Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

How far is the seismic image correct?

EVGENY LANDA and SHEMER KEYDAR, The Geophysical Institute of Israel, Holon, Israel
MOSHE RESHEF, Landmark Graphics Corporation, Tel-Aviv, Israel

Estimation of a velocity model is a


critical step in the sequence leading
to depth interpretation of seismic a)
data. The quality of the depth image
depends heavily on the input veloc-
ity model which itself depends,
among other things, on the input
data, the inversion algorithm, and a
chosen class of models (number of
reflection interfaces, parameteriza-
tion for interfaces, geometry, and
velocities, etc.). The latter is usually
selected from a priori geological
information and/or interpretation of
stacked time sections.
Numerous publications have
been devoted to problems of data
quality and methods for velocity
estimation. Error analysis for esti-
mated velocity-depth parameters is
usually performed under the
assumption that the model class is
correctly selected. In this note we
b)
discuss a question of how an incor-
rect assumption about the number of
reflecting interfaces influences inver-
sion results. For one special case we
propose a criterion for estimated
model correctness based on the joint
use of reflected and refracted wave-
fields.

Synthetic example. To illustrate the


importance of an adequate choice of
the number of reflecting/refracting
interfaces for velocity model estima-
tion, we generated synthetics for a Figure 1. (a) Three-layer model. (b) Zero-offset section with muted shallow
three-layer model with horizontal reflection event.
first and third interfaces and a
curved second interface (Figure 1a). with the correct image (Figure 2c), Figure 3b illustrates an excellent fit
The velocities in all three layers are the mispositioning of the bottom between the traveltime curve for the
constant. For this model we calculat- interface is clear. The continuity of third reflection interface (colored
ed prestack shot gathers and the the event on both sections is of the line) calculated for the resulting
zero-offset section. We decided to same quality. A few CSL (common incorrect model and data. The
study how an incorrect interpreta- surface location) panels for the bot- coherency inversion procedure is
tion of the zero-offset (stacked) sec- tom interface calculated using used to construct the velocity model.
tion influences velocity inversion prestack depth migration are shown As emphasized in the paper “Quo
and depth imaging results. For this in Figure 3a. The overall flatness of vadit inversio?” by Sven Treitel (TLE,
purpose, we omitted a shallow the panels is good. The simple September 1989): “There is a very
reflection event during the time- example above shows that by ana- important corollary; we learn that a
picking process and continued our lyzing the CSL panels we obtain the good fit is a necessary but by no
inversion/migration processing for a “flatness” required for both the cor- means sufficient condition for suc-
two-layer model. The muted zero- rect and incorrect models. cess. By itself, a good fit does not
offset section is shown in Figure 1b. The omission of the first layer guarantee that an inversion is cor-
At this point, a detailed interval introduces lateral velocity variations rect. This occurs, in my opinion,
velocity analysis using iterative into the estimated model and leads more often than we would like to
prestack depth migration was car- to a wrong interface geometry. These think.”
ried out. Figure 2a shows the veloc- results do not depend on the inver- The example described repre-
ity model. Figure 2b is the final sion method, and the estimated sents in fact a situation which may
depth section. When it is compared model satisfies the inversion criteria. arise in many practical cases, espe-

0000 THE LEADING EDGE JULY 1998 JULY 1998 THE LEADING EDGE 919
(if at all possible) to observe on real
a) seismic data.
Simulated common shot gathers
comprise 60 traces in the offset range
0-600 m.
Reflection and refraction (head)
traveltimes were calculated using
ray tracing, and the minimum phase
wavelet was convolved with the cal-
culated traveltimes. Figure 5 shows
one common shot gather with ran-
dom noise added. Refraction and
reflection events are marked.
To estimate the depth of first
interface and the velocity in the sec-
ond layer, we used the first refractor
(H1 in Figure 4). Results of refraction
inversion for the first refractor are
b) correct (Z = 10 m and VH = 900 m/s)
and shown in Figure 4b. For further
model estimation, we combined two
refraction events from interfaces two
and three (H2 and H3 in Figure 5),
assuming that these two events are
refracted from one single interface.
In this way we simulated an erro-
neous time interpretation of first
arrivals on common shot gathers.
Using refraction velocity inversion,
we obtained depth and refraction
velocity for a second refractor. As
c) expected, neither depth (33 m) nor
velocity (VH = 2700 m/s) corre-
sponds to the correct model (Figure
4b). It is also obvious that if we use
this incorrect model for further
depth processing/inversion, we will
obtain an erroneous final depth
interpretation.
Bearing in mind that, in the
described example, in addition to
refraction events from the shallow
interfaces, we have reflections from
the deepest horizon (R in Figure 5).
This enables us to estimate the veloc-
Figure 2. (a) Estimated velocity model is incorrect owing to omission of ity in the deepest layer using this
shallow reflection event during the time interpretation. (b) Prestack depth- reflection event, which means that
migrated image obtained using an incorrect velocity model. (c) Prestack we can make two independent esti-
depth-migrated image obtained using a correct velocity model. Continuity mations of the same parameter,
of events on both sections (b) and (c) is of similar quality, while the mis- namely velocity, using two different
location of the bottom interface on (b) is clear. data sets, i.e., refraction and reflec-
tion events! It is important to empha-
cially in land exploration where shal- ness” of the estimated model in one size that these two events are inde-
low layers can be easily omitted special case which very often occurs. pendent because they are generated
owing to an incorrect interpretation. Here we understand “correctness” by two different interfaces: the
Depth processing, which has become in the sense of the correct choice of refraction is connected to the top of
so popular in the industry during model class (e.g., number of layers). the layer and the reflection to the
the last decade, requires a very good Let us consider a model with three bottom. It is obvious that in the case
knowledge not only of interval shallow horizontal refraction inter- where the model class (in this case
velocities in the deep subsurface but faces at depths of 10 m, 30 m, and the number of layers) is chosen cor-
it also (and maybe first of all!) 100 m, respectively, and one hori- rectly, we should obtain the same
requires adequate information zontal reflection interface at 500 m velocity estimation in the deepest
regarding shallow velocities. (Figure 4a). The velocities in all the layer using the refraction from the
layers are constant and are shown in top or the reflection from the bottom
Model correctness criterion. We the figure. In this example, we ignore of the layer. In the case of layer omis-
would now like to introduce an reflection events from the shallow sion, however, the velocity estimates
objective criterion on the “correct- interfaces which are usually difficult should be different. Using the first

920 THE LEADING EDGE JULY 1998 JULY 1998 THE LEADING EDGE 919
TRUE MODEL ESTIMATED MODEL

a)
a)

b)

b)

Figure 3. (a) Common surface location panels for the


bottom interface calculated using prestack depth migra-
tion. The overall flatness of the panels is good for both
correctly and incorrectly estimated velocity models. (b)
An excellent fit between the traveltime curve for the
third reflection interface (colored line) calculated for Figure 4. (a) Model with three shallow refraction and
the resulting incorrect model and data. The velocity one reflection interfaces. (b) Results of refraction
model was obtained by coherency inversion. velocity inversion where we combined two refraction
events from interfaces two and three, assuming that
these events are refracted from one single interface.
Neither depth nor velocity correspond to the correct
model. Velocity in the deepest layer was estimated
twice using two different data sets; i.e., refraction on
the top and reflection from the bottom of the layer.
Velocity estimates are different (Vr = 2946 m/s and
VH = 2700 m/s) owing to incorrect overburden model.

two layers estimated from refraction events as an over-


burden (interfaces H1 and H2*, and velocities 500 m/s and
900 m/s in Figure 4b) and reflection event R, we estimat-
ed the interval velocity in the third layer by reflection
velocity inversion. The results are shown in Figure 4b.
Velocity estimate Vr = 2946 m/s is different from that esti-
mated using refraction from H2* (VH = 2700 m/s). The dif-
ference is about 10%, and it can serve as a criterion of
model correctness and adequacy.
Again, it is important to note that interior inversion cri-
teria such as maximum semblance for coherency inver-
Figure 5. Common shot gather calculated for the sion, minimum difference between observed and calcu-
model shown in Figure 4a. Refraction and reflection lated traveltimes for traveltime tomography, or CSL
events are marked on the figure. alignment for focusing analysis, cannot be used for this

0000 THE LEADING EDGE JULY 1998 JULY 1998 THE LEADING EDGE 921
a)

b) Figure 7. Common shot gather calculated for model


shown in Figure 6a. Refraction and reflection events
are marked on the figure.

faces in the estimated model. Refraction velocity inversion


applied to the data results in a four-layered model with
velocities of 500 m/s, 700 m/s, 2500 m/s, and 3500 m/s,
respectively (Figure 6b). It is interesting to note that these
velocity estimates coincide with true velocities in the first,
second, and fourth layers, respectively. The depths of
interfaces H3* and H4* are estimated incorrectly due to an
erroneous assumption of the number of layers.
Next, we applied reflection coherency inversion for
velocity estimation in the deepest layer using the three
shallow layers shown in Figure 6b as an overburden and
reflection event R shown in Figure 7. The obtained veloc-
ity value was Vr = 3299 m/s which differs by about 8%
from the estimate obtained using refraction waves VH. This
difference indicates that the four-layered model chosen a
priori for velocity inversion is incorrect.
Figure 6. (a) Five-layered model with four shallow
Discussion and conclusions. Most popular velocity esti-
reflectors and one deep reflector. Velocity inversion in
mation techniques are based on a layered model of the
the third layer leads to the absence of a refraction event
subsurface. Usually a number of reflection/refraction
from interface H2. (b) Results of refraction velocity
interfaces is chosen a priori based on time interpretation
inversion in this case lead to an incorrect decision on
of seismic data. We demonstrated that a wrong assump-
the number of interfaces in the estimated model.
tion of the number of layers in the model may lead to an
Reflection inversion for velocity estimation in the
incorrect final depth image. A global inversion procedure,
deepest layer uses an incorrect overburden model. The
which uses such a priori information, can easily converge
velocity value obtained (Vr = 3299 m/s) differs by about
to a coherent set of depth interfaces with flat panels, all in
8% from the estimate obtained using refraction waves.
the wrong locations. Methods which allow the number of
This difference indicates that the model is incorrect.
layers to change during the velocity inversion process
purpose because the estimated wrong model satisfies have certain advantages. Such methods rely on the abili-
these criteria. Let us now demonstrate the proposed cri- ty to interpret more interfaces in cases where the depth
terion for model correctness in the case of a “hidden” layer image is significantly better than the time image. If we
in the shallow subsurface. Figure 6a shows a five-layered assume that “hidden” layers exist in most of the complex
model with horizontal homogeneous layers: four shallow data sets, additional criteria for the correctness of the
refractors (H1, H2, H3, H4) and one deep reflector R. Veloc- result are required.
ities in the layers are 500 m/s, 700 m/s, 300 m/s, 2500 We propose one criterion for model correctness verifi-
m/s, and 3500 m/s, respectively. Figure 7 illustrates a cation which is based on the joint use of reflection and
common shot gather calculated by ray tracing which con- refraction events for velocity model estimation. In our
tains refraction events H1, H3, and H4, and reflection event opinion, the question we should answer is “How far is the
R. As in the previous example we ignored reflection events seismic image correct?” rather than “How far is the seis-
from the shallow interfaces. Velocity inversion in the third mic image continuous or nice?” LE
layer leads to the absence of a refraction event from inter-
face H2 and an incorrect decision on the number of inter- Corresponding author: Evgeny Landa, email evgeny@iprg.energy.gov.il

922 THE LEADING EDGE JULY 1998 JULY 1998 THE LEADING EDGE 919

You might also like