the geodeticquality of the data is poor (a manual cleaningis frequentlyrequired). In addition, this station is located on the South Americantectonic plate, in the vicinity of the Nazca subductionzone. The motion of Santiago,largely influencedby the elastic couplingwith the subductingplate, couldsignificantly differ from its ITRF 97 linear behavior. Using this station as a reference station can thus influence the
final results. Nevertheless,Santiagois the only station on the South Americanpart of the network providing data of acceptablegeodeticquality for sucha longpe-
has been significantlyimprovedfrom 1995 to 1998. All the data havebeenprocessed with fixed (not only con- riod. Hence its inclusion in the set of reference stations. strained)IGS combinedorbits and InternationalEarth The choiceof the reference stations(Hobart,Kerguelen, Rotation Service(IERS) B Earth OrientationParame- and Santiago)providesa very goodgeographic coverters. Since loose constraints have been applied to the
age,withrespectto theglobalgeometry of ournetwork
positions,the referenceframe after the GPS inversionis (seeFigure1). In orderto assess the sensibility ofthe implicitly determinedby one of the IGS orbits,ITRF 93 results to the geometry of the referencestations,we
till the middle of 1996 and ITRF
94 after.
have also tested an alternative set of reference stations
For our GPS inversionwe haveuseda priori position. s with a weakergeometry. The comparison of theresults providedby the ITRF 96, whenthey exist. The weekly shows negligible discrepancies. Therefore wehaveonly solutionscurrently usedin IGS-type globalprocessings usedthe first set, with the threestationsabove.We introducea smoothingeffect in the time series. Since havetestedtwo strategies to realizethe reference flame our intention is to obtain successivepositionsas discon- with thesereferencestations. The first one is based nectedas possiblein time and independentof geodetic on a Kalman filter, and the secondone usesa sevenconstraints,we have chosento processonly daily solu- parameter transformation.
tions(24-hoursessions). 4.2. Kalman Filter Strategy and Baseline 4. Reference
Frame
Adjustments
A solution is proposed by the GLOBK/GLRED In our daily GPS processing,the a priori stationpopackage developed as an interface withthe sitionsare looselyconstrained(100 m). Thereforethe software
BOUIN
AND VIGNY:
ANTARCTIC
PLATE
MOTION
FROM
GPS DATA
28,283
GAMITsoftware, combining globalandregional GPS we first determine transformation parameters by comsolutions usinga Kalmanfilterliterring,1998].It al- paring day by day our free GPS solution with the ITRF lows to combine individualsessions withstationcoordinates overseveral yearsof measurement, with or withoutstochastic parameters, resultingin stationposition andvelocity estimations. In thisstudyweassume that thebehavior of the sitescouldbe affectedby nonlineareffects. We thereforeseekto obtain time series
correspondingsolution. The seven-parametertransformation is then applied to our daily GPS free solutions, in order to realize a consistent reference
entire duration
of our time series.
frame for the
Unlike
most stud-
ies transformingGPS coordinatesets into a reference frame, the different individual solutionsare weighted matrices of the two sets ofindividualdaily positionsfor eachday of our GPS using full variance-covariance processing. Forcing timeseries positions to linearbe- in the leastsquaresprocess.The station positionsthemhaviorswith high constraintscould producenetwork selvesare not exactly constrained,but the sites can be distortions.To get round this problem, two different more or lessweightedwithin the network for the sevenstrategies arepossible.First, we canestimatean aver- parameter estimation. We realize one transformation aged position andvelocityoverthe entiretimeperiod of the GPS free solutionwith the correspondingseven(4years), foreverystationofournetwork, andthenrun parameter set for each individual daily session. Thus a backwardsolution to obtain individual positionsfor we obtain as many solutionsas 24-hour sessions,which every dayof GPSsession. Theseindividualsessions, for are as disconnectedas possiblefrom the ITI•F constant eachdayof the time series,are relativelyuncorre!ated velocities. We use this strategy in a day by day apif weintroducestochasticparameters in the backward proach,estimatingpositionsfor every day of our series, solution for all the nonfiducialsites. When using this without any estimationof velocities. This transformastrategy for serieslongerthan severalyears,the daily tion strategy offerstwo noticeableadvantagesover the positions obtainedon the reference stations(without Kalman filter method. First, the global network geomstochastic parameters)are tightly constrainedto the etry remainsunchanged,as the seven-parametertransreference (ITRF) values,thus unsuitedto our objec- formation doesnot disturb the baselines. Their length is modified by the scalefactor estimation, the global tive,that is, the study of tectohicimplications. The secondway to proceed,usingthe sameKalman orientation and position of the network is changedby filtersoftware,is to transform the free network into the the rotation and translation parameters, but there are ITRF values,estimatingstation positionsonly, with a no other changeswithin the network. It also provides dailysessionstrategy. We obtain individual solutions globaltransformationparameters,thus reflectingthe forevery24-hoursession,without explicit correlation time evolution of the transformed solution with respect betweenseveral successivesessions,even if it does not to the initial reference frame. Second, the result is reflectthe real geodeticbehaviorof the stations.The relatively independentfrom the choice of the fiducial fiducialstation positionswill be closelyconstrainedto stations. It minimizes the constraint effect brought by theITP•F positionsbut with independentvariationsfor weightingon the comparison,which concernsall the eachday. On the nonfiducialstationsthe resultsshould points within the coordinateset. For this study we have selectedthis referenceframe beverysimilar to those of the previousmethod. The maindifferenceis that this strategyproducesusablere- realization strategy, using a daily seven-parameter suitsonthe referencestationstoo, as their positionsare transformationand Hobart, Kerguelen,and Santiagoas notdependentany longeron ITRF values.In both cases referencestations. One couldbasically achievea similar nevertheless, usinga Kalman filter leadsto adjustments type of analysisusingthe GLOBK/GLORG software, onthe baselines,introducingan internal distortionin as described in its latest release. From the daily cothenetwork.Thesetechniqueswill be especiallysensi- ordinate sets transformed into the reference frame, we tivenot only to the choiceof referencesitesbut alsoto obtain new time seriesallowingus to extract linear vedataoutageat the referencestations.For thesereasons locities and short-term behavior for every site. These rates are estimatedfor eachstation, independentlyfrom wehave not used a Kalman filter method but a method the behaviorsat other sites,by using simplelinear reoftransformation, as described now. gression. 4.3. Seven-Parameter
Transformation
CoordinateSet Comparison
and 4.4.
Reference
Frame:
ITRF
97
Anotherstrategyis to transformsetsof coordinates Another relevantquestionis the choiceof the referfromthe GPSanalysis with a seven-parameter trans- enceframe itself, to transformour free solution. The formation. We have used the CATREF 1.1 software IGS orbits werekept fixed all alongour GPS process-
developed by Altamimi[1997] withtheinitialpurposeing and transformationinto the referenceframe. The ofcombining spacegeodetic solutions in positions or implicit referenceframe definedthroughthe IGS orinbothpositions and velocities to produceITRF so- bits is ITRF 93 from 1995 to June 1996, !TRF 94 after lutions. Setsof coordinates fromvarious spatialtech- this date. The a priori station positionsusedas input niques arecompared usinga seven-parameter transfor- in our GPS processingare from ITRF 96. For consisthe logicalchoicefor our reference frame mation, witha leastsquares adjustment.In ouranalysis tencyreasons
28,284
BOUIN AND VIGNY: ANTARCTIC
PLATE MOTION FROM GPS DATA
shouldbe ITRF 96. Nevertheless,the new realization to the IGS corestationtransformation fromITRF93 ITRF 97 bringsa significantimprovementwith respect to ITRF 94 (13 commonstations). All the time in the ITRF 96;after to ITRF 96, leading to the recommendationof the use tiesarethenimplicitlyexpressed
of ITRF 97 insteadof ITRF 96 [Attamimiet al., 1999], thatwecantransform themintotheITI•F 97using transformation. If especiallyas far as the southern part of the IGS net- the CATREF 1.1 seven-parameter
weneglect thispreliminary transformation onthe riodpriorto June1996,it will not change anything on
work is concerned. The IGS network has been locally densifted,the time seriesusedin the estimation of veloc-
ities are longer, the quality of the IGS orbits abovethe the final transformedresults, but the transformation SouthernHemispherehas been improved. This yieldsa parametervalueson this periodwill be affected, since significantimprovementof the positionsand velocities the additionalimplicit transformationwill be absorbed. of the IGS in and around Antarctica for the ITRF 97 In this situation we observea gap in the transforma. the final transformation of our free tion parameterserieswhenchangingfrom ITRF 93to network(ENS 97) hasbeenrealizedinto the ITRF 97. ITRF 94 in the orbit processing. AlthoughITRF 96 and ITRF 97 are globallyaligned solution. Therefore
in terms of datum definition, a nonzerotransformation can occur between subsets of the two ITRF
4.5. Summary
realizations.
This is alsotrue for the subseto• corestationsincluded
Table 1 summarizesthe processingstrategyandpaitself andthe by all the IGS analysis centersin their orbit process- rameters,includingthe GPS processing
ing, even if the transformationparametersare negligible. We have not applied any globaltransformation prior to the seven-parametertransformationusedfor the comparisonof coordinate sets. To proceedrigorously,we shouldfirst define the appropriatetransformation parametersfrom ITI•F 96 to ITRF 97 for our subsetof stations (this stage is done anyway by the CATREF seven-parameter transformation),then apply
transformation
of the free solution into a reference
frame,as explainedabovein details. Most of theGPS data we have analyzed(exceptDumont) are included in the globalIGS networkprocessing. Nevertheless, the utilization of a processingstrategythoroughlyadapted to revealtectonicimplicationsleadsto significantly improvedresultswith respectto global solutions.Asa matter of fact, our strategyfor analyzingthe GPSdata this transformation to our free solution. This additional andtransforming our resultsinto a consistent reference of transformationis thus included in the global transfor- frame has been adaptedto the geophysicalobjective withoutapmation. To take into accountthe effectof changingthe our study:detectinginternaldeformations externala prioriconstraints. Mor• reference framein the IGS orbit processing (ITI•F 93 to plyingunconsidered ITRF 94, in June 1996), we apply an additionalseven- over, the carefulreprocessing of a subsetof stations This parameter transformationto our free solutionduring makesan improvedcleaningof the data possible. all the periods before June 1996. The parametersare is obtainedboth by adaptingautomaticeditionparamextractedfrom Altamimi et el. [1999]and correspond etersto Antarctic conditionsand by the manualsuper-
Table 1. Data Analysis Summary Description
Parameter
GPS Processing GPS software Sessi.ons Orbits
GAMIT
9.1
24 hours
IGS preciseorbits, fixed
Elevation anglecutoff
IEt{S 15 ø
Phase ambiguities
Free (not resolved)
Earth orientation parameters
Troposphericdry delay model Troposphericwet delay A priori station positions Station position constraints Transformation Reference frame
bulletin
B
Saastamoinen
Estimated every 3 hours ITRF
96
Free network approach
Transformation Into a ReferenceFrame Daily ITRF
97
Strategy
Seven-parameter transformation, full variance/covariance matrix
Software Reference stations
3 (Hobart,Kerguelen,and Santiago)
Velocity estimation
Linear fit on the transformeddaily time series
CATREF
1.1
BOUIN AND VIGNY: ANTARCTIC PLATE MOTION FROM GPS DATA
28,285
vision andtheeditionof peculiardata sets.Thisrepro- proximately 8 cm, followedby the rapid return to the cessing leadsto reincorporation of a significant amount mean value in November1997. This probably reflects ofdata(otherwise rejectedby automaticprocedures).the damageof the radomecoveringthe GPS permaTherefore, althoughit may comefrom the sameorig- nent antenna,whichoccurredduring spring1997. This inalrecords,our final data set is cleanerand contains moreinformationthan the globalsetsof solutions.A considerable increaseof the strengthand precisionof oursolutionis finally obtained. 5. Results
radomehas beenreplacedby a new one by the end of theyear 1997,assuggested by the time seriesvariations. The datefor the changeofthis radome,asgivenby IGS log for Casey,is approximately1 month later than the jump observedin the time series. A possiblescenario, according to B. Twilley (personalcommunication, 2000) is the following:the hole in the radome,coveringone
The 4-year-long time series we have obtained third of its surface,let the snow accumulateover the
(ENS97)allowsusto extractvarioustypesofperiodic antennaitselfandproducedthe first changein the ver-
orshort-termvariations. As an example,Figure3 de- tical component;in Novemberthis snowall melted out, picts thethreecomponents (north,east,andellipsoidaland the heightmeasuredreturnedto its meanvaluebeheight) forthepositionoftheAntarcticstationof Casey forethe changeof the radome(December9, 1997). This (variations arein meters).Fromthesedailypositionspoint underlinesthe necessityof workingwith continuwecanevaluatewith an extremely good accuracya lin- ous and densedata in order to measurevery tiny and eartrendfor the two horizontalcomponents.For clarity slowdisplacements, especiallyon the heightcomponent, theformaluncertainties (between0.3 and 1.2 cm) are and the importanceof gettingreliable and preciseinfornotrepresented.At the beginningof 1997 we observe mation about the status of antennas and receivers. This anabruptvariationof the ellipsoidalheightreachingap- information, which is easily obtained insidepopulated areas,is of critical importancefor remote and unaccessible sites like Antarctica. 009
+
+
007 •+,++½++ •:+-•.:
oo•
•
+
+ +
•- + '•
.o o•
ß
+
+
.o o9
1995
011 ........
009
+
007
oos
+
+
+ +
003
ool
+
+
+ +
++ , %, .•+ + +
+
,
'
ßo ol
of those time
seriesshowsa high scatterof the daily positionsduring 1995 (due to the poor relativequality of the IGS orbits overthe SouthernHemisphere),whichreducesprogressivelyuntil 1997 (improvementof the IGS orbits). In 1998 the introduction of new permanent GPS stations in the Antarctic network improvesthe constrainton the continentaldomainand reducesagain the scatterof positions. Linear trends were fitted, in the least squares sense,to the time serieson the whole period 1995-1998.
+ •.-. •+
The examination
5.1.
Dumont
At Dumont
d'Urville
the velocities
have been extracted
from
020
time seriesincludingthe SCAR 1995and 1996campaign data (10 daysin January1995 and 20 daysin January and February1996;seeFigure 4). Closeanalysisof the recentcontinuoustime series(Figure 5) showsan
0OS
anomalousmotion around March 25, 1998. At this date the largestintraplate earthquakeeverrecordedoccurred in the vicinity. The hypocenter has been located near
.oos
ß 005
•
•
+ ++
.
V Land coast. This Mw=8.1 event occurred to the
West of the Antarctic-Australian-Pacifictriple junction
-0 15
• 21•t 0
the Balleny Islands (62.9øS,149.5øE) off the George
1995.5 1996 0 19•.5 1997,0 1•7,5 1998.0 1998,5 1999,0
(250 kin). It seemsto haveruptureda fault, or a series of strike-slipfault segments,nearly 300 km long, with
a quasi-vertical slip [Nettlesel al., 1998]. The strikeof
Figure3. Timeseries for thethreecomponents in the faults, accordingto the aftershocks,is 287ø, nearly Casey, fromthebeginning of1995till theendof1998' perpendicularto the north-southtrendingfossilfracture (top, middle) horizontal components and(bottom) ver- zones,formingthe most visiblebathymetricfeaturesof ticalvariations. The variations of the dailypositionsthisregion.Althoughthisearthquakeis wellstudiedustiedtotheITRF 97 aregivenin meters.Thevertical ingteleseismic dataand aftershock locations,the lackof component showsthe 8-cm excursioncorresponding to
thedamage oftheradome covering theGPSantenna.recordingstationsand of surfacedisplacementdata in Thevertical lineindicates thedateofthechange ofthe the area introducesa 25 km uncertainty in the location radome.
of the fault rupture. Its depth shouldnot exceed30 km
28,286
BOUIN AND VIGNY: ANTARCTIC PLATE MOTION FROM GPS DATA
are12mm/yrand-15mm/yrforeastandnorthcom-
-0 02
-0,04 • -0 06
ponents, respectively. Fortheperiod aftertheevent,
+
weobtain26mm/yrand-10mm/yrforeastandnorth components, respectively. The uncertainty ontherate
aftertheearthquake is largerthantheuncertainty be-
-o lO
fore,asthetime seriesis shorter.The stepbetween the twolineartrendsgivesusthe valueof the instantaneous coseismic displacement asmeasured by GPS.It reaches 13 mm on the eastcomponentand 4 mm on thenorth
-o12 -o 14
-016 -0,18
-o 20
1995.0
1995,5
1996,0
1996.5
1997 0
1997.5
1998 0
1998.5
1999 0
component, lessthan the predictedvaluesfromthedis-
location models estimation (22.4mmand19.4ram)but quite compatiblein orientation. Thereforethe factthat
the measurements exhibit,despiteroughassumptions -0 02
and high uncertainties,a displacementin the samedi-
+
-0 04
rectionandwith a comparable amplitudethanthepre-
-o O8
dicted displacementclearly indicatesthat this is notan artifact. These horizontal coseismicdisplacements have been removed from the horizontal linear velocitiesat
,
•
-o,o8
• -010 •
-012
Dumont. The correctedvelocitiesfor the wholeperiod
-0 14 -o 18
, +
-0 18
-o2•9;;•
,
are15mm/yrontheeastcomponent and-13mm/yron
+
1995,•
1996 0
the north component. This matcheswell with the linear 1996 5
1997 0
1997 5
1998 0
1998.5
1999
ratesfor the periodbeforethe earthquake (12mm/yr and-15 mm/yr ).
Figure 4. Time seriesfor the horizontalcomponents 5.2. Global Plate Motion in Dumont(variationsin meters)fromthe beginningof 1995 till the end of 1998. The SCAR campaigndata The horizontal velocities, for all the stationsin our (January 1995 and January 1996) are included. The network (includingthe stationsoutsidethe Antarctic trendscorrespondto uncorrectedvelocities(seetext).
but is not yet known with better accuracy.The static Coulombfailure stresschangerelated to this eventhas been calculatedby Toda and Stein [2000]usingfour roughmodelsfor the dis!ocationmechanisms. The nearest GPS station to the assumedepicenterof this event is the new permanent station of Dumont, located at 66.4øS,140.0øF,,approximately600 km from the mainshock. The induced coseismicdisplacementat the site of Dumont has been computed using an elastic disloca-
tion codebasedonthe equationsof Okada[1992],which is veryconsistent with the McGuireet al. [1998]model. The predictedcoseismic displacementaccordingto this model is 22.4 mm along the east component,19.4 mm
alongthe northcomponent, with largeuncertainties due to errors on the main rupture location, the seismicmo-
ment, and to the plane approximationinherentin aI1 dislocation codes(seeFigure6). We attemptedto extract the observedcoseismicdisplacementfrom the GPS time seriesin Dumont, for both north and east compo-
nents,and comparethem to the predictedvalues.The date of March 25, 1998, is indicated on the time series for the horizontalcomponentsin Dumont (Figure 5).
-0,02 I
+Before March 25 1998
i
J0After March 25• 1998
-0,04
-0.06
-o,os
-O.lO
-o,12 -o,14
-o.!6 -0.18
I
o
-0.20 199800
-0,02 •-0,04 •+' •,L•
+
o•
oøo •o
o
+,••••o
• o o ;•
o
,
--o
o
-016 [
•fOre March 25, '9•
' •
-018 [ - o' .%;.• ........
j
I OAfter March 25,1998,•• •
...........
•.;,• .......
" •.; • .... • •..•-
We have evaluated linear trends to obtain horizontal
velocitiesfor the periodbeforethe seismicevent(from Figure 5. Time seriesfor the horizontal components in Dumont. The date of March25, 1998,is clearly the beginning of 1995till March25, 1998,thusinclud- indicated,and the trends are deducedbeforeandafter
ingthe 1995and 1996SCARdata)andfor the period the event.The SCARcampaign data (January !995 following theevent(fromMarch25, 1998,till theendof andJanuary!996), although usedfor computing the theyear1998).The observed velocities beforetheevent preseismicvelocities,are not shownhere.
BOUIN AND VIGNY: ANTARCTIC PLATE MOTION FROM GPS DATA
28,287
DORIS), we infera significant disagreement between them. We think that the GPS-determined velocityis rawvelocity corrected velocity
probably more reliable sinceit is more consistentwith
otherGPSvelocities on the continentandwith the plate modeled coseismic rotation
discussed below.
In Kerguelen,anotherplacewith GPS/DORIS collocation, the evolution of the velocities showsa convergence of the ITRF solutions toward our solution. More
precisely,the ITRF 96 velocityon Kerguelenwas very differentfromthe NNR-Nuvel-IA(NNR-A) prediction. The ITRF 97 velocities come closer to both NNI:L-A and
100km
••
m•yr
our solution,whichitselfremainsuntouchedby the shift from ITRF 96 to ITRF 97. Still, both ore'and ITRF 97 solutionsslightlyand consistentlydiffer from NNR-A. The systematic disagreement betweenthe twogeodetic solutions(ENS 97 and ITRF 97) on all the Antarctic sites is related here to the misuse of the data from
this part of the networkin the global IGS processings Figure 6. Locationof the March 25, 1998,8.1 event conductedby most IGS analysiscenters. Becausethe nearBalleny Islands, with focal mechanism. At Dumontthe uncorrectedvelocity(shadedarrow) and cor- IGS global analysisincludesstationscoveringthe enrected velocity(openarrow) are indicated,aswellasthe tire world, with a greater density of sites in the Northcoseismic displacements (solidsegments)deducedfi'om ern Hemisphere(mainly North America and Europe), the GPS time series and predicted by the dislocation an obvious tendency is to neglect the data of poorer model. geodeticquality from the southern part of the network. One of the interestsof our study is preciselyto focuson the only stations on and around Antarctica. It leads, plate),are summarizedin Table 2. The uncertain- with an appropriateprocessingmethod, to resultsof intiesgivenin this table are of two different types. The creasedquality on these stations. 95%confidencelimits on the velocities correspondto On the stations outside the Antarctic plate, the whitenoiseassumption on the station behavior and are agreementwith the NNR-Nuvel-IA predictionsis rather clearlyunderestimated. We also indicate the uncertaingood. The sites showingdiscrepanciesare located near tiescorrespondingto the differenceson the horizontal the plate boundariesand are affected by local tectonic ratesobtainedby using different ways of tying our free networkto a referencesystem, as describedpreviously. They are larger than the formal uncertaintiesbut reTable 2. Station Velocities V, 95% Confidence flectthe precisionwe can expect from our analysisin Limits assumingWhite Noise (or)and Standard a morerealistic way. Figure 7 showsthese horizontal DeviationsCorrespondingto Realistic Estimate velocitiesfor all the stations of the network, with re-
latederrorellipses.Our solutionis comparedwith the ratespredicted by the NNR-Nuvel-IA model[Argusand
of ConfidenceLimits (A), ENS 97 East
North
Gordon, 1991]and the ITRF 97 velocitieswhenexisting.The ITRF 97 solutionincludesIGS solutions,but
alsosatellitelaserranging,dopplerorbitography and radiopositioning integratedby satellite(DORIS), and verylongbaselineinterferometrydata. For all the stations outside the Antarctic tectonic
plate,the agreement 'betweenour solutionand the ITRF97velocities is excellent.This confirms, fromthe geodetic pointofview,theverygoodqualityofoursolution.Forall the stationsinsidethe Antarcticplate,we observe significant discrepancies betweenoursolution andtheITRF 97. The mostimportantdifference is obtained in Dumont.Wemustpointoutthat theITRF 97 solution in Dumontis derivedfrom DORIS data only partlybecause the newpermanent GPS stationon the
French baseisnotyetincluded in theIGSnetwork, as thedataarenotavailable in realtime. In thissituation, withcollocation by twodifferent techniques (GPSand
Station
1/•
cr
6 8 -40 2
+1 4-1 4-1 4-1
DUMI*
19
4-2
4-10
DUM1 c HOB2
15 15
4-2 4-1
4-10 4- 4
I