Professional Documents
Culture Documents
Self-Consistent
Statistical
Modelsfor the GravityAnomaly,
Vertical Deflections,and Undulation of the Geoid
STANLEY K. JORDAN
Mathematical relationships exist between the anomaly, deflections, and undulation that
constrain the models for the autocorrelation and cross-correlationfunctions of these processes.
In previouswork, constraintsbetween the anomaly and deflectionshave been recognized,but
constraints relating to the undulation have been ignored. In this paper, new models are
proposed that take these constraints into account. One of these new models, the 'third-order
Markov undulation model' is suggestedfor the analysisof inertial navigation systemerrors
causedby uncertaintiesin the gravity field.
During the past severalyears,it has become modelsproposedby Shawet al. and Kasperare
increasinglypopularto treat the anomalyand 'self-consistent,' since the anomaly and deflec-
deflectionsas randomprocesses. For example, tion statistics satisfy compatibility conditions
Nash [1968] used a statistical model of the that are basedon the Vening-Meinesz formulas.
deflectionsto estimatethe inertial navigation However, these models are somewhat incom-
system errors causedby gravity uncertainties. plete,sincethe undulationstatisticsare missing.
Kaula [1957] and Henriksonand Nash [1970] In this paper, the theoretical constraintsbe-
usedstatisticalmodelsfor the anomalyand de- tween the anomaly and undulation statisticsare
flectionsto estimatethe errors in gravimetri- derived.Similarly,the theoreticalconstraints be-
cally computed vertical deflections.Bradley tween the deflection and undulation statistics
[1970] used a statistical model for the deflec- are derived.From thesetheoreticalconstraints,
tionsin the designof a newtechniquefor 'direct 'necessaryconditions'are determinedthat must
recovery'of vertical deflections. Also, statistical be satisfiedin orderfor the anomaly,deflection,
modelsfor geodeticuncertaintiesare required andundulationcorrelationfunctions to be phys-
for optimal prediction,filtering,and smoothing ically reasonable.It is shown that previous
of geodeticmeasurements [Kaula, 1966; Krarup, models fail to satisfy these conditions.New
1969; Moritz, 1970]. statisticalmodelsfor the anomaly,deflections,
In some cases,statistical models for the de- and undulationare proposed.
flectionshavebeenproposedwithout considering In general,statisticalmodelsfor gravity dis-
the associatedanomaly models [Levine and turbancesare classifiedas either globalor local,
Gelb, 1969]. More recently,Shaw et al. [1969] depending on the dimensionsof the area of
recognized that the Vening-Meinesz formulas interest.In the developmentof globalstatistical
provide a constraintbetweenthe anomaly and models,the usual techniquesof spherical-har-
deflections;they used this constraintto derive monic-seriesexpansion'can be used [Meissl,
statistical models for the deflections from theo- 1971]. In this paper, attention is focused on
retical anomalymodels.In particular, Shaw and local disturbances,and the developmentmakes
his co-workersproposedthe 'exponentialanom- use of the 'fiat earth' approximation. This ap-
aly model' and the 'Besselanomaly model' and proximation is reasonable,sincethe correlation
determined the associated deflection models. In distance(1/e point) of the gravity anomalyis
a similar paper, Kasper [1971] proposeda 'sec- small relative to the radius of the earth •[Kaula,
ond-order Markov anomaly model' and deter- 1963]. The approximation is useful becauseit
mined the associated deflection models. The permits closed-formexpressionsfor all the de-
sired correlationfunctions; i.e., seriesexpan-
Copyright ¸ 1972 by the American Geophysical Union. sions are avoided.
3660
STATISTICALI•IODELS FOR TI-IE •EOID 3661
Anotherway to classifystat.is'tical
modelsfor y) = y)
gravity disturbances
is according
to their dimen-
and
sionsin physicalspace.In general,it is desirable
to have three-dimensional
models,sincevaria-
tionsin latitude,longitude,and altitude are all
of interest.The presentpaper is restrictedto
+ y)= -7o y)
zero-mean,stationary,two-dimensional models; In order for the root-mean-square
(rms) values
variations with altitude are not considered. Non- of the deflectionsto be boundedand nonzero,
stationary three-dimensionalstatistical models (4) and (5) imply that
for the anomaly,deflections,
and undulationcan
be obtained from the two-dimensional models
given here by meansof a two-dimensionalPois- < y) < o (7)
son integral [Bellaire, 1971]. at x:y=0
NECESSARYCONDITIONS and
ebg•(x,
y)= E[g(X,
Y)g(X
q-z, Yq-y)](2) qS:v:v
(x, y)= --4'-• •, go
where E is the ensembleexpectationoperator.
It is also well known that the deflections are
ßlog [(x -- u)" q- (y -- v)"]dudv (9)
related to the undulation of the geoid N For this solutionto be bounded,the anomalyacf
[Heiskanen and Moritz, 1967], must satisf.ythe constraint
ON ON
y)= -Oy y)= Ox (s) ff• qSoo(x,
y)dxdy=0 (10)
From (1) and (3) it can be shownthat [Pap- In termsof the analogous thermalproblem,(10)
oulis, 1965] meansthat if a steady-statetemperature distri-
bution existson an infiniteplat.e,the net energy
eb•(x,y) = --(O"/Oy")eb•vN(x,
y) (4) flux to the plate must be zero.In terms of gray-
3662 STANLr•Y K. JOaDAN
1•fo
= go ©_1pCi)oo(p
)Jo(pr)
dp(15) q•2•(r)-- •=(r/D) rl½/ D) (24)
In a later section,four new gravity-disturbance
ß --
modelsare proposedthat satisfy all the neees-
and saw conditions.
Althoughthe main interest in this paper is
with local gravity disturbances,the first two
o©rq•oo(r)
dr=0 (17)on a globalscale,sincetheseconditionsare ap-
necessaryconditions,(7) and (8), are alsovalid
where the primes in (13) and (14) indicate
plied at zero shift. Surprisingly, (10) is also
differentiationby r and the symbol Jo in (15)
valid on a global scale if it is understoodthat
and (16) representsthe zero-orderBesselfunc-
the integrationis over the surfaceof the earth,
tion. In order to satisfy (17), a proposedanom-
rather than over an infinite plane [Hirschorn,
aly acf must dip below the r axis. Since none
1970]. The theoretical constraints between the
of the anomaly models proposed by Kaula
deflectionand undulationstatistics,(4) and (5),
[1959], Hirvonen [1962], Shaw et al. [1969],
are also given by Grafarend [19713].
Vyskocil [1970], and Kasper [1971] satisfy this
requirement, these models must be considered ANOMALY AND DEFLECTION MODELS FROM AN
physically unrealistic. For convenience, the ASSUMED UNDULATION MODEL
models proposed by these authors are noted
As mentionedin the previoussection,one can
below.
construct a theoretically valid gravity-disturb-
Exponential anomaly model [Kaula, 1959; ance model by starting from either the undula-
Shaw et al., 1969]' tion or the anomaly. If the anomaly servesas
the starting point, the associatedundulationacf
r)oo(r)= a•,.e-,/D (18) is given by (9); if the undulation servesas the
Ohio anomaly model [Hirvonen, 196,2]' starting point, the associatedanomaly acf is
given by (6). In either case,the appropriatede-
rkoo(r)
= {-o'/[1 + (r/D)']} (1O) flection autocorrelationfunctionsare given by
(4) and (5).
Besselanomaly model [Shaw et al., 1969]'
From a mathematicalpoint of view, it is
easierto begin with a proposedundulationacf.
STATISTICAL MODELS FOR THE GEOID 3663
•:v•(x,
y)-- --•yy
•:wv(x,
y) (25a)
4w2 f•_• 2\1/2
O '•NN(•I, •2)e--i(•x+•2y)
•1 •2 (32)
•N,(x,y) - --•xx
•N•v(X,
y) (25b) The symmetryconditionsassociatedwith (27)-
(32) are (seethe appendix)
0•
•,(x,y)= --O•O••(x, y) (25c) •(x, y) = •(x, y) (33a•
O•
•(x, y) = -•(x, y) (•)
•e(x,y)= --Oye
•(x, y) (25d)
O•
•.•(x, y) = -•(y, x) (•3c)
•,,(x,y) = --O••(x, y) (25e) •.(x, y) = •(y, x) (•)
The calculation of these deflection correlation The sy•etry conditions(equations26 and
functionsis aided by reco•izing the following 33) greatly reducethe effo• necessary to de-
symmetryconditions(seethe appendix): te•ine all the deflectionand anomaly correla-
tion functions.In particular, only five correla-
•,(x, y) = •e(y, x) (26a) tion functions need tb be calculated expScitly,
for example,•, •, •, .•, and •. All the
•(x, y) = -•(x, y) (•) remainingcorrelationfunctionscanbe obtained
•.•(x, y) = -•(y, •) (•c) from theseby inspection.In spite of thesesim-
plifications,the calculationsinvolvedin (27)-
•.•(x, y) = •.(•, y) (•) (32) are vew tedious.As a checkof thesecal-
culations,it is advisableto determine • two
•..(x, y) = •(y, x) (2•e)
ways, from (27) and (29) and then (inde-
The anomalyacf can be determinedfrom (1) or pendently)from (30).
(6). It remainsto determinethe anomalycross-
correlationfunctions:•, •, •, From the SUMMARY OF FORMULAS•FORTI-IE SPECIALCASE
fiat-earth Stokes'sformula,it followsthat •.v oF AN IsoTgorIc U•DULATION MODEL
can be determined by means of Fourier trans-
forms (appendix) The formulaspresented in the last sectionare
valid for either an isotropicor a nonisotropic
1/2
undulationacf. In practice,an isotropicundula-
•(x, y)= • _ (• + •) tion acf is usually assumed.In this case,the
formulasgivenabovecanbe simplified,as shown
'•NN(•i, •2)e-i(•xx+•2y)
d•l •2 (27) below:
= m•(x, i(•xx+
y)e •2y)
dx dy (28) •e,(x, y) - -- sin 0 cos
-- (1/r)qb•N'
(r)] (35)
The remainingtwo correlationfunctionscan be
determinedfrom (3) and (27), •e•(x,y) = --(cose
0 -- sine (36)
•e(x,y)-- --O-•
•:v(x,
y) (29)
•(r) = •o ••(•):o(•r) (37)
•(x,y)=4w
ig__•o
2 ff©
1/2
ß•(•) = r•(r)Jo(wr) dr (38)
'•N•(w•,w2)e
-•(•+•2•) d(-o1
d(M2 (30)
3664 STANLEY K. JORDAN
r r
undulation model is so called becauseit is par-
ticularly easy to extrapolate aloft. Unfortu-
-- I• r Ko •- 4D•
nately, the time-domainversionsof these auto-
corrdation functionsare not expressiblein terms
of rando•y excited d•erential equations,and
ß[/O(•D)KO(•D)
q-
(52)
this property is often desirable(seethe ne• sec-
tion). The two followingundulation modelsand cos 0 (53)
their associatedanomalymodelsare more useful
4:,:re(x,
y)= x/3-• i q- e-r/z>
becausethey are expressiblein te•s of ran-
domly excitedd•erential equations. y)-- o-.o-N
•:v.(X, %/3•r(1-• •r)e-r/D
sin 0 (54)
1.0
ANOMALY ACF
0.8
0.6
0.4
0,2
SECOND-ORDER MARKOV
UNDULATION
--0.2 I I I I
0 2 4 6 8 10
NORMALIZED SHIFT DISTANCE, r!c g
constant speed ¾. The time-domain deflection The covariancesabsentfrom (66), for example,
and undulation correlation functions are ob- E[z•(t)z3(t)], are all zero.
tainedby makingthe replacement
r -o ¾1tlin Of course, a set of differential equations
the space-domaincorrelationfunctions.For the driven by white noise can also be determined
third-order Markov undulation model, the re- that will generatea random processwith the
sulting time-domain correlation functions are desiredanomalyautocorrelationfunction (47).
givenby However, becauseof the Besselfunctionsthat
appear in (52) and (57), it is not possibleto
4•wv(t)= a•v2(1
-Jr-
fl It[-Jr-•2t•3)e-•'t' (64a) find a set of linear differential equations that
4•(t) = a•"1 + fl {tl)e -•'t' (64b) will generatea random processwith the de-
• •2
4aa(t) a (1 + • It[ __ •2 tae-•'t• (64c) sired anomaly-undulation
and anomaly-deflec-
tion cross-correlationfunctions. Nevertheless, a
,•(t) = [(ana•)/(•3)](1+ • [t[)•te set of equationscan be found that will generate
a randomprocess with correlationfunctionsthat
are approximatelyequalto the desiredcorrela-
wherefl = V/D. The t•e-domain deflection tion functions(equations52, 55, and 56).
processes with these correlationfunctionscan be COMPARISONOF TI-IE CROSS-CORRELATION
representedby the solution to the following FUNCTIONS
randomly excited differential equations if the
identificationsß = z•, 9 = 2•, and N = -- Vz• are The cross-correlationbetween any two ran-
made, domprocessesis a measureof the information
about each processthat is containedin the
• + •z• = •(t) (65a) other. Thus, the cross-correlations
amongthe
•. + •z• = z• anomaly,deflections, and undulationare very
important,sincetheypermitoneto estimateany
• + •z• = •.(t) (65c)
one of these quantitiesfrom another [Kaula,
• + •z• = z• (65a) 1966].In fact, the cross-correlations
canbe used
• + •z• = z• (C•e) to optimally mix anomaly,deflection,and un-
dulation data in order to obtain consistentesti-
where dots indicate differentiation with respect mates of all four quantities.
to time. That is, the solutionto theseequations In order to comparethe various cross-corre-
producesrandomprocesses ,I•, 9, and N having lations,it is convenient to normalizethem and
the correlation properties presented in (64). removethe azimuthalvariations,as shownbe-
The white noiseinputs m_and u• are independent low.
and have spectraldensities4•3(• and 16•'(•.
The initial conditionsfor these differential equa- Deflection-deflection'
tions are zero-mean random variables with the
followingcovariances. [4•,x, y)/(--a•a•sin 20)] (67a)
Undulation-anomaly:
E[z•(t)zx(t)]= 2•a. 2
E[z•(t)z2(t)]= /•a.2 (66a) [rkN•
(r)/ago¾
] (67b)
E[z2(t)z2(t)
] = o',r' Undulation-deflection'
E[za(t)za(t)
] = 81•%'• [4•N•(x,
y)/aga• cos0] (67c)
E[za(t)z4(t)
] = 41•a• Anomaly-deflection:
E[za(t)zs(t)] = 2a•
•[z, (t)z,(t)] = 4• •'
(66b) q•(x,y) (67d)
3ff92 (raft• COS0
E[z4(t)zs(t)]-'
l• These four quantitiesare comparedin Figure 2
3o-92 for the third-order Markov undulation model.
•[z•(t)z•(t)]
- • Note that the undulation-anomaly
correlationis
3668 STANLEYI•. JORDAN
z 1'ø
o i
•-- O.8-
z ANOMALY-DEFLECTION
UNDULATION - ANOMALY
z
UNDULATION- DEFLECTION
O
• 0.4
o 0.2
I
,• 0
0
DEFLECTION - DEFLECTION
•-0.2
N
"• -0.4
0-0.6I
Z
I I I I I ! I I I I
-5 -4 -3 -2 -1 0 1 2 3 4 5
NORMALIZED SHIFT DISTANCE,r/c g
Fig. 2. Normalized cross-correlation functions for the third-order Markov undulation model.
The four curves correspondto (67).
the largest amongthe four and the deflection- conditions' serve as constraints in the selection
deflection correlation is the smallest. Figure 2 of a physicallyreasonablegravity-disturbance
also reveals the rates at which information de- model.Four new gravity-disturbancemodelsare
caysas the shift distanceis increased.Note that examined.The most appealingof theseis the
for small shift distancesthe anomaly provides third-order Markov undulation model. This
better information concerningthe undulation model is both convenientand appropriatefor
than does the deflection,but this situation re- the analysisof the errors in inertial navigation
verses at larger shift distances.Figure 2 also systemsthat are causedby gravity uncertainties.
indicates that the information contained in the In addition, this model is suitablefor usein the
cross-correlationsis negligiblebeyond a normal- optimalprediction,filtering,and smoothing algo-
ized shift distanceof about 5, so that little is rithms developedby Kalman [1960] and others.
gained by includingremote data.
APPENDIX
SUM MARY
Symmetry conditions. The fiat-earth Stokes's
The statistical modelsfor local gravity anom- formula (A1) was used by Shaw e't al. [1969]
alies that have been proposedby Shaw et al. to derive the fiat-earth Vening-Meineszformu-
[1969'] and Kasp'er [1971] do not lead to las.
reasonable undulation autocorrelation functions.
It has been shown that if the anomaly acf is 1
statistically homogeneousand isotropic, then N(x,y)- 2•go
(17) is necessaryfor the undulationacf to be
bounded,and (14) is necessaryfor the anomaly
acf to be boundedand nonzero.These 'necessary ' ff;•,[(x -- g(u, v)(y
u)•'-]- dudv
-- v)•']
(A1)
STATISTICALMODELS FOR TI-IE GEOIO 3669
and an asterisk (•) denotescomplexconjugate. desy, 364 pp., W. H. Freeman, San Francisco,
From (A3), (A5), and (A6), it follows that Calif., 1967.
ß • and • are equal,hencetheir inversetrans- Henrikson, P., and R. A. Nash, Jr., A statistical
analysis of errors in gravimetrically computed
forms are alsoequal vertical deflections,J. Geophys. Res., 75, 4017-
4028, 1970.
•(x, y) = •(x, y) (AS) Hirschorn, R. M., Rotation invariant probability
distributions in geodesy, Rep. TE-41, MIT
Equation27 followsfrom (A3), (A4), and (AS) Meas. Syst. Lab., Cambridge, Mass., June 1970.
and the •version theorem for Fourier trans- Hirvonen, R., On the statistical analysisof gravity
fo•s anomalies, Publ. 37, Isostatic Inst. Int. Ass.
Geod., Helsinki, 1962.
1 Kalman, R. E., A new approach to linear filtering
and prediction problems,J. Basic Eng., $er. D,
y) = 82, 35-45, 1960.
Kasper, J. F., Jr., A second-orderMarkov gravity
ßff:,,,
•o•v(co•,
w,)e
-'(ø'•z+ø"v)
d•ldw,
(A9) anomaly model, J. Geophys. Res., 76, 7844-
7849, 1971.
Kaula, W. M., Accuracyof gravimetrically com-
Equation A8 is one of the desiredsymmetry puted deflectionsof the vertical, Eos Trans.
conditions.From (3) it can be shownthat A GU, 38, 297-305, 1957.
3670 STANLEY K. JORDAN
Kaula, W. M., Statistical and harmonic analysis Nash, R. A., Jr., The estimation and control of
of gravity, J. Geophys.Res., 66, 2401-2421,1959'. terrestrial inertial navigation system errors due
Kaula, W. M., Determination of the earth's gravi- to vertical deflections, IEEE Trans. Automat.
tational field, Rev. Geo,phys. Space Phys., 1, Contr., 13, 329-338, 1968.
507-551, 1963. Papoulis, A., Probability, Random Variables and
Kaula, W. M., Global harmonic and statistical Stoc.hasticProcesses, 583pp., McGraw-Hill,New
analysis of gravimetry, in Gravity Anomalies' York, 1965.
Unsurveyed Areas, Geophys. Monogr. Set., vol. Papoulis,A., Systemsand Transformswith Appli-
9, edited by I-I. Odin, pp. 58-67, AGU, Wash- cationsin Optics,474 pp., McGraw-Hill, New
ington, D.C., 1966. York, 1968.
Krarup, T., A contribution to the mathematical Rose,R. C., and R. A. Nash,Jr., Direct recovery
foundation of physical geodesy, Publ. 44, Dan. of vertical deflectionsusing an inertial navi-
Geod. Inst., Copenhagen, Denmark, 1969. gator,IEEE Trans.Geosci.Electron.GE-10(2),
Levine, S. A., and A. Gelb, Effect of deflections April 1972.
of the vertical on the performance of a terres- Shaw, L., I. Paul, and P. Henrikson, Statistical
trim inertial navigation system, J. Spacecr. models for the vertical deflectionfrom gravity-
Rockets, 6, 978-984, 1969. anomaly models, J. Geophys. Res., 74, 4259-
Meissl, P., A study of covariance functions related 4265, 1969.
to the earth's disturbing potential, Rep. 151, Vyskocil, V., On the covariance and structure
Dep. of Geod.Sci., Ohio State Univ., Columbus, functionsof the anomalousgravity field, Stud.
Ohio, 1971. Geophys.Geod.,14, 174-177,1970.
Moritz, I-I., Least-squaresestimation in physical
geodesy,Rep. 130, Dep. of Geod. Sci., Ohio (ReceivedNovember27, 1971;
State Univ., Columbus, Ohio, 1970. revised April 10, 1972.)