Professional Documents
Culture Documents
Refraction Methods
Refraction Methods
Introduction
Seismic refraction has been well established for the investigation of the geomorphic
underground. Seismic refraction delivers information about the structure and the
velocity of the underground. However the quality of the results strongly depends on
the interpretation of the data and therefore on the methods used.
The standard refraction interpretation tools are well known. You may distinguish
between simple 1-dimensional direct inversion tools like the intercept time method,
more sophisticated 2-dimensional methods like the General Reciprocal Method
(GRM, Palmer, 1986) and forward modelling tools (raytracing methods, Cerven et
al., 1977). The recently used higher data coverage also allows to combine forward and
inversion tools within the so called refraction tomographic methods (Nolet, 1987).
In the following some of the different interpretation tools are shortly described and
the practical use of the tools for geomorphic refraction data will be discussed.
Figure 1: scheme of the seismic refraction basics, forward shot at A and receiver at
Y, reverse shot at B and receiver at X
Normally the interpretation of the seismic refraction data will be restricted to the
interpretation of the first arrivals. Additionally only the kinematic part, the
traveltimes of the refracted waves, are used for the standard interpretation. These
data deliver information about the depth and the velocity of the refractor R.
1
Picking the first arrivals
The first interpretation step consists of picking the traveltimes of the first arrivals.
This interpretation step must be performed for all subsequent interpretation
methods. You may distinguish between fully automatic, semi automatic and manual
picking.
Standard automatic picking algorithms can be distinguished between one and
multichannel algorithms. A onechannel algorithm (Hatherly, 1982) uses a method
consisting of a test-sequence based on the characteristics of expected signals specifying
a first-break when the sequence is successfully passed the first time. The multichannel
algorithm from Gelchinsky and Shtivelman (1983) is based on spatial correlation
properties of the signals.
Recently published papers mainly deal with backpropagation neural network
approaches (e.g. Chu and Mendel, 1994) whereby the network must be trained.
The automatic algorithms work well for noisy-free data. A network approach may
also give good results for noisy data but the training of the network may be quite
timeconsuming and difficult. It might be a good choice for large datasets with similar
signal characteristics.
In any case the automatic pick result must be manually inspected.
The best method for small noisy datasets is manual picking.
There are some problems to be considered when picks are made:
1. the same phase must be picked
2. polarity changes must accounted for
3. high velocity precursors (e.g. due to a small scale near surface inhomogeneity
of high velocity) may lead to an incorrect pick (see Figure 2).
4. often the vertical gradient for the uppermost layer is quite high. Therefore it is
sometimes necessary to follow the phase of the direct wave to larger distancies
than specified by the turn-over point (see Figure 3).
2
Figure 3: real seismic data showing a high vertical gradient within the uppermost
layer. On the left the interpreted velocity-depth distribution is shown with a
velocity increasing from 250 to 800 m/s within the uppermost layer
The picked traveltimes for all shots (Figure 4) form the database for the subsequent
interpretation.
3
2D-direct inversion methods wavefront reconstruction
The most famous and widely used representative of the 2D-direct inversion methods
is the General Reciprocal method (GRM) described by Palmer (1986). The method is
derived by geometrical considerations and many other well-known methods such as
timeterm-method, plusminus-method, intercept-time method and reciprocal
method. Because of the simple geometrical base there is a vast restriction of model
complexity.
The results of the backtracing process are the coordinates xR and zR of the refractor
and the traveltimes tAR and tBR of the backprojecting wavefronts. The new refractor is
automatically constructed at those points where the sum of the downward
traveltimes is equal to the reciprocal traveltime.
The refractor velocity vr is determined from the mean of the slopes of the forward
and reverse wavefronts at the new calculated refractor points.
4
The 2-dimensional direct inversion methods need an assignment of the picks to the
individual layers. If several shots shall be used for the inversion a combination of the
picks is necessary as well (see Figure 5).
Figure 5: combination of the assigned traveltime picks to one forward and reserve
traveltime branch for layer 2. The traveltime branches show some gaps which have
been filled by an interpolation resulting in significantly different forward and
reverse total traveltimes.
If gaps of the assigned traveltime picks due to large shot increments exist, an
interpolation is necessary for the combination of the traveltime picks into a complete
traveltime branch. Such an interpolation may lead to significantly different forward
and reverse total traveltimes which strongly influence the inversion result. Therefore
such an interpolation should be avoided especially within complex regions or a
manual interpolation should be done in such a way that the resulting forward and
reverse traveltimes match (within a few milliseconds).
In any case the inversion result should be input to forward modelling techniques.
Figure 6 shows an example of the wavefront inversion.
Figure 6: result of the wavefront inversion (upper panel) comparison of the real
data with the calculated ones based on the inverted model (lower panel), the total
timedifference is about 4.45 ms.
6
Forward modelling
Whereas the first application (the control of an inverted model) is a fast and easy to
handle process which should be used as a standard, the second application may be
very time consuming and requires experience. On the other hand, in some cases such
time intensive interactive forward modelling might be the only possibility to get a
reliable result especially with sparsely covered data.
The following figure shows the result of the interactive forward modelling. The
starting model was derived using the wavefront inversion method (see chapter
above). Some new layers as well as lateral velocity changes and vertical gradients
have been included. In comparison to the wavefront inversion result a much better
prediction of the real data was achieved especially within the distance range between
500 and 650 m.
7
Figure 7: result of the interactive forward modelling interpretation (upper panel)
comparison of the real data with the calculated ones based on the inverted model
(lower panel), a nearly perfect match with a total mean timedifference of about 0.86
ms was achieved.
8
Tomography
In the case of the 2D refraction vertical tomography all sources and receivers
are located along one line at the surface. In order to allow for high data coverage
within the medium vertical velocity gradients should be present and a curved
raytracing for the calculation of the traveltimes must be used. A standard approach
for the inversion is based on SIRT (simultaneous iterative reconstruction
technique). The curved rays are calculated using a finite difference approximation of
the Eikonal equation (see forward modelling). A start model must be defined. No
assignment of first break time to traveltime brancs (i.e. layers) is necessary.
The start model should contain a quite strong vertical velocity gradient and the max.
velocity variations should be large enough (e.g. 200 % of the original values) in order
to enable strong vertical gradients at those positions where an interface is assumed.
A smoothing in the horizontal direction is often useful because of the normally quite
large receiver increments.
Sometimes it is useful to force the first iteration to generate a new model even if the
resulting residuals are larger than for the starting model.
The following figure shows an example of the tomographic inversion for the same
dataset as for the wavefront inversion. The starting model was a simple lateral
homogeneous model with a velocity of 300 m/s and a vertical gradient of 100 1/s.
The wavefront-inversion result of Figure 6 is given for comparison (upper panel
dashed line). There is a good correspondence between the two results. The lower
panel of the Figure shows the observed traveltime curves together with the calculated
ones based on the inverted model. Such a comparison gives good quality control for
9
the tomography. The prediction of the real data is better than for the wavefront result
(2.42 ms in comparison to 4.46 ms for the wavefront inversion see Figure 6).
10
Summary
Different methods for the inversion of seismic refraction data have been discussed. It
has been shown that all methods are suitable for interpretation and that a
combination of the different methods (especially the additional use of forward
modelling) may give you a more accurate result and interpretation with respect to
minimizing the residuals.
Literature
Cerven, V., Molotkov, I.A. and Psencik, I., 1977, Ray method in seismology,
Universita Karlova, Praha.
Chu, C.-K.P., and Mendel, J.M., 1994, First-break refraction event picking using
fuzzy logic, IEEE Transactions on Fuzzy Systems, 2, 255-266.
Gelchinsky, B., and Shtivelman, V., 1983, Automatic picking of first-arrivals and
parametrization of traveltime curves, Geophysical Prospecting, 31, 915-928.
Hatherly, P.J., 1982, A computer method for determining seismic first-arrival times,
Geophysics, 47, 10, 1431-1436.
Moser, T.-J., 1991, Shortest raypath calculation of seismic rays, Geophysics, 56, 59-
67.
Palmer, D., 1986, Refraction seismics, in Helbig and Treitel (eds): Handbook of
geophysical exploration, Seismic Exploration, 13, Geophysical Press, London-
Amsterdam.
Thornburgh, H.R., 1930, Wavefront diagram in seismic interpretation, Bull. Am. Ass.
Petr. Geol., 14, 185-200.
11