Вы находитесь на странице: 1из 12

Geophysical Prospecting, 2011, 59, 176187 doi: 10.1111/j.1365-2478.2010.00900.

x
Comparing gravity and gravity gradient surveys
Gary J. Barnes

, John M. Lumley, Phill I. Houghton and Richard J. Gleave


ARKeX Ltd, Newton House, Cambridge Business Park, Cowley Road, Cambridge CB4 0WZ, UK
Received November 2009, revision accepted April 2010
ABSTRACT
Noise levels in marine and airborne full tensor gravity gradiometer surveys together
with conventional land, marine and airborne gravity surveys are estimated and anal-
ysed in gridded form, resulting in relations that detail how these different survey
systems can be compared analytically. After defining survey parameters including
line spacing, speed and instrument bandwidth, the relations estimate the noise levels
that result on either grids of gravity (gz) or gravity gradient (Gzz) as a function of the
spatial filtering often applied during geological interpretation. Such comparisons are
believed to be a useful preliminary guide for survey selection and planning.
Key words: Gravity gradiometry, Instrumentation noise, Survey comparisons, Sam-
pling, Survey design.
I NTRODUCTI ON
Interpreters of gravity and gravity gradient data invariably
study gridded forms of survey measurements so that a variety
of filtering can be easily applied to emphasize anomalies of
different sizes and depths. In general, it is known that filtering
to longer wavelengths improves the detectability of anomalies
but sacrifices overall resolution. If data are over-filtered, then
due to the loss of resolution, positional accuracy is compro-
mised and multiple isolated anomalies can even be mistaken
as a single entity. On the other hand; without adequate filter-
ing, the noise level can be too intrusive and mask the signal
completely. The noise level therefore dictates the useable res-
olution for interpretation and consequently ultimately deter-
mines the positional accuracies of any identifiable anomalies.
Knowledge of the expected noise on gridded data is therefore
crucial when planning a survey to ensure that the geological
features of interest can be detected and resolved. To aid this
understanding, we estimate and analyse the noise from land,
marine and airborne gravimeters and gravity gradiometers
showing how, given a set of survey parameters, noise from
the raw acquisition system impinges on the final gridded data.
Once in gridded form, the impact of the noise from the dif-

E-mail: gary.barnes@arkex.com
ferent survey systems can be directly compared showing their
relative benefits as a function of the filtering applied to the
gridded data. The formulae derived here are intended to be
used as a general rough guide for survey planning and instru-
ment selection. Relying on various approximations outlined
below, they should be used appropriately within limits and
only as a precursor to more comprehensive numerical feasi-
bility modelling.
SURVEY DATA NOI SE
To facilitate the analysis for both gradiometer and conven-
tional gravimeter surveys, when viewed in the space domain,
the noise from the measurement system is assumed to be un-
correlated. This can be partly justified by considering the
two domains of interest; the time domain of the measure-
ment system and the space domain for the survey and the
signal. Low-frequency instrument noise appears correlated in
the time domain whereas the signal appears correlated in the
space domain. When survey noise is viewed in the space do-
main, the total noise power is unchanged but the spectrum
becomes whiter (less correlated) since the ordering of the data
is partially randomized.
The changing of domains is generally not enough to ex-
clude low-frequency noise from the space domain and so
176 C
2010 European Association of Geoscientists & Engineers
Comparing gravity and gravity gradient surveys 177
various methods are employed to remove the noise from the
rawdata. Land gravity surveys return to common stations and
repeat measurements allowing one to estimate and remove the
drift variation that occurred in-between. Marine and airborne
surveys achieve similar corrections through the use of tie lines
where drift can be corrected for by minimizing the differ-
ences at the intersection points. The heterodyne nature of the
full tensor gradiometer measurement system eliminates much
of the low-frequency noise in the first place (Metzger 1977).
In well designed surveys, especially those measuring multiple
components, much of the low-frequency noise can be removed
by advanced methods that attempt to model the drift variation
with time (ARKeX Ltd 2008).
The following sections will attempt to estimate the noise
levels on the raw measurements for each survey system under
consideration.
NOI SE PER MEASUREMENT POI NT
Airborne full tensor gradiometer
The full tensor gradiometer consists of three distinct mod-
ules known as gravity gradient instruments each measuring
components of differential curvature. For example, a module
aligned in the horizontal plane would have in-line output
(Gxx Gyy) / 2 and cross-line output Gxy (note that the full
tensor gradiometer actually adopts a tilted coordinate system
where each module axis makes the same angle to the vertical).
By having three orthogonal modules in the full tensor gra-
diometer one can sum the three in-line outputs and the signal
cancels, (Gxx Gyy) +(Gyy Gzz) +(Gzz Gxx) =0. When
constructing this quantity, known as the in-line sum, what is
left forms a useful estimate of the overall full tensor gradiome-
ter noise. It relies on an accurate orientation and calibration
of each module so that the real gravity gradient signals can-
cel exactly; any remaining signal causes over-estimates in the
noise levels. To mitigate this possibility, the survey chosen for
noise analysis was based on its generally low signal levels due
to very little terrain relief in the area. Often, noise analysis is
performed on long survey lines flown straight at high altitude
thus providing ideal conditions and attenuated signal. We,
however, chose to analyse a complete survey to give a better
picture of what to expect during a real situation.
Figure 1 shows actual noise power spectral densities (PSDs)
estimated from a survey consisting of almost 250 lines parti-
tioned into three equally sized groups according to their av-
erage turbulence level. The three series therefore correspond
to the low, medium and high turbulence conditions experi-
enced during the survey. The turbulence measure is defined
as the vertical acceleration power over a 5 seconds window
and can vary considerably along the line. Only lines that met
the quality control criteria where used in this analysis, leav-
ing those with excessive noise levels (often due to sections of
high turbulence) as rejected. Such lines, being out-of-spec, are
generally reflown in a commercial survey. The statistics built
Figure 1 Averaged output channel noise
from full tensor gradiometer survey data
for three levels of turbulence as shown in
the legend quantified as (mean level, me-
dian level). Black line shows an empirical
noise model fitted to the data.
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
178 G. Barnes et al.
Figure 2 Noise level on Gzz (black curve) calculated from an equivalent source model inverting on full tensor gradiometer measurements
replaced with white noise (red curve).
up in Fig. 1 therefore represent the noise levels in a selected
favourable subset of all the lines flown during a survey. The
overall turbulence level of these lines is generally less than the
average survey but representative of a trouble free survey
where the terrain is flat and the weather is fairly benign. The
black line is a model of the noise level derived by fitting all
the data to an empirical relation,
PSD( f ) = w +a

f
c

f
c
b

e
f/ fc

,
(1)
with fitted parameters: w=12, a =2.33, b =8.56, c =0.375,
f
c
= 4 10
3
.
This relation is accurate over the range 1300 mHz where
sufficient data were available to fit the model. Noise at fre-
quencies greater than 300 mHz is normally outside the mea-
surement bandwidth of interest and below1 mHz the unavail-
ability of long enough lines in this survey (having line lengths
up to 65 km) prohibited accurate estimates. Figure 1 shows
that the full tensor gradiometer noise is largely uncorrelated
even in the raw measurements before tie-line levelling or drift
corrections and full tensor processing.
During full tensor processing, all the six measurement chan-
nels of the full tensor gradiometer, each having noise charac-
teristics as shown in Fig. 1, are combined together to produce
an enhanced estimate of the gravity gradient. We find that
this is most successfully performed using equivalent source
techniques that invert the measurements into a single density
model (Dampney 1969; Li 2001). The resulting density model
can be used to predict a variety of gravity gradient and gravity
components either back at the original measurement points,
or projected onto level grids. By forward calculating Gzz back
to the original locations of the survey, one can consider (for
the purposes of this analysis) the full tensor gradiometer as
an equivalent Gzz gradiometer that measured this data series.
To estimate the noise on this equivalent Gzz, the differential
curvature components on a survey are replaced with random
numbers making up pure white noise with 1 E/

Hz power
spectral density. Using this data, the equivalent source inver-
sion is performed in exactly the same way as in the real data
case (a procedure that has been verified to honour genuine ge-
ological signals) to yield an equivalent source density model of
the noise. In this process, noise from different measurements
become averaged together and attenuated on the basis of non-
harmonic variation making the inverted model less affected by
measurement noise. Forward calculating Gzz from this model
and analysing the noise power (Fig. 2) shows that the noise
on the equivalent Gzz at low frequency is seen to be reduced
by a factor 2.1 when compared to the noise level on the in-
dividual measurement channels. At higher frequencies (above
0.1 Hz), the equivalent Gzz noise power falls rapidly due to
the rejection by the equivalent source inversion of physically
impossible high frequency power. Such power is attenuated
by the upward continuation of the potential field from the
Earth to the aircraft.
Figure 3 shows the noise levels on the differential curvature
components and the equivalent source calculated Gzz for the
real data of the survey. This situation is less transparent than
the previous purely numerical analysis since there is a real sig-
nal present in the data. However, by using a noisy survey, ap-
plying the terrain correction directly on the measurements and
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
Comparing gravity and gravity gradient surveys 179
Figure 3 Power spectral densities of terrain corrected and spatially detrended full tensor gradiometer data (top series) and equivalent source
calculated Gzz (blue series).
also spatially detrending each data set with a 6
th
order polyno-
mial function in x and y, the affect of the signal can be reduced
leaving a data set where a lower limit on the noise reduction
factor can be made. With reference to Fig. 3, at frequencies
between 0.040.1 Hz, this factor is approximately 1.8. Below
this range the noise estimate is compromised by the signal
power that dominates the power spectral density and above
this range the factor appears much larger due to the upward
continuation as discussed above. This figure fromthe real data
therefore partially supports the conclusion fromthe numerical
analysis.
Consequently, in the following analysis, we will consider
the full tensor gradiometer as an equivalent instrument mea-
suring Gzz with a noise level consistent with this enhanced
gravity gradient level. Lumping all the channels into an equiv-
alent component obviously ignores one of the key advantages
of the full tensor gradiometer having measurements of the
full tensor but it considerably simplifies the noise compari-
son with the gravity systems. Figure 4 shows the cumulative
noise spectrum for the enhanced Gzz component as a func-
tion of bandwidth estimated from full tensor gradiometer sur-
vey data. Typical bandwidths used in practice vary between
0.20.3 Hz (equivalent to 325217 m wavelengths along the
line at 65 ms
1
) or up to 0.4 Hz (163 m) for highly detailed
surveys with tight line spacing.
Over the measurement bandwidth, the noise from the
full tensor gradiometer is often conveniently approximated
as white recognizing the above factors that combat low-
frequency noise both in the acquisition system and the subse-
quent processing. For a white noise power spectral density, P
over a bandwidth, f
bw
, the rms variation on each measurement
point is given by
n
full tensor gradiometer
= P

f
bw
. (2)
The comparison in Fig. 4 shows that for rough calcula-
tions, a white noise approximation is adequate but tends to
over-estimate the total noise power slightly. In the following
analysis, although the full tensor gradiometer noise is assumed
to have a white spectrum, the total noise power is evaluated
using the more accurate model in equation (1).
Marine full tensor gradiometer
The time domain full tensor gradiometer noise amplitude for
a marine survey is assumed to be the same as in the airborne
case. There will be an equivalent set of noise characteristics but
this time driven by sea-state rather than turbulence. Overall,
preliminary evidence is suggesting that mapping the average
noise statistics (equation (1)) from the airborne to the marine
system is over-estimating the noise but until we have acquired
sufficient marine data to analyse, we will consider this as ac-
ceptable. The biggest difference appears in the space domain,
where due to the differing acquisition speeds (65 m/s versus
2.5 m/s), the marine case has significantly higher resolution
along the lines. Analysis of the marine full tensor gradiometer
noise will be presented elsewhere in due course.
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
180 G. Barnes et al.
Figure 4 Noise as a function of bandwidth for full tensor gradiometers equivalent Gzz measurement. Blue series: full tensor gradiometer noise
model (Fig. 1); Red series: white noise approximation, 10 E/

Hz.
Land gravity
Whilst several field land gravimeters can measure the value of
gravity with accuracy suitable for microgravity surveys (Seigel
1995), for a standard realistic survey constrained by a cost and
time budget, the instruments precision is rarely the limiting
factor. For a carefully acquired and corrected land gravity
survey, a more typical level of rms noise per station is
noi se = 0.05 mGal. (3)
This figure does, however, vary considerably from one sur-
vey to another. Due to the proximity, the terrain correction
can often be the dominant factor in the land gravity accu-
racy. In a survey over an area covered with sand dunes rising
to 100 m, this uncertainty alone was estimated to contribute
0.1 mGal to the total error (Hallinan, Gheit and Reid 2002).
When interpreting data over a large area, it is often the case
that the land gravity measurements derive from an amalga-
mation of surveys of different qualities and vintages making
the noise characteristics change from place to place. Many
errors in legacy data derive from inaccurate height measure-
ments causing 0.3086 mGal per metre errors in the free air
correction and consequently a much larger uncertainty than a
modern GPS assisted survey. Due to the large range of qual-
ities, when comparing the different survey scenarios, we will
split the land gravity performance into three categories; legacy
Table 1 Uncertainties in land gravity surveys
Category Accuracy (mGal)
Legacy 0.5
Terrain limited 0.1
Carefully acquired and corrected 0.05
(pre-GPS), terrain correction limited and a carefully acquired
and corrected modern survey, see Table 1.
Airborne gravity
For moving gravity surveys, the noise does not approximate
well to a flat spectrum because of the rapidly increasing noise
power with frequency. In this case, either available data, or a
suitable theoretical model that is able to predict the noise as a
function of bandwidth must be used. In the case of airborne
gravity, we have found that using a model for GPS derived
acceleration noise used by van Kann (2004) approximates
well to available survey data (Fig. 5). Airborne gravity noise
is analysed more comprehensively for the Sander AIRGrav
system as a function of time domain filtering in Elieff and
Ferguson (2008). Here it is seen that for a repeatedly flown
test line, the GPS limit over-estimates the noise for filtering
times greater than roughly 70 seconds and underestimates
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
Comparing gravity and gravity gradient surveys 181
Figure 5 Model for airborne gravity noise
based on theoretical GPS limit. Data points
extracted from Argyle et al. (2000).
for time constants less than 70 seconds. For the purposes of
this paper, we adopt the GPS model as a convenient approx-
imation for a general airborne gravity system relying on GPS
derived corrections and assume its validity for time constants
<100 s.
Marine gravity
For marine gravity, we have constructed a noise model by
analysing survey line / tie line intersection errors as a function
of time domain line filtering (Fig. 6). The survey, employing
a LaCoste and Romberg marine meter, covered roughly 50
lines and experienced sea-states between 24 with an average
of 3, which is reasonably typical. At the intersection locations,
differencing measurement pairs cancels the signal leaving an
estimate of the noise in the data (the difference is actually

2 times the noise). The data were line levelled before this
analysis to prevent bias errors skewing the real rms noise
estimates.
The roll-off seen in the mis-tie data at short filter time con-
stants (Fig. 6) is due to a fixed low pass filter (T
c
300
seconds) applied at a preprocessing stage. Data before this
stage were not available to us. Such a filter is typical in ma-
rine gravity corresponding to a wavelength of approximately
750 m at 2.5 ms
1
. For time constants, T
c
, greater than
roughly 5500 seconds, the increase in the apparent noise is
due to the error in cancellation of the signal, which being
heavily filtered (>14 km) along the survey and tie lines, has
Figure 6 Model for marine gravity noise based on mis-tie analysis.
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
182 G. Barnes et al.
accumulated signal from different parts of the survey. For the
noise model, data in these extreme limits were excluded from
the parameter fit giving a relation that can be used to predict
the noise over a useful range,
noi se(T
c
) = 0.3 +0.16e
6.310
4
Tc
. (4)
Similar but earlier analysis based on the LaCoste and
Romberg meter reported larger errors ranging from0.831.16
mGal (10 minute filtering) over a variety of weather conditions
(Valliant 1983). These figures most likely reflect the lack of
good navigation data preventing accurate eotvos corrections.
In an evaluation of the BGM-3 meter, Robin and Watts (1986)
deduced an accuracy of 0.38 mGal (180 second filter) under
force 1 (optimal) conditions. The consensus of the 2002 SEG
workshop on Gravity Noise Workshop, Salt Lake City, Utah,
USA concluded repeatability <0.2 mGal for low sea-state and
0.6 mGal for average conditions.
As with the other survey systems, it is difficult to find suit-
able published analysis and coupled with the different ways of
quoting noise in survey data, it is often impossible to deduce
the underlying accuracy of the original measurements. In the
analysis outlined later, we will use our above noise models but
values from other sources could easily be substituted.
Acquisition sampling frequency
In all the above cases, after the measurement bandwidth has
been defined either by a frequency bandwidth or a filter time
constant, the effective sampling frequency must be matched
to prevent over-sampling and subsequent over-averaging of
the noise in further analysis. With a bandwidth f
bw
, a valid,
independent new measurement point will be provided at a
sampling frequency given by the Nyquist condition,
F
s
= 2 f
bw
. (5)
NOI SE ON GRI DDED AND FI LTERED
SURVEY DATA
To assess the noise levels originating fromdifferent acquisition
systems and survey patterns, one must analyse the data in
its final gridded form. Once gridded, gravity and gradient
data can be transformed into different components, surveys
acquired at different heights can be upward continued to a
common height and spatial filtering can be readily applied.
After these processes, the rms noise level as a function of
spatial wavelength can be compared for the various survey
scenarios.
After filtering to the bandwidth and accordingly redefin-
ing the sampling frequency equation (5), the acquisition noise
data are assumed to be transposed onto a grid with equal
x and y pitch such that on average, one measurement point
contributes to one grid cell. For land gravity surveys having a
regular array of station points, this array simply becomes the
data grid. For airborne and marine acquisition, the presence of
survey lines and tie lines make the situation more complicated
because of the highly anisotropic nature of the sampling often
with high density along the lines and relatively sparse sam-
pling in-between the lines. The use of one point per grid cell
on average in these cases is therefore an approximation that
has to be taken with caution since in the severe case, where
the line spacing is much greater than the inter-sample point
spacing, the actual measurement point density could vary con-
siderably across the grid. Consequently, when focusing ones
attention on the shortest wavelengths in such a grid (those
approaching the Nyquist limit), some small features may not
actually be resolvable due to the lack of local sampling in a
particular area (between survey lines, for example). In other
areas (where survey and tie lines cross, for example), there
will be an abundance of measurement points supporting the
shortest wavelengths. When trying to find extremely localized
anomalies, one hopes that they lie under these intersections
and can even analyse profile rather than gridded data. If the
sampling is very non-uniform, it is advisable to view distribu-
tions of the sample density in conjunction with the data grids.
Often, survey data are gridded by methods such as minimum
curvature in conjunction with filters that tend to even out the
anisotropic sampling. Such grids consequently produce data
bandwidths lower than that predicted by the above one point
per cell on average statement. Here, we want to extend the
analysis to bandwidths based on the average sample density
to give noise estimates for higher resolution grids even if they
have areas of under and over sampling.
The spacing of valid independent measurement points along
the survey and tie lines is given by
d
L
=
v
F
s
, (6)
where v is the speed of the vehicle and F
s
is the bandwidth
limited sampling frequency (equation (5)). If the survey lines
and tie lines are spaced with separations S
x
and S
y
respectively,
then the grid pitch, resulting in one measurement per cell
on average is given by
=

v
F
s

1
S
x
+
1
S
y

1
. (7)
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
Comparing gravity and gravity gradient surveys 183
With these assumptions, the rms noise on each grid point is
the same as that on each bandwidth limited sample point; the
gridding process is essentially a rearrangement of the original
data points. If this uncorrelated rms noise level is n, the cor-
responding power spectral density for the grid, P is white and
obtained using Parsevals theorem,
n =

k
Nq
kx,ky=0
P
2
dk
x
dk
y
,
P =
n
k
Nq
=
n

, (8)
where k
Nq
=

is the Nyquist spatial frequency of the grid in


the x and y directions and the two dimensional power spectral
density has amplitude units of E/

m
2
(read as eotvos per root
wavenumber area).
In the case of the full tensor gradiometer, by capturing the
entire gravity gradient tensor at each measurement point, one
can reconstruct the field more accurately due to the associated
information regarding howthe field is changing in different di-
rections (Barnes et al. 2008). Equivalently, with gradient infor-
mation, one can widen the line spacing without suffering alias-
ing (While, Biegert and Jackson 2008). In this analysis, when
evaluating the performance of full tensor gradiometer surveys,
the smallest grid pitch that the full tensor gradiometer derived
data can support is therefore assumed to be half that given
in equation (7). As described above, the noise level attributed
to the full tensor gradiometers equivalent Gzz already ac-
counts for the reduction due to the combination of multiple
measurements so therefore the PSD on the grid (equation (8))
must remain unchanged. To resolve this, the rms noise on
each full tensor gradiometer grid point, n is doubled. The ef-
fect of the full tensor gradiometers enhanced interpolated Gzz
is then only to allow the extension of its bandwidth to twice
the original Nyquist frequency.
The pitch, defines the resolution of the grid where the
shortest wavelength supported in all directions is given by

Nq
= 2. A rectangular grid however defines a rectangular
Fourier space that supports higher frequencies in the diagonal
directions; the shortest wavelength being

2 located at the
corners of the Fourier space. When referring to the resolution
of data on a grid, it is more meaningful to talk in terms of a
single radial wavelength whose value applies in all directions.
Accordingly, the grid should be filtered to this radial wave-
length making the bandwidth consistent and easier to specify.
After such a process, the rms noise on the grid, n
g
as a function
of radial wavelength (k = 2/., 2) can be calculated
using Parsevals theorem, this time in radial coordinates,
n
g
() =

1
4

2
/

0
P
2
2 kdk
=
n

.
(9)
Note that the factor of 1/4 in equation (9) is required to
scale the circular integral over the positive frequencies only in
accordance with the definition of the power spectral density.
When interpreting potential field data, one often applies a
degree of high-pass filtering to remove the regional field and
other long wavelength variations that would otherwise swamp
the features of interest. This is especially pertinent for gravity
data whose spectrum is dominated by the long wavelengths
forcing one to interpret a gravity residual or even to derive
derivatives of the field. In the Fourier domain, the longest
wavelength component is defined by the survey size. To ac-
count for this and also to allow the provision for estimating
the noise level on a band-pass filtered grid, we define a length
L being the maximum feature size of interest bounded by the
survey extent. Modifying equation (9) accordingly gives
n
g
() =

1
4

2
/

2
/
L
P
2
2 kdk
= n

2

1
L
2

.
(10)
GZZ AND GZ TRANSFORMATI ONS
AND UPWARD CONTI NUATI ON
To compare the relative performance of gravity and gravity
gradient surveys, either the gravity data must be transformed
to gradient data or vice-versa to allow a like for like compari-
son. In the Fourier domain, these transformations are readily
performed using the differential operator k to convert gz to
Gzz or 1/k to convert Gzz to gz. Augmenting with upward
continuation is accomplished by incorporating the exponen-
tial factor e
kz
where z is the upward continuation distance.
The following equations give the noise levels (n
g
) on grids of
gz and Gzz deriving from uncorrelated noise distributions (n)
of gz and Gzz,
Gzz grid noise from Gzz measurements or gz grid noise from
gz measurements:
n
g
(L, ) =

1
4

2
/

2
/
L

P e
kz

2
2 kdk
=
n
2z

e
4z/x
(x +4z)
x

x=
x=L
, (11)
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
184 G. Barnes et al.
Gzz grid noise from gz measurements:
n
g
(L, ) = 10
4

1
4

2
/

2
/
L

P ke
kz

2
2 kdk
=
10
4
n
4z
2

e
4z/x

3x
3
+12zx
2
+24
2
z
2
x+32
3
z
3

x
3

x=
x=L
,
(12)
gz grid noise from Gzz measurements:
n
g
(L, ) =

1
4

2
/

2
/
L

P
1
k
e
kz

2
2 kdk
=
n
10
4

0,
4z
x

x=
x=L
,
(13)
where (a,z) is the incomplete Gamma function defined by
(a, z) =


z
t
a1
e
t
dt, (14)
and the factor 10
4
appearing in equations (12) and (13) reflects
the change in units between mGal and eotvos.
COMPARI SON EXAMPLE
The three equations (11)(13) allow graphs to be constructed
that compare the noise levels of five survey scenarios; airborne
full tensor gradiometer, airborne gravity, marine full tensor
gradiometer, marine gravity and land gravity. Although the
comparisons between land gravity and the marine systems
are a little academic, all five cases are placed side by side to
provide insight into their relative merits.
Figures 7 and 8 compare the noise for the five systems as a
function of resolution for grids of gravity and gravity gradient
data based on a survey extent of 80 km (defining the longest
wavelength, L, in the Fourier decomposition). The parameters
used for instrument noise, bandwidth, survey line spacing etc.
are listed in Table 2 and represent an example case where the
data from the different surveys are to be compared. Measure-
ment noise levels were estimated using the models described
earlier. The x axes of the graphs represent the wavelength,
, to which the grid has been filtered and would normally
be chosen by the interpreter who is interested in analysing a
particular feature in the data. The y axes then give the esti-
mate of the noise level that remains on the grid after such a
filter, defining a bandwidth L to , has been applied. If the
remaining noise level falls sufficiently below the amplitude of
the signal of interest, then the feature can be analysed with
Figure 7 Comparison example of gridded gz noise for different survey scenarios as a function of bandwidth L = 80 km to .
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
Comparing gravity and gravity gradient surveys 185
Figure 8 Comparison example of gridded Gzz noise for different survey scenarios as a function of bandwidth L = 80 km to .
Table 2 Survey parameters used for the comparison example
Line Tie-line Upward
Speed spacing spacing Noise Measurement continuation
(ms
1
) (m) (m) (rms) bandwidth (m)
Air full tensor gradiometer 65 500 2500 4.0 E 0.2 Hz 0
Air gravity 65 500 2500 0.85 mGal (GPS limit) 70 seconds 0
Marine gravity 3.0 500 2500 0.45 mGal 200 seconds 200
Marine full tensor gradiometer 3.0 500 2500 4.0 E 0.2 Hz 200
Land gravity 500 500 grid 0.5, 0.1, 0.05 mGal 200
a level of confidence. Knowing the bandwidth and amplitude
of the target signal, curves like these can therefore be used to
give a quick estimate of how well a survey system might be
able to resolve it.
When comparing one survey system with another, it is
sometimes useful to identify the wavelength of the filter at
which the two perform equally well. These cross-over wave-
lengths generally occur at different points when viewing grids
of Gzz or gz. Tables 3 and 4 list the cross-over values for this
particular set of example survey parameters. The cross-over
between airborne full tensor gradiometer and GPS limited air-
borne gravimetry occurs at 35 km when considering grids of
gz. This means that when grids deriving from the gravity sur-
vey are filtered to leave only wavelengths greater than 34 km,
they will have less noise power than those deriving from the
full tensor gradiometer measurements. When viewing grids of
Gzz, the cross-over occurs at 39 km. If the target anomalies
necessitate interpreting grids with lighter filtering (preserving
wavelengths less than 35 kmfor gz grids or less than 39 kmfor
Gzz), then a full tensor gradiometer survey starts to become
favourable. Viewing data as Gzz suppresses the long wave-
length information and can have similar results to high-pass
filtering gz data to remove regional fields. This is equivalent
to decreasing the value of L and as this is done the cross-over
on the gz comparison moves to longer wavelengths (for ex-
ample, 42 km when L = 60 km). This happens because the
gridded data noise level is more focused on the short wave-
lengths where the gradiometer has its advantage. To make
the gravimeter data comparable one needs to apply heavier
low-pass filtering hence increasing the cross-over wavelength.
Conversely, when interested in the longer wavelengths (deep
or regional geology, for example), then larger surveys and less
high-pass filtering is necessary, both meaning larger values of
L. One then typically interprets gz grids of the data and as
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
186 G. Barnes et al.
Table 3 gz noise cross-over wavelengths taken from Fig. 7, showing where one system offers lower noise than another. Example: gz grids from
marine full tensor gradiometer have lower noise than 0.1 mGal land gravity when the grid is filtered to wavelengths less than 4 km
Airborne full Marine full Land Land Land
tensor Airborne tensor Marine gravity gravity gravity
gradiometer gravity gradiometer Gravity 0.5 mGal 0.1 mGal 0.05 mGal
Airborne full tensor gradiometer <35 <4 <5.5
Airborne gravity >35
Marine full tensor gradiometer All All <28 <38 <4 <1
Marine gravity >4 All >28 All
Land gravity 0.5 mGal >5.5 All >38
Land gravity 0.1 mGal All All >4 All All
Land gravity 0.05 mGal All All >1 All All All
Table 4 Gzz noise cross-over wavelengths taken from Fig. 8, showing where one system offers lower noise than another. Example: Gzz grids
from marine gravity have lower noise than airborne full tensor gradiometer when the grid is filtered to wavelengths greater than 7.5 km
Airborne Marine full Land Land Land
full tensor Airborne tensor Marine gravity gravity gravity
gradiometer gravity gradiometer gravity 0.5 mGal 0.1 mGal 0.05 mGal
Airborne full tensor gradiometer <39 <7.5 <11.5 <3.0
Airborne gravity >39
Marine full tensor gradiometer All All <33.5 <49 <9 <3.5
Marine gravity >7.5 All >33.5 All
Land gravity 0.5 mGal >11.5 All >49
Land gravity 0.1 mGal >3.0 All >9 All All
Land gravity 0.05 mGal All All >3.5 All All All
Table 5 Survey point densities and shortest recoverable wavelengths
corresponding to the survey parameters shown in Table 2
Point density Shortest recoverable
(km
2
) wavelength (km)
Air full tensor 14.8 0.26
gradiometer
Air gravity 1.1 1.95
Land gravity 4 1
Marine gravity 4.7 0.71
Marine full tensor 320 0.06
gradiometer
L increases even more emphasis is placed on the long wave-
lengths where the gravimeter derived data have their advan-
tage. The cross-over wavelength with the gradiometer derived
gz data then decreases.
The point densities and corresponding shortest recoverable
wavelengths (based on the average point density as described
earlier) are listed in Table 5. The shortest recoverable wave-
lengths for the full tensor gradiometer cases have accounted
for the expected resolution enhancement provided by measur-
ing the full tensor. The marine full tensor gradiometer case
having line spacing of 0.5 km 2.5 km yielding a shortest
recoverable wavelength of 60 m illustrates the anisotropy of
the sampling discussed earlier. Clearly, the value of 60 m sig-
nal resolution is skewed by the high resolution along the lines
but will not be provided at locations in-between survey lines.
Indeed, to reap full benefit froma marine full tensor gradiome-
ter survey, both grids and survey lines should be analysed to
make full use of the resolution.
CONCLUSI ONS
Analysing noise fromgravity and gravity gradient instruments
and making assumptions about how the noise translates into
gridded form has allowed us to derive relations that com-
pare a variety of survey scenarios. The relations estimate the
noise levels on the grid as a function of the bandwidth de-
fined by filtering commonly performed by the interpreter. The
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187
Comparing gravity and gravity gradient surveys 187
comparison accommodates upward continuation when com-
paring land / marine with airborne surveys and can either be
made in terms of gravity (gz in mGal) or gravity gradient (Gzz
in eotvos). The quality of the data on these grids is highly
dependent on the noise accompanying the original measure-
ments and also on the parameters that define the survey de-
sign. Such factors together with the complications of compar-
ing systems that fundamentally measure different quantities
has always left the analyst uncertain as to the relative bene-
fits of one system over another. It is commonly understood
that gravity surveys are suitable for regional or deep geology
and gradiometer surveys are better for high resolution. The
equations derived above were intended to quantify this by
offering a sufficiently simple algebraic comparison allowing
one to assess the merits of the different survey scenarios easily
and quickly. In forming this analysis, approximations and as-
sumptions had to be made concerning the noise characteristics
of the systems and the design of the survey. It is hoped how-
ever, that the expressions derived will be sufficiently useful to
form a rough guide to survey design and a precursor to more
comprehensive numerical feasibility modelling.
REFERENCES
Argyle M., Ferguson S., Sander L. and Sander S. 2000. AIRGrav
results: A comparison of airborne gravity data with GSC test site
data. The Leading Edge 19, 11341138.
ARKeX Ltd 2008. Gravity survey data processing. Patent:
WO/2008/093139.
Barnes G., Barraud J., Lumley J. and Davies M. 2008. Advantages of
multi-tensor, high resolution gravity gradient data. 78
th
SEG meet-
ing, Las Vegas, Nevada, USA, Expanded Abstracts, 35873590.
Bell R.E. and Watts A.B. 1986. Evaluation of the BGM-3 sea gravity
meter system onboard R/V Conrad. Geophysics 51, 14801493.
Dampney C.N.G. 1969. The equivalent source technique. Geophysics
34, 3953.
Elieff S. and Ferguson S. 2008. Establishing the air truth from
10 years of airborne gravimeter data. First Break 26(11), 7377.
Hallinan S., Gheit A.B. and Reid A. 2002. Terrain noise. SEG Work-
shop on Gravity Noise, Salt Lake City, Utah, USA.
Li Y. 2001. Processing gravity gradiometer data using an equivalent
source technique. 71
st
SEG meeting, San Antonio, Texas, USA,
Expanded Abstracts, 1466.
Metzger E.H. 1977. Recent gravity gradiometer developments. AIAA
Guidance and Control Specialist Conference, Hollywood, Florida,
USA, 306315.
Seigel H.O. 1995. A Guide to High Precision Land Gravimeter Sur-
veys. Scintrex Ltd.
Valliant H.D. 1983. Field trials with the LaCoste and Romberg
straight-line gravimeter. Geophysics 40, 611617.
Van Kann F. 2004. Requirements and general principles of air-
borne gravity gradiometers for mineral exploration. In: Airborne
Gravity 2004 Abstracts from the ASEG-PESA Airborne Gravity
Workshop (ed. R.J.L. Lane), pp. 15. Geoscience Australia. ISBN
1920871136.
While J., Biegert E. and Jackson A. 2008. Interpolation of gravity and
gravity gradient data by using the generalized sampling expansion:
Theory. Geophyiscs 73, 111121.
C
2010 European Association of Geoscientists & Engineers, Geophysical Prospecting, 59, 176187

Вам также может понравиться