GEOPHYSICAL RESEARCH LETTERS, VOL.19,NO.

14,PAGES 1487-1490, luly24,1992

PRECISEDETERMINATION OF EARTH'S CENTER OF MASS USING MEASUREMENTS FROM THE GLOBAL POSITIONING SYSTEM

Yvonne Vigue, Stephen M. Lichten, Geoffrey Blewitt, Michael B. Heflin, and Rajendra P. Malla

Jet Propulsion Laboratory, California Institute ofTechnology

Abstract.GlobalPositioning System (GPS)datafroma ranging (SLR)andverylongbaseline interferometry (VLBI) worldwide geodetic experiment were collected duringa 3 data.TheSV5 scale andorigin arefromSLR, andtheorigin week period earlyin 1991. We estimated geocentric station ischosen to be thegeocenter. coordinates using theGPSdata,thus defining a dynamically SinceGPS satellites movein high-Earth orbits,they are determined reference frameorigin whichshould coincide with relatively insensitive to errors in thegravity field,particularly theEarthcenterof mass,or geocenter. The 3-weekGPS avforrelatively short data ares of a fewdays or less [Bertiger et erage geocenter estimates agree to 7-13cmwithgeocenter esal., 1986]. Witheach ground sitetracking 5-6 GPSsatellites timates determined fromsatellite laserranging, a well-estabthegeocenter canbe estimated quickly(in a lished technique. The RMS of daily GPSgeocenter estimates simultaneously, few days or less). Use of GPS with short arcs to quickly were 4 cm for x andy, and30 cm for z. determine thegeoeenter wouldin principle enable timevariaIntroduction

Thesatellites of theGlobalPositioning System (GPS)can beused to precisely measure globalgeodynamical quantities

such as changes in Earth'srotationvector and rotationrate,
crustal motion, and location of the Earth's center of mass.A

recent opportunity to testthese capabilities was provided by the1991G._PS I_ERS (International Earth RotationService) and G._eodynamics experiment (GIG), whichtookplaceJan 22- Feb13 1991. Although only 15 GPSsatellites wereoperational,the GPS constellation will include 24 satellites

tions overa few daysor lesstobe measured for thefirsttime, without complications of re-estimation of thegravity field. In thispaper,we present the firstdecimeter-quality GPS solutionsfor the geocenter which we obtainedfrom 21 daysof data.We discuss notonlyaccuracy by comparing with other independent data,butalsoshort-term (daily) precision of the estimates andthepotential for further improvement. Over100 globally distributed tracking sitesparticipated in theGIG'9! experiment. Only datafrom the 21 RogueGPS receiver sites (Fibre 1 andTable1) wereusedin our analysis. The Rogue receiveris described by Srinivasanet al.

when complete andsubstantial improvement in theaccuracy ofgeodetic measurements is expected in the nextfew years. TheEarth orientation andbaseline results recently published

[1989]. Thesites cover +79to -35latitude, butonly4 of

these 21 sites werelocated in thesouthern hemisphere. Data from Jan 31 and Feb 13 were not includedin this analysis since an inputerrorin antenna heightwas discovered. Data fromsatellite PRN6 wasnotused because it wasspinning.
Estimation Strategy

(Herring et al., 199!, Heftinet al, 1992,Lichten et al., 1992, Lindqwister et al., 1992) showthatevensuchlimitedGPS data provided centimeter-level accuracy formeasurements of Earth orientation variations andground station baselines. Thelocation of the Earth's center of mass, or geocenter, is afundamental measurement which has geophysical andpracticalsignificance. The definitionof a terrestrial reference frame requires specification of theorigin, scale, andorientation of a coordinate system. Since themass distribution of the Earth is constantly changing, it is important to measure

The data collectedduringthe GIG'91 experimenthave been processed using theGIPSY (GPSInferredPositioning SYstem) software [Lichten andBorder,1987;Sovers and Border,1990]. A standard estimation technique is usedin thisanalysis, consisting of single-day are solutions.Three
sites are fixed as fiducial sites in the SV5 reference frame, while all other station locationsare estimated. A fiducial network can establish a well-defined terrestrial reference frame

the time variations of thegeocenter relative topoints onthe Earth's crust.Since geodetic measurements areoften made with different techniques and instruments inslightly different terrestrial reference frames, precise measurement of theorigin,scale, andorientation of these frames relativeto onean-

other is needed in order to combine different geodetic data.

with2 or 3 receivers fixedto theirknown coordinates [MaIla and Wu, 1989]. The fixed vectors between fiducialsites define orientation andscale in accordance with SV5 [Murray

adopted for this studyare Oeocenter estimates presented inhere are based ona compar- et al., 1990]. The fiducia!s Troms, Norway; Pinyon, California (USA); andWettzell, ison with the SV5reference frame [Murray etal.,1990] oriare estimated. The satellite orbits are estimated about a

gin. TheSV5 frameis constructed fromboths..atellite l_aser

Germany. Satellite orbits, non-fiducial station locations, and 3 components of thegeocenter offset from thenominal origin

Copyright 1992 by theAmerican Geophysical Union.

Paper number92GL01575

dynamically defined originwhich determines theEarth's

centerof mass,or geocenter. The 3 estimatedgeocenter

parameters represent a translational offset between the origin

of thereference frame andthedynamical origindefined by the
1487

0094-8534/92/92 GL-01575503.00

1488

Vigue etal.' Precise Determination ofGeocenter Using GPS

90i I /FA,

' II.ALL i ..... I __y

.....

40

30

.............................. ..2...1..d.a...y.s: Stations Combined

' '1 '

'

oL JOLDLG''-u _ ou _ SCR / MA

St St) _

10

-30 -d0 -

0-

..... .... .......

. .....
30

a 4o

..... .... :.: ................. .................. I..................... ............... )[ ..... L..:.. .........
.......
35 25

-20
-30
-40

[., geoc X Xmean = -8.3 cm, rms =4.0 crn! /i/ ......... 'g;:%"' ....... '?'"/ii';/i'ii'-::"i'5':':F"//i"Fi'F:'"/:'
20

-90 -180

I -120

i -60

I 0

! 60

I 120

....

I 180

L__

, ,

40

45

Dayof Year(1991)

lo.gitude(deg) Fig. 1' GIG'91 GS Rogue Receiver cations

GPS orbits.Thusgeocentric station locations aredetermined

Fig. 2: GPSGeocenter Estimates - x,y comp.

forx andy components, respectively; these values represent

thedifference between theGPSsolution for thegeocenter and the nominalvalues,which are basedon yearsof SLR data andarebelieved to be accurate to better than5 cm [Eanes et at., 1990]. Thedailyrmsscatter about themean for thex and transmitter/receiver clocks were also estimated. Details on the y components of the geocenter are 4.0 cm and 4.4 cm, reestimation strategy aregiven byLichten [1990a, 1990b]. spectively. This daily rms scatteris a measureof internal An alternativeno-fiducial techniquedemonstrated by precision of the technique presented here,andcanbe considHeftineta!. [1992]involves estimating all station coordinates eredan indication of the potential to measure variations in the andsatellite positions withveryweakconstraints (~1 km). geocenter overone day. This defines a rigidnetwork whichhasan ill-conditioned oriThe estimates for thegeocenter z component from3 weeks entation butisnevertheless constrained by thedata tothegeoof GIG'91 data are shownin Figure 3. This component centerthrough the dynamics of the GPS satellites.To define shows largervariations thanthex andy components, butwe geocentric coordinates, thenetwork canbe translated, rotated, do not believethat these results showactualdaily center of andscaled intothereference frame of choice (e.g.,SV5).The mass variations. The 3-week weighted mean offset from 3 estimated translational parameters arethegeocenter offset nominalfor the z component is 7.7 cm, while the dailyrms fromtheorigin of thisframe [B!ewitt etal., 1992]. scatteris 30 cm. It is likely that the estimationof this component is muchmoresensitive to the uneven distribution

forall non-fiducial sites. Earth orientation parameters, 3 solar radiation coefficients persatellite, GPScarrier phase biases, random walk zenithtroposphere delaysfor eachsite, and

Results and Discussion

of fiducial and non-fiducial

stations between the northern and

Figure2 shows daily x andy geocenter component estimates as offsets from the nominal values. The nominal values

usedfor eachcomponent of the geocenter comefrom the globalSV5 station location solutions. The x andy components of thegeocenter exhibitvariations whichappear to be negatively correlated. An erroranalysis predicts thiscovariant behavior, and it is thought to be dueto theasymmetric global distribution of sites. The weightedmeanvalues (3-week weighted GPS averages) are 8.3 cm and 13.4 cm,
TABLE 1:G!G'91 GPS RogueReceiver Sites

southern hemispheres, and to the uneven 1991 GPS constellation. Althoughthe z componentformalerrors are are indeed largerthanthose for thex andy components, the dailyz geocenter formalerrors areaboutone-halfaslarge as the observed rms scatter. Additional systematic error

100 ........... I

'" I

I ......

i"'

--

21 days:Stations Combined X,Y PolarMotion Estimated

Go-Algonquin, Canada '"

NALL-Ny Alend, Norw'ay ..............

CANB-Canberla, Australia PINY-Pinyon, California(USA) FAIR-Fairbanks, Alaska(USA) PGC1 Victoria, Canada GOLD-Goldstone, California (USA) SANT-Santiago, Chile HART-Hartebeesthoek, South Africa SCRI-LaJolla,California (USA) HONE-Honefoss, Norway TROM-Tromso, Norway JPLM-Pasadena, California (USA) WETB-Wettzell, Germany KOKB-Kokee, Hawaii(USA) USUD-Usuda, Japan KOSG-Kootwijk, Netherlands YAR1-Yarragadee, Australia
MATE-Matera,Italy MADR-Madrid, Spain YELL-Ye!lowklfife, Canada

-60

..............................

'10020

25

30

35

40

130 4era
45

Dayof Year(1991)
Fiff. 3: GPS Geocenter Estimate- z component

Vigue etal.: Precise Determination of Geocenter Using GPS

contributions for thez component (ascompared to x andy)

!489

non-eclipsing satellites [Schutz et aI., 1990].The 2-dayarc

reduced the variationsand anotherfiduciaI networkslightly increased them. For 3-dayarcs, theresults degraded further.

may bedue totheweak coverage in the southern hemisphere caseyieldedmixed results.One fiducial networkslightly
during the experiment. For instance, Hartebeesthoek, a southern hemisphere site whosedatawere geometrically

important, wasoperating less than 12hours perdayduring

this experiment. Simulations showthatthesigmas for x, y, and z geocenter estimates arenearly thesame when using a full 24-sate!lite constellation and full ground coverage between thenorthern andsouthern hemispheres [MaIlaet al.,

Table2 shows a summary of thegeocenter estimates obtainedusingthe technique presented in this paperas compared withthose in Blewittet aI. [1992]wherea 7-parameter

transformation (3 translations, 3 rotations, anda scale factor) is estimated.Tx, Ty, and Tz shownin Table 2 are the 3

offsets of thegeocenter fromthereference frame 1992]. Thez component also appears to exhibit a disconti- translational origin;the valuesfrom eachtechnique agreevery well in a nuity near February !, which isnotwellunderstood. Many
alternative analysis techniques were testedin an attempt to determine what systematic errors may have causedthe
statistical sense. The differences are much less than the corre-

discrepancy in thez geocenter estimates between thefirstand

second half of the experiment.Omittinga questionable stationor satellite did notchange thispattern. Variations in the estimation strategy were studied in an effortto observe changes in the geocenter estimates. Different fiducial networks yielded little variation in the results,

sponding 1-c formaluncertainties of the estimates. Both techniques useGIG'91 databut useslightlydifferent strategies in obtaining thegeocenter offsets.It is important to note thatBlewittet al. [1992]uses 12 sites collocated with the International TerrestrialReference Frame (ITRF) [IERS, 1991]. Since SV5 is intended to bealigned withITRF, it is notsurprising thattheresults arein close agreement.
GravityFieldSensitivity Analysis

although thedifferent networks we trieddid notresult in a

wide variation of thefiducialnetwork geometry. A sensitivity

analysis showed thata 3 cmfiducial error in all components

(x,y,z) would giveanerror of approximately 9-15cmin the geocenter, although this error would beexpected primarily as
anoverall biaswith smaller day-to-day variations. The"3 cm fiducial error" refers to independent 3-cmerrors in each of x,

y, z components for eachof 3 fiducials, with no error

cancellationassumed.However, results of Blewitt et al.

[1982] inidicate that fiducial errors are more likely tobeatthe

levelof 1-2 cm. Both the fiducial and non-fiducial siteswere

notwell-distributed in z duringGIG'91, thusmagnifying errors in the z-component. With a full constellation of 24 GPS and equal north-south hemisphere coverage, theratio of

The useof GPSto determine the geocenter from the dynamical information in theorbits maybe a powerfultechnique sincethe large numberof GPS satellitesenablesfull geographical coverage of theglobe witha relatively smallnumber of ground stations (10-20) in a dayor less.This enables rapid determination of the geocenter with GPS data. In addition, therelatively highaltitude of GPSsatellites (morethantwice thedistance from thecenterof the Earthas LAGEOS) makes the GPS solutionsmuch lesssensitiveto drag and gravity

fiducial togeocenter error iscloser to !:1, according Malla

andWu [1989].

Polar motion (Earthorientation) parameters wereheld

fixedin anothercase,with only slight degradation in the

mismodeling. To quantify this,we examined the effectof gravityerrors on theGPSgeocenter estimates. Our analysis computes the error which is introduced by estimating the geocenter with GPS data while leavingthe gravity fieldunadjusted. Theeffectof notestimating certain parameters can be quantifiedthrougha sensitivity,or
consider analysis.In this type of error analysis,the filter computes the sensitivity the estimated parameters to those parameters not beingestimated.Let vectorYc contain the

geocenter estimates. We alsoestimated Earth rotation (UT1lyrc) stochastically (every 12hours) producing littlechange. Longer orbitarclengths werealsostudied. In these cases, eclipsing satellites' orbits wereestimated assingle-day arcs to minimize orbit modeling error. Eclipsing satellites aresatelliteswhose orbit orientation is such that the satellites enter a

considered (unestimated) parameters, and X"'; are the computed filterestimates andcovariance (not including effects fromtheconsidered parameters). Thenthesensitivity matrix,
S is defined as

shadowed region wherethesun's radiation effects on the spacecraft arereduced (penumbra) or completely blocked (umbra) by theEarth. In general, eclipsing satellite orbit errors have been found tobesignificantly larger than those for

s = o(x- x)
Oyc

O)

where S is calculated from the measurement partials(see

Bierman [1977]).Thetotal error covariance, including the

Fiducial andSevenParameter Transformation Techniques
GPS estimates of the GPS estimates of the

TABLE 2: Comparison of Geocenter Estimates Between

formal errorscomputed in the filter as well as the error contribution fromunestimated (considered) parameters, is

geocenter offsetusingfiducials

geocenter offset byestimation of a 7 Parameter Transformation where Pcis thea priori covariance for theconsider parameTx = -7.5 cm +/- 2.6 cm

Pcon= P + SPc ST

(2)

Gx=-8.3 cm +/- 2.7 cm

ters. Usually, Pcisa diagonal matrix, buta more realistic representation sometimes utilizesa full matrix in orderto account

Gy= 13.4cm +/- 2.4 cm

Ty = 13.0cm +/- 2.5cm

Tz = -14.8 cm +/- 13.8 cm

C,z = -7.7cm +/- 13.7cm

for correlations betweenconsidered parameters. For our

analysis, the consider parameters are simply thegravity coef-

1490

Vigue etal.' Precise Determinh[ion ofGeocenter Using GPS

Geophys. Res. Lett., 19, No. 9, 853-856, 1992. Eanes, R., B. Schutz, B. Tapley and M. Watkins, Earth
Orientation Results from CSR 89L02, IERS Technical

ficients.A verysimple andsomewhat abbreviated analysis by Bertiger et al. [1986]indicated thattheeffectof gravity errors on GPS orbits determined from less than 24 hours of

International Terrestrial Reference FrameUsingGPS,

datawasanorderof magnitude smaller thanothererrors such

as data noise, fiducial errors,etc. One would infer from this

that other GPS-inferredquantities,suchas the geocenter, would alsobe relativelyunaffected by gravityfield errors. Howeverwe re-analyzed thisusingthe full GEM-T2 covari-

Note 5, p.51, Observatoire de Paris,1990. Heftin, M., W. Bertiger, G. Blewitt, A. Freedman, K.

Hurst,S. Lichten,U. Lindqwister, Y. Vigue, F. Webb, T. Yunck,andJ. Zumberge, GlobalGeodesy UsingGPS WithoutFiducial Sites,Geophys. Res.Lett., 19, 131-134,
1992.

ance[Marshet al., 1990]for Pcwiththeglobal tracking network for GIG'91, in order to better understand the effect of

Herring, T., D. Dong, and R. King, Sub-milliarcsecond

leavingthe gravityfield fixedwhile estimating a neworigin (geocenter). The new calculations showthattheerrorsintroducedin the GPS geocenter estimates from holding the gravity field fixedareonly 0.05 mm, 0.7 mm, and0.5 mm in the

Determination of PolePosition Using Global Positioning

SystemData, Geophys. Res. Lett., 18, pp. 1893-!896,
1991.

IERS, Annual Report for 1990, IERS Central Bureau,

Observatoirede Paris, 1991.

x, y, andz components. These errorsare lessthan1% of other observed errorsand are clearly insignificant in the
overallGPS geocenter errorbudget.
Summaryand Conclusions

Lichten,S., S. Marcus,andJ. Dickey, Sub-DailyResolution

of Earth RotationVariationswith Global Positioning

System Measurements,Geophys.Res. Lett., 19, 537540, 1992.

Lichten, S., Towards GPS Orbit Accuracy of Tens of

Centimeters, Geophys. Res.Lett., 17, 215-218,1990a.

Lichten,S., Estimationand Filteringfor High-Precision GPS PositioningApplications,Manus. Geod., 15, 159-176,
1990b.

The mean GPS geocenter estimatefrom a 3-week 1991 experiment agreesto 7-13 cm in each componentwhen comparedwith an SLR (SVS) reference measurement. Individualsingle-day GPSsolutions exhibited rmsvariations of 4 cm in x andy and30 cm in z. Geocenter estimates made with a standard fiducial strategy and with a fiducial-free7parametertransformation producesimilar answers, with formal errorsof 2-3 cm in x and y, and 15 cm in z. The

Lichten, S., and J. Border, Strategiesfor High Precision Global Positioning System Orbit Determinatin, J. Geophys. Res., 92, 12751-12762,1987. Lindqwister, U., A. Freedman, and G. B lewitt, Daily
Estimates
1992.

of the Earth's Pole Position with the Global

PositioningSystem,Geophys.Res. Lett., 19, 845-848,

Malla, R., and S. Wu, GPS Inferred Geocentric Reference

incomplete GPS conste!lation and very limited southern hemisphere ground dataarebelieved to be factors limiting the precision andaccuracy of dailyestimates fromthe199! data
set. With satellite launches in the near future, the GPS

Frame for Satellite Positioning and Navigation, Bull.

Geod., 63, 263-279, 1989. Malla, R., S. Wu, and S. Lichten, Geodetic ReferenceFrame

constellation will grow by about50%. Additionalsouthern hemisphere data shouldimprovethe z component of the

for Geodynamics, J. Geophys. Res., Solid Earth: ModellingCrustalDeformation Special Section, In Press,
1992.

geocenter particularly. GPS geocenter accuracy on a daily

basiswith a full constellation and uniformglobalnetworkis

Marsh, J., F. Lerch, B. Putney, T. Felsentreger,B. Sanchez, S. Klosko, G. Patel, J. Robbins, R.

expected to be betterthan5 cm; incorporation of datafrom low earthorbiters suchas TOPEX is expected to strengthen thesolution even more[Mallaet al., 1992].Further GPSand SLR intercomparisons couldpotentially uncover systematic errors thatmayexistin eithertechnique.

Williamson,T. Engelis,W. Eddy, N. Chandler, D. Chinn, S. Kapoor, K. Rachlin, L. Braatz, andE. Pavlis, The GEM-T2 Gravitational Model,J. Geophys. Res., 95 (B13), 22043-22071,1990.

Murray, M., R. King,andP. Morgan, SV5:A Terrestrial

Reference Frame forMonitoring Crustal Deformation with the GlobalPositioning System, (Abstract),EOS Trans,
AGU, 71, p. 1274, 1990.

Acknowledgements. The work described in thispaper wascarried outby theJetPropulsion Laboratory, California Institute of Technology, under contract with theNational Aeronautics andSpace Administration. We thank themany

Schutz, B., C. Ho, P. Abusali, andB. Tapley,CASAUN0 GPS Orbit and BaselineExperiments, Geophys.Res.
Lett., 17, 643-646, 1990.

participants whohelped make GIG'91a success, andYoaz

Bar-Sever and Ron Muellerschoen, who contributed to

Sovers, O., andJ. Border,Observation ModelandParameter Partials for the JPL GeodeticGPS Modelling Software

"GPSOMC",JPL Pub. 87-21, Rev. 2, Jet Propulsion

Laboratory, Pasadena, CA, 1990.

special software used in thiswork.

References

Srinivasan, J., T. Meehan,and L. Young,Codeand

Codeless Ionospheric Measurements withNASA'sRogue

GPS Receiver, Proceedings Inst.ofNavigation GPS-89

Conference, !989, 451-454 (Inst. of Navigation).
Geoff Blewitt, Michael Heflin, Steve Lichten, Rajju Malla, andYvonne Vigue,JetPropulsion Laboratory, Mail Stop.
238-600, 4800OakGrove Drive,Pasadena, CA 91109.
(Received March10,1992; accepted May 8, 1992.)

Bertiger, W., S. Wu, J. Border, S. Lichten, B. Williams, and J. Wu, High Precision GPS Orbit Determination UsingMarch1985 Demonstration Data,AIAA 24th Aerospace Sciences Meeting, paper AIAA-86-0089, 1986.
Bierman, G., Factorization Methods for Discrete Sequential Estimation, V128, Academic Press, New York, 1977. Blewitt, G., M. Heflin, F. Webb, U. Lindqwister and R. Malla,Global Coordinates withCentimeter Accuracy in the