You are on page 1of 18

75

Chapter 4
Starting Models For Refraction
Inversion
4.1 - Summary
The algorithms of the generalized reciprocal method (GRM) are applied to a set
of reversed traveltime data for a two layer model with a synclinal refractor
interface, in order to generate a family of starting models. Each starting model
shows much the same depths as the original model, but each has a narrow zone
in the refractor with an anomalous wavespeed. The traveltimes through each of
the starting models differ from those for the original model by less than a
millisecond. If any were used as starting models for tomographic or model-based
inversion, then the final result would show only minor differences. This example
demonstrates the non-uniqueness of model-based inversion.
In order to address the issues of non-uniqueness with model-based inversion, it
is recommended that a range of starting models, such as those which can be
generated with the GRM, be used.
Alternatively, other approaches, which aim to resolve these ambiguities, can be
employed. In many cases, the minimum variance criterion of the GRM can
resolve whether lateral variations in the refractor wavespeeds are genuine, or
whether they are artifacts of the inversion algorithm. In addition, it is proposed
76
that the amplitudes of the refraction convolution section can indicate where there
are genuine changes in the wavespeed of the refractor, because the amplitudes
are a function of the contrasts in wavespeeds between the refractor and the layer
above.
4.2 - Introduction
The inversion of seismic refraction data with model-based or tomographic
methods consists of deriving a starting model of the subsurface with standard
algorithms, and then testing it by comparing the computed traveltimes of the
model with the observed data. If there are differences, then the model is
adjusted until an acceptable agreement is achieved. Commonly, several
iterations may be required.
While most geophysicists are satisfied to generate a model which reproduces the
observations, there is the fundamental theoretical reality that an infinite number
of solutions can reproduce the data (Oldenburg, 1984; Treitel and Lines, 1988),
although not all of these solutions will be geologically plausible. This non-
uniqueness becomes more significant where the data are inaccurate and
incomplete as is often the case with field data, and where models, which do not
fit the data precisely, are accepted.
The issues of non-uniqueness are not usually considered with shallow refraction
tomography. The non-uniqueness includes the wavespeeds in both the
overburden and the refractor and they are often inter-related. One compelling
example is the somewhat paradoxical situation of the poor determination of
wavespeeds in the refractor, despite the fact that over 90% of traveltimes are
from that layer (Lanz et al, 1998, Figure 8). This situation is at variance with the
experiences of most seismologists using more traditional methods of refraction
77
processing, and it is probably related to the use of a linear wavespeed function
with a very high gradient in the upper layer.
Previous studies (Hagedoorn, 1955) have demonstrated the ambiguities in
determining the wavespeed stratification within a single layer above the refractor.
Even in the absence of undetected layers, generally known as hidden layers
within the blind zone and reversals in wavespeed, it is not possible to accurately
specify the mathematical function which describes the wavespeed in the
overlying layer. As a result, there is a large range in the depths to the refractor
computed with the various mathematical functions which can be fitted to the first
arrival traveltime data with acceptable accuracy.
Palmer (1992, Appendix 2) has demonstrated, that when the refractor interface is
sufficiently irregular in relation to its depth, the generalized reciprocal method
(GRM) (Palmer, 1980; 1986), can significantly improve the accuracy of the depth
computations for a wide range of mathematical functions in the upper layer. The
mathematical functions include wavespeed reversals and transverse isotropy,
which are not adequately addressed with other approaches.
This study examines the non-uniqueness in the determination of the wavespeeds
in the refractor. In these cases, the non-uniqueness is usually related to the
starting model for the inversion process and in turn, to the selection of the
inversion algorithm used to generate that model.
I demonstrate that a range of geologically plausible starting models can be
readily generated from the one set of reversed traveltime data with the algorithms
of the GRM, and that each of these models fits the data to an acceptable
accuracy of a few milliseconds. I conclude that the selection of the initial starting
model is critical with model-based methods of refraction inversion. I further
conclude that the issues of non-uniqueness, which currently are not adequately
examined with most model-based methods for inverting refracting data, can be
78
addressed by testing a family of starting models which can be generated for
example, with the GRM. Finally, I propose the use of two methods for resolving
ambiguities, namely, the minimum variance criterion of the GRM and the use of
amplitudes in the refraction convolution section.
4.3 - Inversion Of A Two Layer Model With The GRM Algorithms
Figure 4.1 shows a simple two layer model with isotropic homogeneous seismic
wavespeeds separated by a synclinal interface. It represents an obvious step for
increasing the complexity of the interpretation model over the simple two layer
case with plane interfaces. The dips of the sloping interfaces are 9.2, which
are relatively large. This model was used to generate the traveltime data shown
in Figure 4.2, which in turn were processed or inverted using the two algorithms
of the GRM for computing time-depths and refractor wavespeeds.
Figure 4.1: Two layer model with a synclinal interface.
The time-depth t
G
, at G is given by equation 4.1, viz.
t
G
= (t
AY
+ t
BX
- t
AB
- XY/V
n
)/2 (4.1)
79
where A, X, G, Y, and B are collinear, A and B are source points, X and Y are
detectors and G is midway between X and Y, t
AY
is the traveltime from A to Y, t
BX
is the traveltime from B to X, t
AB
is the reciprocal time, the traveltime from the
source at A to the source at B, and V
n
is the wavespeed in the refractor.
Figure 4.2: Traveltimes generated for two layer model with a synclinal interface
shown in Figure 4.1. The station spacing is 5 m.
Figure 4.3 shows the time-depths computed for XY values from zero to 30 m in
increments of 5 m, which is the detector spacing. Each set of time-depths shows
the synclinal structure of the refractor, although there are minor differences in
detail around the hinge point at station 12.
80
Figure 4.3: Time-depths computed for the synclinal model in Figure 4.1 for a
range of XY values. The reciprocal times have been systematically decreased
with increasing XY value, in order to separate each set of graphs for clarity.
The second function computed with the GRM is the refractor wavespeed analysis
function t
V
, given by equation 4.2, viz.
t
V
= (t
AY
- t
BX
+ t
AB
)/ 2 (4.2)
Two parameters can be derived from this function. The first is the wavespeed in
the refractor V
n
, from the reciprocal of the gradient, ie
d/dx t
V
= 1 / V
n
(4.3)
81
Figure 4.4: Wavespeed analysis function computed for the synclinal model in
Figure 4.1 for a range of XY values.
Figure 4.4 shows the wavespeed analysis function for the same range of XY
values used in Figure 4.3. Each set of graphs for a given XY value shows the
same wavespeed in the refractor of 2820 m/s, except for a short interval around
the hinge point at station 12. Here the wavespeed ranges from as low as 2000
m/s to as high as 4800 m/s.
The second parameter determined from the wavespeed analysis function is the
intercept of t
V
at the source point, which is the time-depth t
A
at a distance of XY
from the source point, ie
82
t
A
= t
V
|
x=0
(4.4)
For this two layer model, the time-depths presented in Figure 4.3 can be
converted into depths z
G
, with equation 4.5, viz.
z
G
= t
G
/ DCF (4.5)
where the DCF, the depth conversion factor relating the time-depth and the
depth, is given by:
DCF = V V
n
/ (V
n
2
- V
2
)

(4.6)
or
DCF = V / cos i (4.7)
V is the average wavespeed above the refractor and
sin i = V / V
n
(4.8)
Figure 4.5: A summary of the starting models which can be generated from the
traveltime data for the synclinal model in Figure 4.1. The region with the variable
wavespeeds near the hinge point of the interface is an artifact.
83
Figure 4.5 is a summary of the range of depth models which can be generated
with the XY values from zero to 30 m. Although the depth sections reproduce the
synclinal structure of the original model, there is an additional segment in the
second layer with wavespeeds from 2000 m/s to 4800 m/s which is not present in
the original model. This additional segment could represent a weathered dyke or
a shear zone for the low wavespeed cases or an unweathered dyke or a silicified
shear zone for the high wavespeed cases. Therefore, all models are geologically
both plausible and significant. Nevertheless, they are artifacts generated by
equation 4.2, the refractor wavespeed analysis algorithm.
4.4 - Time Differences Between Starting Models
Figure 4.6 shows the time-depths for the range of XY values from zero to 30 m
plotted without the vertical separation obtained by changing the reciprocal time
t
AB
, in equation 4.1. This presentation, which emphasizes the subtle variations
between different XY values, shows that the time-depth values are identical for
the planar sloping surfaces, but diverge by less than 2 ms in the vicinity of the
hinge point.
The smaller values are associated with the XY values which are less than the
optimum of 15 m and in turn are associated with the zone of lower wavespeeds
in the refractor. Although there is a slightly higher DCF computed with equation
4.6 in this narrow region, there is still a reduction in depth at the hinge point, and
in turn a small decrease in traveltimes in the upper layer. This decrease
approximately compensates for the slight increase in traveltimes in the region of
lower wavespeeds in the refractor.
The larger time-depth values are associated with the XY values which are
greater than the optimum of 15 m and in turn are associated with the zone of
84
higher wavespeeds in the refractor. Although there is a slightly lower DCF
computed with equation 4.6 in this narrow region, there is still an increase in
depth at the hinge point, and in turn a small increase in traveltimes in the upper
layer. This increase approximately compensates for the slight decrease in
traveltimes in the region of higher wavespeeds in the refractor.
Figure 4.6: Time-depths computed for the synclinal model in Figure 4.1 for a
range of XY values. The reciprocal times are identical for each XY value, and it
results in an emphasis of the subtle variations between different XY values.
Therefore, while there are small differences in depths at the hinge point, they are
matched with compensating changes in the wavespeeds in the refractor. The
85
final result is that there are few significant differences in the traveltimes in final
depth models.
Figure 4.7: Refractor wavespeed analysis function computed for the synclinal
model in Figure 4.1 for a range of XY values. The reciprocal times are identical
for each XY value.
Figure 4.7 is a similar presentation in which the wavespeed analysis function in
equation 4.2 is presented with identical reciprocal times for all XY values. Again
the aim is to emphasize the subtle variations between each set of values. It can
be seen that the maximum difference between the values computed for XY
86
values of zero and 30 m is 2.2 ms and that maximum difference between the any
set and those computed with optimum XY value of 15 m is less than 1.1 ms.
Figure 4.7 also demonstrates a fundamental problem in determining wavespeeds
in narrow intervals of the refractor with seismic refraction methods. The variation
in wavespeed in the refractor of 2000 m/s to 4800 m/s is very large and
geologically significant. However, these variations in wavespeed do not result in
commensurately large changes in traveltimes. An inspection of Figure 4.4 shows
that a single wavespeed can be fitted to each set of points with an accuracy of
better than a millisecond.
The significance of Figures 4.6 and 4.7 is that the time differences between each
model of the refractor computed with the selected range of XY values are subtle
and are generally within 1 ms of that computed with the optimum XY value of 15
m. These differences are typical of the acceptable residuals for most model-
based or tomographic methods of inversion.
4.5 - Agreement Between Starting Models And Traveltime Data
The small time differences between the various models as shown in Figures 4.6
and 4.7 suggest that each model should closely honor the original traveltime
data. Such a result is in fact the norm with the GRM, because the algorithms
seek to separate or analyze the traveltimes into the source point and detector
time-depths, together with the traveltime in the refractor, while still preserving the
original traveltime data. This is demonstrated by the simple addition of equations
4.1 and 4.2, viz.
t
AY
= t
G
+ t
V
+ XY/V
n
(4.9)
From equations 4.3 and 4.4, it is readily shown that
87
t
V
= t
A
+ AG/V
n
(4.10)
Equations 4.9 and 4.10 can be combined to obtain
t
AY
= t
G
+ t
A
+ AY / V
n
(4.11)
4.6 - Discussion
This study illustrates some of the inherent problems of non-uniqueness with
determining wavespeeds in the refractor. Using a simple model and the GRM
algorithms, it is possible to generate a family of starting models, each of which
has much the same depths to the refractor as the original model but each of
which includes a narrow zone in the refractor with an anomalous wavespeed.
Even with the noise-free model data used in this study, the time differences are
generally less than one millisecond, which is the error commonly assigned to the
measurement of traveltimes from field data, and which is within the range of
acceptable residuals for tomography. Therefore, if any were used as starting
models for tomography, then there would be minimal differences with the final
result of the inversion process. Furthermore, all of these models are geologically
meaningful and hence cannot be readily discarded.
(As an aside, geologically meaningless models can also be generated with the
GRM, simply by increasing the XY value in the wavespeed analysis function in
equation 4.2, until negative wavespeeds are obtained with equation 4.3.)
Frequently, the algorithms of the standard reciprocal method (SRM) (Hawkins,
1961), which is a special case of the GRM with a zero XY value, are used to
generate starting models. These algorithms are probably the most commonly
used throughout the world for shallow seismic refraction investigations, because
88
of their simplicity and robustness. These algorithms, which are also known as
the ABC method in the Americas (Nettleton, 1940; Dobrin, 1976), Hagiwara's
method in Japan (Hagiwara and Omote, 1939), and the plus-minus method in
Europe (Hagedoorn, 1959), can be viewed as simple extensions of the
slope/intercept method (Ewing et al, 1939), whereby computations are extended
from the source points to each detector location (Palmer, 1986).
However, this study demonstrates that the starting model generated by any
single method, such as an SRM analysis of the traveltime data need not
necessarily converge to the correct model. Therefore, in order to address the
issues of non-uniqueness, it is recommended that model-based methods of
inversion test a family of starting models such as those which can be readily
derived with the GRM.
The results over the Elura orebody (Hawkins and Whiteley, 1980) demonstrate
the significance of artifacts. The claim, that the massive sulphide orebody was
characterized by a low wavespeed, attracted considerable debate (Emerson,
1980), and was at variance with laboratory tests on hand specimens (MacMahon,
1980). An alternative analysis with the GRM indicated that the low wavespeed
was probably an artifact which coincided with an increase in the depth of the
regolith over the orebody (Palmer, 1980b). Many of the qualitative aspects of the
model study above can be recognized in the Elura case history.
The inability of model-based inversion methods to recognize artifacts can also
have important legal implications. There are instances where the combination of
the SRM and ray tracing is a contractual requirement of major geotechnical
investigations, in order to obviate claims for compensation by construction
companies for unexpected variations in site conditions.
Numerous model studies and case histories (Palmer, 1980; Palmer, 1986,
Palmer, 1991) demonstrate that the minimum variance criterion of the GRM is
89
frequently able to resolve whether lateral variations in wavespeeds in the
refractor are genuine or are artifacts. For the model study above, the wavespeed
analysis function in Figure 4.4, shows that the optimum XY value of 15 m, that
the measured wavespeed is same as the model and that no artifacts are
generated.
However, the effective application of the GRM is not always possible, often
because the detector interval is too large. In these cases, alternative methods
are required. Other studies demonstrate that the amplitudes of the refraction
convolution section (Palmer, 2001a; Chapter 5) can indicate where there are
genuine changes in the wavespeed of the refractor. These amplitudes are a
function of the contrasts in wavespeeds between the refractor and the layer
above, and therefore provide another approach which is independent of the
traveltime data.
4.7 - Conclusions
The inversion of seismic refraction data with model-based methods or
tomography consists of deriving a starting model of the subsurface with standard
algorithms, and then testing it by comparing the computed traveltimes of the
model with the observed data. If there are differences, then the model is
adjusted until an acceptable agreement is achieved. Commonly, several
iterations may be required.
However, a simple model study illustrates the inherent problems of non-
uniqueness with this approach. The GRM is able to generate a family of starting
models, all of which are geologically meaningful and all of which are compatible
with the original traveltime data. If any were used as starting models for
tomography, then there would be minimal differences with the final result of
inversion.
90
In view of the significance of the starting model, it is recommended that model-
based methods of inversion test a range of starting models such as those which
can be readily generated with the GRM. In general, the models which can be
derived with the GRM tend to be compatible with the original traveltime data.
In many cases, the minimum variance criterion of the generalized reciprocal
method (GRM) can resolve whether lateral variations in the refractor wavespeeds
are genuine, or whether they are artifacts of the inversion algorithm.
In those cases where the effective application of the GRM is not possible, then
alternative methods are required. It is proposed that the amplitudes of the
refraction convolution section (Palmer, 2001a; Chapter 5) frequently can indicate
where there are genuine changes in the wavespeed of the refractor.
4.8 - References
Dobrin, M. B., 1976, Introduction to geophysical prospecting, 3rd edn.: McGraw-
Hill Inc.
Emerson, D. W., 1980, The geophysics of the Elura orebody, Cobar, NSW: Bull.
Aust. Soc. Explor. Geophys., 11, 347.
Ewing, M., Woollard, G. P., and Vine, A. C., 1939, Geophysical investigations in
the emerged and submerged Atlantic Coastal Plain, Part 3, Barnegat Bay, New
Jersey section: Bull. GSA, 50, 257-296.
Hagiwara, T., and Omote, S., 1939, Land creep at {Mt} {Tyausa-Yama}
(Determination of slip plane by seismic prospecting): Tokyo Univ. Earthquake
Res. Inst. Bull., 17,118-137.
91
Hagedoorn, J. G., 1955, Templates for fitting smooth velocity functions to seismic
refraction and reflection data: Geophys. Prosp., 3, 325-338.
Hagedoorn, J. G., 1959, The plus-minus method of interpreting seismic refraction
sections: Geophys. Prosp., 7, 158-182.
Hawkins, L. V., 1961, The reciprocal method of routine shallow seismic refraction
investigations: Geophysics, 26, 806-819.
Hawkins, L. V., and Whiteley, R. J., 1980, The seismic signature of the Elura
orebody: Bull. Aust. Soc. Explor. Geophys., 11, 325-329.
Lanz, E., Maurer H., and Green, A. G., 1998, Refraction tomography over a
buried waste disposal site: Geophysics, 63, 1414-1433.
MacMahon, B. K., 1980, Discussion in Emerson, D. W., ed., The geophysics of
the Elura orebody, Cobar, NSW: Bull. Aust. Soc. Explor. Geophys., 11, 346.
Nettleton, L. L., 1940, Geophysical prospecting for oil: McGraw-Hill Book
Company Inc.
Oldenburg, D. W., 1984, An introduction to linear inverse theory: Trans IEEE
Geoscience and Remote Sensing, GE-22(6), 666.
Palmer, D., 1980a, The generalized reciprocal method of seismic refraction
interpretation: Society of Exploration Geophysicists.
Palmer, D., 1980b, Comments on "The seismic signature of the Elura orebody":
Bull. Aust. Soc. Explor. Geophys., 11, 347.
92
Palmer, D., 1986, Refraction seismics - the lateral resolution of structure and
seismic velocity: Geophysical Press
Palmer, D., 1991, The resolution of narrow low-velocity zones with the
generalized reciprocal method: Geophys. Prosp., 39, 1031-1060.
Palmer, D., 1992, Is forward modelling as efficacious as minimum variance for
refraction inversion?: Explor. Geophys., 23, 261-266, 521.
Palmer, D., 2001a, Resolving refractor ambiguities with amplitudes: Geophysics
66, 1590-1593.
Palmer, D., 2001b, Model determination for refraction inversion: Geophysics,
submitted.
Treitel, S., and Lines, L., 1988, Geophysical examples of inversion (with a grain
of salt): The Leading Edge, 7, 32-35.