You are on page 1of 11

CALCULATION OF THE GRAVITY FIELD USING

VERTICAL GRADIENT OF THE THE FOURIER TRANSFORM *


BY

B.

N. P. AGARWAL

**

and

T. LAL

***

ABSTRACT AGARWAL,B. N. P. and T. LAL., 1972, Calculation


Gravity Field Using Fourier Transform, Geophysical

of the Vertical Gradient Prospecting 20, 448-458.

of the

Fourier transform techniques have been used to calculate the theoretical filter (amplitude) response function of Nth order vertical derivative continuation operation. The amplitude response functions of the vertical gradient and its continuation follow from the same. These response functions are subsequently used to calculate the weighting coefficients suitable for two dimensional equispaced data. A shortening operator has been incorporated to limit the extent of the operator. For comparative study, some of the developed coefficient sets and the one presented in this paper are analysed in the frequency domain and their merits and demerits are discussed.

In gravity interpretation, the importance of vertical derivative maps of various orders in delineating the structures of interest has been established beyond doubt. Amongst these, the second derivative method has been put to maximum use. The first derivative (gradient) method, inspite of certain subtle advantages over the second derivative (Evjen 1936; Baranov 1953 ; and Henderson 1960) has received little attention. And except Baranov (1953) and Henderson (1960) not much work has been done to develop practical schemes for the computation of the gradient. This state of affairs seems to be largely due to the fact that the mathematical analysis needed for gradient computation is more complicated than that needed for the second derivative (Baranov 1953) * The mathematical analysis of Baranov (1953) and Henderson (1960) requires the use of upward continuation integral which is valid only for z < o and is singular at .z = o, with the z-axis positive downwards. Because of this singularity, the differentiation of the integrand under the integral sign, to
* Received September 1971. ** Department of Applied Geophysics, Indian School of Mines, Dhanbad, Presently at EER Division, C.W. & P.R.S., Khadakwasla, Poona-24, India. *** Department of Geophysics, Banaras Hindu University, Varanasi-5 India. India.

VERTICAL

GRADIENT

OF GRAVITY

449

obtain the gradient at z = o, can not be carried out. The method of Baranov (1953) essentially consists of removing the aforesaid singularity by expressing the average gravity function g(r) in form of an even fourth order polynomial which is continuous for all values of ZY, the radius of the circle. This makes the integrand continuous and therefore differentiable under the integral sign for all values of z. In Hendersons (1960) method, the gradient is calculated at different heights above the plane z = o, i.e. for z < o, by evaluating numerically the integral obtained by differentiating the integrand of the upward continuation integral once with respect to z. The various gradient values thus obtained are then extrapolated for z 3 o to achieve the gradient on and below the plane of observation. In this paper, the Fourier transform technique has been utilized to derive the theoretical filter (amplitude) response functions of the gradient and its downward continuation operations. Four sets of weight coefficients for gradient and its downward continuation at unit depth have been calculated using the corresponding amplitude response functions in the cut off frequency region. The amplitude response curves of Baranov and Hendersons coefficient sets are presented along with those of the derived coefficient sets. The merits and demerits of the various coefficient sets are discussed. Both the methods of Baranov and Henderson assume the average gravity value over circles of different radii being considered as infinite point average. However, in practice we have the observations of gravity field at a finite number of points. This fact has been taken care of in this analysis which involves the use of discrete point values of the gravity field.
THEORY

Consider a Cartesian system of coordinate with the z-axis positive downwards. Let g(x, y, - / z / ) and g(x, y) be the fields at two planes z = - 1z 1 and z = o related by the upward continuation integral (Dean 1958)

mm
&>Y>-l4) =
-m

ss

1.2 l/zx

(a + p2 + 22)3/2

* g(x-~,y-p)~da*dp.

(1)

-m

Equation (I) is a two dimensional convolution and hence represents a filtering operation. The theoretical filter (amplitude) response of upward continuation, w&b v, - I 2 I ), is obtained by Fourier transform of the weighting function

wu(xJ yJ-- 11) =

(%2

12 l/zx y2 +

22)3/2

450 as

B. N. P. AGARWAL

AND

T. LAL

where i = 1/ - I, and 427~ and v/2x are frequencies unit distance in x and y directions respectively.

measured filter

in cycles per response of

It can be easily proved (Dean 1958) that the theoretical downward continuation W,(ti, v, 1z 1 ) is Wd(z4, v, 1 z 1 ) = eJZIVUa+va. A more general relation than equation (I) can be written fields at two different levels using equations (3) and (4) as g(x, y, z) = -$

(4) for the gravity

f f G(u, v) - exp [J vz& + 02 + i(ux + vy)] * du * dv, -m -m where G(u, v) is the Fourier transform of g(x, y).

(5)

Expression (5) has certain advantages over expression (I) which will be discussed below. It is known that the integral (I) is only valid for z < o. By substituting .z = o without precaution, we find that &, y, o) = o but the true value is g(x, y, o) = g(x, y). The wrong result is explained (Baranov 1953) by the fact that the integrand becomes singular for z = o at the origin, and if z, dcand p are all zero, the function to be integrated has infinite value, indicating a discontinuous behaviour of the integrand at the origin. The principal conclusion from this is that we cannot differentiate under the integral sign for calculating the gradient at z = o. Now, when we substitute i = o in equation (5), we obtain g(x, y, o) = g(x, y), which indicates that the integrand is continuous for all values of z and therefore differentiable under the integral sign. Thus the main advantage of the filter analysis lies in the fact that-though the expression (I), relating the two fields at different levels in space domain, is discontinuous at the origin and is valid only for negative Z, i.e. .z < o-the transformed equation (5) in frequency domain is continuous at all points lying outside the singularity of the anomalous mass. Differentiating 3 equation (5) N times with respect to Z, we have [Z 1/z@+v2+i(ux+yv)]du dv.

aN

g(x, y, z) = 5

-f _s G(u, v) - (z@+v2)N/2.exp m m

VERTICAL

GRADIENT

OF GRAVITY

4.51

Therefore, the theoretical amplitude response function WN(U, ZI, z) of the Nth order vertical derivative continuation can be written from equation (6) as W&u, 0,~) = (u2 + v2)N/2 * exp (2 1/u2 + ~2) Equation (7) indicates that the theoretical amplitude operations, i.e., upward and downward continuation, various orders, are circularly symmetric in nature. (7) response functions of all vertical derivatives of

By substituting z = o and N = z in equation (7), we obtain the theoretical amplitude response function of the second derivative as (8) which is same as given by Dean (1958). Similarly, the amplitude response function of the gradient can be obtained by putting z = o and N = I in equation (7) as
W,(u,

21,0) = 242+ 112

Wl(z4, v, 0) = Ilu2 + v2. (9) The advantage of the calculation of the gradient over the second derivative in gravity or magnetic interpretation can be seen from equation (9) which indicates that the amplitude response of the gradient increases linearly with increase in radial frequency whereas for second derivative it increases as square of radial frequency. Thus the gradient is less sensitive to observational errors (high frequencies) than the second derivative.
DESIGN OF OPERATOR

To achieve any desired operation on equispaced gravity or magnetic data, it is desirable to know the corresponding set of weight coefficients which can be calculated from the weighting function W(X, y, z) by taking the inverse Fourier transform of the filter response function W(U, YJ, Z) as 7 f W(u, v, z) - e+i(ux+vu) -du * 0%. -m -m (IO) can be rewritten as nay, z) = c2 7 7 W(U, ZI,z) . e+z(umAx+vnA~) . du * dv, w(x, y,z) = k2 (10)

Equation ~(mhx,

(II)

-IL. -"y

where uuz and vr, are the cut-off frequencies form of I+(%, v, z) is specified to be zero. For case, it is conveniently chosen as the Nyquist interval, i.e. uuz = 7~lAx and vy = rc/AY. Thus
n/b

beyond which the Fourier transdiscrete weights, as in the present frequency or 0.5 cycles per data equation (II) becomes (12)

~(mAx,

nay, Z) =

;?

-x/As -x/Ap

J y(f4

v, 2) - e +2(umAx+vnW . du. dv.

452

B.

N.

P.

AGARWAL

AND

T.

LAL

As the amplitude response function operations, we have w(m,n,.z) = f

happens to be even in u and w for all

; ; W(~,v,z)cosm~.cos~v.d~.dv. 0 0 The amplitude

(13)

with Ax = Ay = I (unit of grid spacing). from equation (7) turns out to be wr(m,n,o) for the vertical
w~(m,n,

response functions

=; and
-$

.~ j~1/u2+v2Cosmu.CosnZI.du.dZI 0 0

(14)

gradient
I) =

s" s" 1/d II 0

21.2. &a+"a~~~nzu~

cosw

*duedv

(15)

for the gradient

continued

downwards

for one unit of depth.

Integrals (14) and (15) have been evaluated numerically using Simpsons rule of integration. The weight coefficients thus obtained are presented in tables (ra) and (za). From equations (14) and (15) one can calculate any
TABLE I

Two dimensional
a)

vertical gradient coefficients


zkhout
2 + .o5og37 .oog405 .007687 -

Calculated
I

shortening
3

operator
4 + .or5ogo .oooggg .oorg45 .oorrgz .ooog24 .000577 5 .012512 .002018 .ooogrq .000854 .000577 .000468

m==
92 = 0 I 2 3 4 5 + + + -

0. 2.402261 .434812 .o5og37 .037196 .015ogo .012512 -

.434812 .o78993 .oog4o5 .007272 .oooggg .002018

.o37rg6

-.007272 .oojog5

-.003og5 .001g45 .ooog14

-.002228 .0011g2 .000854

b) m= 0

Calculated
I

with
2

shortening

operator
3 4 5

P&=0
I 2 3 4 5

+
+

2.402261 .420080 .044268 .026856 .008222 004525

.420080 .073701 .007883 .000521 .ooo6g4

+ -

.044268 .007883 .005765 .oorgog .000890 .oooz6g

.026856 .005052

+
-

.008222 .000521

.004525 .ooo6g4

.oorgog
.001114 .000431 .ooorgo

.ooo8go
.000431 .000231 .000082

.oooz6g
.000190 .000082 .000031

+
-.

-.005052

TABLE

Two dimensional

vertical gradient continuation:


a) Calculated
3 7
-1.ooggo

downward continuation coefficients : Depth of One unit of grid spacing


without
4 5 6 8

shortening 9

operator
IO

12=0

--2z666gz
1.36134

w%%+

--4+x841

2.94074

-1.93418 .o7331 .517o5 .18584

48.40207 -22.66692

.23561

7.46332 2.78072

-2.78072

I.35839
49070 .o8go5
.04612

.79684 .Izgo6 ---03593


.02054 -.01367

.36395 .0272g .01732

.26813

.777o2 --.20668

.50061 -.13282

.c+I4

9.678;4 4.95841

.oo457
.01732 .0124g .oog30

2.94074 .I3907 .o8gog


.o5274 .03618
.0272g

1.35839 .796%
.06336 .04612 .o5274 .o3353 .02408

.99912 .49070 .z828g .06336 .03618 .02408 .01641

-.61676 .16330 -.05846 .02808

.z828g .I3907 .07950 .oogjo


-.00711 .00526 --00433 .oo337

-.oI65g

I.93418 1.36134

.517o5

.18584

.o4713 -.02306 .01321 -.00878

..00760 .02808 .02306

.36395
.03593 .00597
.0124g .oo8gj .00711 .00542

.00597
--00457 .oo337
.0032g -.0027g -.0027g

1.ooggo

.Izgo6 .og614

2 3 4. 5 6 7 8 g .02054 .oI65g .01321, .01367 .01054 .00878

.01054 -.00760 .00542 -.0043i

IO

.777o2 .61678 .50061

b) Calculated
3 -4.51716
.22go2 -

.26813 .20668 .16330 .13282

.o7331 .05846 .04713

.00216

with
4 5 6

shortening

operator
7 8

2 2.48653 .66652

9 .gIg88

IO

n=o .01gq -

48.40207

--22.43659
.I9495
-

g.z8gI5
1.2-53 .42848

.243o7 .08320 .o3847 .02015

.5%98 .oIIg6
.00704 .I5314

.37111

-2.64135

--22.43%9 9.28915

7.31216 2.64135

.gIg8z

.op8g

--.233o6 .06082 -.02084 --.09741 .03319 -.01520 .00787 .0238i.01286 .00723


.oogzg

3 4

4.51716

1.22453 .66652

-1.48166 .3gI68 .13610 .o615g .03360

--00493

.3gI68

2 -

2.48653 1.48166 .gIg88 -

.42848 .22go2 .13610 .08320 .10653 .o615g .03847 .02381 .01520

8
.oogzg
.00563

.58398 .37111

.243o7 .I5314

.oyz8g

.09741

.10653 .05626 .03360 .02015 .01286 .00787

.oIIg6 .00723

.I4172 -.03700 .01250 -.00563 .00285 -.00160 .00088 --.00050 .ooo25

--

.00493
.00285 -

.00270 -.00161 .ooogo -.00052 .00026 .00459 .00270 .00160 .oo445 .00263 .00161 .00088 .oo445 .00261 .00162 .ooogo .00050 --00459 .00263 --.00162 .ooog1 -.00052 .00025
-.00012

-.00012

IO

.23306 .I4172

.06082 .03700

.oyjIg .02084 .01250

.00004

454

B. N. P. AGARWAL

AND

T. LAL

number of coefficients in each set, thus giving rise to an operator of infinite extent. But in actual practice, we use only a finite number of coefficients in each operation. Therefore, it is essential to shorten the operators presented jn tables (Ia) and (za). This is achieved by multiplying the weight coefficients by a function defined as :

I+Cos

x(m2 + n2y2 _____ 1 II


(X2 + yzp/z

for Ifn I\<x


1 lnl for or I Ifll>Y <y 1m [ > X

(Fuller rg67), where X and Y are the tolerated x and y (or m and n) extents in data intervals. Let zd(m, 1z,z) be the shortend operator, then we have W(m, n, 2) = s(m, n) * W(m, n, 2). (17)

The coefficients thus calculated for the vertical gradient and gradient continuation for .z = I unit of grid spacing are presented in tables (rb) and (zb). The smoothing and shortening of the infinite extent operator coefficients modify the amplitude response. Hence, to get a clear picture of the exact nature of these derived operators and some of the existing operators, it is essential to calculate the amplitude response for matching with the theoretical amplitude response using the relation N wc(u, v) = w(0, 0) + 2 i; z0(m, 0) * cos mz4 + 2 C ~(0, n) . cos nv +
m=1 *-l

+ 4 f; i zu(m, 12)~~0~mz4~cosnv, m-1 I=1 where w~(u, V) is the calculated amplitude maximum frequency range of investigation x<f4+j-x;

(18)

response of a particular set. The (Nyquist frequency) is given by (19)

--x<v<+n.

For the sake of comparison, the derived coefficient sets and the coefficient sets of Baranov (1953) and Henderson (1960) are analysed using equations (18) and (rg). The results are discussed below separately for each type of data operation.
DISCUSSION OF THE RESULTS

Vertical Gradient The theoretical and calculated amplitude responses of Baranov (Ig53), Henderson (rg6o), and the derived coefficient sets (table I) are shown in

u-

Fig.

I.

Theoretical amplitude vertical gradient.

response of

Fig. z. Amplitude gradient COeffkkntS

response of vertical Of Baranov (195-j).

Fig. 3. Amplitude gradient coefficients

response of vertical of Henderson (1960).

Fig. 4. Amplitude unsmoothed vertical

response of derived gradient coefficients.

u-

Fig. 5. Amplitude

response of derived

smoothed

vertical

gradient

coefficients.

Fig. 6. Theoretical amplitude response of vertical gradient downward continuation of one unit of grid spacing.

Fig. 7.. Amplitude response of vertical gradient continuation coefficients of Baranov (1953).

Fig. 8. Amplitude response of vertical gradient downward continuation coefficients of Henderson (I 960).

Fig. g. Amplitude response of derived unsmoothed vertical gradient downward continuation coefficients.

Fig.

IO.

Amplitude

response

of derived smoothed vertical ation coefficients.

gradient

downward

continu-

VERTICAL

GRADIENT

OF

GRAVITY

457

figures I to 5. It can be seen from figure 2 that the response of Baranovs coefficient set does not show the circular nature of contours though the maximum amplitude is nearly the same as that of the theoretical response. The amplitude response of Hendersons coefficient set (figure 3), to some extent, shows the circular pattern of the contours. However it possesses a directional effect as it approximates the theoretical response more closely in the 45 direction. The responses of the derived coefficient sets (figures 4 and 5) show a close match with the theoretical amplitude. The contours are slightly distorted near the cutoff region.

Gradied

dowrtward contimaation

The theoretical and calculated amplitude responses of Baranov, Henderson, and the derived coefficients sets (table 2) for gradient downward continuation for unit depth are shown in figures 6 to IO. The amplitude response of Baranov and Henderson (figures 7 to 8) are very similar to each other and show the directional alignment at an angle of 45. The magnitude of the amplitude response of these coefficients sets are very small in comparison with the theoretical, indicating that these are not suitable approximations of the gradient downward continuation operation. The amplitude response of unsmoothed derived coefficient set (figure 9) shows the approximately circular nature of the contours which are not smooth in nature. The amplitude res- ponse of the smoothed and shortened derived coefficient set (figure IO) has smooth circular contours. The magnitude of responses, in both cases, are close to the theoretical response.
CONCLUSION

Fourier transform technique has been used to remove the singularity of the upward continuation integral which is valid for z < o. A more general expression, relating the gravity fields at two different levels and valid for all values of z, has been obtained in frequency domain. The theoretical amplitude response function of Nth order derivative continuation has been derived which in turn yields the amplitude response functions of the vertical gradient and its continuation. These response functions are further used to develop the weight coefficients for the gradient and its continuation without and with the use of a shortening operator to limit the extent of the desired operation. The weight coefficients of Baranov (195x), Henderson (1960), and the set reported in this paper for the gradient and its downward continuation to one unit of grid spacing have been analysed in the frequency domain. The analysis reveals that the derived coefficent sets provide an improvement over Baranov and Hendersons coefficient sets. Also Hendersons coefficient sets show better response than Baranovs coefficient sets.
Geophysical Prospecting, Vol. 20 30

458

AGARWAL

AND LAL, VERTICAL

GRADIENT

OF GRAVITY

It is also possible to analyse the second derivative downward continuation coefficients of Henderson (1960) using the present method which would give an idea about how closely his set approximates the true operation,
ACKNOWLEDGEMENT

Authors are thankful to Prof. Jagdeo Singh for critically cript and for the valuable suggestions.
BIBLIOGRAPHY

reading the manus-

BARANOV, V., 1953, Calcul du gradient vertical du champ de gravite ou du champ magnetique measure a la surface du sol, Geoph. Prosp. I, 17r-rgr. DEAN, W. C., 1958, Frequency analysis for gravity and magnetic interpretation, Geophysics 13, 97-127. EVJEN, H. M., 1936, The place of the vertical gradient in gravity interpretation, Geophysics I, 127-136. FULLER, B. D., 1967, Two dimensional frequency analysis and design of grid operators, Mining Geophysics, Vol. z, (Theory), 658-708. HENDERSON, R. G., 1960, A comprehensive system of automatic computation in magnetic and gravity interpretation, Geophysics 25, 569-585.

You might also like