Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Journal of Applied Geophysics 42 Ž1999.

35–45

Automatic 1D interpretation of DC resistivity sounding data


E.A. Muiuane ) , L.B. Pedersen
¨ 16, S-75236 Uppsala, Sweden
Department of Geophysics, Uppsala UniÕersity, VillaÕagen
Received 27 May 1998; accepted 16 April 1999

Abstract

In the last few years, much of the work carried out on DC resistivity has mainly concentrated on 2D and 3D techniques
for data acquisition and interpretation. However, when the resistivity changes are smooth, 1D techniques can be used to
interpret the data. In the present paper, we present an automatic 1D inversion scheme for DC resistivity sounding. The
inversion scheme is based on an iterative least squares procedure ŽILSQP. with Singular Value Decomposition ŽSVD.. In
order to reinforce the convergence of the inversion scheme towards a global minimum, the ILSQP is combined with a
logarithmic parameterization of the unknown model parameters, and splitting of the data set into parts. The data are then
inverted stepwise, i.e., starting from the first data part which corresponds to the smallest electrode spacings. As the inversion
continues more data parts, corresponding to increasingly larger electrode spacings, are included in the inversion until the
whole data set is inverted. In this way, the shallower part of the model becomes well estimated first, and as the inversion
advances and more data parts are included, the deeper structure is better resolved. The solution of the resistivity inverse
problem by standard least squares procedure with SVD allows us to distinguish between well and poorly resolved linear
combinations of model parameters. For a given inversion step, the construction of models that give a better data fit can be
done with truncation of eigenvectors belonging to the least resolved combinations. Thus, as the iteration process advances
those singular values can be activated one by one. This approach has been tested on synthetic data representing some
layering that is assumed common in shallow studies. The model studies suggest that there is a strong coupling between the
resolving power of DC resistivity data and their random errors. For high to moderate data quality, the resolution power of
the scheme is generally good, depending upon the degree of complexity of the model. For larger error levels, however, there
is a loss of resolution. The inversion scheme is suitable for fast interpretation of data that are collected in shallow studies in
connection with environmental, hydrogeological and geotechnical investigations if the lateral changes in the study area are
smooth. The method may also be used as a first interpretation prior to a 2D or 3D survey. q 1999 Elsevier Science B.V. All
rights reserved.

Keywords: Electrical resistivity; Automatic inversion; Data splitting

1. Introduction Že.g., Langer, 1933; Slichter, 1933. . From that


time until the late 1980s, the methodology of
The DC resistivity inverse problem was in-
field survey and character of data from the
vestigated for the first time in the early 1930s
measurements remained unchanged. From the
late 1980s and early 1990s until today, there has
)
Corresponding author. Fax: q46-18-501110; E-mail: been considerable improvement in data collec-
em@geofys.uu.se tion and interpretation.

0926-9851r99r$ - see front matter q 1999 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 6 - 9 8 5 1 Ž 9 9 . 0 0 0 1 5 - 4
36 E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45

In the field of data collection, automatic mea- and a description of the real world in terms of
suring systems using computer controlled smooth models is thus not quite appropriate.
multi-electrode arrays are becoming increas- Smooth models will fit the data equally well as
ingly popular Ž Van Overmeeren and Ritsema, models with a few distinct layers.
1988; Sorensen and Sorensen, 1995. . Major ad- Linearized iterative solutions of the 1D geo-
vantages of these systems over conventional DC electrical problem requires an estimate of the
resistivity systems are that with a single cable initial solution for starting the iterations. How-
layout, soundings and profilings over a profile ever, this might cause the solution to become
defined by the electrode array can be simultane- somewhat dependent on the initial guess. There-
ously performed by combining different elec- fore, research and development is directed to-
trodes. Moreover, measurements can be done ward inversion algorithms which are capable of
over an area of complex geology. resolving the pertinent structures regardless of
On the other hand, over the same period, the the initial input parameters or automatically.
attitude towards the interpretation technique has The approach developed by Zohdy Ž1989.
changed profoundly. An excellent review work was one step forward in this regard. In Zohdy’s
on the developments in the interpretation tech- inversion scheme a layered model is obtained
niques of DC resistivity sounding data is given directly from a digitized sounding curve without
by Christensen Ž 1986. . The development of the defining a preliminary guess of model parame-
interpretation technique of DC resistivity sound- ters. Dahlin Ž1993. applied this approach to
ing was profoundly marked by three events; interpret large amounts of resistivity data.
namely Ž1. the linear filter theory Že.g., Ghosh, In the present work, we apply a modified
1971a,b; Christensen, 1979., Ž2. the widespread version of standard iterative least squares proce-
use of digital computers and Ž 3. the application dures ŽILSQPs., which includes a logarithmic
of general linear inverse theory. With the latter, reparameterization of the unknown model pa-
the concept of automatic inversion and analysis rameters and a search strategy in parameter
became popular, in which computer programs space in order to reinforce the iteration proce-
resulting from this theory were able to find the dure. Automation of DC resistivity data inter-
best fitting model automatically, provided they pretation is essential considering its increasing
were given a reasonably accurate initial model application in shallow studies in connection with
andror a priori information Ž e.g., Johansen, environmental, hydrogeological and geotechni-
1977; Jacobsen, 1982. . In MT inversion for cal investigations.
example, the most commonly used algorithms
in the automatic layered inversion are those of
Constable et al. Ž 1987. using a smoothened 2. Outline of the method
layered model and the Fischer scheme Ž Fischer
et al., 1981. for constructing models with a The resistivity inverse problem is non-linear
small number of layers. and as such, is usually solved using ILS-
However, it can be argued that real earth QPs which perform the quantitative adjustment
structure, especially in sedimentary environ- of the model parameters Ž r i ,d i . and yield the
ments, often is layered with rather well-defined optimum parameter set Ž r io ,d io . minimizing
resistivities that can be well differentiated from the weighted sum of squares of data residuals
resistivities of other layers, even though each ŽJohansen, 1977. .
formation may show variations on a variety of As stated above, the robustness of any inver-
scales. Observations from borehole investiga- sion scheme based on linearized iterative solu-
tions suggest that electrical resistivity often tions is highly dependent upon the strategy of
varies abruptly from one formation to the next, defining the initial model for the iteration pro-
E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45 37

cess to start as well as the search strategy in thicknesses to stay within predefined limits Ž Eqs.
parameter space. In the next sections, we de- Ž1a. and Ž 1b...
scribe in detail each of several steps followed in rm y rmin
our approach. x m s log Ž 1a .
rmax y rm
and
2.1. Stripping the earth from top to bottom d m y d min
xUm s log Ž 1b.
d max y d m
The usual strategy in linearized iterative solu-
tions of the 1D geoelectrical problem is to start where rm and d m are the resistivities and thick-
with an estimated solution and to solve the nesses, respectively, and x m , xUm are the corre-
forward problem to obtain the predicted data. A sponding parameters in the logarithmic space.
perturbation is sought which, when added to the With this parametrization, it is ensured that all
estimated solution, yields a model which repro- model parameters will fall within their bounds,
duces the observed data. As pointed out above, i.e., rmin - rm - rmax and d min - d m - d max . The
this causes the solution to become somewhat bounds should not be chosen too close but also
dependent on the initial guess, and in some not too far from the true range of parameters. In
cases leads to instability in the iteration process. such cases, it may happen that convergence is
We have found a very simple and practical way impeded and bad data fit may result.
to circumvent this problem.
We divide the total data set into subsets, 2.3. Strategy for finding initial models
starting with the smaller electrode spacings
ŽABr2’s., i.e., resolving the upper part of the The initial model for the start of inversion is
found directly from the data, i.e., from the first
model first. When a best fitting model is found
data part. It is taken as a uniform halfspace
the next data segment is added and the iteration
whose resistivity is defined as follows:
process continues by using the previous best
fitting model as a starting model. As we will 1 n
x 1 s Ý yimeas Ž2.
demonstrate in Section 3, the splitting approach n is1
gives stability to the inversion procedure.
where
y meas s log rap
meas
.
2.2. Reparameterization of model parameters
The best fitting uniform halfspace is then
used to define a new initial model with two
We reparameterize the unknown model pa- layers. In the initial phase, the resistivities of the
rameters in logarithmic space. Logarithmic pa- upper layer and of the underlying homogeneous
rameters are advantageous in the sense that halfspace are set at the same value. The layer
changes can easily be constrained to be less boundary is chosen to be midway between the
than a specified threshold and thus, preventing maximum and minimum values of ABr4’s in a
them to become too large or small. Another logarithmic sense ŽEq. Ž3...
benefit of using logarithmic parameters is that
xU1 s Ž z max y z min . r2 Ž3.
only positive values of resistivities and thick-
nesses will be allowed. in which
Current approaches may introduce layers that z min s log Ž ABr4. min
during the iteration process become extremely
and
thin with extreme resistivities. The logarithmic
parameterization will force all resistivities and z max s log Ž ABr4. max .
38 E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45

ABr4 is the estimated depth of investigation important aspect of the inversion scheme is to
ŽBarker, 1989. . The best fitting two-layer model find models that are continuously giving im-
is subsequently used as the initial model with proved fits without being trapped in local min-
three layers. The thickness of the extra layer ima, and that such initial models for a given
Žlayer number two. is taken to be one third of number of layers and data already fit the data
the total logarithmic thickness. This process can well. The output of the ILSQP consists of the
be repeated as many times as needed for calcu- optimum parameter set Ž r io ,d io . minimizing the
lating a best fitting n-layer model. weighted sum of squares of residuals, with the
For the general case of an m-layered earth, weights as estimated errors Ž Eq. Ž 5.. .
the definition of new initial models is done by n < yimeas y yipred < 2
1
using the previous best fitting model and in- Qs Ý Ž5.
creasing the number of layers by one. The nym is1 si2
thickness of the extra layer, i.e., layer number where
m y 1 is defined as:
y meas s log rap
meas
, y pred s log rappred ,
xUmy1 s Ž z max y z min . rm. Ž4.
n and m are the number of data and model
The inversion scheme works in such a way parameters, respectively, n y m is the number
that data fit with an extra layer is always better of degrees of freedom and si is the standard
than or equal to the previous best fitting model error on measurement number i. Another impor-
without the extra layer since the previous best tant aspect of this approach is that the addition
fitting m-layered model has exactly the same of an extra layer in the final stage of the con-
resistivity–depth distribution as the new initial struction process described above may, in some
Ž m q 1.-layered model. cases, lead to estimates of model standard devi-
For a given number of layers, convergence of ations that are exceedingly high. The final model
the iteration process to improve the fit is rein- is found if either of the following conditions is
forced by initially choosing the number of de- met.
grees of freedom Ži.e., the number of singular Ž Ø . The addition of an extra layer does not
values in the Singular Value Decomposition improve the data fit defined by Eq. Ž5. . In this
ŽSVD. of the Jacobian matrix. to be equal to case, the previous best fitting model without the
that of the previous m-layered model, i.e., 2 m extra layer will be taken as the final model.
y 1 and incrementing that number in steps of Ž Ø . The addition of an extra layer leads to
one to the maximum number of degrees of one singular value being below a given thresh-
freedom of the present model, i.e., 2 m q 1. old, thus leading to model standard deviations
that are too large to be of practical use, while at
2.4. The final model the same time, the data fit is improved signifi-
cantly. In this case, the model with one very
The solution of the resistivity inverse prob- small singular value will be taken as the final
lem by standard least squares procedures with model.
SVD allows us to distinguish between well and
poorly resolved linear combinations of model
parameters. For a given inversion step, the con- 3. Illustration of the inversion scheme
struction of models that give a better data fit can
be done with truncation of eigenvectors belong- 3.1. General
ing to the least resolved combinations, and as
the iteration process advances, those singular In order to illustrate the principles and merits
values can be activated one by one. The most of the inversion scheme, two models with some
E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45 39

ideal resistivity distributions of geological rele- to test the sensitivity of the inversion under field
vance were chosen. These models represent two conditions. The noise levels investigated were 2
special cases in the application of the DC resis- and 5%. According to Dahlin Ž1993., the 2%
tivity technique. In general, models of type A noise level is reasonable for areas with small
Žbelow. are not difficult to resolve, whereas the resistivity contrasts and relatively low resistivi-
resolution of those of type B is more difficult ties at the surface, whereas, the 5% noise level
and the convergence is slow. In addition, we is typical for sites with high resistivity contrasts
consider two more complex models consisting and higher electrode contact resistances.
of five layers. In all cases, the layer thicknesses Throughout this study the bounds of the model
are distributed logarithmically with a density of parameters were set at the following values:
1 per octave. rmin s 1 V m; d min s 1 m; rmax s 1000 V m;
d max s 50 m.
Model A. Increasing resistiÕity with depth
r 1 s 30 V m d1 s 4 m
3.2. Modelling results
r 2 s 100 V m d2 s 8 m
r 3 s 500 V m
3.2.1. Model A. Increasing resistiÕity with depth
(ideal case for DC resistiÕity technique)
Model B. Resistor embedded between two con-
Models with increasing resistivity with depth
ductors
are quickly resolved with DC resistivity tech-
r 1 s 30 V m d1 s 4 m
nique. We included a model of this type in
r 2 s 500 V m d2 s 8 m
order to investigate the performance of this
r 3 s 100 V m
approach for the simplest case. A 2% Gaussian
noise is added to the synthetic data. In the initial
Model C. FiÕe-layer model (resistiÕity increa-
stage, the whole data set was inverted in one
sing with depth with a resistiÕe upper layer)
step, i.e., without data splitting. Fig. 1 shows
r 1 s 100 V m d1 s 4 m
r 2 s 50 V m d2 s 8 m
r 3 s 100 V m d 3 s 16 m
r4 s 250 V m d 3 s 32 m
r 5 s 500 V m

Model D. FiÕe-layer model (resistiÕity decrea-


sing with depth)
r 1 s 500 V m d1 s 4 m
r 2 s 250 V m d2 s 8 m
r 3 s 100 V m d 3 s 16 m
r4 s 50 V m d 3 s 32 m
r 5 s 30 V m

The model responses were calculated using


SELMA ŽSimultaneous Electromagnetic Mod-
elling and Analysis. by Christensen and Jacob-
sen Ž1997. . The data set consists of 30 data
points distributed logarithmically with a density Fig. 1. Model A. Resistivity increasing with depth; 2%
of 10 per decade. Gaussian distributed noise Gaussian noise added to the data. Resistivity–depth distri-
was added to the calculated responses in order bution, no data splitting.
40 E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45

the result of inversion. The data fit is quite good sion results. Therefore, it was not investigated
Žlog Q s y0.1.. The derived model parameters any further.
are close to their true values except the resistiv-
ity of the intermediate layer which is slightly 3.2.2. Model B. Resistor embedded between two
smaller than its true value. For this particular conductors
case of resistivity–depth distribution, the data The inversion starts with a 2% Gaussian noise
splitting Žnot shown. apparently does not pro- data set with a splitting interval of 1 decade per
duce any significant improvement in the inver- data segment, i.e., 10 data points per data seg-

Fig. 2. Model B. Resistor embedded between two conductors; 2% Gaussian noise added to the data. Ža. Apparent resistivity
curve for the best three-layer model determined by the inversion, data split into three parts. Žb. Resistivity–depth
distribution; final model, data split into three parts. Žc. Resistivity–depth distribution; data split into six parts.
E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45 41

ment. The results of inversion are shown in Fig. result of the inversion is shown in Fig. 3. The
2a,b,c. Fig. 2a shows the curve of apparent data set is split into three parts. As in the
resistivity vs. ABr2. The inversion starts with a previous example, we can note that the inver-
homogeneous halfspace Ž dashed horizontal line. sion starts with a homogeneous half space with
with a resistivity of ca. 40 V m and the first a resistivity of ca. 40 V m and the first data
data part is fitted. When the best fitting model is part is inverted by improving this model andror
found the second data part is added and the adding more layers. After the best fitting model
inversion continues by adjusting the model pa- is found, the next data part is added and the
rameters andror adding more layers. This pro- inversion continues as explained previously un-
cess is repeated until the whole data set is til the whole data set is inverted. The data fit
inverted. The data fit is quite good. In Fig. 2b, Žlog Q s 0.06. is slightly above the threshold
the resistivity–depth distribution is shown. but still acceptable. However, the parameter
Shown are also the initial model, the intermedi- resolution is poor. Although the resistivities of
ate initial models with two layers and the final the upper and lower homogeneous half spaces
three layer model. We notice that the resolved can be resolved perfectly, the thickness and the
model parameters are quite close to the true resistivity of the intermediate resistive layer de-
ones. viates considerably from that of true model.
In the next step, we decrease the splitting This well-known high resistivity equivalence is
interval to half a decade per data segment, i.e., an intrinsic problem of this model.
five data points per decade. The resulting resis- By inverting the data set with a splitting
tivity–depth distribution is shown in Fig. 2c. interval of half a decade, no further significant
We can see that there are no major differences improvements are noticed in the inversion re-
between this and the previous result ŽFig. 2b.. sults Žnot shown..
The only difference is that the resistivity of the
initial model Ž homogeneous half space. in this
case is slightly smaller than in the previous
example with three data parts.
These results suggest that the data splitting
approach is a fast and stable procedure. This
might be due to the fact that in this approach the
data segments are smaller and, therefore, the
model construction strategy as described above
produces initial models which are not too far
from the true ones for the corresponding data
part. On the other hand, by inverting the whole
data set in one step, the starting model is too
poor so that divergence occurs.
We can conclude that by splitting the data set
into parts, the iteration process will be stable
and, therefore, the convergence to a global min-
imum will be faster. We also see that a splitting
interval of 1 decade or less can be taken as a
reasonable rule of thumb, since the produced
inversion results ŽFig. 2b and c. are comparable. Fig. 3. Model B. Resistor embedded between two conduc-
In the following, we shall see the effect of tors; 5% Gaussian noise added to the data, data split into
adding 5% Gaussian noise to the data. The three parts. Resistivity–depth distribution.
42 E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45

3.2.3. Model C. FiÕe-layer model (resistiÕity model. Fig. 4a shows the curve of apparent
increasing with depth with a resistiÕe oÕerbur- resistivity data vs. ABr2. The data fit is quite
den) good. The derived resistivity depth distribution
For this case, we will show only the results is shown in Fig. 4b. The parameters of the first
of the inversion when the data set Ž 2% Gaussian two layers are almost perfectly resolved. How-
noise. is split into six parts, i.e., five data points ever, the remaining ones deviate considerably
per data segment. For larger splitting intervals, from those of the true model except the resistiv-
e.g., 1 decade per data segment, the inversion ities of the fourth layer and the underlying half
scheme is not able to resolve the five-layer space which are close to being optimal.

Fig. 4. Model C. Five-layer model Žresistivity increasing with depth with a resistive overburden.. Ža. Apparent resistivity
curve for the best five-layer model determined by the inversion, data split into six parts. Žb. Resistivity–depth distribution,
data split into six parts.
E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45 43

Fig. 5. Model D. Five-layer model Žresistivity decreasing with depth.. Ža. Apparent resistivity curve for the best five-layer
model determined by the inversion, data split into three parts. Žb. Resistivity–depth distribution, data split into three parts.

3.2.4. Model D. FiÕe-layer model (resistiÕity tervals, e.g., half a decade Žnot shown., do not
decreasing with depth) show any significant improvement.
The results of inversion when the data set
Ž2% Gaussian noise. is split into three parts
Žone decade per data segment. are shown in Fig. 4. Discussion and conclusions
5a and b. The data fit is perfect but the five-layer
structure is not recovered. However, some model A simple approach to 1D automatic inversion
parameters are perfectly resolved, e.g., the resis- of DC resistivity sounding data has been pre-
tivity of the upper layer and the lower halfs- sented. In the present inversion scheme, we use
pace. Inversion results with shorter splitting in- the standard least squares procedures with SVD
44 E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45

and reinforce convergence by applying other setting local bounds on each layer. This possi-
measures of robustness such as splitting up of bility was not explored in the present work.
the data set and logarithmic parameters with The comparison of inversion results with dif-
bounds. The bounds to be set in the inversion ferent models suggests that the performance of
process should be chosen not too close nor too the present algorithm is model dependent.
far from the expected values otherwise, geologi- The present 1D inversion scheme is very fast
cally unrealistic results may result. and ease to implement. Therefore, it can be used
The initial model is found directly from the for fast surveys in areas where lateral variation
data. The inversion starts with a uniform half is slow or as a first interpretation tool where the
space whose resistivity is the logarithmic aver- density of data is insufficient for a 2D or 3D
age of the apparent resistivity for the initial part. interpretation.
We have found the general rule that in order the
present algorithm to be successful, the data
splitting interval should be 1 decade or less. The References
optimum splitting interval seems to depend on
the type of resistivity–depth distribution perti- Barker, R.D., 1989. Depth of investigation of collinear
nent to each case as is apparent from the results symmetrical four electrode arrays. Geophysics 54,
of the models C and D. The no splitting case in 1031–1037.
many cases fails to yield reasonable results Christensen, N.B., 1979. Fast Hankel Transforms.
because the construction of the initial models GeoSkrifter, 12, Aarhus University, 68 pp.
Christensen, N.B., 1986. The geolectrical sounding method;
involves the whole data set, and therefore, poor
a historical review of interpretational techniques and
initial models may result. Poor initial models interpreter’s evaluation of the method. In: Moller, J.T.
generally, lead to instabilities in the inversion ŽEd.., Twenty-five Years of Geology in Aarhus, Geo-
process. logical Essays, ISBN 8787529483. GeoSkrifter 24,
The model construction strategy is closely Aarhus University, pp. 91–102.
related to the use of the SVD algorithm, whereby Christensen, N.B., Jacobsen, B.H., 1997. Simultaneous
Electromagnetic Layered Modeling and Analysis
the maximum number of layers can be clearly ŽSELMA.. Instructions Manual, Department of Earth
identified by the occurrence of very small singu- Sciences, University of Aarhus.
lar values. When this happens no more layers Constable, S.C., Parker, R.L., Constable, C.G., 1987. Oc-
are introduced until eventually, during the stage cam’s inversion: a practical algorithm for generating
of adding more and more data to be inverted, smooth models from electromagnetic data. Geophysics
52, 289–300.
the smallest singular value goes above a given
Dahlin, T., 1993. On the automation of 2D resistivity
threshold. surveying for engineering and environmental applica-
We have seen that this approach works well tions. PhD thesis, Department of Engineering Geology,
for data of good to moderate quality. For noisy Lund Institut of Technology.
data, e.g., 5%, there is a degradation of parame- Fischer, G., Schnegg, P.A., Peguiron, M., LeQuang, B.V.,
ter resolution. 1981. An analytic one-dimensional magnetotelluric in-
version scheme. Geophys. J. R. Astron. Soc. 67, 257–
Provided that the input parameters, such as 278.
the lower and upper limits of the model parame- Ghosh, D.P., 1971a. The application of linear filter theory
ters are chosen properly, the present inversion to the direct interpretation of geoelectrical resistivity
scheme can yield a layer sequence agreeing sounding measurements. Geophysical Prospecting 19,
with the measured data as well as with the 192–217.
Ghosh, D.P., 1971b. Inverse filter coefficients for compu-
geological concept. Notice that in our approach,
tation of apparent resistivity standard curves for hori-
we use global bounds. In some cases, it might zontally stratified earth. Geophys. Prospect. 19, 769–
be possible to have estimates of the true param- 775.
eters. This would, possibly, give us a chance of Jacobsen, B.H., 1982. Inversionsteori. Grundlag, teknik og
E.A. Muiuane, L.B. Pedersenr Journal of Applied Geophysics 42 (1999) 35–45 45

anvendelse. GeoKompendier No. 19, Aarhus Univer- Sorensen, K.I., Sorensen, K.M., 1995. Pulled array contin-
sity, 257 pp. uous vertical electrical sounding ŽPACVES.. Proceed-
Johansen, H.K., 1977. A manrcomputer interpretation sys- ings of the SAGEEP’95, Orlando, FL, pp. 893–897.
tem for resistivity soundings over horizontally stratified Van Overmeeren, R.A., Ritsema, I.L., 1988. Continuous
earth. Geophys. Prospect. 25, 667–691. vertical electrical sounding. First Break 6, 313–324.
Langer, R.E., 1933. An inverse problem in differential Zohdy, A.A.R., 1989. A new method for automatic inter-
equations. Am. Soc. Math. J. 39, 14–28. pretation of Schlumberger and Wenner sounding curves.
Slichter, L.B., 1933. The interpretation of the resistivity Geophysics 54, 245–253.
prospecting method for horizontal structures. Physics 4,
307–322.

You might also like