Anda di halaman 1dari 21

J. Geodynamics Vol. 23. No. 2, pp.

129-149, 1997
Copyright Q 1996 Elscvier Science Ltd
Printed in Great Brimin. All rights rese.cvcd
PII: 80x4-3707(%)ooo2&9 0264-3707197 $17.00+0.00

GLOBAL BATHYMETRY DERIVED FROM ALTIMETER


DATA OF THE ERS-1 GEODETIC MISSION

G. RAMILLIEN and A. CAZENAVE


UMR 5566 (CNES-CNRS), 18AvenueEdouard Belin, 3 1 055 ToulouseCkdex,France

(Received 7 March 1996; accepted in revised form 3 July 1996)

Abstract-A global 2-D spectral inversion of dense geoid data of the ERS-1 Geodetic Mission
is presented to compute the ocean-floor topography in the waveband 15-500 km. The model
accounts for flexural compensation with spatially varying elastic thickness of the lithosphere.
The inverted 2-D bathymetric signal is then completed by wavelengths longer than 500 km
using either the ETOPOJ dataset or a bilinear interpolated grid from direct shiptrack
measurements (NGDC data). For the inversion a global marine geoid grid computed from the
ERS-1 data using the collocation method was used. An analytical transfer function beween 2-D
spectra of geoid and bathymetry is obtained from a simple lithospheric flexure model, which
depends upon several parameters such as mean ocean depth, crustal thickness, elastic thickness,
density distribution of sea water, crust and mantle. An analysis of sensibility of the prediction
filter with respect to these parameters provides a hierarchy between them, the elastic thickness
being the most critical. The elastic thickness vs age variations is taken into account using the
half-space cooling model. It is assumed that the elastic thickness follows an isotherm. mo cases
are considered: 450C and 600C isotherms. Validation of the predicted bathymetry is made
through comparison with long profiles of NGDC original soundings and ETOPOJ data in two
selected areas: Central Pacific and North Atlantic oceans. Results show that the predicted
bathymetry is generally closer to NGDC data than ETOPOJ data, which are smoother, and that
the 600C isotherm case gives lower root-mean square values. Copyright 0 1996 Elsevier
Science Ltd

INTRODUCTION
One of the main characteristics of a planet is its topography. It gives information on several
geophysical processes which act near the surface or at depth. Whereas the continental
topography can be easily measured by direct observations (using airborne photogrammetry or
stereoscopic images from satellites), the remaining 70% of the Earths surface covered by
oceans is unevenly surveyed, the bathymetry being measured by echo sounding along ships
tracks only. Today vast oceanic areas are still unexplored, especially in the southern oceans,
where the seafloor topography of these regions is poorly known.
With the systematic mapping of oceanic areas by satellite altimetry it has become possible to
measure undulations of the mean sea surface, which approaches the marine geoid. At the
shortest wavelengths (C 500 km), the marine geoid reflects the seafloor topography and its
isostatic compensation. Inverting the geoid data (provided that the compensation mechanism is
129
130 G. Ramillien and A. Cazenave

known) allows the prediction of the bathymetry.


Recently the European ERS-1 satellite has completed a Geodetic Mission (GM) whose
objective was to realize a global mapping of the oceanic areas with a cross-track resolution of
8 km at the equator, a value to be compared to previous altimetry missions (e.g. 80 km for
ERS-1 35-day repeat mission, - 150 km for Geosat Exact Repeat Mission, - 250 km for
Topex-Poseidon). Such a dense coverage allows us to compute the high-frequency component
of the seafloor topography.
Bathymetric prediction from satellite altimetry had been performed in several previous
studies (e.g., Lazarewicz and Schwank, 1982; Baudry and Calmant, 1991; Jung and Vogt, 1992;
Calmant, 1994) but the lack of cross-track resolution of available altimeter data limited these
studies to methodological developments. With the declassification in the early 1990s of the
Geosat Geodetic Mission (GM) data south of 3OS, Smith and Sandwell (1994) computed
altimeter-derived topography maps over the southern oceans and demonstrated the considerable
interest of using altimetry for mapping the high-frequency component of the bathymetry.
In this paper, we adapted the Baudry and Calmant (1991) approach to predict the bathymetry
globally, using ERS-1 GM data. To validate our bathymetric prediction, we compared it to in
situ data such as NGDC (National Geophysical Data Center) ship track measurements as well
as to the ETOPO-5 grid.

ERS-I DATA

The remote-sensing ERS-1 satellite, launched by the European Space Agency in July 1991,
conducted a dense altimetry mapping of oceanic areas from April 1994 to March 1995,
providing gravity field measurements with a uniform and high resolution. This geodetic phase
consisted of two successive 16%day cycles. Starting in April 1994, the first cycle provided a
spacing between satellite tracks of 16 km at the equator. In September 1994 this phase was
shifted 8 km west and the second 168-day cycle was completed by mid-March 1995. Data of
both cycles ensure a global coverage of the oceans with a resolution of 8 km at the equator in
both along- and cross-tracks directions. Details concerning the processing of the ERS-1 GM
data can be found in Cazenave et al., (1996). The processed data (i.e. geophysical corrections
applied and residual orbit errors reduced through cross-over analyses) have been further used by
Mazzega ef al., (1996) to compute global high-resolution grids (mesh size of l/16) of mean sea
surface (or geoid) and gravity anomalies through the least-squares collocation method
(Heiskanen and Moritz, 1967). The main advantage of the collocation over a classic
interpolation process is the possibility of taking errors into account (e.g. orbit error,
oceanographic signal, geophysical correction errors, etc.) through covariance functions. The
collocation method also allows the simultaneous computation of geoid and gravity anomalies in
the same inversion.

PREVIOUS STUDIES

Different algorithms have already been proposed to predict seafloor topography from
altimetry measurements. Spectral methods, based on the Fourier decomposition of signals, have
been first applied along 1-D satellite profiles crossing seamounts in the North Pacific (Dixon et
al., 1983; Dixon and Parke, 1983), or over a tiny 2-D zone around a single seamount from
gridded geoid data (Baudry and Calmant, 1991). These authors used an iterative spectral method
assuming all physical characteristics of the lithosphere (in particular lithospheric flexure) are
Global bathymetry of ERS-1 Geodetic Missiondata 131

known and constant in their inversion. Declassified Geosat GM data over the southern oceans
below 30s and SEASAT measurements allowed Smith and Sandwell (1994) to predict
bathymetry in a large oceanic area. They considered the uncompensated waveband (wavelengths
less than 160 km) of the gravity signal to compute the high-frequency component of the
bathymetry, then completed it by the low-frequency component estimated directly from ship
track measurements. Isostatic compensation was determined empirically from shiptrack
bathymetry and gravity profiles.
Recently, Calmant (1994) proposed a new iterative approach based on a least-squares
inversion to compute a seamount topography from combined gridded geoid and gravity data.
This formalism allows in siru depth measurements to be included and provides realistically
constrained bathymetry.

THEORYOF INVERSION
At wavelengths less than about 500 km, the observed geoid is the sum of two signals: the
seafloor topography plus its isostatic compensation effects (e.g. Watts and Daly, 1981). At longer
wavelengths, geoid anomalies may reflect deeper processes, in particular mantle convection.
Thus, the first step is to filter out the geoid grid from wavelengths longer than 500 km. This is
performed by using a 2-D filter based on inverse methods, which allows a smooth cut-off
(Tarantola, 1987). The filtered geoid (high-frequency component) is simply the difference
between the initial gridded geoid and the filter.
The computed bathymetry b, is obtained by completing the short-wavelength geoid-derived
bathymetry b, with wavelengths longer than 500 km (contribution b,). The latter are estimated
from low-pass filtered ETOPO-5 or NGDC grids. So the predicted bathymetry is simply the sum
of two terms:
b&)=b(r)+b,(r). (1)
r is a position vector of a geographical point, located at (x,y) in the plane. In the spatial domain,
the geoid height n(r) is not related linearly to the bathymetry b(r):
n(r)=z(r)**b(r), (2)
where ** denotes the bi-dimensionnal convolution product. In equation (2), the impulse
response z(r) acts as a Greens function. It contains geophysical information on the isostatic
compensation. At short wavelengths (< 500 km), it is assumed that compensation is ensured
through lithospheric Aexure (Fig. 3). It is difficult to handle z(r) directly for computing a 2-D
convolution product. Noise in the input data n(r) can strongly affect the prediction, because it
is amplified by the response function z(r), mainly in the short-wavelength domain, which gives
rise to unrealistic oscillations (Diament, 1985). This problem may be overcome if the 2-D
Fourier transform of equation (2) is considered (Dorman and Lewis, 1970):
iV(k)=Z(k)S(k). (3)
N(k) and B(k) are the 2-D Fourier transforms of n(r) and b(r), respectively. k is the spatial
wavenumber vector, and its norm depends on the directional wavelengths A,, AYalong the x and
y axes, respectively:
k2=l# (4a)
k,,=2~&,, (4b)
132 G. Ramillien and A. Cazenave

The linear reversed equivalent expression of equation (3) gives the Fourier transform of the
bathymetric signal as a function of short geoid undulations:

B(k)=Z-(&V(k). (5)
In this equation, the transfer function Z depends only on the norm of the wavenumber vector
because of radial symmetry in the bi-dimensional Fourier domain (it has the same shape in each
k direction). This function is often defined as the Hankel transform of z(r) (Bracewell, 1978).
The short-wavelength component of the bathymetry b(r) is then:
b(r)=Re[Fl- [Z-(k).N(k)]] (6)
Re and FI- denote real part and inverse Fourier transform, respectively. The predicted filter
Z- is the inverse of an analytical admittance function Z, deduced from the elastic flexure model,
presented in Fig. 1. The model simply consists of three underlying layers: ocean layer, elastic
lithosphere and ductile asthenosphere. Z depends upon several physical parameters of this three-
layer model, i.e. mean ocean depth h,, crustal thickness T,, and density distribution assumed
constant in each layer: pw for the sea water, pC for the oceanic crust, pi for the lithosphere and
pa for the asthenosphere. In the following model, we assume that p~=p~=p m, the mean upper-
mantle density.
Lithospheric flexure is taken into account in the prediction using the filtering flexural function
Q(k) which relates the spectrum B(k) of the topography b(r) to the spectrum W(k) of associated
Moho deflection w(r) (Banks et al., 1977):
W(k)=@(k)dI(k). (7)

Fig. 1. Schematic view of the lithospheric flexure model.


Global bathymetry of ERS-1 Geodetic Missiondata 133

The mechanical equilibrium of elastic thin plates equation yields:

Y (PC- Pw)
Q(k)= - (8)
Dlk14- I

where y is mean gravity, and D, the flexural rigidity, is related to the elastic thickness T,
through:

E.T;
O= 12(1 - 9)
(9)

where E is the Youngs modulus (= lOI N/m) and y is the Poisson ratio (=0.25)
Flexure studies have shown that T, is proportional to the square root of plate age at the time
of loading and coincides with an isotherm in the range 450-6OOC (Watts, 1978; Calmant et al.,
1990; Wessel, 1992). We consider these two isotherm values and further discuss which case
better agrees with direct depth measurements. T, has been computed through the classical
expression for the half-space cooling model (e.g. Carslaw and Jaeger, 1959):

where T is the considered isotherm, T,,, is mantle temperature, K is thermal diffusivity and t is
plate age. We used T,,,= 1250C and K= 1 mm2 s-l. Age t was estimated from the age grid of the
ocean seafloor of Miiller et al. (1996).
At the mid-ocean ridge the young lithosphere is thin and the value of the elastic thickness is
small, so Q(k) tends to a constant value. This case corresponds to the local Airy compensation.
The general expression of the admittance function is (Dorman and Lewis, 1970; Baudry and
Calmant, 1991):

2lrG
Z(k)= -_[GW+@WM41,
31k, (11)

where G is the gravitational constant.


E,(k)=@, - pw).e-kho (124
E,(k)=&,, -pc).e-lk(*o+~) (12b)
The above terms account for the density contrasts at the seafloor and Moho interfaces,
respectively.
Figure 2 presents the predicting filter Z- vs spatial wavenumber k for different elastic
thickness values. Z- acts as a weighting function in the Fourier decomposition of the 2-D geoid
signal. The amplitude of the predicted topography increased when T, decreases. If T, increases,
Z- decreases toward values of the uncompensated case, which corresponds theoretically to an
infinite value for T,. Z- tends exponentially toward infinite values when k increases. In this
case, the inversion becomes unstable at high frequencies, and very short artificial wavelengths
appear in the predicted bathymetry. So the predicting function Z- needs to be stabilized in the
high-frequency waveband.
134 G. Ramillien and A. Cazenave

Method of stabilization
An efficient procedure for stabilizing the inversion consists of multiplying the predicting filter
Z- by a stabilizing function Q(k) which allows the values of Z- in the high frequency domain
to gently tend to zero when a cut-off wavenumber k, is reached. Q(k) is assumed to have a
Wiener filter expression (Forsberg and Solheim, 1988):

Q(k) = [ 1 + (k/k,j41 (13)


and
Z,- (k)=Z-~ (k).Q(k,, (14)
where ZS- is a stabilized version of Z~- .
Considering the mesh size dr of the input geoid anomaly grid, the resolution of the computed
bathymetric grid, A,, must verify A,<2 dr. Here, dr is l/16, so we have chosen A, around
15 km.
An iterative stabilization of the predicting filter could be used (Baudry and Calmant, 1991)
but this approach gives too slow convergent results when large oceanic areas are considered.

Wavelengths

0.06 0.08 0.10 0.12 0.14

Wavenumber (l/km)

Fig. 2. Linear predicting filter Z- vs wavenumber k (or wavelength: upper scale), drawn for different
elastic thicknesses T,. All these curves tend exponentially to infinity when k increases toward important
values. They are also asymptotic to the T,=Inf. curve (case of uncompensation) in the high-frequency
domain.
Global bathymetry of ERS- 1 Geodetic Mission data 135
(a)

1
..-.-.D = 5.1. e23 Nm
- - --D = 2.1, ~24 Nm

-I
0 100 200 300 400 500 600 700
Distance along the profile (km)

-16 1

Tc=2km

5
._
2 -24 _
% Te = 10 km
8

-28 -
Te = 30 km

-30 ~~~.I...I....1.~~~I~~~~~....~~~~~I..~~
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45
Wavelength (l/km)

Fig. 3. (a) Profile of predicted bathymetry crossing the mid-Atlantic ridge from west (35N) to east
(30N), plotted for different values of the flexural rigidity D. The corresponding NGDC profile is also
plotted. (b) Theoretical curve of the sensibility of Z- to various values of T, (or D) vs wavenumber k.
136 G. Ramillien and A. Cazenave

Sensibility analysis
Variations of the physical parameters appearing in the flexural model (T,, T,, h,, p,,, pc, p,,,)
may affect the numerical values of the predicting filter, hence the amplitude of the predicted
bathymetry. Considering an infinitesimal variation dZ developed as a linear combination of
small variations dp, of the parameters pk, the coefficients of the development are the partial
derivatives of Z- with respect to each parameter:

&
az~-
I
U-(k)= aPL(ki. dpk> (15)
c
k= I

The coefficients of the development (or sensibility coefficients) can be computed since the
analytical form of Z is known. They are as follows.

Table 1. Coefficients of sensibility

Range Maximum of
Extremal variation variation*
Parameter Curve Min Max of the parameter (bathymetry)

Flexural rigidity 0 ix 10 I8 1.e26 N.m > 1 x lo9 m (!)

Ocean depth 0 1x10 4000 m 0.4 m

Crustal thickness 0 0.25 6000 m 150m

Water-crust density 0 Infinity 2OOOkgm- r1OOOm


contrast

Crust-mantle density 0 1.2 1OOOkg me3 120m


contrast

* For a IO-cm geoid anomaly.


Global bathymetry of ERS- I Geodetic Mission data 137
138 G. Ramillien and A. Cazenave

w w
Y
% a f a R t-
Global bathymetry of ERS- 1 Geodetic Mission data 139

-0
140 G. Ramillien and A. Cazenave
Wavelengths < 500 km

t,...1,...1....1,,..I.,..I....l,...l..,,I,,,,1
500 1000 1500 2000 2500 3000 3500 4000 4500 5000
-2000 I, 8,. , -1 , , , I 7 , I . IT, 1 I I I
. Predicted
-3000 - bathymetry 0) -

-6000 -

-7000. 1 * 3 3 1 * 1 - 3
500 1000 is00 2000 2500 3000 3500 4000 4500 5000
-2000

-3000

3 -4000
5
8 -5000
CI
-6000 1
-7000 t -- 4 1 * 11 * * * * 11 0 *. 10 I I I I I I c 2 1 - * -3 13 I..
moo 1500 2000 2500 3000 3500 4000 4500 5000
-2000500,.,.,,...,....,...,,..,.,. ~ , , , , , , , , , , , , , .., ,

-3000 _ ETOPO-5
(d) -

-6000 _

-7000 1 j 1 3 1 3 1 3 3 3 I, I B
500 1000 1500 2000 2500 3000 3500 4000 4500 5000

Distance along the NGDC profile (km) b

Fig. 8. (a) Short-wavelength geoid profile (less than 500 km); (b) predicted bathymetric profile using the 600C isotherm model; (c) NGDC profile: and (d) -
ETOFO-5 profile. These profiles are interpolated along a NGDC shiptrack west-east profile crossing the mid-Atlantic ridge at 38N.
Table 2. Root-mean square analysis in the North Atlantic Ocean

Profile length Root-mean square Root-mean square Root-mean square Root-mean square
(number of NGDC of NGDC minus of NGDC minus of NGDC minus of NGDC minus
points) ETOPO-5 (m) predicted predicted predicted
bathymetry using a bathymetry bathymetry
constant D (m) considering the considering the
600C isotherm 600C isotherm
and ETOPO-5 low and NGDC low
frequencies (m) frequencies (m)

11,557 366.00 285.15 266.63 233.47


3613 313.14 213.87 199.8 1 188.55
4079 408.92 292.74 287.98 280.10
7664 465.24 303.67 290.38 32 1.02
2508 577.23 388.24 371.71 402.02
4652 295.68 197.90 190.50 182.36
1890 282.92 224.46 228.5 1 208.53
7797 207.07 174.60 163.79 179.36
2320 346.63 243.17 242.66 237.18
5012 338.38 271.01 266.02 250.27
11,941 329.73 260.39 243.58 220.45
511 316.64 268.03 216.64 141.92
2005 327.04 346.94 285.66 270.12
1890 282.92 214.79 2 19.63 208.53
678 352.48 243.38 245.18 246.19
1803 310.21 194.64 143.76 186.85
10,134 299.12 184.17 183.24 197.56
7797 207.06 195.38 171.51 179.44
10,706 251.50 179.17 167.72 176.53
3169 282.49 195.56 I9 I .84 179.36
7664 465.24 343.14 335.18 321.02
2207 513.02 433.12 404.82 389.1 I
4652 295.68 200.63 197.32 192.36
3169 282.49 195.55 191.84 179.52
3920 402.75 333.09 308.76 300.81
1016 447.18 274.44 5 1.49 274.90
2353 490.80 359.5 I 343.47 346.94
4560 558.29 339.83 335.22 335.23
3911 46 I .73 295.38 288.68 272.37
2013 384.96 238.05 38.61 228.30
1029 328.92 212.59 2 10.46 219.18
2062 500.28 380.14 326.62 325.38
4079 408.9 I 290.19 284.02 280.09
5180 406.1 I 248.06 234.98 232.63
8399 439.05 281.22 276.92 257.55
2894 392.52 230.53 726.60 221.18
2944 392.89 271.81 269.15 266.91
2545 338.73 2 18.98 713.04 206.45
2211 347.82 254.49 243.96 260.79
2442 439.66 311.07 280.73 270.78
3338 413.61 261.27 51.02 278.53
3613 313.13 199.54 191.70 253.76
2073 365.70 261.60 237.81 239.76
2315 349.97 313.41 272.13 247.26
6267 406.04 264.84 257.14 246.90
17,256 316.79 240.27 2 18.94 220.16
2985 373.01 376.02 235.38 247.26
9363 215.31 143.07 201.30 238.07
15,043 554.03 232.88 232.42 231.52
2985 442.08 378.13 350.23 235.83
Mean: 372.16 m Mean: 265.20 m Mean: 249.73 m Mean: 246.01 m
Deviation: 87.56 m Deviation: 64.63 m Deviation: 56.05 m Deviation: 54.36 m
Minimum: Minimum: Minimum: Minimum:
207.06 m 143.07 m 190.50 m 141.92 m
Maximum: Maximum: Maximum: Maximum:
577.23 m 433.12 m 371.71 m 402.02 m
Global bathymetry of ERS- 1 Geodetic Mission data 143

Table 3. Root-mean square analysis in the Middle Pacific Ocean

Profile length Root-mean square Root-mean square Root-mean square Root-mean square
(number of NGDC of NGDC minus of NGDC minus of NGDC minus of NGDC minus
points) ETOPO-5 (m) predicted predicted predicted
bathymetry using a bathymetry bathymetry
constant D (m) considering the considering 600C
600C isotherm isotherm and
and ETOPOJ low NGDC low
frequencies (m) frequencies (m)

2687 334.60 323.76 316.01 319.02


3395 361.90 295.26 295.75 293.45
4282 187.12 166.84 170.52 164.45
1502 356.86 386.30 385.82 380.49
757 366.45 253.40 254.94 249.46
2258 277.55 205.08 204.44 202.97
1160 165.28 103.57 124.35 91.41
8635 244.30 392.84 262.39 274.62
846 94.21 54.07 45.58 45.60
1301 122.46 105.64 122.14 91.14
3067 128.07 95.67 100.61 84.20
2810 149.33 165.75 167.88 156.64
3045 196.57 158.63 160.77 157.18
6415 158.04 223.66 125.16 125.15
1810 188.74 154.88 160.41 151.66
7577 144.92 142.34 152.03 141.24
3534 170.04 131.49 130.85 129.74
6272 163.82 137.26 139.13 134.97
3261 170.75 130.46 133.34 132.14
462 1 211.59 170.73 174.52 136.22
4279 242.09 217.04 212.30 213.99
1395 162.93 119.18 129.43 122.76
26,553 270.27 209.57 197.50 184.55
3075 168.02 130.97 129.72 129.24
1725 245.26 220.95 217.85 217.15
5207 165.74 136.54 134.63 134.11
5271 233.48 223.87 213.11 206.9 1
4543 252.5 1 285.01 279.21 272.99
5966 335.29 312.00 311.34 307.02
19,536 242.16 195.70 172.60 171.86
7785 247.37 233.65 237.03 235.98
3421 296.40 273.36 262.06 258.40
2539 226.50 195.74 190.93 186.95
1198 395.84 347.03 347.46 346.38
2422 256.67 227.32 221.74 220.28
7187 248.19 227.50 229.62 227.42
1470 185.52 136.78 136.69 133.49
2503 160.38 121.38 117.86 117.41
2510 170.31 130.52 129.90 128.15
4235 175.59 160.12 157.25 156.13

Mean: 221.83 m Mean: 197.54 m Mean: 191.37 m Mean: 185.82 m


Deviation: 73.43 m Deviation: 80.04 m Deviation: 73.02 m Deviation: 75.83 m
Minimum: 94.21 m Minimum: 54.07 m Minimum: 45.58 m Minimum: 45.60 m
Maximum: Maximum: Maximum: Maximum:
395.84 m 392.84 m 385.82 m 380.49 m
144 G. Ramillien and A. Cazenave

Flexural rigidity (D)

&fg-Jd3
Z(k)= E*(k) r&k).
c w
(164

Ocean depth (h,)

g(k)=- IklZ(k). (16b)


0
Crustal thickness (T,)

g(k)=- qQJ(k)E*(k). (16~)


c

Water-crust density contrast (p, - p,)

(164

Comparison on 50 NGDC profiles in the North Atlantic Ocean

Prediction using a constant D ( = 1 .e25 N.m ) ~


0 Prediction using the 450 C isotherm
0 Prediction using the 600 C isotherm

400 1 /

_ .._i~ii_il - L-_-_ /pi j _/ .~ L -~ ~__ L___. _..-L-_ 1 I 1

100 150 200 250 300 350 400 450 500 550 600
RMS of NGDC minus ETOP05 (meters)

Fig. 9. Root-mean square (r.m.s.) of NGDC minus predicted bathymetry plotted vs r.m.s. of NGDC minus
ETOPO-5 data, for 50 profiles in the North Atlantic.
Global bathymetry of ERS-1 Geodetic Missiondata 145

Crust-mantle density contrast (p,,,- p,)

Assuming n(r) as a pure frequency geoid anomaly of amplitude n, and wavenumber k,:

n(r) =no cos (kor), (17)

the variation of a given parameter (the others being assumed constant), produces a bias:

(18)

with

N(k)=n,.tT(k=k,,) (19)

where S is the Dirac function.


So the maximum bias related to a single parameter p has a simple expression:

Comparison on 40 profiles in the Central Pacific Ocean

+ Predictionusing a constant D ( = 1~25 N.m )


0 Prediiion usingthe 450 C isotherm
l Predictionusingthe 600 C isotherm

RMS of NGDC minus ETOP05 (meters)


Fig. 10. Root-mean square of NGDC minus predicted bathymetry plotted vs r.m.s. of NGDC minus
ETOPO-5 data, for 40 profiles in the Central Pacific.
146 G. Ramillien and A. Cazenave

_I!. "5
db,,, = - no

iz2 1
dP k=b
dp.

Figure 3(a) shows different estimates of the bathymetry for different values of the flexural
rigidity D (or elastic thickness 7J. Curves of sensibility to D as a function of k are shown in Fig.
3(b). Both figures show that the sensibility to D decreases when T, (or D) and k increase.
Variations of D do not affect strongly the short-wavelength geoid undulations.
Table 1 presents a synthetic comparison of the different sensibility coefficients as well as the
range of variation they cause in the prediction. The flexural rigidity D appears as the most
critical parameter because its spatial variations yield the largest bathymetry amplitudes. Because
of their exponential increase in the high frequency domain, values of the water-crust density
contrast affect the short-wavelength undulations of the prediction too. The other parameters do
not give important errors, their coefficients dropping gently toward zero as k increases.

RESULTS AND VALIDATION

The whole oceanic domain has been divided into adjacent 5 x 5 cells. Then spectral
inversion has been performed in each cell with a size mesh of l/16 x l/16. In each cell, a mean
age value has been estimated from the age grid and T, has been computed using equation (10).
The mean depth h,, for each cell has been estimated from the ETOPO-5 database. Other
parameters T,, pw, pc, and p,,, have been kept constant over the whole oceanic domain (with
T,=7000 m, p,=lOOO kg rnm3, p,=2700 kg mm3 and p,,,=3350 kg me3). Thus, from one cell to
another only T, and h, vary. In order to avoid edge effects (Gibbs phenomena), an overlap of
0.5 has been considered between adjacent cells, and an apodization procedure was applied at
the edges of each cell using a cosine function, allowing the signal to tend softly to zero. It was
sometimes necessary to remove a low-degree polynomial surface in each cell to avoid
diffraction effects (very short instabilities) in the Fourier decomposition of the geoid signal.
A discrete 2-D Fast Fourier Transform (FFI) algorithm (Cooley and Tuckey, 1965) was used
to compute the harmonic decompositions of b(r) and n(r). The long-wavelength component of
the topography was estimated from the ETOPO-5 database.
Bathymetric prediction has been performed for two cases: T, following the 450C [case (a)]
or 600C [case (b)] isotherm. Sediment effects on the predicted bathymetry were neglected in
this global inversion. Indeed, they could not be adequately modelled using a linear filter. This
problem will need further investigation. A global map of the predicted bathymetry [case (b)] is
presented in Fig. 4.
Validation of the predicted bathymetry has been made by comparing with direct shiptrack
measurements. We used for this purpose the NGDC database. Our predicted bathymetry has also
been compared to ETOPO-5. Two different oceanic zones have been selected in the North
Atlantic ocean (lON-50N, 6OW-2OW) and in the Central Pacific ocean (O+OS, 13OW-
9OW). NGDC shiptrack coverage is presented in Fig. 5(a) and (b) for each area. The
l/16 x l/16 computed bathymetry and ETOPG-5 bathymetry grids are presented on in Figs
6(a) and (b), and 7(a) and (b), respectively, over these two areas.

NGDC data
Shiptrack bathymetry is commonly measured through two-way travel time of an acoustic
pulse emitted by an onboard echo-sonar system with a good sampling resolution (often less than
Global bathymetry of ERS-1 Geodetic Mission data 147

1 km). National Geophysical Data Center has recently released its bathymetric data bank on CD-
ROM (National Geophysical Data Center, 1988a, b), where the original soundings are gathered
per oceanographic institution. Most of these data are two-way travel time records before the
mid-1980s, whereas some of them are more recent measurements based on multibeam swath
mapping.
Errors may affect original sounding data in several ways. Partial absence of measurements on
a profile, which corresponds to a flat signal. Travel time errors can also occur because sound
velocity in water is not accurately known, since it depends upon water salinity. Old data are
based on celestial navigation (southern oceans) and do not benefit from recent and more precise
satellite navigation systems (e.g. GPS).
Coverage of the NGDC shiptracks is not homogenous [see Figs 5(a) and 5(b)]. The spatial
distribution of measurements is important near coastlines, islands and areas of geophysical
interest such as mid-oceanic ridges and major fracture zones. Strategic areas and important
overseas trading roads are also well covered, but there is a dramatic lack of depth measurements
in several oceanic areas [i.e. South Pacific, Fig. 5(b)].

ETOPO-5 grid
This dataset was obtained by digitizing contours of the GEBCO charts (e.g. Canadian
Hydrographic Service, 1981). ETOPO-5 is quite smoothed inside several oceanic areas giving
undetailed maps of the seafloor (Smith, 1993). These areas correlate to unexplored NGDC
zones. Step-like depth variations, multiples of 500 and 1000 m, are sometimes observed, which
are artifacts of the contour digitization (Smith, 1993). By comparing ETOPOJ data and NGDC
original soundings, large biases are noticed: important differences in depth estimation may exist,
frequent shifts of marine features being observed in ETOPO-5.

Comparisons between the ERS-I -derived bathymetry and NGDC and ETOPOJ data
There is a good agreement between modelled bathymeuy and original shiptrack data. Such an
example is shown along an NGDC profile, crossing the mid-oceanic ridge at 38N, in the
Atlantic Ocean (Fig. 8). Because NGDC profiles contain high-frequency signal (the sampling
resolution is generally smaller than l/16), they show more details of the seafloor. The ETOPO-
5 corresponding profile proves to be smoother.
In order to validate efficiently our bathymetric prediction, a statistical analysis over a large set
of profiles has been processed in each selected oceanic area: 50 profiles in the North Atlantic and
40 profiles in the Central Pacific. In the latter area, the selected profiles cross the mid-oceanic
ridge in the average east-west direction, which gives a representative spatial variation of the
parameter D (or elastic thickness T,) with age. It was necessary to remove the mean depth of
each profile before comparison to avoid bias at low frequency in the root-mean square (cm.&)
estimate between NGDC and ETOPO-5 or predicted bathymetry. Statistics of these comparisons
are reported in Table 2 for the North Atlantic Ocean and Table 3 for the Central Pacific.
In general, the predicted bathymetry is closer to NGDC data (mean r.m.s. of 200 and 270 m
for Pacific and Atlantic zones, respectively) than ETOPO-5 (r.m.s. are in the range 100-500 m)
(Figs 9 and 10).
Because the Pacific Ocean presents slower variations of D with age, the predicted bathymetry
is more sensible to the choice of the isotherm value. Statistics for the two cases, 450C and
600C isotherms, are also reported in Table 3. Better agreement is observed with the 600C
isotherm model (191 m r.m.s. for the 450C model, and 185 m r.m.s. for the 600C model).
148 G. Ramillien and A. Cazenave

Comparisons with ETOPO-5 data indicate frequent n&location of topographic features. In


some cases, the smooth ETOPO-5 does not detect any structure whereas satellite data reveal the
presence of volcanoes. To avoid the effect of these kinds of mismatch which can generate
important r.m.s., it is necessary to ignore profile portions having dramatic lack of details.
Geographical shifts can be corrected empirically, but this operation proves to be difficult for
long profiles because the shift is not constant.
Because of the lack of precision of ETOPO-5 data, even at long wavelengths, we also
completed the predicted short-wavelength bathymetry by the low-frequency (>%I km)
component of NGDC data. For this purpose, we interpolated NGDC profiles over the well-
covered test area (Atlantic Ocean) using a bilinear gridding algorithm (Smith and Wessel, 1990)
and extracted the low-frequency signal (~500 km) (Table 2).

Comparison of the predicted bathymetty with the Smith and Sandwell results
Smith and Sandwell (1994) also considered a spectral method to compute the bathymetry in
the southern oceans, using gravity data derived from GEOSAT GM. Their method is somewhat
different from ours, as discussed in the section on Previous Studies.
We compared our predicted bathymetry to Smith and Sandwells one over the South Pacific
Ocean (6OS~OS 16OW-14OW). The r.m.s. difference between the two grids is about
100 m. In the area of comparison, discrepancies appear essentially around high topography
features, and may result from the difference in modelling the isostatic compensation.

CONCLUSION

Detailed maps of the seafloor topography have been computed from altimetry data taking into
account the spatial variations of the elastic thickness of the oceanic lithosphere. Behaviour and
influence of each parameter of the flexure model have been discussed. Comparison with NGDC
shiptrack measurements shows better agreement for the 600C than the 450C isotherm.
Because of bias in mean depths and mislocation of marine features, comparison with ETOPO-5
gives larger r.m.s. values.
The global bathymetry map presented in this paper may be used for various applications, in
particular for studies in marine geophysics.

REFERENCES

Banks R. J., Parker R. L. and Huestis S. P. (1977) Isostatic compensation on a continental scale:
local versus regional mechanisms. Geophys. J. R. Astl: Sot. 51,431-452.
Baudry N. and Calmant S. (1991) 3-D Modelling of seamount topography from satellite
altimetry. Geophys. Res. Lett. 18, 1143-l 146.
Bracewell R. (1978) The Fourier Transform and its Applications, 2nd edn. McGraw-Hill, New
York.
Calmant S. (1994) Seamount topography of least-squares inversion of altimetric geoid heights
and shipbome profiles of bathymetry and/or gravity anomalies. Geophys. J. Znt. 119,
428-452.
Calmant S., Francheteau J. and Cazenave A. (1990) Elastic layer thickening with age of oceanic
lithosphere: a tool for prediction of the age of volcanoes or oceanic crust. Geophys. J. Int.
loo, 59-67.
Canadian Hydrographic Service (1981) General Bathymetric Chart of the Oceans (GEBCO),
Global bathymetry of ERS-1 Geodetic Missiondata 149

5th edn. Hydrographic Chart Distribution Office, Ottawa, Canada.


Carslaw H. S. and Jaeger J. C. (1959) Conduction of Heat in Solids. Oxford Science
Publications, Oxford.
Cazenave A., Schaeffer I., BergC M., Brossier C., Dorninh K. and Gennero M. C. (1996) High
resolution mean sea surface computed with altimeter data of ERS-1 (Geodetic Mission) and
Topex-Poseidon. Geophys. J. Int. 125,696794.
Cooley J. W. and Tukey W. (1965) An algorithm for the machine computation of complex
Fourier series. Math. Comput. 19, 297-301.
Diament M. (1985) Influence of method data analysis on admittance computation. Annales
Geophysicae, 3,785-792.
Dixon T. H., Naraghi M., McNutt M. K. and Smith S. M. (1983) Bathymetric prediction from
Seasat altimeter data. J. geophys. Rex 88, 1563-157 1.
Dixon T. H. and Parke M. E. (1983) Bathymetry estimates in the southern oceans from Seasat
altimetry. Nature 304.
Dorman L. M. and Lewis B. T. R. (1970) Experimental isostasy, 1, Theory of the determination
of the Earths isostatic response to a concentrated load. J. geophys. Res. 75,3357-3365.
Forsberg R. and Solheim D. (1988) Performance of FFT methods in local gravity field
modelling. In: Chapman Conference on Progress in the Determination of the Earths Gravity
Field, 1OO- 103.
Heiskanen W. A. and Moritz H. (1967) Physical Geodesy. W. H. Freeman, London.
Jung W. Y. and Vogt R. (1992) Predicting bathymetry from Geosat-ERM and shipbome profiles
in the South Atlantic Ocean. Tectonophysics 210,235-253.
Lazarewicz A. P. and Schwank D. C. (1982) Detection of uncharted seamounts using satellite
altimetry. Geophys. Res. Lett. 9, 385-388.
Mazzega P., BergC M., Cazenave A. and Schaeffer, P. (1996) Maps of the mean sea surface and
corresponding gravity anomalies from ERS-1 Geodetic Mission. J. geophys. Res. (sub-
mitted).
Muller R. D., Roest W., Royer J. Y., Gahagan L. and Sclater J. G. (1996) A digital age map of
the ocean floor. J. geophys. Res. (in press).
National Geophysical Data Center (1988a) ETOPO-5 bathymetry/topography data. Data
Announc. 88-MGG-02 Nat. Oceanic and Atmos. Admin., U.S. Dept Commer., Boulder,
Colorado.
National Geophysical Data Center (1988b) GEODAS CD-ROM Worldwide Marine Geophysical
Data, updated 2nd edn. Data Announc. 93-MGG-04, Nat. Oceanic and Atmos. Admin., U.S.
Dept Commer., Boulder, Colorado.
Smith W. H. F. (1993) On the accuracy of digital bathymetric data. J. geophys. Res. 98,
9591-9603.
Smith W. H. F. and Sandwell T. (1994) Bathymetric prediction from dense satellite altimetry and
sparse shipbome bathymetry. J. geophys. Res. 99,2 1, 803-2 1, 824.
Smith W. H. F. and Wessel P. (1990) Gridding with continuous curvature splines in tension.
Geophysics 55,293-305.
&rant& A. (1987) Inverse problem Theory: Methods for Data Fitting and Model Parameter
Estimation. Elsevier, Amsterdam.
Watts A. B. (1978) An analysis of isostasy in the worlds oceans, 1, Hawaiian-Emperor
Seamount chain. J. geophys. Res. 83,5989-6004.
Watts A. B. and Daly S. F. (1981) Long wavelength gravity and topography. A. Rev. Earth
Planet. Sci. 9,415-488.
Wessel P. (1992) Thermal stresses and the bimodal distribution of elastic thickness estimates of
the oceanic lithosphere. J. geophys. Res. 97, 14, 177-14, 193.

Anda mungkin juga menyukai