AN

INTERACTIVE
COMPUTER/GRAPHIC-DISPLAY-TERMINAL
SYSTEM
FOR INTERPRETATION
OF RESISTIVITY
SOUNDINGS
*
BY

H. K. JOHANSEN

**

ABSTRACT
JOHANSEN,

for Interpretation

H. K., 1975, An Interactive

of Resistivity
Soundings,

Computer/Graphic-Display-Terminal
Geophysical Prospecting
23, 449-458.

System

A fast computer-procedure
giving the apparent resistivity
curve as well as the partial
derivatives
with respect to the layer-parameters
is presented. It is based on the linear
filter method developed by D. P. Ghosh in 1971.
The sampling frequency is IO points per decade, and 3 decades are covered. The maximum relative error is less than 10-3, and in most cases orders of magnitude
smaller.
The computation
time on a CDC 6400 for one curve given in 30 points ranges linearly
from .17s for a two-layer case to .36s for a ten-layer case.
The procedure is used to plot master curves interactively
on a graphic display terminal
(Tektronix
4010)
connected to the CDC 6400. By trial-and-error
adjustments
a set of
layer-parameters
is found, giving essentially the measured curve.
INTRODUCTION

The linear filter method developed by Ghosh (1971, 1,971) has opened a
new range of possibilities for the interpretation
of resistivity soundings. It
is well suited for computerization.
Because of the small number of operations
needed the execution time is almost negligible. One can afford to calculate
master curves based on purely guessed layer-sequences and just throw them
away if they do not fit the data.
In his paper Ghosh points out that a sampling frequency of three points
per decade is a good compromise when the tools are a desk calculator and
logarithmic paper, because a relative accuracy better than .02 is achieved using
only nine filter coefficients. However, when working with an electronic computer
there is no reason not to use about hundred coefficients, if this proves to be
advantageous.
The direct interpretation
method is to calculate the resistivity transform T
from the field data and then derive the layer-parameters
(resistivities pj and
* Presented at the 35th meeting of the European
Association
of Exploration
physicists, Brighton,
June 1973.
** Laboratory
of Geophysics, Finlandsgade
8, DK-8200 Arhus N, Denmark.

Geo-

H. K. JOHANSEN

450

thicknesses dj) from T. In order to obtain T by the filter method, one has
to extrapolate
the measured curve both to the left and to the right, thus
using data that were not really measured. It is also necessary to smooth the
data. Both operations possibly introduce errors in the interpretation,
and a
weak indication
of a nearly suppressed layer might get lost, -too. Since our
aim is to extract the maximum amount of information
from the measurements
it might be preferable to take the indirect approach:
A guess of the layer-parameters
z) T(pj, do) is calculated, and papp
coefficients,
3) Trial-and-error
adjustments
of
curve agrees with the measured

I)

(pi, do) is made.

(~3, dj) found by convolution
(pj, dj) are made until
data.

with the filter

the calculated

papp-

Notice that the tinprocessed data are used for comparison only. The influence
of the interpreters personal judgment is thus reduced to the absolute minimum :
to decide whether the model is acceptable or not.
Step 3 may be carried out by an iterative least-squares programme, but the
method remains indirect in principle.
The relative accuracy of the field measurements is about .03. We wouldlike
the inaccuracies of the calculated curve to be an order of magnitude
smaller
so that their influence on the interpretation
can be safely neglected. In order
to achieve this a smaller sampling interval and hence a larger set of filtercoefficients must be used.
SUMMARY

In the Schlumberger

OFTHE

arrangement

FILTERMETHOD

we have

pas(s) = s2 ; T(h)Jl(hs)Adh
0
where s is half the electrode spacing and T(A) is the resistivity
(Koefoed 1968, 1970, Ghosh 1971, 197Ia).
Introducing

transform

new variables x and y by

x = In(s) and y = In(+)

and substituting

(1)

(2)

them into (I) we get

~a&)

= i T(YM-m

y)dy

(3)

(x)) exp (2x).

(4)

where
b(x) = Jl(exp

INTERACTIVE

TERMINAL

We note the following

SYSTEM

property

FOR

RESISTIVITY

INTERPRETATION

of the convolution

pas(4 = i T(Y + S)b(x -m

which holds for any shift S.

integral

Y-

F(f)

= G(f)

(3) :

WY

Now let F, G and H denote the Fourier-transforms

ively. Then

451

(5)

of pas, T and b respect-

. H(f).

(6)

Since all information

about the layer sequence is contained in T alone
e
same
function for all cases (pj, &). The violent oscillaH(f) = F(fMf)
is th
tions of b(x) for large x imply that H exists only in a generalized sense. Hence
the following indirect approach (Ghosh 1971) to the determination
of H is
chosen :
The particular

pair of functions
P(X) = exp (3X)/(1 + exp (2x))52
T(y) = Q ~XP (-

Y) =P (-

exp (-

(7)
~1)

satisfy (3).
Their Fourier transforms
are found by numerical integration,
modulus Qdf) and phase O(f) of the quotient F/G determine H.
For a given sampling interval Ax the Nyquist frequency is fp
and the corresponding sine-response function is given by
c(x)

's"Qcf,cm

I/fN

(a(f)

zxfx)df.

(A+)

to sample C(X) at positions

(9)
with the nodes

@(fN)

because it makes the sampling points coincide asymptotically

of C(X), thus reducing the length of the filter.
The apparent resistivity
approximately
by
pas(iAx) -

pas in the sampling

maz
C T((i -j)Ax
j =Gnh

points x( = iAx is then given

+ S)C(jAx -

where jnzgn and j nta5 are chosen so that the filter

larger indices can be neglected.
Geophysical

Prospecting,

Vol.

23

= ~/(zAx),

(8)

As observed by Koefoed (1972) it is convenient

shifted to the left by the amount

and the

S)

coefficients

(10)
with smaller or
29

H. K. JOHANSEN

4.52

ALGORITHM FOR CALCULATION OF T

The resistivity transform T(p, . . . . ., pi; dr, . . . . ., dnr-1; h) for a N-layer
stratified earth (see fig. I) can be calculated recursively (Koefoed 1970).
For a layer (pi-I,

dN-1) on top of a substratum

1 -

TN-I(~)

pN-1

kN-

1 eXp

(pN) the transform

2dN- oh)

(-

I + kN.e I exp (-

is

(11)

ZdN- lh)

where
kN-1 =

layer

(PN-1

(12)

dl

91

+ PN).

PN)/(~N-1

/--T(h)
i

layeri

rTi+l

j-TN-1
layer

(A)

(A)

N-l
QN-1

dN-1
A

layer
(infinite

Fig.

I.

QN
substratum)

A N-layer horizontally stratified earth with definition of symbols used in the

formulas.

The transform Tj for layer (pj, dj) on top of the sequence (p~+l, . . . . . , pN;
d,+l, . . . . . dN-1) with transform TJ+I is given by

w,(h) + Tj+$)
T(A) = I + Wj(A)Tj+,(A)/p;

j=N-z,N--3

,...,

2,1;

(13)

where
W,(V
Recursive

application

I - exp (- zd+)
= Pj I + exp (- zd+) *

(14)

of (13) and (14) yield T = Tl(A).

ASYMPTOTIC EXPRESSIONS

We find easily that

(15)

INTERACTIVE

TERMINAL

SYSTEM

FOR RESISTIVITY

INTERPRETATION

453

and hence
h-+co
A --f o .

Tj(h) --f pj
Tj +1(A)

(16)

The way in which these limits are approached depends on the specific set
of layer-parameters
considered. Let us look at the simple two-layer case:
For large A we have

TN-I(~)
The relative

M p~-1(1

departure

from the asymptotic

A = I TN-&)

is smaller than some small number

h > Together

zkN-1 exp (-;zd~-IA)).

(17)

value

plv-1 I / plv-I

(1s)

E if only
ln(a/z)/(zdlv-

1).

(19)

with
A$ = exp (-

this gives an upper bound

we get i < - 5.

iAx)

on i. If E = 10-2,

(20)
= I and Ax = (In IO)/IO,

&,-I

Something similar holds in the general case for T = TI and pl, although
the expansion of T is more complicated than (17). Experience has shown that
we are on the safe side if i c - 9.
For small A we have

There are two extreme

cases corresponding

TN-I(~)

M PN(I -

TN-I(~)

M ~N(I

to kN-1 w -l I :

(~N/~N-I}~~N-I)

for

PN

>

PN-1

(22)

and
+{p~-I/~N}~~N-I)

forp~<p~-l.

(23)

Since the resistivity ratios can take on all positive values, we have to compromise.
Let us consider (23) say. If PN-~/PN 5 10~ and dii-1 < Io3, then A 5 IOs2
if only A 5 10-l.
This happens if i > IOO.
Experience shows that i > IOO is sufficient also in multi-layer cases.
CUTOFF AND ACCURACY

Apart from the sampling frequency there are two factors affecting the
accuracy: the tails of the filter must be cut off at some point, and the coefficients can only be calculated with a limited precision.

H. K. JOHANSEN

454
The problem is how
at a relative accuracy
Let us first consider
For convenience we
in (10).

to compromise on these two aspects when we are aiming

better than 10-3.
the cutoff problem.
write TtL-j) and C(J) for T((i -j)Ax
+ S) and C(jAx-S)

We split the sum into three parts:

p,s(Q = { ;
+ *i + i } T(-5) C(j) ; i = 0, I, 2, . . . .
j--m
i =la
!I +1

(24)

Now let us choose j, and jz so that T can be approximated

by its asymptotic
values pi and pl in the first and third term of (24) in all cases.
Then
(25)
where
C$~I) =

i
C(f) and C$~S)=
I--m

i C(5)
1=1%

(26)

The analysis of the asymptotic

behaviour indicates that j, = -100
and
and
Cp)
have
been
found
by
applying
the
jz = 40 is what we need. C$-loo)
Euler transformation
to the tails of the filter.
To what precision should the coefficients be known if we want to handle
resistivity ratios up to p~/pl N 10~ ? Consider the part of the curve where
PM N pl. There are terms T * C in the sum (25) with T w pN = 10~ Q m.
The round-off error in such a term should be no more than N I o/o0 of pl.
Since the order of magnitude
of the largest coefficients is I, all coefficients
must be given with N 8 decimals. It is difficult to judge the absolute precision
of a two-fold numerical integration, and there is no guarantee that the decimals
given are accurate. However, we carry them all in our calculations, just to
reduce round-off errors.
The actual precision of the filter was tested in the same way as suggested
by Ghosh: The two functions
e3X
d) = P1 + b2 - d (I + e2x)3/2
(27)
and
T(x)

form a transform

PI

pair resembling

(~2 -

PI) exp

a two-layer

(--

exp

(-

4)

(28)

case.

Applying the filter to T yields a peonv to be compared with p. In fig. 2 the

result is shown for pl = I and p2 = IO. The error Ap = pconv - p is seen to
be very small, indeed.

8
0.00,

0.01

0.1

10

100

1000

Fig. 2. Accuracy test. peonv is the convolution

of the resistivity
transform
given in
equation (28) in the text, with pi = I and pi = IO. The other curve shows the difference
between peonv and the exact function pesaet given in equation (27). The relative error is
in this case less than one part in 106.
TABLE

The filter
decade.

i
-99
-98
-97
-96
-95
-94
-93
-92
-91
-90
-89
-88
.-87
-86

-45
-84
-83
-82

-81
-80
-79
-78
-77
-76
-75
-74
-73
-72
-71
-70
-69
-68
-67
-66
-65

coefficients

to a san@ling frequency of
Ax = (ln IO)/IO,
j, = -100,
jz = 40, S = --I.py,g458,
6.174~~ - 5, C!) = 6.0goglo - 4

C(j) . 108

-12484
12726
-12975
13231

-13494
13765
-14043
14330

-14625
14930
-15244
15567
-15901

16246
6602
16971
--I7352
17746

-I

-18154

18577
-19015
19469

-19941
20429

-20936
21463
-22oog

22577
-23166
23779

-24416
25079
-25768
26487
-27235

corresponding

-64
-63
-62
-61
-60
-59
-58
Y-57
-56
-55
-54
-53
-52
-51
-50
-49
-48
-47
-46
-45
-44
-43
-42
-41
-40
-39
-38
-37
-36
-35
-34
-33
-32
-31
-30

C(1) . 108
28016
-28830
29680
-30568
31496
-32467
33484
-34549
35666
-36838
38069
-39363
40724

ccn . 108
-110741
117248

-27
-26
-25
-24
-23

-124303

132085
-140461

-159826

-22
-21
-20

171917

-182946
199955

---I9

-209469
239052
-234543
304916
-234124
453990
-106745
899282
550573
2442523
3250077
7926675
13023345
25610307
41150741
64zy8og
72803988
36118538

-18

-17

-42156

-16

43666
-45259
46940
-48717
50596
-52587
54697
--56936
59314
-61845
64540
-67414
70484
-73767
77284
-81057
85111
-89475
94183
-99267
I04775

---I5
-14
-13
-12
-11
-10
1;
1:

-5
-4
-3
-2
-1

-100406442
0
I
2

3
4
5

-242172543
20052460
444506381
-489348908
wG%m398
-137791072

points per
C!-loo) =
C(l) . 10s

-29
-28

149959

IO

IO
II
12

13
14
15

16
17

18
19
20
21
22

23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39

61285163
--29362w
15817356
-9504597

6226174
-435i505
3198475
-2441493
1920840

-1548505
1273595

-1065148
903512
-775750

673079
-589375
520264
-462558
413891
-372478
336951

-306251
279543
-256168

235594
-217394
201216
---I86773

I73826
-162176
151657
-142126
133463
-125568

H. K. JOHANSEN

456

TABLE

Exewtion
no. of layers

times vewus mumber of layers for the CDC 6400.

0.174

IO

0.217

0.234

0.271

0.364

CP time (set)

0.191

When p2 is increased Ap grows proportional

to pi. Thus when the resistivity
ratio is p2/p1 = 104, the maximum
relative error is still smaller than IO-~,
which is satisfactory.
A comparison with the results of J. C. van Dam (1965) showed complete
agreement within the limits indicated above. The filter coefficients are given
in table I. The computation
time of the present method depends only on the
number of layers. Characteristic values for the CDC 6400 are given in table 2.
They are seen to be almost negligible.
APPLICATIONS

At our Laboratory
we have at our disposal a graphic display terminal
(Tektronix
4010) connected to the Aarhus University
computing
centers
CDC 6400 via telephone. We are using this facility in the following interactive way:
The measured resistivity values are displayed with error bars indicating
the estimated accuracy.
z) A (skilled) guess of the layer-parameters
is entered by the interpreter.
3) The corresponding p app-curve is computed and displayed.
4) Trial-and-error
adjustments
of the layer-parameters
are made until the
theoretical papp-curve agrees with the measured data.
I)

IO

100

AU2 1000 m

Fig. 3. An example of the Tektronix

4010 graphical display illustrating
the trial-anderror procedure. The inclined bars indicate the measured values i 5 %. The layer parameters for the three consecutive guesses are given in table 3.

INTERACTIVE

TERMINAL

SYSTEM FOR RESISTIVITY

INTERPRETATION

457

An example of this procedure is shown in fig. 3. The parameters are listed

in table 3.
It is sometimes desirable to let the computer
carry out the tedious
step 4 by an iterative least squares procedure. To this end the partial derivatives
of papp with respect to the layer-parameters
are needed. Recursive formulae
for them are given in the appendix.
TABLE

Layer parameters for the three cwves shown in

I

curve

fig. 3.

layer
no.
I
2

3
4

27

I5
40

20

160

15
40

160

3.5

3.5

27

I.5

160

21

3.5

APPENDIX
THE

PARTIAL

DERIVATIVES

OF T

In order to get a recursive relationship for the partial derivatives

ferentiate (13) : For convenience we will refer to the layer-parameters
meaning either pk or dk, in cases where it does not matter which one
talking about.
Consider first eq. (II).
We put a = exp (- zhd~-I),
b = kN-1
R = 4p~-1 * b/(1 + b)2.

we difas pk,
we are
a, and

Then
~TN-I

3PJJ
~TN-1
~
+N-1

- R * pit-l/(pN-1

= TN-JPN-I-R

+ pN121

(29)

* PN/(PN-1 -~PIv)~~

(31)
Since the pk for k = N, N we have

JTj
apk

3Tf
-

3Tj+1

~JT~+I
ajk

I, . .: . . j + I enter

k=N,N-I,

TJ only through

. . . . . j +I.

T~+I,

(32)

458

JOHANSEN,

INTERACTIVE

TERMINAL

SYSTEM

Also
aTj
--3d3 -

3T3
3w3

4P3a
(I +a+

(33)

where
a = exp (Putting

zd31).

I + W3T3+,/p~ = c, we have
JT3
-=-3P3

3T3 w3
a w3 p3 + 2(w3

+ T3 +I) W3T3 +I p3 - 3/c2

(34)

and
3T3
3T3+1

iJTi

__
aw3

= (I -

= (I -

w,/p;yc2

T;+Jp;)/c2

These expressions hold for j = N - 2, N - 3, . . . . . , 2, I; so that recursive

application gives the partial derivatives of T = TI.
Since the function C(X) in (IO) is independent
of the layer-sequence we
have simply :
(37)
where Tea-3) = T((i - j)Ax

+ S) and C(3) = C(jAx

S).

REFERENCES
GHOSH, D. P., 1971, The Application
of Linear Filter Theory to the Direct Interpretation
of Geoelectrical
Resistivity
Sounding Measurements,
Geophysical
Prospecting
rg,
192-217
GHOSH, D. P., rg7ra, Inverse Filter Coefficients for the Computation
of Apparent Resistivity Standard Curves for a Horizontally
Stratified Earth, Geophysical Prospecting
19, 769-775.
KOEPOED, O., 1968, The Application
of the Kernel Function in Interpreting
Geoelectrical
Measurements,
Geoexploration
Monographs,
Series I, No. 2, Gebriider Borntraeger,
Berlin-Stuttgart.
KOEFOED, O., 1970, A Fast Method for Determining
the Layer Distribution
from the
Raised Kernel Function, Geophysical Prospecting
18, 564-570.
KOEFOED, O., Igp, A Note on the Linear Filter Method of Interpreting
Resistivity
Sounding Data, Geophysical
Prospecting 20, 403-405.
VAN DAM, J. C., 1965, A Simple Method for the Calculation
of Standard Graphs to be
Used in Geo-electrical
Prospecting,
Geophysical Prospecting
13, 37-65.