American Oil and Gas Reporter - July 2019 - 70

SpecialReport: Reservoir Characterization

FWI Images Complex Subsalt Features
By Zhigang Zhang,
Jiawei Mei,
Feng Lin,
Rongxin Huang
and Ping Wang
HOUSTON-Full waveform inversion
(FWI) goes beyond refraction and reflection tomography techniques that use
only the travel-time kinematics of the
seismic data by incorporating more information provided by the amplitude and
phase of the seismic waveform.
The result is a high-resolution velocity
model that can be used to resolve complex
geological features, improve reservoir
imaging and identify potential geohazards.
The highly detailed models also can be
input directly to pore pressure prediction
flows, time-lapse monitoring and reservoir
characterization analysis.
Subsurface images derived from FWIbased modeling provide interpreters with a
much higher degree of confidence with respect to reservoir delineation and subsequent
well planning. However, using FWI to
update velocity models that contain complex
salt bodies with high velocity contrasts has
been a challenge. It is even harder if largescale salt geometry errors are part of the
starting velocity model for FWI.
At the BP-operated Atlantis Field in
the Gulf of Mexico's subsalt Miocene
play, FWI recently was applied to wideoffset ocean-bottom node (OBN) 3-D
data acquired with a usable low-frequency
signal down to 1.6 hertz. The unmodified
FWI velocity model was then used to reverse-time migrate the seismic data. According to a paper presented by BP at the
Society of Exploration Geophysicists'
2017 Annual Meeting, the FWI-based
velocity model generated "some of the
best subsalt images ever seen at Atlantis

and clearly revealed several major interpretation errors in the legacy salt model"
(SEG 17738630, "Salt Model Building
at Atlantis with Full Waveform Inversion,"
by Xukai Shen, Imtiaz Ahmed, Andrew
Brenders, Joe Dellinger, John Etgen and
Scott Michell).
The Atlantis Field FWI work stressed
the importance of low frequencies, full
azimuths and long offsets of the OBN
data. Encouraged by the success at Atlantis,
some aspects of FWI algorithms have
been revisited to minimize the cycleskipping and amplitude discrepancy issues
that are common in the presence of salt
and salt misinterpretations. The newly
developed FWI algorithm, called "timelag FWI" uses travel-time misfit measured
in frequency-dependent time windows as
the FWI cost function. It is devised to
minimize the negative impact from the
amplitude discrepancy and cycle skipping
between the recorded data and modeled
synthetic data.
Furthermore, time-lag FWI uses a
cross-correlation coefficient between
recorded data and synthetic data as a
weight function in gradient computation
to promote higher quality travel-time
measurements. The effectiveness of this
approach has been demonstrated using a
full-azimuth streamer dataset in a field
with complex shallow salt bodies in the
Keathley Canyon area in the Central Gulf
of Mexico.
Salt Model Building
Building an accurate salt velocity
model is a critical step in subsalt imaging.
Standard practice for salt model building
is to use tomography, sometimes combined
with shallow diving-wave FWI, to first
build the best possible sediment velocity
model. Then, a sediment-flood migration

FIGURE 1
Full-Azimuth Towed Streamer Dataset in Keathley Canyon Study Area

70 THE AMERICAN OIL & GAS REPORTER

is used for top-of-salt (TOS) interpretation,
followed by a salt-flood migration for
base-of-salt (BOS) interpretation. In the
next step, a number of salt scenario tests
usually are required to resolve the salt
geometry, especially for complex areas.
This procedure is not only labor-intensive
and time-consuming, but it is also prone
to human misinterpretation.
FWI is considered the most promising
data-driven tool to automatically build
velocity models by iteratively minimizing
the difference between recorded data and
modeled synthetic data. Many successful
examples have been reported using FWI
to update shallow sediments, gas pockets
and mud volcanoes. However, FWI had
not been used successfully to update salt
structures until the Atlantis Field project
using low-frequency, full-azimuth and
long-offset OBN data.
Two major issues that cause FWI to
fail are cycle skipping and amplitude discrepancies between synthetic data and
recorded data, particularly when there are
sharp contrasts and large-scale
geobodies/salt bodies in the model. When
phases, or travel-time differences, between
synthetic and recorded data are large, the
oscillatory pattern of the seismic signals
introduces local minima. FWI may attempt
to fit the synthetic data to a wrong cycle
in the recorded data. In addition, the amplitude distribution of recorded seismic
data is usually very different from that of
synthetics. If this difference is not appropriately addressed, it will be interpreted
as velocity error by FWI, which leads the
inversion to erroneous velocity models.
Many methods have been proposed to
solve cycle-skipping and amplitude discrepancy issues. The travel-time difference
of seismic signals is more linearly related
to the velocity error than the waveform
difference, and any discrepancies in the
amplitude difference are inconsequential.
Alternatively, synthetic and recorded data
can be brought in phase by applying a
matching filter to one of the two datasets.
New FWI Algorithm
The time-lag FWI algorithm integrates
a travel-time cost function with a frequency-dependent time window for travel-time measurements and a cross-correlation coefficient-based weight function
to promote travel-time measurements of
high quality. As demonstrated on a full-

American Oil and Gas Reporter - July 2019

Table of Contents for the Digital Edition of American Oil and Gas Reporter - July 2019