SPECIAL

M a r i nSECTION:
e and o
Mfaf rs ihnoer e
a ntde cohf n
fosh
log
rey technology

$VLPSOLHGZRUNRZIRUDFFXUDWHWLPHWRGHSWKFRQYHUVLRQXVLQJ
'JULGWRPRJUDSK\
FABIO MANCINI, Woodside
Downloaded 04/09/13 to 76.8.139.2. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

T ime-to-depth conversion is the process where seismic

velocities are scaled accordingly to check shots and well
logs so that geological markers and seismic horizons match
Because most of todays seismic projects are run
through PSDM, an elegant way to derive the velocity vol-
ume for time-to-depth conversion is to use the same tool
in depth. The nal stacked seismic volume, if imaged in used for PSDM velocity model building, the 3D grid to-
the depth domain, is stretched to time with the migration mography. Tomography transforms the residual move out
velocity, or a smoothed version of it (Jones, 2009), before (RMO) picks or curves into depth errors for all the osets
redepthing with the calibrated velocity model. There are in the CDP gathers. These depth errors are inverted into
many dierent techniques used routinely in the industry (see traveltime errors and corrected by minimizing the dier-
Etris at al., 2001, for a detailed summary), by scaling either ences between the observed traveltimes and those comput-
average velocities or interval velocities with some derived ed by ray tracing for the current guess of the model (Jones,
correction functions. These functions can be simple scalars, 2010).
depth variant scalars, or more complex time and space variant The well-seismic mis-ties can be used in the same way.
functions, which need to be interpolated between wells. In In this case (well-tie tomography), because it is a vertical
any case, the depthing process is quite laborious because update, it is sucient to ray-trace only the normal ray (Fig-
the derivation correction functions for each well is time- ure 1b), which makes the process much faster compared to
consuming. The application of these correction functions conventional tomography, which uses rays for a large-angle
may produce results that look somehow discontinuous, range (Figure 1a). This process can be run comfortably on
with jumps in velocity at the horizons where the corrections
have been applied, and, when working on average velocities,
interpolated between wells, they could produce unrealistic
interval velocity in the presence of thin layers.
In areas of complex geology, it is not unusual to have
residual depth mis-ties between seismic data and well mark-
ers even when more advanced techniques for velocity model
building and imaging, including VTI and TTI anisotropic
prestack depth migration (PSDM), are applied. There can
be several reasons for this. For one, when VTI and TTI
analysis are performed, the output Thomsen parameter ,
which controls the depthing, is often gently smoothed over
large intervals, although geologically constrained, to avoid
sudden discontinuities. Then generally further tomographic
iterations are run, which update only the velocity, leaving
the anisotropy values unchanged, thus introducing residual
shifts in seismic depthing. Another reason for the presence of
residual depth mis-ties is that, in reality, even the best tomo-
graphic velocity model is only a low-frequency representation
of the true subsurface velocity. Seismic tomography always
produces relatively smooth velocity models; their resolution
limit is given by the size of the tomographic inversion grid
cells. Although resolution has improved signicantly in re-
cent years, the cell size is still in the order of the few hundred
meters in x, y, and z, so that, in eect, tomography updates
only the longer spatial wavelengths of the velocity model
(Jones, 2010). To add to this, a damping factor is usually
applied which tends to smooth the results to avoid instabili-
ties. Recent advances in full waveform inversion have shown
velocity models with shallow high-frequency details which
were never seen before. The fact that we are not able to ac- Figure 1. (a) Shooting rays for conventional tomography, and (b)
count for this variability leads to errors in seismic imaging well-tie tomography. Only the normal ray is needed to minimize
the depth mis-ties, compared to a large fan of rays to account for the
and its depthing. oset aperture of a full CDP gather, making the process much faster.

430 The Leading Edge April 2013

 Marine and offshore technology

a single workstation for projects of medium size. In our of approximately 900 km 2 on a 16-CPU machine with 49
case study, each iteration took only a few hours for an area Gb of memory.
In a way, this process is similar to that described by
Zdraveva et al. (2010), in which the authors used well/seis-
mic mis-tie maps to derive detailed values of , because the
vertical velocity, V0, is linked to the NMO velocity by the
Downloaded 04/09/13 to 76.8.139.2. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

relationship: Vnmo2 = V02(1 + 2). The dierence here is that I

take the process outside the PSDM velocity building phase
with the only scope of ne tuning the seismic depthing. In
fact, given horizon grids in the time-migrated domain and
well markers, the process can run without any seismic data.
The workow is basically a conventional tomography veloc-
ity-model building routine, where instead of RMO picks,
mis-tie values are inverted. In more detail, starting with the
initial velocity model:

1) Depth-convert target horizons from time-migrated do-

main
2) Calculate depth mis-ties for each horizon and grid them
into mis-tie maps
3) Perform 3D grid tomography to minimize depth mis-
match at all horizons
4) Depth-convert target horizons with new velocity model
and calculate new depth mis-ties

Steps 3 and 4 are repeated until we are satised with the

residual errors. As in conventional velocity model building,
we decrease the grid cell size of the inversion and increase the
sensitivity of the tomography as we progress with the itera-
Figure 2. (a) Input velocity model, after PSDM, and (b) the tions. Residual mis-ties after each iteration are analyzed to
nal velocity model after well-tie grid tomography. The color scale verify convergence of the solution both in terms of rms error
is the same. Most of the changes are focused in the high-velocity and minimal and maximal mis-tie values.
carbonates and in the low-velocity layer underneath. The initial
PSDM velocity was mildly smoothed to avoid artifacts in the Pluto case study
time-to-depth conversion due the sharp velocity changes in the
water-bottom canyons. The Pluto Field is located in permit WA-350-P in the North-
ern Carnarvon Basin, Australia, and
was discovered in April 2005. Almost
all of the hydrocarbon resources are
within the upper Triassic, Jurassic,
and lower Cretaceous sandstones be-
neath the regional early Cretaceous
seal. This is a key area for Woodside;
its operated Pluto LNG Project start-
ed production in 2012.
Imaging of the Pluto 3D seismic
data has been historically dicult
because of the complex overburden.
In particular, (a) the water bottom
steeply slopes from the shelf to the
platform, with depth changing from
100 m to 1100 m. Water velocity is
likely to change and cannot be treat-
ed as a uniform body. (b) Numer-
ous canyons run through the water
bottom, which has a rugose appear-
Figure 3. Structural framework with the ve horizons used in the project, from the top: water ance. (c) Fast carbonates are present
bottom, H1, H2, H3 and H4. in the overburden and tend to pinch

April 2013 The Leading Edge 431

 Marine and offshore technology
Downloaded 04/09/13 to 76.8.139.2. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

Figure 4. (a) Mis-tie maps for the shallower H1 horizon, and (b) the target horizon H4. H1 mis-ties range from 38 m to +10 m with an
rms error of 27 m. H4 mis-ties range from 108 m to 65 m, with an rms error of 90 m. A similar pattern can be seen with the area of largest
mis-ties matching the area of maximum dip of the water bottom (c).

distribution. Five key horizons were

chosen to tie the wells, and the third
and fourth horizons (H2 and H3,
yellow and purple in Figure 3) lo-
cally become closely spaced, with a
minimum separation of 50 m, which
would create diculties to convention-
al depthing methods which use average
velocities. In fact, when the layers used
for correction are closely spaced, small
perturbations in the interpolation and
gridding of the correction factors away
from the wells between layers can pro-
duce an unrealistic interval velocity,
which translates into erroneous layer
thickness.
Figure 5. Summary of the results of the time-to-depth conversion: Minimal and maximal Because I am dealing with an
mis-tie and rms errors are represented for each of the ve horizons. Depthing with the new isotropic velocity model, there is no
velocity reduced mis-ties to less than a depth sample, 5 m, for all horizons at every well. In the
columns named Input in Figure 5, we can see the initial mis-tie errors for all ve horizons, with a surprise to see large errors at target
maximum mis-tie of 108 m and an rms error of 90 m for all the nine wells (green bars). There is also depth (H4, in green in Figure 3). For
a large variability between minimum and maximum errors (red and blue bars) for each horizon. the H4 horizon in particular, the dif-
ference reaches the order of 4050 m
out in proximity of the water-bottom slope (Figure 2a and for wells relatively close to each other (4 km distance, Figure
Figure 3). Because of the complex overburden, traveltime to- 4). This variability is already partially present in the shallow H1
mography seems to struggle to nd a correct structural solu- horizon (dark blue in Figure 3). Figure 4 shows the mis-tie maps
tion. The PSDM project was performed in 2007. Anisotropy for H1 and H4. A similar pattern in both maps can be seen, with
was not taken into account, so no well scaling was attempted the largest mis-tie area roughly aligned with the dipping part
during velocity model building. The PSDM velocity model, of the water bottom, shown in Figure 4c. This is a conrma-
shown in Figure 2a, was derived after ve iterations of 3D tion that during velocity model building, the tomography
grid tomograph. It is quite detailed with the presence of sig- struggled to capture the details of the overburden velocity.
nicant geological structures. A complex zone of high-veloc-
ities (carbonates) is evident with velocity inversions observed Tomography application
below the carbonates. Given the size and high variability of the mis-tie values, I
Figure 3 shows the structural framework and the lo- performed four iterations of grid tomography to build the
cation of the nine wells used, which have a good spatial new velocity model. The nal results are summarized in
432 The Leading Edge April 2013
 Marine and offshore technology

the columns named Tomography

in Figure 5. After application of to-
mography, mis-tie errors are reduced
for all ve markers at each well to
approximately one depth sample, 5
m, both in terms of rms error and,
Downloaded 04/09/13 to 76.8.139.2. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

probably more importantly, in the

values of the minimal and maximal
mis-ties, which could be considered
our error bar. Looking at the output
velocity volume, Figure 2b, we can
see that the correction was strongly
focused on the fast shallow carbon-
ates which were too fast in the input.
The overall velocity changes were in
the order of 10%, which is not a sur-
prise considering that anisotropy was
not accounted for in the original ve-
locity model building.
One question at the start of the
project was how the tomography
would update an already detailed
velocity model given the relatively
sparseness of the input data (nine
wells and ve horizons). In other
words, would we see artifacts linked
to the horizon positions and the
shape of the gridded mis-tie maps?
Careful analysis of the output vol- Figure 6. Composite line through four of the nine wells used for the seismic volume depth-
ume reassured us that the changes converted with the initial (a) and the nal (b) velocity volumes. Depth markers are represented by
were smooth and distributed through small disks of the same color of the horizons and indicated by the arrows. The size and variability
the entire volume. No imprint of the in mis-ties are evident in the input volume. In the output the horizons intersect all the markers.
input horizons could be seen. It looks
like the tomography has operated within the high-frequen- the tomographic inversion, the nal velocity (green curves)
cy limits of the input model, without creating new small- matches the sonic values (red curves) with a good degree of
scale features that did not exist before. The smoothness of accuracy. The input velocity volume (blue curves) is generally
the solution provides the added advantage that small errors too fast. Again, it is important to notice that the tomography
in the horizon interpretation, like possible mispicks during has not altered the basic shape of the velocity curve, but only
autotracking, have little or no eect on the results, because adjusted it to match the well sonic.
the corrections are not hard-wired to the horizon. A compos-
ite line intersecting four of the nine wells is shown in Figure Conclusions
6, for the data depth-converted with the input (Figure 6a) A simple and fast method to perform the nal time-to-
and the nal velocity model (Figure 6b). The well markers depth conversion is 3D grid tomography, in which the in-
are displayed as disks of the same color of their correspond- version aims to minimize only the vertical mis-ties between
ing horizon and are indicated by arrows. After tomography, wells and seismic data rather than the traveltime errors at
the horizons intersect all the markers; while before, the mis- all osets in a common image gather. Given that most of
tie between markers and horizon is evident. After further todays seismic projects are run through PSDM, it seems
quality control (QC), it was comforting to notice that no only natural to continue to shape the velocity model using
small-wavelength structural artifacts, such as spurious pull- the same tool used in the PSDM velocity building cycle.
ups or pull-downs, were present in the seismic volume after Only the normal ray needs to be shot, making the process
depth conversion with the nal velocity. fast so that it can easily be run on a single workstation.
For the nal QC, I extracted the velocity values at the This methodology avoids lengthy processes of correction
well locations for the input and nal volumes and compared using externally derived functions and/or transformation
them with the smoothed sonic logs. Figure 7 shows the re- of velocities from interval to average and back. Because of
sults for two wells, one with the smallest initial mis-tie at tar- the way it works and its execution speed, it is well suited
get, 38 m for Pluto-1, and the other with the largest initial for updating models each time new wells are drilled and
mis-tie, 108 m for Pluto-2. Although well logs did not enter included in the mis-tie analysis.
April 2013 The Leading Edge 433
 Marine and offshore technology

In this example, I started from a suboptimal situation, be-

cause the PSDM model building was only isotropic and the
overburden velocity was complex, leaving large residual mis-
ties with important lateral variability. Still the results were
robust, with tomography producing a smooth model, which
tied all the markers for all wells within a velocity depth sam-
Downloaded 04/09/13 to 76.8.139.2. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/

ple. Each of the four iterations managed to gradually reduce

the mis-tie values. Careful QC proved that no small-scale
structure or imprints of the mis-tie maps were present in the
output model. The resulting velocity also closely matched the
smoothed sonic logs.

References
Etris, E. L., N. J. Crabtree, and J. Dewar, 2001, True depth conver-
sion: More than a pretty picture: CSEG Recorder, 26, 1122.
Jones, I. F., 2009, Tutorial: Time conversion of depth migrated
data: First Break, 27, 5155, http://dx.doi.org/10.3997/1365-
2397.2009013.
Jones, I. F., 2010, Tutorial: Velocity estimation via ray-based tomogra-
phy: First Break, 28, 4552.
Zdraveva, O., A. Bakulin, and Y. Liu, 2010, Building geologically
plausible anisotropic models using well data and horizon-guided
interpolation: First Break, 28, 7780.

Acknowledgments: I thank Woodside for permission to publish this

Figure 7. Seismic velocity from the input volume, blue, and the paper and Paradigm for help and feedback in running the 3D grid
output volume, green, compared with the smoothed sonic log, red,
for wells Pluto-1 and Pluto-2, which show dierent initial mis- tomography.
tie values. The output velocity matches the sonic log with a good
degree of accuracy. Corresponding author: [email protected]

434 The Leading Edge April 2013