Geophys. J. R. astr. SOC.

(1985) 82,81-103

Inversion of seismic refraction data in planar dipping

structure

B . Milkereit Institute f o r Geophysics, university of Kiel, Kiel, West Germany

w. D. Mooney and W. M. Kohler us Geofogical Survey,

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

345 Middlefield Road, Mail Stop 77, Menlo Park, CA 94025, USA

Accepted 1984 December 26. Received 1984 December 26; in original form 1984
March 21

Summary. A new method is presented for the direct inversion of seismic

refraction data in dipping planar structure. Three recording geometries, each
consisting of two common-shot profiles, are considered: reversed, split, and
roll-along profiles. Inversion is achieved via slant stacking the common-shot
wavefield to obtain a delay time-slowness (tau-p) wavefield. The tau-p
curves from two shotpoints describing the critical raypath of refracted and
post-critically reflected arrivals are automatically picked using coherency
conjuntamente
measurements and the two curves are jointly used to calculate velocity and
de este modo
dip of isovelocity lines iteratively, thereby obtaining the final two-dimensional
velocity model.
This procedure has been successfully applied to synthetic seismograms
calculated for a dipping structure and to field data from central California.
estrechamente-espaciados
The results indicate that direct inversion of closely-spaced refraction/wide-
logrado
aperture reflection data can practically be achieved in laterally inhomogeneous
structures.

Introduction
Investigations of the deep structure of the continental crust commonly utilize the seismic
mientras bastante
refraction method. Whereas the gross velocity structure ofesfuerzo
han sido en aumento
the crust is rather well determined
in many regions, there has recently been an increasing effort to obtain higher resolution of
the crustal structure in regions of geologic importance, such as fault zones, geothermal areas
emparejado por
and other areas of possible exploration interest. This interest has been matched by techno-
mayor que
logical advances that have made it possible to deploy long (1 0 km and greater) seismic arrays
with large numbers of sensors at an interstation spacing measured in tens of metres. Such
de este modo
profiles contain both near-vertical and wide angle reflected waves and refractions, thereby
representación
rendering obsolete the traditional distinction between ‘reflection’ and ‘refraction’ seismology.
densificar
Denser sampling of the wavefield has also made possible the application of new automated
data interpretation procedures t o what was previously a trial and error forward modelling
process.
82 B. Milkereit, W. D. Mooney and W. M. Kohler
To date. many of the new procedures have been based on the transformation of the
observed wavefield into the delay proporciona
time-slowness (tau-p)pantalla
domain. The most important
advantage of the approach is that it provides a convenient display of the information needed
to obtain velocity as a function of depth.
In this paper we continue the investigation of the travel-time inversion of tau-p trans-
formed refraction data. We describe a new method for determining the p-tau(p) curve of
principal arrivals (critical raypath) in the transformed wavefield, and demonstrate a method
in which lateral variations in structure are resolved. We first describe these procedures
entonces
theoretically, and then apply ellos
them to synthetic and real data.

Theory

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

inclinado apilamiento
Slant stacking is a procedure that transforms the entire observed wavefield @ ( T x) , in the
time-distance domain into a wavefield $ ( t a u . p ) in the tau-p plane, with p being the
stacking or horizontal slowness and tau the intercept time or delay time (McMechan &
Ottolini 1980; Brocher & Phinney 1981a; McMechan, Clayton & Mooney 1982). The quality
of the inversion of the $ (tau, p ) wavefield is directly related to the quality of the transfor-
por lo tanto, nosotros
mation of the seismic data, hence, we describe tau-p transformations in detalle
detail. In preference
to integral representation, we have written equation (1) for discrete, finite aperture data as a
sum over N seismograms, which is a satisfactory procedure when a sufficient number (A‘) of
seismograms with dense geophone spacing is available. Slant stacking may be written:

with @(T,x): seismic true amplitude trace at distance x, and time T = tau + p x ,
f ( x ) : geometrical spreading correction, and
x : horizontal range.

Slant stacking essentially is a summing along lines of constant ‘step out’. The method can
be regarded as a generalization of velocity filtering techniques for determining arrival
slowness and intercept time (Chapman 1981).
To avoid spatial aliasing at the lowest horizontal phase velocity, the geophone spacing A x
must satisfy (Stoffa etal. 1981a):

Ax Q V(min)/2f(max)
with V(min): lowest horizontal phase velocity,
f(max): maximum frequency.

Therefore, the transformation ( 1 ) is restricted to high quality, densely spaced reflection and
refraction data. Stoffa et al. (1981a) and Stoffa, Diebold & Buhl (1981b, 1982) applied
the slant-stacking technique to marine wide-aperture CMP-data. A coherency measurement
aseguran
(semblance; Taner & Koehler 1969) ensures that coherent arrivals across a subarray will be
stacked and transformed.
The slant stack has also been modified by a number of other authors performing a plane
wave decomposition for spherical excitation; Muller (197 l), Chapman (1978,198 l ) , Phinney,
Chowdhury & Frazer (1981), and Henry, Orcutt & Parker (1980) introduce a convolution
operator into the simple slant-stacking algorithm. Kennett (1 98 1 ) showed the possibility of
presenting the plane wave reflection coefficients in the slowness-intercept time domain as
an element of an approach for a combined travel-time and amplitude inversion. Treitel,
 Inversion of seismic refraction data 83
Gutowski & Wagner (1982) computed the plane wave decomposition using the angle of
en cambio
incidence (at the surface) instead of the horizontal slowness p , and applied deconvolution to
the transformed seismic data.

Inversion of slant-stackeddata
ventajoso
Compared with interpreting the original T-x data, slant stacking is advantageous because it
desenreda las triplicaciones completa
unravels triplications, and since it is a transformation of the who!e record section, there are
no problems with hidden low-velocity layers. In the tau-p domain it is possible to dis-
tinguish between ‘principle arrivals’ (refracted and critically reflected) and their multiples,
and pre-critically and critically reflected arrivals are separated.
If there is no low-velocity zone we can construct the velocity-depth function directly
from the critically reflected and refracted wavefield because this tau (p) curve of ‘principal

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

arrivals’ is a single valued monotonic function (except in the case of extreme lateral varia-
tions). One inversion method, a tau-sum recursive inversion scheme, was described by
Diebold & Stoffa (1981) for horizontal layers and a set of n observed p , tau ( p ) values:
n-1
tau (p,) =2 C zi( u ; ~- p;)l12
i=l

with p , : observed apparent (horizontal) wave slowness,

tau (p,): intercept time of pn,
zi: thickness of a layer with velocity ui,
ui: velocity in layer i.
The tau-sum recursion for travel-time inversion solves for the layer thicknessvia the equation:
n-2
[tau (p,)/2] - C zi(ui2 - pi)1”

(Diebold & Stoffa 1981, equation 15). We note that when a [ p , , tau ( p , ) ] pair are related to
post-critically reflected arrivals, the calculated layer thickness z , - = 0, i.e. a velocity
discontinuity is indicated.
Clayton & McMechan (1981) presented an alternate iterative method for inversion of
slant-stacked refraction data. Their method is based on the technique of wavefield continua-
tion for which the inverted velocity-depth curve is extracted directly from the input
$(tau, p ) wavefield. In their method, convergence of the inversion is determined when the
.output wavefield images the same velocity-depth function as was input to the continuation.
The method has recently been applied to several field data sets by McMechan et al. ( I 982).
Carrion & Kuo (1 984) computed velocity profiles by finding the critical path (principal
arrivals) in the tau-p domain with an energy approach. Since their minimization procedure
is based on the use of relative energies, they do not take into account phase shifts along the
p-tau ( p ) curve. Velocity-depth and density-depth profiles for horizontally stratified media
were calculated from plane wave reflection coefficients in the tau-p domain by Carrion, Kuo
& Stoffa (1 984).
Our approach is to determine the curve of ‘principal arrivals’ directly from the slant-
stacked wavefield $ (tau, p ) . When the velocity at the surface ( u l ) is known, then p1= l/ul,
debiso a
and the intercept time tau ( p l ) = 0. Due to the slant stack the wavefield is sampled in p and
tau we have:
p i + < p i with p i + = p i - A p ( A p : sample interval in p ) , and
tau ( p i + 1) > tau (pi),with tau (pi+1) = tau ( p i ) + A taui.
84 B. Milkereit, W. D.Mooney and W. M. Kohler

tau

.
tau(p,) -- -
a I
P3 pz P, P
Figure 1. Schematic of a $ (tau, p ) wavefield and the curve of principal arrivals (curved dashed Line). Suc-

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

cessive tau ( p i ) values are determined by an L,-norm correlation beginning with p , .

In order to determine the curve of ‘principal arrivals , the ‘time shift’ A taui from trace to
trace has to have been determined from the slant-stacked wavefield $(tau,p) (Fig. 1). We
note that each $ (tau, p ) trace consists of the complex plane wave reflection coefficient
comvolucionado
R ( p ) to the delta function excitation convolved with the source time function S ( t ) , we have
(Kennett 1981; Brocher & Phinney 1981b):

(tau, P ) = s (t)*Rb).
desde cuenta
Since we do not take into account the condition at the source (i.e. line source, point source,
etc.) when we apply the plane wave decomposition to the observed wavefield $ ( T , X) we
recuperar debido a
cannot recover the true reflection coefficient for one ray parameter R ( p i ) . Due to the limits
of integration this is true for a truncated seismic aperture in the T-X domain as well
(Durrani & Bisset 1984). On the other hand, each slant-stacked seismogram $ (tau, p i ) will
contain the same source time function S ( t ) .
With the starting information p1= l / u l ( u l = the velocity at the surface) and tau (pl) = 0,
we may compute the ‘time shift’ A tau by cross-correlation
surgir
between two slant-stacked traces
$(tau, p i + and J/ (tau, p i ) . Some errors may arise in this procedure, as McMechan (1983)
puntiagudo
pointed out that ‘ . . .p is usually assumed to be the reciprocal of the measured apparent
pulso la forma
phase velocity across an array. This assumption is correct only if the arrival pulse shape is
stable with distance’. For example, the phase velocity is not equal to the group velocity for
rama
the postcritical reflected travel-time branches. A more stable (and more time-efficient) pro-
cedure is the application of the LI-norm (Claerbout 1976) instead of the cross-correlation in
the presence of progressive phase change. Jannsen, Voss & Theilen (1 985) used the L l-norm
(equation (4)) for the interpretation of seismic data in a medium with intrinsic attenuation.
Thus, we determine the A tau for which:

IJ/(tau,pi)- J / ( t a u , p i + , ) ( + m i n f o r A t a u ~ [ O ,Atau,,,] (4)

W

with: W: length of the correlation window (set approximately equal to the source time
function S ( t ) ) , and
A tau: time shift between the li/ (tau, pi) and J/ (tau, p i + trace, for 0 < tau < taumax.

Since the wave shape changes a little from trace to trace (decreasingp), small changes in
the shape of the correlation operator at each step leads to a decrease in the bias. Once the
p-tau ( p ) curve has been determined, its inversion for plane layers is straightforward (e.g.
equation (3)).
 Inversion of seismic refraction data

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

i 2 4 6
RANGE (KM)

3.5 2.5 V (kmlsec)

0
9
Y

G
w
EO
3 2
a
I-

80.24 0.28 0.31 0.35 0.39 0.43 0.47 0.51 0.55

P (SECIKM)

V (KM/SEC)
3.00 4.00

Figure 2. (a) Seismic record section (trace scaled by true amplitude times distance) recorded by the USGS
in the Imperial Valley, southern California. Amplitude maxima on the first arrival curve are marked ‘a’
and ‘b’. (b) Slant stack of the time-distance data. Short horizontal bars indicate the p-tau ( p ) curve deter-
mined by the automatic picking procedure (see text). (c) Velocity-depth curve obtained by inverting the
p-tau ( p ) curve via equation (3).
86 B. Milkereit, W. D. Mooney and W.M. Kohler
Application of the picking method
The application of this method of determining (‘picking’) the p-tau@) curve to real data is
illustrated with seismic refraction data in a common-shot geometry (Fig. 2). These data were
recorded by the US Geological Survey in 1979 in the Imperial Valley of southern California
(Fuis et al. 1984). The seismograms are plotted in true amplitude, multiplied by distance.
The first arrivals show amplitude maxima, labelled ‘a’ and ‘b’, and secondary, multiply-
refracted arrivals following the first arrivals (located 0.6 s behind the first arrivals at 7 km
range). This record section has been slant stacked using plane-wave decomposition.
A reliable seismic model can be obtained from the refracted and post-critically reflected
wavefield when the slant stack is done for a sufficient number of equally spaced ray para-
meters if the length of the profile has been chosen to ensure that all ray parameters needed

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

for inversion are within the observed data (Brocher & Phinney 1981a). Throughout this
paper the transformed wavefield $ (tau,p) is sampled for 80 ray parameters. The p-tau(p)
curve of principal arrivals corresponding to refracted and critical reflected arrivals has been
automatically picked using the procedure described above. starting with tau ( p l ) , = 0 s,

0
0 1 2 4 7
RANGE (Km)

-0.00 2.00 3.00

~ ,
4.00

Figure 3. Ray-theoretical synthetic seismogram calculated using the indicated velocity-depth curve
(simplified from Fig. 2c). Only principle arrivals have been calculated. Plotting format as in Fig. 2(a). The
amplitudes ‘a’ and ‘b’ in Fig. l(a) have been reproduced, which indicated that the p-tau@) picking
procedure and the inversion work well.
 Inversion of seismic refraction data 87

11 / -

Y10
I
t
Y
n

20

22
Figure 4. Dipping-layer velocity model used for the calculation of synthetic record sections. There are two
velocity discontinuities which produce reflected phases referred to as R , and R,. respectively. The three
shot points used to produce synthetic data in these geornetricies are; roll-along, split, and reversed profiles.

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

p1 = 0.55 s km-' (corresponding to a surficial velocity of 1.8 km s-l). The picking procedure
is stable, even along portions of the wavefield where phase changes occur, for example, near
p values of 0.39 and 0.28skm-' (Fig. 2b). This p-tau(p) curve has been inverted to a
velocity-depth curve (Fig. 2) using equation (4). The velocity-depth function shows a high
velocity gradient and two prominent velocity discontinuities at about 0.9 and 1.6 km depth.
Velocities greater than 3.5 km s-' are not properly resolved due to the limited profile length.
In order to evaluate the validity of the velocity-depth function obtained, synthetic seismo-
grams were calculated for a slightly simplified velocity-depth function (Fig. 3). These
seismograms were calculated using ray theory and include only the primary reflected and
refracted P-waves. Comparison of Figs 2(a) and 3(a) shows that the inverted velocity-depth
function provides travel times that fit the observed data. Furthermore, the synthetic seismo-
grams show amplitude maxima at about 2.5 and 5.4km that closely match those in the
observed data. This agreement indicates that the picking algorithm provides a satisfactory
determination of the desired p-tau ( p ) curve of principal arrivals. Having derived a method
of automatic picking of the p-tau(p) curve, we apply this algorithm to seismic data in
planar dipping structure.

Extension of p-tau inversion to dipping layers

Based on a geometrical formulation of the seismic travel time for refracted and critically
reflected arrivals, an inversion scheme for plane dipping layers was suggested by Ocola
(1972). Cunningham (1974) solved the travel-time problem for repeated single-ended
refraction profiles, or 'roll-along' profiles, while Johnson (1 976) inverted split-spread refrac-
'
tion data. Diebold & Stoffa (1981) extended the inversion of slant-stacked seismic data to
dipping layers in cases of a CMP-gather where moderate dips < 3" can be easily overlooked.
They also demonstrated deformation of a p-tau ( p ) curve due to dipping layers for single
shot or receiver geometry, but did not take into account amplitude information.
Following Ocola (1972) and Diebold & Stoffa (1981), the travel time T for a seismic
event with an apparent wave slowness p n for a fixed source (or fixed receiver) and a stack of
n layers with constant velocity ui can be written:
n-1
T(p,) = p,x + 1 z i [cos (ai) + cos (bi)]/ui
i= I
with p n : apparent horizontal slowness due to dip,
x: horizontal range,
zi:layer thickness below the shotpoint,
ai, bi: angles from vertical (rays going up or down), and
a, = sin-' (ul pn).
 88 B. Milkereit, W.D. Mooney and W. M. Kohler
After performing a slant stack for the apparent ray parameter pn, the observable delay time
tau ( p n ) becomes:
n-1
tau ( p n ) = 2 zi [cos (ai)+ cos ( b i ) ] / u i . (5)
i= 1

We solve equation (5) recursively for the layer thickness zi starting at the surface, where
p 1= l / u l and tau(pl) = 0. The relationship of the quantities x, zi,a[ and bi are described in
the Appendix.
Equation (5) provides a powerful tool for the inversion of dipping structures. We
emphasize that zi, the thickness of a constant velocity layer directly below the shotpoint, is
directly related to tau(pi+l), the true layer velocity ui, and the dip of the layer; it is no

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

longer the depth of a bottoming point of a ray in a horizontally inhomogeneous medium
(see Appendix).
The procedure to pick the p-tau(p) curve automatically in the tau-p domain from the
slant-stacked wavefield and to invert this curve can be easily extended to dipping layers. We
interpret the observed wave slowness p (used for the slant stack) as an apparent wave
slowness due to dip. There are three possible inversion schemes for seismic refraction data:
roll along, split spread, and reversed profile geometries.
To test the inversion in laterally-varying media we have calculated ray-theoretical synthetic
seismograms in a velocity model consisting of three layers, each with a positive velocity
gradient (Fig. 4). (Since Diebold & Stoffa (1981) showed that the tau-sum travel-time
inversion (equation 2) is perfectly adequate for zero dip velocity gradient models, we d o not
attempt to justify the extention of equation (5) to velocity gradients.) The interface
between the second and third layers has a dip of 4.6" to the left; since the velocity is
constant along the top and bottom boundaries ( e g 2.3 km s-' above the first boundary, and
2.65 km s-' below it), the velocity gradient varies laterally in the layer, decreasing from right
to left. Similarly, the velocity gradient increases from right to left in the third layer.
Velocities are linearly interpolated between the boundaries. The first boundary has a
velocity contrast of 0.35 km s-', the second boundary of 1.O km s-'. Synthetic seismograms
have been calculated for shot points at three locations in the model (Fig. 4) in order to have
data to test the three recording geometries considered. Each is described in the Appendix.
The synthetic seismogram sections were calculated using the ray theory approach of
McMechan & Mooney (1980); in this method, ray paths are calculated through the medium
by numerical integration of the eikonal equations (Cerveny, Molotkov & P5enEik 1977). By
. calculating complex transmission and reflection coefficients at boundaries, geometrical
spreading between successive rays, and noting phase shifts or caustics, the amplitudes of
geometrical rays may be obtained. The computed record section (Fig. 5a) includes refractions
and pre- and post-critical reflections. The impulse response of the model has been convolved
with a simple source function.
We begin the examination of the inversion for laterally-varying structures with an applica-
tion of one-dimensional inversion of the synthetic data for shotpoint 1. Throughout this
section we are interested in the principal tau-p curve. We note that dipping structures will
cause deformation of travel-time curves in the T-X and tau-p domain, and low p values
may be out of observation range for common shot gathers (i.e. in shooting downdip). There-
fore we do not show low p value slant stacks. The slant-stacked section (Fig. 5b) has been
automatically picked using a starting value of tau (pl) = 0, u1 = 1.8 km s-' ( p l = 0.55 s km-').
The display is normalized for the maximum in the $ (tau, p ) array. Shading has been applied
to the traces for positive values greater than the picked tau value. Since no multiples or con-
verted phases were computed in the synthetic section there are no phases in the tau-p
 0 2 4 5.5
RANGE (Kin)
5.0 4.0 3.0 2.5 2.0 1;a

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

---7,
0.20 0.24 0.28 0.33 0.37 0.42 0.46 0.51 0.55
P( Sec /Kin)

R1,TO 1 I -

0-L7T----- , -~
O0.20 0.24 0.28 0.33 0.37 0.42 0.46 0.51 0.55
P( Sec/ Km)

(d) V(Km/Secl
0.00 1;25

T I

r0.J
0
0

Figure 5. (a) Synthetic record section for shot point 1 (Fig. 4) to the right. Seismograms within this and
all other synthetic sections are plotted in the format of true amplitude times the source-receiver distance.
Precritical reflections from the two velocity discontinuities in the model (Fig. 4) are labelled R ,and R,.
(b) Slant-stacked record section of (a). Light horizontal bars mark the automatic picks for the p-tau ( p )
curve and positive swings within each seismogram have been shaded for time greater than the tau ( p ) pick.
The portions of the p-tau (p) wavefield corresponding to pre-critically reflected energy are labelled R ,
and R , on this (and other) slant-stack sections. (c) Comparison of automatically picked p-tau ( p ) curve
(crosses) with exact p-tau ( p ) curves (solid lines) determined from a ray tracing through the velocity
model. (d) Comparison of true velocity-depth curve beneath shot point 1 with the curve obtained via
inversion assuming no lateral velocity variations. Solid line is input model, dashed line inverted model.
Both the absolute velocities and the velocity gradients have been over-estimated below 0.3 km depth.
90 B. Milkereit, U! D. Mooney and W. M. Kohler
plane for delay times greater than the picked curve; this makes the p-tau section appear
much simpler than the corresponding section for real data (Fig. 2). Clear phases off the
principal branch of the p -tau curve are the two pre-critical reflections labelled R and R 2 .
The accuracy of the automatic picking algorithm in this case can be easily assessed: the
travel-time response of the model (shotpoint 1) has been transformed into a p-tau curve and
is compared with the picked value (Fig. Sc). The agreement is excellent in most areas. The
largest discrepancy occurs near p = 0.37 s km-' where, due to post-critical reflections, phase
changes occur in the (tau, p) wavefield.
The p-tau (p) curve has been inverted to a velocity-depth curve using the one-dimensional
formula (equation 3). The result (Fig. 5d), when compared with the model velocity-depth
function beneath the shotpoint, nearly correctly estimates the depths of the layer boundaries,

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

but clearly over-estimates the velocity gradients below the first boundary (0.3 km depth). In
addition, the first-order discontinuity at the second boundary has been modelled as a high
gradient zone. These discrepancies between a flat layer interpretation and the dipping layer
model motivated the application of the inversion method (Appendix) that will correctly
model structures with dips.

Inversion in laterally-varying media

(1) ROLL-ALONG GEOMETRY

We first apply the inversion formulae (Appendix) to seismic refraction/wide-aperture reflec-

tion data in a roll-along geometry. The synthetic seismograms (Fig. 6) are for shotpoints 1
and 2 of the laterally varying velocity model (Fig. 4). The shotpoint separation is 2 km and
seismograms are calculated every 50m. The 4.6" dip on the second boundary gives a 0.2 km
shallower depth below shotpoint 2 as compared to shotpoint 1. This difference is apparent
in the synthetic record section in a shift of the high-amplitude reflection branch, R 2 , to
shorter ranges for shotpoint 2 than for shotpoint I .

RANGE (KM)

G
w
ro,

Q 2 4 5.5
RANGE (KM)

Figure 6. Synthetic seismograms for shotpoints 1 and 2 for propagation t o the right in Fig. 4 (updip
direction). Plotting format as in Fig. 5(a). These two sections provide synthetic 'roll-along' data.
 Inversion of seismic refraction data 91
V (KM/SEC)
5.0 3.0 2.5 2.0 1.8

0.28 0.33 0.37 0.42 0.46 0.51 0.55

P (SEC/KM)

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

V (KM/SEC)
3.0 2.5 2.0 1.8

R2

R,
3 0133 0137 0:42 0:46 0 I1 0.55
P (SEC/KM)

V (KMISEC) V (KMISECj

-x a-_
N

- +zo -f B

The two time-distance sections have been slant stacked (equation 1) to obtain J/ (tau, p )
record sections (Fig. 7). The automatic picking and display is as previously described for
Fig. 5(b). Following the derivation of the inversion for roll-along geometry (Appendix), the
two shotpoints record equal values of observed apparent wave slowness, p , but unequal
values of tau@> due to the dip. The inversion proceeds through the 80p-tau ( p ) values,
iteratively calculating the depth for equal p given the unequal tau(p) values (Appendix,
equations A13 and A14).
The comparison with the actual velocity-depth functions beneath the shotpoints indicates
an excellent agreement between the model and the inversion result (Fig. 7b). In addition to
92 B. Milkereit, W. D. Mooney and W. M. Kohler

Y
-c

I 10
55
RANGE Ikm) RANGE (km)

Figure 8. Synthetic seismograms for shotpoint 2 for propagation t o the left and right (up- and downdip,
respectively). Plotting format as in Fig. 5(a).

V (KM/SEC)
4;O 3,O 2.0 1.8

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

0.24 0.28 0.33 0.37 0.42 16 51 0.55
P (SECIKM)
V (KMISEC)
4.0 3.0 I

0.20 0.24 0.28 0.33 0.37 0.42 0.46 0.51 0.55

P (SECIKM)

V (KMISEC) DIP (DEGREE)

( b)
0.00 2.50 3.75 5.00 -5.00

2
0
J
i

Figure 9. (a) Slant-stacked record sections for the synthetic seismograms of Fig. 8. Plotting format as in
Fig. 5(b). (b) Result of combined inversion of the p-tau ( p ) curves of (a) above, using the split-spread
solution. The calculated model (light lines) closely approximate the true model (heavy lines), but the
agreement is not as good as for the roll-along geometry (Fig. 7). Dip is indicated in terms of the dip of
isovelocity lines with depth; this function reaches a maximum at 1 km depth, at the 3.0-4.0km s-'
interface (cf. Fig. 4).
 Inversion of seismic refraction data 93
determining the differing depths to the second boundary correctly, the inverted velocity-
depth functions avoid two deficiencies noted in the one-dimensional inversion (Fig. Sd): the
velocity gradients have not been overestimated and the first-order discontinuity has not been
modelled as a high gradient zone. As will be seen, when compared to the other acquisition
geometries, the inversion in roll-along geometry provides the best fit to the input mo-del
because interpolating the p or tau ( p ) values in the inversion process is not necessary.

(2) SPLIT-SPREAD GEOMETRY

The inversion procedure for a split-spread can be evaluated by applying it to synthetic

seismograms to the left and right of shotpoint 2 (Fig. 8). The effect of the dip on the second
boundary in the velocity model (Fig. 4) is to shift the high amplitude reflection (R,) to

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

greater range on shotpoint 2, left, as compared to shotpoint 2, right. These synthetic
sections have been slant stacked, automatically picked using the surficial velocity of
1.8 km s-' ( p = 0.55 s km-'), and displayed with shading for tau ( p ) values greater than the
tau ( p ) pick (Fig. 9). As described in the derivation of the inversion procedure (Appendix),
split-spread observations show differing p values but equal intercept times (tau ( p ) ) for
refractions which bottom at a given isovelocity line. The inversion therefore steps through
tau(p) from tau(p) = 0 to tau(p) = max on shotpoint 2, right. At each step we find the p
value corresponding to an equal tau ( p ) value on the shotpoint 2, left, slant stack. Since the
automatic picking method provides discrete p , tau ( p ) values and will not, in general, result
in equal tau ( p ) values for the two slant stacks, it is necessary to interpolate the shotpoint 2,
left, p-tau(p) curve to find the value of p corresponding to each p-tau(p) value on
shotpoint 2, right, curve. Any appropriate interpolation procedure can be applied; we have

0 2 4 5.5
RANGE (Krn)

5.5 4 2
RANGE (Krn)
Figure 10. Synthetic seismograms for shot point 1 for propagation to the right in Fig. 3, and for shot
point 3 for propagation to the left. These two sections provide synthetic reversed data.
94 B. Milkereit, W. D. Mooney and W. M. Kohler
used a linear interpolation. We note that since the inversion requires equal tau ( p ) values on
both slant stacks, the inversion can only proceed to the smaller of the two maximum tau ( p )
values on the two slant stacks. Geometrically, this is equivalent to the observation that a dip
on an isovelocity line can only be determined if the line is sampled by both record sections
on the split-spread.
The combined inversion of the p-tau ( p ) curves, using equation ( S ) , results in a velocity-
depth function beneath the shotpoint and an estimate of the dip on the isovelocity lines
(Fig. 9b). The inversion has accurately determined the velocity at most depths but has some-
what 'smeared' the first-order discontinuity at the second boundary. The inverted dip closely
follows the correct trend but is a rather rough function, unlike the input model. Finer
sampling in p and tau, and a higher-order interpolation might improve the agreement.

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

(3) REVERSED-PROFILE GEOMETRY

The inversion procedure for a reversed profile can be evaluated using synthetic seismograms
for shotpoint 1 (right) and shotpoint 3 (left) (Fig. 4). In these reversed profiles (Fig. 10) the
effect of the dipping structure can be seen in both the differing intercept time for the R ,
reflection and in the substantially larger amplitudes on this reflection at distances less than
2 km from shotpoint 3 as compared to the same distance from shotpoint 1 . A well-known
relationship that applies to reversing profiles is that reciprocal times are equal; this is true in
the present case when comparing the times at the far offset ends of the two synthetic
sections. This relationship was used in the derivation of the inversion procedure (Appendix).
These synthetic sections have been slant stacked and automatically picked, and the result-
ing p-tau ( p ) curve is indicated with shading (Fig. 1 1). The most noticeable difference in the
two p-tau(p) curves is the lower amplitude of the curve for values of p less than about
0.26 s km-' for shotpoint 3. This is due to the effective low-velocity gradient encountered
below the second boundary in the down-dip direction; successively refracted rays do not
encounter significantly higher velocities as the ray turning point moves to the left due to the
dipping boundary. Just the opposite situation holds for the up-dip direction: an effective
high velocity gradient is encountered.
A second effect of the dip is to cause the apparent velocity to be higher than the true
velocity in the up-dip direction. This corresponds to a lower apparent p value. However, as
can be seen in the slant stack for shotpoint 1 (Fig. 11) the linear p-scale is non-linear in
velocity, and so velocities greater than 4.5 kms-' are only represented by five traces in the
slant stack. Thus, despite the clear separation of refracted and reflected branches beyond
4 k m in the synthetic seismogram for shotpoint 1 (Fig. lo), these branches are not well
separated in the corresponding slant stack; they should separate better at low p values.
For the inversion of a reversed profile, it is necessary to use a pair of p-tau ( p ) values from
the two sections whose reciprocal times are equal, corresponding to the geometrical situation
of refractions or reflections from the same isovelocity line in the model. Finding equal
reciprocal times requires an interpolation in both p and tau. Since treCp= tau ( p ) + x p , we
choose a p and tau(p) value on the first slant stack, calculate rrecp, and then find two
p-tau ( p ) picks on the second slant stack that bracketed the desired trecpvalue. The desired
point on the second p-tau ( p ) curve can then be found by interpolation of p and tau ( p ) .
The combined inversion of the p-tau(p) curves using equation (5) results in the two
velocity-depth functions beneath the shotpoints (Fig. 11). Both velocity models are in
agreement with the input model for the depth to the second boundary. The dip of 4.6" can
easily be calculated from the two depth estimates and the known separation between shot-
points. The velocity below the second boundary has not been correctly inverted due to the
undersampling in p .
 Inversion of seismic refraction data
V (KMISEC)
5.0 4;O 3,O 2.0

0.20 0.24 0.28 0.33 0.37 0.42 0.46 0.51

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

P (SEC/KM)
V (KMISEC)
3.0 2.5

0.20

(b)
0.24 0.28

V (KMISEC)
0.33 0.37
P (SK'KM)
1
0.42
1

0.46

V (KMISEC)
0.51

Figure 11. (a) Slant-stacked record sections for the synthetic seismograms of Fig. 10. Plotting format as in
Fig. 5(b). (b) Result of combined inversion of the p-tau ( p ) curves of (a) using the reversed profile solu-
tion. The calculated model (light lines) closely approsimates the true model as deep as the second velocity
discontinuity including the difference in depth due t o dip. At greater depth, the error is larger due to the
short length o f t h e profiles; for shot point 1 in particular the third layer is poorly sampled in ray parameter
(five values).

These three examples have illustrated inversion of synthetic data in three geometries. The
roll-along geometry seems t o be the most suitable since it is not necessary to interpolate
either p or tau ( p ) values between the slant-stack sections.

Application to field data

Seismic refraction/wide-angle reflection data recently collected in the eastern Central Valley,
California, are ideally suited for the application of the roll-along inversion procedure. The
data were collected with a close group interval in sign bit mode and show clear arrivals t o at
 B
a
a
s

DISTANCE (KM)
Figure 11. Seismic record sections recorded for the USGS in the Central Valley, California, in sign-bit format with 23 m
group interval. Five explosive shots have been summed to make each record section. The roll-along inversion solution
(Figs 6 and 7) may be applied to this data. Arrows indicate the cross-over t o the seismic basement with an apparent
velocity of 5.8 k m s-!.
Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023
P V (KMISEC) V (KMISEC)

sL" s
L"
3
a
+

0.15 0.19 0.23 0.28 0.32 0.37 0.41 0.46 0.50 0.15 0.19 0.23 0.28 0.32 0.37 0.41 0.46 0.50
;a"
P (SECIKM) P (SEC/KM) E'
3

(b) SP 870 V(KMISEC) SP 928 V (KMISEC)

W
Figure 13. (a) Slant-stacked record sections for the field data of Fig. 12. Plotting format as in Fig, 5(b). (b) Result of combined inversion of the 4
p-tau ( p ) curves of (a) using the 'roll-along' solution. A difference of 150 m in the depth to seismic basement (velocity greater than 6.0 km S - ' )
has been resolved.
Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023
98 B. Milkereit, W. D.Mooney and W. M. Kohler
least lOkm offset (Fig. 12). The sign-bit recording mode makes it possible to have a large
number of channels (1024) but yields very low dynamic range (0-10 digital counts in this
case). Seismic basement has a clear dip in this area; this is evidenced by the greater cross-over
distance to the basement refractor for shot point (SP) 928 as compared to SP 870. The
inversion procedure was applied to these data to resolve the lateral change in structure.
The record sections in Fig. 12 were slant stacked into the p-tau domain via equation (2)
using a semblance threshold of 0.3 (Stoffa et al. 1981a) to obtain a better signal-to-noise
ratio from this sign-bit data. The p-tau (p) curve has been automatically picked from the
sections (Fig. 13a) using an initial (surface) velocity of 2.0 km s-'. The inverted velocity-
depth curves (Fig. 13b) show a great similarity within the sedimentary section. The velocity
contrast at the basement is from about 3.1 to 6.0 km s-', with the depth to basement below
SP 928 about 150 m greater than below SP 870. This calculated difference is in good agree-

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

ment with the difference in depth determined by drill-hole measurements (Bartow 1983).
The velocity gradient in the basement is poorly resolved by a profile of this length.

Conclusions
The processing of seismic refractionlwide-aperture reflection data for laterally-varying
velocity structures is an important aspect of data interpretation. We present a technique for
processing densely spaced refraction data collected in three linear geometries (roll-along,
split-spread, and reversed profiles), that is both automatic and stable for reasonable dips (up
to about 8"). The technique consists of three steps: the slant stacking of the two time-
distance wavefields to intercept-slowness wavefields, the picking of the p-tau (p) curve of
principal arrivals using an L l-norm coherency measurement, and iterative downward stripping
of the p-tau (p) curves. The resulting model consists of planar dipping isovelocity layers
between shot points. The method has been successfully applied to all three types of geo-
metries using synthetic data for a simple planar dipping structure. The most accurate results
are for 'roll-along' geometry, and the application to real data have been illustrated.
The inversion scheme obtains, from the refracted and critically reflected arrivals, dip-
corrected interval velocities which may be used as input data for the stacking of seismic
reflection data. The results in all cases are compatible with those obtained from conventional
processing, but the present method is significantly more objective and is particularly well
suited to the automatic processing of large volumes of data. The present approach does not
take into account the possible presence of low-velocity zones. The ease and practicality of
the method encourage its further development and application.

Acknowledgments
Critical reviews by T. Brocher, M. Springer, P. A. Spudich, D. A. Stauber, G. A. McMechan
and an anonymous reviewer are appreciated. We thank C. Wentworth and A. W. Walter for
assistance in obtaining the sign-bit data tapes.

References
Bartow, J . A,, 1983. Map showing configuration of the basement surface, northern San Joaquin, Cali-
fornia, USgeol. Surv. Misc. Field Studies Map MF-1430, scale 1 : 250,000.
Brocher, T. M. & Phinney, R. A., 1981a. Inversion of slant stacks using finite-length record sections,
J. geophys. Res., 86,7065-7012.
Brocher, T. M. & Phinney, R. A., 1981b. A ray parameter-intercept time spectral ratio method for seismic
reflectivity analysis, J. geophys. Rex, 86,1865-1813.
 Inversion of seismic refraction data 99
Carrion, P. M. & Kou, J. T., 1984. A method for computation of velocity profiles by inversion of large
offset records, Geophysics, 49, 1249-1258.
Carrion, P. M., Kuo, J. T. & Stoffa, P. L., 1984. Inversion method in the slant stack domain using ampli-
tudes of reflection arrivals, Geophys. Prospect., 32, 375-391.
Eervenq, V., Moltokov, 1. A. & PSenElk, I., 1977. Ray Method in Seismology, Charles University, Prague.
Chapman, C. H., 1978. A new method for computing synthetic seismograms, Geophys. J. R. astr. Soc.,
54,481-518.
Chapman, C. H . , 1981. Generalized radon transforms and slant stacks, Geophys. J. R. astr. SOC., 66,
445-453.
Clayton, R. W. & McMechan, G. A., 1981. Inversion of refraction data by wave field continuation,
Geophysics, 46, 860-868.
Claerbout, J. F., 1976. Fundamentals of Geophysical Data Processing, with Applications to Petroleum
Prospecting, McGraw-Hill, New York.
Cunningham, A. S., 1974. Refraction data from single ended refraction profiles, Geophysics, 39,

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

292-301.
Diebold, J. B. & Stoffa, P. L., 1981. The traveltime equation, tau-p-mapping, and inversion of common
midpoint data, Geophysics, 46,238-254.
Durrani, T. S. & Bisset, D., 1984. The Radon transform and its properties, Geophysics, 49, 1180-1187.
Fuis, G. S., Mooney, W. D., Healy, J. H.,McMechan, G. A. & Lutter, W. J., 1984. A seismic refraction
studies of the Imperial Valley region, California, J. geophys. Res., 89, 1165-1189.
Henry, M., Orcutt, J. A. & Parker, R. L., 1980. A new method for slant stacking refraction data,
Geophys. Res. Lett., 7, 1073-1076.
Jannsen, D., Voss, J. & Theilen, F., 1985. Comparison of methods to determine Q in shallow marine
sediments from vertical reflection seismograms, Geophys. Prospect., in press.
Johnson, S. H., 1976. Interpretation of split-spread refraction data in terms of plane dipping layers,
Geophysics, 41,418-424.
Kennett, B. L. N., 1981. Slowness techniques in seismic interpretation, J. geophys. Res., 86, 1 1 575-
11 584.
McMechan, G. A., 1983. p-x imaging by localized slant stacks of T-x data, Geophys. J. R. astr. SOC.,72,
213-221.
McMechan, G. A., Clayton, R. W. & Mooney, W. D., 1982. Application of wavefield continuation to the
inversion of refraction data, J. geophys. Res., 87,927-935.
McMechan, G . A. & Mooney, W. D., 1980. Asymptotic ray theory and synthetic seismograms for laterally
varying structures: theory and application to the Imperial Valley, California, Bull. seism. SOC.Am,
70,2021-2035.
McMechan, C. A. & Ottolini, R., 1980. Direct observation of a p-tau curve in a slant stacked wavefield,
Bull. seism. SOC.Am., 70, 775-789.
Muller, G., 1971. Direct inversion of seismic observations, Z. Geophys., 37, 225-235.
Ocola, L. C., 1972. A nonlinear least-squares method for seismic refraction mapping - part 11: model
studies and performance of REFRAMAP method, Geophysics, 37, 273-287.
Phinney, R. A., Chowhury, K. R. & Frazer, L. N., 1981. Transformation and analysis of record sections,
J. geophys. Res., 86,359-377.
Stoffa, P. L., Buhl, P., Diebold, J. B. & Wenzel, F., 1981a. Direct mapping of seismic data to the domain
of intercept time and ray parameter - a plane wave decomposition, Geophysics, 46,255-267.
Stoffa, P. L., Diebold, J. B. & Buhl, P., 1981b. Inversion of seismic data in the tau-p plane, Geophys.
Res. Lett., 8,869-872.
Stoffa, P. L., Diebold, J. B. & BUN, P., 1982. Velocity analysis for wide aperture seismic data, Geophys.
Prospect., 3 0 , 2 5 4 7 .
Taner, M. T. & Koehler, F., 1969. Velocity spectradigital derivation and application of velocity functions,
Geophysics, 34, 859-881.
Treitel, S., Gutowski, P. R. & Wagner, D. E., 1982. Plane wave decomposition of seismograms, Geophysics,
47,1375-1401.

Appendix: inversion of p-tau ( p ) curves for planar dipping structure

Two spatially dense seismic sets @ ( t ,x ) ~and @ ( t ,x ) ~
are available from two shotpoints
observed along a single profile line. Both data sets are slant stacked for the same number and
range of ray parameters. Beginning with the starting surface velocity information p l = l / u l
100 B. Milkereit, W. D.Mooney and W.M. Kohler
and tau (pl) = 0, we automatically determine the curves of the principal arrivals [ p , tau ( p ) ]1,
from the slant stacked wavefields $(tau, p)l and $(tau, p ) 2 (see text). Both sets of discrete
Lp, tau ( p ) l l , values are the input for an iterative inversion scheme to compute the layer
thickness, velocity and dip.
Fig. A1 shows a model with n dipping layers and with the raypath. travel time and delay
time geometry for rays critically reflected or bottoming at the nth boundary. Source or
receiver positions are located at the surface at A, B, C, D and E. The intercept time tau (p,)
for an apparent wave slowness p n can be expressed in terms of the layer thicknesses, z j , and
the velocities, ui, below the shotpoint. Angles ai and bi against the vertical describe the
raypath and angles yi describe the dip. We have,

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

and, at the bottoming point (last layer),

a n - 1 = a n - 1 Y n = bn - 1 + Y n (A21
u, = u, - /sin 01, -1 . (A3 1
At the bottoming point of the critical reflection at the nth boundary, substituting
b, - = n - 1 - 2y,, equation (Al) can be separated into two terms:

Term I describes the layer thickness z , - and the dip y , of the nth iayer while term I1
contributes the raypath from the surface to the IZ - 2 layer and back to the surface.

SPLIT-SPREAD GEOMETRY
For split-spread geometry (shotpoint at position C and receivers at A and E; Fig. Al), all
angles ai and bi for i = 1, 1.1 - 1 can be calculated using Snell's law recursively starting at the
surface. At the receiver A a plane wave with the lower apparent slowness p n a A(updip) is

Ign' "a"D I

I I I

Figure A l . Cross-section showing recorder and shot point locations (A, B, C, D and E) with geometry of
ray paths and of the planar dipping-layer structure and of travel times for up- and downdip observations.
 Inversion of seismic refraction data 101
observed and at receiver F a plane wave with the higher apparent slowness p,, (downdip) is
observed, and p,, A f p , , F. The intercept times at shotpoint C are equal:

tau (P,, A) = tau @I*, F). (A51

When the velocity at the surface u1 is known we obtain the take-off angles at the surface
a l and b l with
a l = sin-' (ulp,, E), b l = sin-' ( u l p n , A ) . ('46)
The slant stacked data set for shotpoint C to A will contribute the updip angles b j and the
data set C to E will contribute the downdip angles a j . We then solve equation (Al) or (A4)
using both curves of principal arrivals [ p , tau(p)]l,2 to determine the unknown layer thick-
ness z , - of the n-1 layer below shotpoint C and the velocity u, with dip y n of the nth

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

layer. Below we present a processing flow chart for an n-layer model with two data sets
[p,, tau ( p n ) l l . and a split-spread geometry (Fig. A2).

REVERSED PROFILES

For reversed profiles (shotpoints at A and D, separated by a distance X,) a plane wave from
shotpoint A with an apparent downdip slowness p n , c will be observed at receiver C with the

Start with layer 2

Initialize the inversion

Takeoff angles
at the surface

Raytracing through the

uppermost n - z layers

(A41 sum = s u m t ( 2 , /v,)(cos(a,)+ coscb,))

I I

Snell's law at
the ( 1 + 1 1 ' ~interface

Next layer

Dip of the interlace (A21

Critical angle at
the interface (A21

Velocity contrast (A31

Thickness 01 the
(n - 1) laver (A41

Figure A2. Processing flow chart for an N-layer model with the two data sets [ p l , , , tau(pl,,)] and
[ p 2 , , , tau ( P ? , ~ ) obtained
] in a split-spread recording geometry.
102 B. Milkereit, W. D. Mooney and W. M. Kohler
delay time tau (pn c). The reciprocal time (time along a given refractor from one shotpoint
to the other, Fig. Al) will be

trec,A= tau ( P n , C ) +xlPn,C. ('47)

At receiver B a plane wave from shotpoint D bottoming at the same boundary is observed
with an apparent updip slowness p n , B and the delay time tau (pn,B).
The reciprocal time will be

trec, D = tau (Pn. B) + xl Pn, B = tau b n , C) + xl Pn,C = t r e c , A (A81

but neither the apparent slownesses nor the associated delay times are the same:

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023

Pn,C #Pn, B and tau (Pn,C) # tau (pn,B). (-49)

Again, we obtain the angles ai and bi recursively for i = 1, n - 1 starting at the surface with:

al = sin-' ( u l p n ,c), bl = sin-' ( u l p n , B). (A10)

The data set A to C will contribute the downdip angles ai and the data set D to B will
contribute the updip angles bi we need to solve equation (Al) or (A4) for the unknown
layer thicknesses z , - of the n-1 layer beneath shotpoints A and D and the velocity u,
with dip yn of the nth layer.

R O L L - A L O N G P R 0 FILES

For the roll-along technique (shotpoints at positions A and B, separated by a distance X2) a
plane wave from shotpoint A with an apparent slowness pn,c will be observed at receiver C
with the delay time tau ( p n ,c )(Fig. Al). At receiver D a plane wave from the same interface
from Sjlotpoint B is observed with an apparent slowness p n , and the delay time tau (pn, D).
The observed apparent slowness p n is not directly related to the layer velocity u, :
Pn,C=Pn,DfI/Un, n'l. (A1 1)

Planar dipping structure requires different delay times for the same apparent slownesses p n
from the two shotpoints:

tau (Pn, C) # tau ( P n ,D).

Rewriting equation (A4) for shotpoints A and B:
n-1Z
tau (pn,c ) = 2(cos ai + cos bi)
i=l ui.

n-1 zB,i
tau (pn,D) = -(COSai + cos bi) (A141
j=l U..i

where zA,i and z B , iare the thicknesses af the ith layer below shotpoints A or B respectively.
The data sets A to C and B to D will only contribute the downdip angles ai. Starting at
the surface with
al = sin-' ( u l p n ,c) = sin-' ( u l p , , D) (A15)
 Inversion of seismic refraction data 103
we obtain the angles ai for i = 1, n - 1 using Snell's law recursively. Additionally the total
depth to the nth layer beneath shotpoint A and B is given by:

with
yn = tan-' [ ( Z B - zA)/xZ>l. (A1 7)
We compute the angle bi, using relations (A2) and (A3), for a range of possible dip angles,
y,. For each trial y , , we solve equation ( A l ) or (A4) t o obtain Z A and ZB, (A16). The
correct value of y, is the one for which equation (A17) is also satisfied.

Downloaded from https://academic.oup.com/gji/article/82/1/81/551363 by guest on 12 May 2023