You are on page 1of 14

An Integrated Pre-Stack Depth Migration Workflow Using Model-

Based Velocity Estimation And Refinement


Andy Furniss *

Abstract:

The advent of velocity-sensitive depth imaging techniques such as pre-stack depth migra-
tion has created a demand for more accuracy in velocity model building. Traditional methods that
use seismic processing velocities are limited by many assumptions inherent in the Dix formula and
the distortion caused by complex ray paths. Techniques based on model ray-tracing allow more
accurate estimates of interval velocity to be made, which, when constrained by an interpreters
geological knowledge of an area, have the ability to produce superior velocity models for depth
conversion or pre-stack depth migration.

The role of the interpreter in any depth imaging project is often understated because of a
lack of understanding of velocity modelling and depth imaging techniques. The purpose of this
paper is to introduce some of these concepts to interpreters in an attempt to promote more active
involvement in velocity modelling projects and for them to not simply rely on the processing ex-
perts to build geologically consistent velocity models.

Velocity estimation techniques such as stacking velocity inversion and coherency inversion
are-equally applicable to depth conversion projects as they are to pre-stack depth migration. Coupled
with map migration techniques, these methods allow the interpreter to reduce the uncertainty of
conventional depth conversion and ultimately reduce the geological risk of prospects.

Pre-stack depth migration and tomographic model refinements are an essential and power-
ful tool to produce the optimum seismic data quality. These methods are described to give the
explorationist a practical understanding of the workflows involved
in depth imaging.

Introduction: domain, from simply improving our initial ve-


locity models using model-based ray-tracing
Recent advances in computer-based in- techniques through to the ultimate in depth im-
terpretation systems have significantly reduced aging, pre-stack depth migration (PSDM).
the cycle time for exploration projects while im- Whether we interpret on time migrated data and
proving the accuracy of interpretations in the scale our maps to depth or choose to interpret on
migrated-time domain. However, as every inter- depth migrated data, there are many tools avail-
preter knows, the final test for any interpretation able that can assist in building more accurate
lies in the ability to produce an accurate progno- depth models. Whatever approach we take, our
sis in the depth domain. It is inadequacies in this accuracy in depth is wholly dependent on our
step of converting an interpretation from the time ability to produce an accurate interval velocity
to the depth domain that are commonly the cause model and this can only be achieved by taking a
of poor predictions, leading to increased drilling model-based approach using the interpreters
costs or at worst, creating prospects that simply knowledge of the region to constrain that model.
do not exist. Model refinement techniques that rely on auto-
mated tomographic approaches should be used
There are several approaches that can be in conjunction with model-based approaches as
taken to improve our predictions in the depth they have been shown to give erratic and unreli-
able results when used as the sole method of
1
GEOHORIZONS Vol . 5 No. 1
July 2000
model building or refinement. This paper de- maps are vertically scaled to depth using an av-
scribes some of the available technologies that erage velocity map or a series of interval veloc-
can be applied to produce better velocity models ity maps. This process is adequate in areas of
without labouring on the theory behind the meth- low dip without lateral or vertical velocity varia-
ods. Interpreters are encouraged to play an ac- tions or faults but unfortunately, these areas are
tive role in the building of velocity models and rarely of interest to us in the search for hydrocar-
not to leave this Critical step of the interpreta- bons. Map migration is a ray-traced depth con-
tion process to the local velocity expert who version technique that produces more accurate
may not have all the geological constraints or depth maps and also compensates for the spatial
models at hand. positioning errors that are left uncorrected by time
migration. Significant lateral errors may be ob-
From Conversion to Inversion served where there is very little velocity varia-
tion in the overburden and only map migration
Traditionally, the interpretation cycle for techniques can accurately reposition this data.
a project consisted of mapping key horizons on
time migrated seismic data followed by depth The third flaw lies in the actual imaging
conversion using a combination of well and/or of the seismic data in the time domain. Although
seismic velocities to derive a regional velocity we are taught that time migration will image seis-
field. While application of this technique has re- mic reflections in their correct spatial position,
sulted in the discovery of some of the worlds most of us are aware that this is not the case where
largest oil and gas fields, it is flawed for a num- lateral velocity variations, velocity gradients and
ber of reasons. scattering distort the seismic ray paths. Conse-
quently, our seismic gathers do not stack or mi-
Table 1: An example illustrating how a 1% error in ve- grate correctly, leaving us with a distorted and
locity creates the same depth error as 1% error in two inferior image of the subsurface in the time do-
way time. While a 20 ms mistie in time would be deemed
unacceptable, a 30 m/s velocity mistie would be frequently
main. If we can produce an accurate velocity-
overlooked depth model, PSDM will give us a far superior
image of the subsurface by correctly positioning
traces during migration. By scaling the depth
migrated data back to time, calibrating veloci-
ties to well data then rescaling back to depth, we
not only have the optimum image quality of our
seismic but also a seismic line or volume that
can be interpreted in the depth domain. Adding
Firstly, an explorationist may spend many additional well control simply means re-calibrat-
months interpreting structural horizons in fastidi- ing the velocity field and rescaling the data to
ous detail and then discard that precision by depth.
spending only a matter of days or a week build-
ing the velocity model for depth conversion.
Table I emphasizes the importance of producing
accurate velocity models. For the example given,
a 1% error in two-way-time gives the same 30
meter error in depth as a 1% error in average
velocity. Few interpreters would accept a 20 ms
mistake in their interpretation while a 30 m/s er-
ror in velocity is often overlooked.

The second flaw in the interpretation Fig 1. A typical workflow used for building and refining
workflow is the inaccuracy caused by traditional interval velocity models for use in pre-stack depth
depth conversion methods where time structure migration

2
GEOHORIZONS Vol . 5 No. 1
July 2000
While PSDM is often seen as a panacea shows that interval velocity, Vintn, can be calcu-
for seismic imaging it must be remembered that lated for the nth interval where Vrmsn-l, tn-l, and
the process itself is dependent on the raw input Vrmsn, tn are the root-mean-square velocity and
gathers and the accuracy of the velocity-depth travel times to the n-1th and nth layers respectively.
model used for the migration. PSDM has pro-
duced spectacular results in most of the petro- V2rms tn - V2rms t
Vint = n n-1 n-1
leum producing basins around the world and will n (1)
add value to any interpretation project by improv- tn - tn-1
ing the seismic image. However, its limitations
must be recognised and there are many cases The Dix transform from RMS velocity
where data imaging problems will not be com- to interval velocity is based on many assump-
pletely resolved by this process. tions and is frequently applied in areas where
those assumptions are invalid. Interpreters com-
Figure 1 outlines a workflow that will not monly use stacking velocity as input into Equa-
only improve the interpretation of seismic data tion (1) assuming they approximate to RMS ve-
but also provide more accurate depth predictions. locity. However, stacking velocity will only ap-
The workflow replaces traditional depth conver- proach RMS velocity in areas with no structural
sion with depth inversion by using sophisticated dip, no lateral or vertical velocity gradients, and
velocity modelling and refinement tools and when common midpoint (CMP) gathers have
PSDM. With this new workflow, interpretation very restricted offsets. In the real world, where
is finished. Advances in computer hardware and atleast one of these assumptions is invalid, seis-
improved software algorithms are rapidly put- mic raypaths will show non-hyperbolic moveout
ting this technology into the hands of the inter- causing stacking velocity and RMS velocity to
preter. Whereas the desktop workstation was differ. This produces inaccurate results from the
once used only for creating time interpretations Dix equation. In 2D a simple correction to re-
and maps, it is now capable of producing depth move the effect of structural dip, q, can be ap-
migrated data, the input model for which is the plied using the formula from Levin (1971) as
interpretation. shown in Equation (2).

Model-Based Interval Velocity Estima-


tion Vrms
Vstack = (2)
Cosq
For many years, the process of depth con-
verting time structure maps relied on the use of The Levin equation shows that for areas
average velocity maps hand contoured from well with structural dip, stacking velocity will always
data. This method is perfectly adequate in areas be faster than RMS velocity. In 3D, the dip cor-
where there is sufficient well control, velocity rection becomes more complicated and terms to
variation is small and the structure does not vary account for the dip azimuth must also be intro-
significantly away from the control points. Ob- duced. Despite these corrections, the Dix trans-
viously, in new exploration areas there is insuf- form will still produce incorrect results as ray
ficient well control to allow this accurate map- bending still occurs due to velocity variations
ping and therefore additional velocity control within layers and at layer boundaries.
must be employed. Velocities derived during the
processing of seismic data provide substantial The Dix equation itself is based on sev-
additional control away from wells but are sel- eral assumptions that are rarely honoured in the
dom accurate representations of the earth veloc- real world. Equation (1) assumed that the travel
ity. The most common approach to using this times to the two layers bounding a given inter-
velocity information is to use the Dix equation val are essentially identical apart from the travel
(Dix, 1955), which relates interval velocity to time within the layer itself. When the layers are
root-mean-square (RMS) velocities. Equation (1) not parallel, have structural dip or have long off-

3
GEOHORIZONS Vol . 5 No. 1
July 2000
sets, the Dix equation will give erroneous results.
Similarly, when the time interval in the Dix equa-
tion. dt, becomes small, the resulting interval
velocity becomes very large, and the results be-
come unstable. A cut-off should be applied to
the isochron such that values of dt below some
threshold value are not used in the Dix trans-
form. A dt threshold of 200 ms is typically used
in areas with dips between 5 and 15 degrees, re-
ducing as the dip decreases.

Several model-based techniques exist


that can estimate the interval velocity of a layer Fig 2a. Principles of coherency inversion Normal incidence
from the travel time through it, giving more ac- raypaths are modelled by velocity inversion methods to
curate results by avoiding the Dix equation and determine zero-offset travel time through the model
its many assumptions. To accurately estimate
interval velocity, travel times must be calculated inversion is a layer-stripping approach and there-
by CMP ray tracing through an interval veloc- fore requires an interval velocity and depth model
ity-depth model and comparing these modelled for all layers above the interval under investiga-
travel times with those actually recorded by the tion. For the layer being modelled, a range of
CMP gathers. interval velocities is given for each CMP along
the line. At each CMP, local normal incidence
Stacking velocity inversion and coher- ray migration is used to migrate the time inter-
ency inversion (Landa et. al 1999) are two such pretation to its correct position in depth. Normal
processes but differ in the type of input data used incidence ray tracing is than performed on the
for the comparison. Coherency inversion uses interval velocity model to calculate the raypath
actual pre-stack CMP gathers in the time domain and its travel time through the model (Figure 2a).
to correlate with the modelled travel time curves, This horizon depth-migration and ray tracing
whereas stacking velocity inversion correlates the procedure is than repeated for each constant in-
modelled Raypath with a hyperbolic traveltime terval velocity in the given range.
curve implied by the stacking velocity.
Semblance is calculated at each CMP to
Coherency Inversion measure the correlation between the recorded
CMP gathers and the modelled traveltime curve
The key advantage of coherency inver- for each interval velocity tested (Fig 2b). The
sion over Dix-based methods is that it uses ray- highest semblance value represents the velocity
traced modelled traveltime curves to compare that successfully flattens the CMP gathers im-
with actual travel times recorded from the earth. plying that the model honours the velocities along
Lateral or vertical variations in velocity, refrac- the actual raypath through the earth. At each
tion according to Snells Law, and structural dip CMP, the moveout corrected gathers are dis-
within the model are all accounted for by ray trac- played (Figure 3a) and a histogram of semblance
ing and we are therefore no longer limited by the against velocity can be observed for QC (Figure
hyperbolic moveout assumption of the Dix equa- 3b). A semblence display is created for the hori-
tions. Coherency inversion also provides a sta- zon (Figure 4) that is then used to pick the opti-
tistical estimate of uncertainty in the velocity mum interval velocity at each CMP. The veloc-
model (Landa et.al., 1991) ity profile is used to ray migrate the horizon to
depth and the process is continued for the next
Coherency inversion beings with the in- layer in the model.
terpretations of key velocity horizons in the
stacked time (unmigrated) domain. Coherency Coherency inversion uses curved rays and

4
GEOHORIZONS Vol . 5 No. 1
July 2000
accounts for vertical and lateral velocity gradi-
ents allowing modellec travel time curves to ac-
curately match those recorded in the
CMP gathers.

Fig 2b. Principles of coherency inversion. The modelled


zone offset travel are correlated against recorded CMP
gathers to determine the optimum interval velocity
within a layer Fig 4 : When run in continuous mode, coherency inver-
sion and stacking velocity inversion produce a semblance
display along a given horizon. By intepreting the maxi-
mum semblance(blue), the interpreter builds up a velocity
profile this is used to construct the velocity depth model.

Stacking Velocity Inversion

Stacking velocity inversion uses a simi-


lar process to coherency inversion but does not
use pre-stack gathers for the correlation of mod-
elled travel time curves. Instead, stacking veloci-
ties are interpolated to the key horizons identi-
fied on the stack section. These velocities are then
used to calculate a hyperbolic moveout curve that
is correlated with the modeled travel time curve
(Figure 5). The outputs from stacking velocity
inversion are similar to coherency inversion and
provide an estimate of interval velocity along
each horizon. The main assumption of this pro-
Fig 3a. The result of coherency inversion are a set of cess is that the stacking velocity measured from
CMP gathers flattened according to the interval velocity the data represents the hyperbola that best fits
model used. the data. Clearly, this may not be a valid assump-
tion in areas of poor signal to noise ratio or at
greater depths where moveout discrepancy is
more difficult.

Both coherency inversion and stacking


velocity inversion are equally applicable in the
3D world although full 3D ray tracing is required
to accurately model rays from varying offsets and
azimuths. The output from 3D coherency inver-
sion or 3D stacking velocity inversion are maps
Fig 3b. The optimum interval velocity will produce flat
of interval velocity for each layer. These maps
gathers and will have the highest semblance versus can be used directly as an initial model for PSDM
interval velocity or can be calibrated to well data and used for
depth conversion of time structure maps.

5
GEOHORIZONS Vol . 5 No. 1
July 2000
migration algorithms use simplifying assump-
tions to improve computation speeds and this
results in residual errors particularly where there
are lateral velocity variations. Also, time migra-
tion algorithms do not take into account any ve-
locity variations and simply collapse hyperbolic
diffraction curves to the minimum travel time.
These velocity variations may be the result of
intra-formational vertical or lateral velocity
Fig 5: Stacking velocity inversion uses a hyperbolic gradients or cross-fault juxtaposition of differ-
travel time curve calculated from the input stacking ent formations.
velocity model to correlate with modelled travel times.
The correlation is then used to calculate semblance and
determine the optimum interval time

Travel time inversion techniques can be


successfully used to build initial velocity mod-
els for depth conversion or depth migration. The
model based approaches, particularly those us-
ing pre-stack CMP gathers, give more accurate
and reliable estimates of interval velocity than
methods based on the Dix equations.

Map Migration
Fig 6: Schematic illustrating how reflections are
The traditional method of depth conver- mispositioned by time migrated and how image rays can
sion in which time migrated maps are vertically be used to correct for this error(After Fagin, 1991)
scaled to the depth domain, has many inherent
problems caused by lateral velocity variations
that result in horizons being spatially misplaced. Following time migration, the residual
Additional problems include holes in the time spatial error at a given CMP may be calculated
structure grids caused by faults, pinchouts or using image rays (Hubral, 1977). These rays can
unconformities. All of these problems can be be used to position the reflector to its true spatial
successfully resolved using a model-based map position as shown in (Figure. 6). Time migration
migration approach. algorithms works by collapsing a hyberbolic dif-
fraction curve on a CMP gather to its point of
Map migration is a zero-offset 3-D in- minimum travel time, which will generally be at
version and is one of the most powerful tools the apex of the diffraction. The misplacement of
that can be applied to most structural interpreta- the reflector is caused by the assumption that the
tions. Given accurate time structure maps and crest of the diffraction curve lies vertically be-
an accurate layered interval velocity model, map neath the CMP surface location. In any situation.
migration transforms maps from time to depth where the true raypath is bent or refracted by
while correcting the spatial error caused by velocity gradients or contrasts, this assumption
raypath bending and velocity variations within is invalid and the reflector will be mispositioned
and between layers. (Figure 7)

The need for map migration is often over- Any model with structural dip or veloc-
looked by interpreters who assume that their in- ity variation will benefit from map migration
terpretation is correctly positioned if it is picked which can be applied in the same time that it takes
on time-migration data. Unfortunately, most time to perform the more conventional vertical scal-

6
GEOHORIZONS Vol . 5 No. 1
July 2000
ing depth conversion. Figure 8 compares the re-
sults of a standard depth conversion using ver-
tical scaling (Figure 8a) with a map migration of
the same event (Figure 8b). The difference map,
shown as Figure 8c shows depth shifts of up to
100m as a result of the map migration. Figure 8d
is a displacement map showing the lateral move-
ment of grid modes from the migrated time do-
main to the depth domain with displacements of
up to 130 m observed.

Many mapping packages now offer map


migration as an option for depth conversion. Any
prospect that has even moderate velocity varia-
tions, structural dips greater than 3 degrees, or Fig 8(a): Depth structure map produced by vertical
faulting should use map migration techniques in scaling of migrated time depth structure map
place of the conventional scaling methods.

Pre-Stack Depth Migration

The ultimate goal of any seismic


programme is to provide an accurate image of
the subsurface that will enable a reliable geologi-
cal interpretation. Few seismic surveys ever re

Fig 8(b) : Depth structure map produced by image ray


map migration.

Fig 7: Schematic showing focussing distortion and


positioning error caused by time migration. Depth Fig 8(c) : Difference map between scaled and map
migration corrected in both migrated structure.

7
GEOHORIZONS Vol . 5 No. 1
July 2000
ally achieve the desired level of imaging for many
acceptable. New algorithms and techniques can
reasons including distortion of the seismic sig-now process seismic data faster (cheaper) than
ever before and produce superior images of the
nal by noise, earth filtering effects or simply poor
acquisition geometry. As discussed above, the subsurface. One such technology that is being
data represented in the migration time domain rapidly accepted world wide, is the use of pre-
are seldom the best image of the subsurface and stack depth migration (PSDM). Until only re-
reflectors are often spatially mis-positioned. cently, PSDM was reserved for the biggest super
There is an obvious trade-off between precision computers and the process took weeks to com-
and cost during time domain processing of seis- plete a single 2D seismic line. Table 2 shows a
mic data that stems from the immense number comparison of how this technology can easily
of computations required to process a seismic be applied today in a matter of hours using an
survey. This trade-off is normally achieved by inexpensive desk-side workstation. With the re-
making simplifications and assumptions in some cent increases in processing speed, a projection
is made 5 years from now showing that PSDM
of the computing algorithms that often results in
a less than optimal product. will be in the hands of every interpreter and will
eventually become the accepted standard for any
Advances is computing technology mean seismic interpretation project.
this trade-off should no longer be necessary or
So why is PSDM becoming the preferred
process? As discussed, time-migration algo-
rithms result in events being spatially mis-posi-
tioned and the results have the obvious draw-
back of being represented by a two-way travel
time. Depth migration provides an image in
depth, but more importantly, pre-stack algorithms
avoid the many assumptions and simplifications
that cause mis-positioning of events in the time
domain. The reflections are not only positioned
correctly but also, as a by-product of this their
continuity and discontinuities affecting them
(faults) are better imaged.

Figures 9a and 9b respectively, show a


comparison of a post-stack time migrated sec-
Fig 8(d) : Displacement map showing lateral movement tion and a pre-stack depth migrated section. The
of subsurface points when map is migrated to the depth
domain. The maximum displacement shown is 130m.
primary objective in this prospect was the tilted
Note the significant lateral shift in areas with dip. fault block to the right of the well track, indi-
cated by the arrow. The well was planned on the

Table 2: Illustrative runtimes for depth migrating processing. Computer hardware and software advances will soon
make pre-stack depth migrated data the interpreters main tool.
Process Desk top workstation Runtimes(illustratives only)
1990 1995 Today 5 years from now ?
2 D Post Stack Depth Migration 5 hrs 1 hr 15 min <1 min
2 D Pre-Stack Depth Migration Several days 4 hrs 1 hr 2 mins
(current running
time on 8 CPU
parallel machine)
3D Post-Stack Depth Migration(25 sq Km) N/A 1 day 4 hrs 30 mins
3D Pre-Stack Depth Migration(100) sqKm) N/A N/A 4 weeks 2 days
(4 CPU paralleled machine)
8
GEOHORIZONS Vol . 5 No. 1
July 2000
post-stack time migrated section to intersect the
fault block well away from the bounding fault. While PSDM will generally give the in-
However, during drilling, the well missed the terpreter the best possible with todays technol-
primary objective and encountered the fault as ogy, one must remember that depth migration
seen or the PSDM data that was completed after algorithms are far more sensitive to the input
the well was drilled This is a classic example in velocity model than their time-domain equiva-
lents, making accurate velocity modelling a criti-
cal factor in the successful application of the
method. PSDM is often seen as a remedy for all
seismic imaging problems but the method can
only help where an accurate velocity field in the
subsurface can be determined.

A common misconception is that PSDM


seismic volumes will perfectly tie well data when
interpreted. This is not the case for a very simple
reason - the velocity model used for PSDM is
designed to give the best seismic image quality
Fig 9(a): Comparison of Post Stack time migrated data
and Pre-Stack depth migrated data. A well was drilled,
in exactly the same way that stacking velocities
based on the post-stack time migrated data, to test the and RMS velocities do in the time and migrated-
titled fault-block indicated by the arrow. The well failed time domains respectively. Velocities from well
to intersect th eprognosed reservoir and following the bores effectively measure the vertical component
dry hole, pre-stack depth migration was performed on of the velocity field, whereas seismic velocities
the line. typically measure the horizontal component. In
any situation where there are vertical or lateral
velocity variations anywhere along the seismic
raypath, the two velocities will not be coincident.
Consequently, the seismic reflectors correspond-
ing to a well marker will not tie the seismic events
precisely. To correct for this, the depth migrated
volume is scaled back to the lime domain using
the PSDM velocity model, the interval velocity
model is then calibrated to the wells, and the
volume is scaled (not migrated) back to depth
where it will tie the well markers.
Fig 9(b) : With the correct velocity model and accurate
A second misconception is that PSDM
positioning in depth, it is clear to see why the well failed.
The original primary target was not intersected by the
is only of benefit in structurally complex areas.
well bore that passed straight through the fault plane
This is certainly not the case. Any situation that
adjacent to the prospect. has imaging problems caused by ray bending or
velocity problems will benefit from PSDM. Im-
which velocity contrast across the fault resulted aging problems associated with, for example,
in severe ray-bending and consequent mis-posi- shallow gas accumulations, fault shadows, over
tioning of data in the migrated time domain. thrusts, reefs or a hard seafloor can all be suc-
cessfully resolved given an accurate velocity
Amplitude preserving 3D PSDM algo- model, although they may not be considered to
rithms are now available that not only improve be structurally complex. Because PSDM process-
structural interpretations, but will also have a ing can now be completed in the same time as a
dramatic effect on stratigraphic inversion tools conventional time processing workflow, depth
and AVO analysis. imaging is being used increasingly in explora-

9
GEOHORIZONS Vol . 5 No. 1
July 2000
tion areas, where until recently, only time imag-
ing would have been considered.

The third misconception is that PSDM is


a complex process that should be left in the hands
of the specialist processing geophysicist. Whilst
the mathematics behind the PSDM algorithms
may be daunting, the process should never be
applied without substantial input from the inter-
preter. Indeed, some of the best PSDM results
have been created by interpreters because they
usually have a better understanding of the earth
model than their processing geophysicist coun-
terparts. This knowledge of what is geologically
reasonable is the key to building a robust, accu-
rate velocity model that will provide outstand-
ing results in the depth domain.

Velocity Model Refinement Fig 11. Depth CRP gather showing residual delays
caused by depth migration using an incorrect velocity
The quality of a velocity model used for model. The delay dz is determined and used as input to
PSDM can be assessed by analysing the output tomography.
common reflection point (CRP) gathers. For a
given CRP gather, all rays will have sampled the based and grid-based global tomography. The
same point on the subsurface irrespective of their new velocity model resulting from the updates
source - receiver geometry, and it would be flat is then used to re-run the PSDM, after which the
if the correct velocity model was used for migra- process is repeated until the depth CRP gathers
tion. are flat. Figure 10 shows a typical tomography
workflow to refine the velocity depth model.

Fig 10. Typical work flow for refining velocity models


by global tomography

Any residual delay in the CRP gathers Fig 12: Schematic showing how the error in travel time
not only degrades the migrated image but also at reflection point A is the result of an accumulation of
implies that the spatial position of the reflectors errors, dti, within each layer where the ray has trav-
will be incorrect because of an incorrect model. elled. The objective of tomography is to determine dti,
the error in travel time within each layer, and derive
By analysing these residual delays, the model is from that, the error in velocity, dvi; and the error in
refined through a number of techniques ranging depth dzi, each parameter is then used in the tomography
from hyperbolic delay corrections to horizon- equation to update the model.

10
GEOHORIZONS Vol . 5 No. 1
July 2000
Tomography is a sophisticated tool that the same way as normal move out velocity is used
attempts to correct errors in the velocity depth to flatten gathers before stacking. This residual
model by analysing the residual delays after RMS velocity is then transformed to a residual
PSDM. Tomography is a global approach that interval velocity, which is used to update the ini-
can translate an error in time at one location to tial velocity model. It is important to remember
an error in depth or velocity at any point in the that this method makes all the Dix assumptions
model through which the ray has travelled. These and therefore where the delays are non-hyper-
errors are solved simultaneously by making bolic, the gathers will still not flatten correctly
changes to the velocity and depth model across and a tomographic method should be applied. It
the entire section. is also important to note that any remaining er-
ror in the overburden will propagate down to the
Migration with an incorrect velocity will next layer and consequently the method is best
result in depth delays, dz on the CRP gather (Fig- applied in areas with simple geology with no
ure II). The first step in tomography is to scale strong lateral velocity variations.
the depth-migrated gathers to tune and where
errors exist, the time at far offset, t, will differ Global Depth Tomography
from the time at zero offset, t This delay, dt, is
0
measured for each CRP and used as input to the Horizon-based and grid-based tomogra-
tomography algorithms. phy are global approaches that solves a simulta-
neous set of equations to produce update param-
The tomographic approach attributes the eters (depth and velocity) for the model. The two
delay, 4t that was measured from the CRP gather, techniques differ in their inputs to the algorithms.
to an accumulation of errors within each layer During horizon-based tomography, the interpreter
along the raypath. The objective is to obtain the analyses delays at each CRP along the line for
error in velocity and the error in depth within each horizon in the model and these are input to
each layer from the error in travel time within the tomography along with the interpreted geo-
each layer. For example, in Figure 12, the residual logical horizons in depth. Grid based tomogra-
time delay observed at point A is an accumula- phy is an automated approach where the delays
tion of velocity and depth errors in each layer are picked by the computer along small, coher-
along the raypath. By updating the velocity-depth ent segments of the data interpreted on the depth
model in all the layers, the residual travel time migrated section. An important difference in the
error at point A can be minimised. An important two approaches is the way in which the model is
feature of global tomography is that errors are updated. In horizon-based tomography, both the
simultaneously solved using least squares to depth interpretation and the interval velocity
minimise the error in travel time across the whole model are updated whereas grid-based tomogra-
model. phy only updates the velocity section.

Hyperbolic Update of Velocity Grid tomography is useful in a number


Hyperbolic updating is a technique used of situations where a horizon-based method may
to flatten gathers following PSDM with an in- not be applicable. An example of this may be in
correct velocity model. This method is based on very complex structures or on poor data areas
the Dix formula and although it is not strictly a where it is too difficult to interpret continuous
tomographic approach, it can be successfully reflectors through the model. In all cases, grid
applied to update velocity models where the re- tomography requires a good initial velocity
sidual delay approximates a hyperbola. model, derived from seismic processing veloci-
ties, for it to produce reliable results.
Depth CRP gathers are scaled to time and the A typical workflow for horizon-based to-
residual delay is measured. By using this delay mography begins by analysing the residual depth
in the Dix equation, we can calculate an RMS delays along the key horizons. Figure 13a shows
velocity error that will flatten the gather in much a typical semblance display where the velocity-

11
GEOHORIZONS Vol . 5 No. 1
July 2000
grating the data, the resulting gathers are flat (Fig-
ure 13c) and the semblance display shows no
residual error indicating that the velocity depth
model for that layer is now correct (Figure 13d).
Note the improved image quality when the data
is migrated with the correct model.

Fig 13 a. Depth migrated CRP gathers before tomogra-


phy. An incorrect velocity model causes residual delays,
shown by curvature on the gathers and high semblence
values away from zero line on the semblence plot.

Fig 13(c): Depth migrated CRP gathers after horizon


based global tomography. The CRP gathers are now flat
and the maximum semblence falls along the zero delay
line.

Fig 13(b) : Horizon delay analysis. The semblence


display indicates where there are residual depth delays
along a given horizon. In this example, there are
significant delays remaining on layer 4 indicating the
model needs updating.

depth model is incorrect. Where the model is


correct (i.e. where the gathers are correctly flat-
tened), the maximum semblance, indicated by
blue colours, lies along the zero delay line. Posi-
tive and negative delays are represented by maxi-
mum semblance by lying away from the zero line.
Fig 13(d): The horizon semblance display after tomogra-
These delays are picked by the interpreter from phy shows zero delays remaining on Layer 4. The depth
the semblance displays along each horizon (Fig- migrated image is significantly improved on the initial
ure 13b) and are used as input to the tomogra- migration shown in Figure 13(b)
phy. After running the tomography and re-mi-
12
GEOHORIZONS Vol . 5 No. 1
July 2000
For grid tomography, it is not essential
to have horizons mapped across the whole sec-
tion. Instead, small segments of coherent data are
interpreted along the line and these are used to
analyse the delays for input to the tomography.
The software application automatically picks the
residual delays around the small horizon seg-
ments without any intervention form the user.
These delays are used as input to the tomogra-
phy which attempts to minimise the errors across
the whole section by updating the velocity field
only.

Fig 14(b): Velocity model after one pass horizon based


tomography. Note the significant changes in velocity to
layer 3 and 4. Elsewhere on the section, few changes
have been made to the model.

tomography. While both models produce flat


CRP gathers, the grid based approach introduces
non-geological artefacts in the interval velocity
model.

While automated grid-based tomographic


techniques are useful in very complex data ar-
eas, they should generally be avoided until the
model is very close to a solution; i.e. depth de-
lays are small. There are two schools of thought
about how good the initial model must be for
tomography to converge on a solution. One
Fig 14(a): Effects of Global tomography on the velocity
model a) initial velocity model used as input for pre-
school advocates using a very crude initial model,
stack depth migration. relying on numerous iterations of tomography to
A common question that is asked is,why update and correct the model. The other school
not only use grid-based tomography that is more refines the model as much as possible before run-
automated than horizon-based tomograpy ? The ning PSDM and only uses tomography to make
answer is very simple. Grid based tomography minor adjustments. Experience with tomography
attempts to solve the errors in a statistical sense has shown that the algorithms typically converge
without any constraint on the geological model. on a solution within 3 iterations providing the
Without that constraint, grid-based tomography initial velocity model is within 15% of the final
can produce very erratic velocity models and solution. Beyond that, the model either will not
geologically implausible results, which in turn converge or converges on a wrong solution, re-
produce false structures or poor imaging in depth. sulting in incorrect depth images. Therefore, it
It is recommended that horizon-based tomogra- is recommended that the model is refined as
phy is used whenever possible, only using grid- much as possible before running PSDM.
based tomography for final, minor refinements
to the model. Figure 14a shows an inital velocity Emerging Tomographic Techniques
model, and the results following both horizon
based tomography (Figure 14b) and grid-based The one obvious flaw in the tomography

13
GEOHORIZONS Vol . 5 No. 1
July 2000
workflow is the fact that PSDM must be run be-
fore and between model updates. Even where PSDM is not required, in-
terpreters should be using tools such a coherency
Interactive tomography is an emerging inversion and stacking velocity inversion to more
technology that will speed up the whole tomog- accurately determine interval velocity and avoid
raphy workflow by allowing the interpreter to the many problems inherent in the transforma-
perform all of the model updates tion of seismic velocities by the Dix equation.
prior to running the PSDM Map Migration should always be used for depth
conversion as method produces more accurate
By creating local depth CRP panels results than scaling methods and accounts for the
around each horizon, residual interval velocity lateral mis-positioning of events introduced by
can be analysed and interpreted. Global tomog- time migration algorithms.
raphy can be run using these delays and the whole
process repeated until the local gathers around The key message of this paper is that the
each horizon are all flat and no residual delays interpret geological knowledge of the region is a
are observed. Only when the interpreter is satis- critical constraint any velocity model building
fied that the optimal model has been derived, or depth imaging project and role of the inter-
PSDM is applied to the data. This process will preter should not be understated. Automated
result in at least a three-fold reduction in the time model building and refinement techniques that
required for the PSDM workflow as it required requires interpreter input should be treated with
only one pass of depth migration on the data. considerable caution as they routinely produce
unrealistic models of the geology.
Conclusion
References
The end objective of any exploration-
mapping project is to provide a reliable under- Dix, C.H., 1955, Seismic velocities from surface
standing of the shape and form of the subsurface measurements: Geophysics, 20,68-86.
in depth. Conventional mapping techniques have
relied on interpretation of data in the time do- Fagin, S.W., 1991, Seismic modelling of geo-
main followed by a depth conversion, frequently logic structures: applications to exploration prob-
using an oversimplified velocity model. lems. Geophysical Development Series Volume
2 Soc. Expl. Gophys.
The interpreter now has many tools in
hand that allow accurate imaging of the earth in Hubral, P., 1977, Time migration-some ray-theo-
the depth domain. PSDM is considered the ulti- retical aspects : Geophys Prosp. 25,738-745
mate tool for seismic imaging for a good reason
- the quality of the resulting images is not only Landa, E Thore, P., Sorin, V, & Korcn, Z., 1991,
superior to that in the time domain, but the seis- Interpretation of velocity estimates from coher-
mic data is also correctly positioned in space al- ency inversion : Geophysics, 56(9), 1377-1383
lowing the interpreter to work directly in the same
domain as the well data. The workflow used for Levin, F.K. 1971, Apparent velocity from dip-
PSDM implicitly results in an interpretation of ping interface reflections: Geophysics 36, 510-
both the key structural horizons in depth and an 516
accurate velocity model in the same timeframe
as that required for conventional time process-
ing.

14
GEOHORIZONS Vol . 5 No. 1
July 2000

You might also like