Professional Documents
Culture Documents
Integrated Velocity Model Estimation For Improved Positioning With Anisotropic PSDM
Integrated Velocity Model Estimation For Improved Positioning With Anisotropic PSDM
net/publication/264858820
CITATIONS READS
15 123
1 author:
Tom Dickens
ExxonMobil
24 PUBLICATIONS 185 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Tom Dickens on 19 September 2016.
T here are many geologic settings We call the process of building a model
Downloaded 09/19/16 to 158.26.2.162. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/
where anisotropic migration is neces- that integrates all data integrated veloc-
sary to obtain accurate seismic images. ity model estimation.
While this is well known, stable aniso- In this section we present our inte-
tropic parameter estimation has posed grated velocity model estimation solu-
a serious challenge. Seismic data, tion. In particular, we will discuss
though extensive in coverage, cannot some hurdles that can make integra-
constrain the anisotropy parameters tion difficult and our solution to over-
alone (Tsvankin and Thomsen, 1995). come them. Our early attempts at
The set of parameters is better con- integrated velocity model estimation
strained by integrating the seismic led to some significant successes in
information with certain types of well improved imaging. However, we
data. However, the well data are gen- found that interpreters could not use
erally sparse, so the parameters are these images, because they usually had
only constrained at a few locations. structural artifacts associated with
Nonuniqueness is obviously a funda- velocity artifacts in the model. We
mental issue in our estimation prob- found that these artifacts were the
lem. To produce stable and consistent result of the following problems:
estimates for the anisotropy parame-
ters, we have developed an approach • apparent inconsistencies between
that incorporates the following as- the various sources of velocity
sumptions: information (this issue has gener-
ally been eliminated by including
• All available data should be inte- anisotropy)
grated into the model building • instability in the methods used to
process. convert measured data to interval
• The model should only be as com- velocity (e.g. Dix conversion mag-
plex as the data requires. nifying the errors in the measured
• The model should be geologically RMS velocity)
reasonable. • localized velocity anomalies due to
• A priori geologic assumptions can the sparse coverage (and hence,
be used to constrain the model only local influence) of some data
where the data are inadequate. Figure 1. Interpretive workflow for integrated types (e.g. well data)
model estimation. • unnecessary variability added to
There are three key requirements the model where the values were
for implementing our approach. First, we need a way to not well constrained by the data
assess the consistency of the model with many different
data types. Second, we must be able to easily adjust the A common solution for integrated model estimation is
model to improve the fit. Third, we need a model repre- to use tomography. Each data type of interest is incorpo-
sentation that allows values to be defined at sparse locations rated in the tomographic cost function along with geologic
and for those values to be spread through the rest of the constraints and appropriate weights. Tomography does
model in a way that is consistent with any geologic assump- address some of the issues listed above. In particular, tomog-
tions we have made. raphy measures the model’s fit to the data in the data
In this paper, we will present our solution for meeting domain. This avoids the instability that is always associated
these requirements. We will also present our model build- with converting a field measurement to the model domain
ing procedure and three anisotropic prestack depth migra- (i.e. interval velocity). So, for example, check-shot traveltime
tion (APSDM) examples from marine sedimentary basins. measurements are integrated into the model by comparing
For each data set, we were able to fit all the available data forward-modeled with measured check-shot times. This is
while producing geologically reasonable models. We also as opposed to converting the check-shot measurements to
observed improved positioning accuracy compared to interval velocities and comparing those velocities to the
isotropic depth images. model’s velocity. Also, with attention to parameter “tweak-
ing,” the difficulty with sparse coverage of some data types
Integrated velocity model estimation. Surface seismic reflec- can be largely overcome in tomography.
tions, surface seismic direct arrivals, well data, and prior geo- We tested tomography as an approach to integration, and
logic information can be used to constrain a subsurface found that it often produced large improvements in imag-
velocity model. These various sources of subsurface veloc- ing beneath velocity anomalies. However, geologic inter-
ity information have different strengths and weaknesses. preters did not use these images, because they contained
Integrating all available velocity information into a veloc- structural artifacts. These artifacts were caused by variabil-
ity model increases the accuracy of the model by offsetting ity in the velocity model that was not necessary to fit the
the weakness of one data type with the strengths of another. data, but was added by the tomographic process. The arti-
Figure 7. Slices from vertical velocity and η models for Area B. The veloc-
ity follows a compaction trend, with a slight dip following the water
bottom, while η dips more steeply, following a shaly sequence.
with the water bottom trend (not local structure). The fit to
a subset of the available well data is shown in Figure 2. Figure 9. Well path overlying anisotropic PSDM result for area B. The
well, whose path was chosen based on the migrated data, encounters the
Figure 5 shows that to fit all the data available, the ver- top and base of the reservoir within ±10 m of the depths predicted from
tical velocity values generally did follow a compaction trend. the seismic data.
However, the trends for the anisotropy parameters can be
seen to be more consistent with structure (Figure 5 only was used as the extrapolation direction for both velocity and
shows the values for η, but the values for δ are similar in anisotropy parameters; however, it became apparent that it
magnitude and distribution). Since the interpolation/extrap- would be difficult to fit the seismic data while simultane-
olation was biasing the anisotropy to follow the compaction ously selecting δ and η to correlate well with each other in
trend, this structural trend was solely due to efforts to match depth, as one would expect on physical grounds.
the data—particularly the OCS data. Another interesting Changing the extrapolation direction for anisotropy
observation is that the anisotropy values appeared to qual- parameters to follow structure, we found that indepen-
itatively match the shape of the gamma logs filtered to seis- dently picked η values correlated well with regions of non-
mic frequencies, at least to the depths constrained by the zero δ and with structure. Maps of the δ and η parameters
check-shot data. This suggests that the anisotropy values are clearly show their areas of largest magnitude closely corre-
not just structurally controlled, but are lithology dependent. spond with a massive shale formation, while vertical veloc-
This is not particularly surprising (Alkhalifah and Rampton, ity again follows a compaction trend. Figure 7 shows a slice
2001), but, again, this dependence appeared solely from try- through the velocity and η models. Figure 8, a plot of the
ing to match check-shot, sonic log, and seismic data. difference between predicted and observed OCS travel-
Figure 6 shows the final APSDM result for area A, com- times, illustrates that we were able to fit the OCS data out
pared to a previous isotropic poststack depth migration. to large offsets with the structurally interpolated anisotropy
Gamma ray logs are overlaid on the section to illustrate the models. (Note the extremely poor OCS fit obtained with an
accuracy of positioning of sands adjacent to the salt body. isotropic velocity model.)
Note that even the steepest sands are properly positioned. The APSDM data for Area B were used to plan the loca-
In contrast, in a previous isotropic PSDM result, vertical tion of a well. Figure 9 illustrates the positioning accuracy
depth errors of 1000 ft were present. we were able to obtain by incorporating anisotropy into the
Model building at Area B, another marine sedimentary velocity model. The figure shows the highly deviated well
basin setting with an OCS survey, presented additional chal- path and the relatively steeply dipping reservoir (~ 30°); the
lenges because it has very few well locations. Since δ can migrated section matches the formation tops marked on the
only be quantified at well locations, we were forced to well path to within ±10 m, while previous work based on
extrapolate it over relatively large distances, using either isotropic prestack time migration led to depth errors of as
compaction or structural trends. Initially, the water bottom much as 80 m. It is clearly advantageous to include