Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.

org/page/policies/terms
DOI:10.1190/leedff.2025.44.issue-3

Special Section:
Near-surface geophysics in the energy transition

ISSN 1070-485X
March 2025 · Volume 44, No. 3
 We Give You The Best Image™
· True Amplitude consistency in all the modules.
· VTI, TTI and Orthorhombic Anisotropy.
· OVT Gathers.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

· All imaging modules use Topography

· Converted waves, OBC, VSP geometry.
· Gamma migration scans.
· Anti-aliasing: triangles, boxcar,
multiple bandpass.
· Diffraction Imaging.
· Grid based tomography with horizon constrains
· 360 degrees Wide Azimuth Tomography using
Extended Gathers.
· Updates using OVT Gathers.
· Targets, grid of targets, sparse target lines.
· Single and multiple parameter residual
velocity estimation.
· Automatic or manual residual velocity picking.
· Anisotropic parameter and velocity updates.
· Updates from topography.
· Automatically
points.
generated back-projection

· Layer freezing or global updates.

DOI:10.1190/leedff.2025.44.issue-3

· Global and local error visualization and QC.

· Automatic Vrms estimation.
· Optimization
nodes.
of data distribution on parallel

· Job monitoring. Add and subtract nodes

during runtime.

A Complete Depth Imaging System · Adaptive run-time parameter optimization.

· 5-D Beam Forming using PWD filters.
Z-Terra Software Solutions Modern 3D/4D Visualization and
Imaging Intuitive Graphical User Interfaces
ZTK Kirchhoff PSTM and PSDM. Anisotropic VTI, TTI and Orthorhombic Traveltimes. OVT Gathers. · Intuitive Graphic Interface on each module.
ZTK-DI Diffraction Imaging in Kirchhoff time and depth. Specularity Gathers. · Base Map Viewer for input data and grid
definition.
ZRTM Two Way Reverse Time Wave Equation Migration.
ZFBM Gaussian Beam and Fast Beam Migration. · Beams and rays QC and visualization tools.
Z5D 5-D Interpolation using Azimuth Move Out (AMO). · Overlay of migration operator on input data
visualization.
ZSM Wave Equation Multiples Imaging Migration.
ZFA Spectral Decomposition and Super Resolution Thin Layers Inversion. · 3-D/4-D synchronized visualization, input data,
gathers, stack, velocity model, travel-times,
Velocity Model Building Base Map.
ZTomo Grid based tomography constrained by horizons. Anisotropic updates.
ZTomo-360 Wide azimuth tomography using Extended Gathers.
ZBT Beam Tomography. High Resolution Tomography using Beam Migration.
ZESTANI Anisotropic Velocity Model Building starting from well logs.
ZMod Acoustic Modeling, Wave-Equation Illumination.
ZFWI Full Waveform Inversion in Data Domain. 281.945.0000
ZEOS Visualization and Velocity Model Building. Contact us today at [email protected].

HOUSTON THE HAGUE RIO DE JANEIRO BUENOS AIRES KUALA LUMPUR

17171 Park Row, Suite 247 Stationsplein 6, 5th Floor Av. Nilo Pecanha 50 Av. Lima 575, 8th floor Dialog Tower, No. 15, Jalan PJU 7/5,
Houston, TX 77084 Voorburg 2270 JB Edificio De Paoli, Sala 1616 Buenos Aires C1073AAK Mutiara Damansara, 47810 Petaling Jaya,
USA The Netherlands Rio de Janeiro, RJ Brazil Argentina Selangor Darul Ehsan, Malaysia.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms
DOI:10.1190/leedff.2025.44.issue-3

OUR HISTORY FUELS

OUR FUTURE
For over 90 years, we forged a legacy of scientific innovation under the CGG banner.
Now, as Viridien, we build on that foundation, enhancing our core business capabilities while
creating new market opportunities.
Our remarkable history fuels our vision for a new era of growth, and at the heart of it are our
people. Every day, our talented professionals find new ways to solve complex challenges and
drive technological progress.
Together with our clients, we are redefining what’s possible – advancing technologies that
make a difference, today and for decades to come.

viridiengroup.com

SEE THINGS DIFFERENTLY

 EVERYBODY’S
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

TALKING ABOUT IT.

WE’RE DOING IT.
DUG Elastic MP-FWI Imaging
DOI:10.1190/leedff.2025.44.issue-3

Conventional Workflow

DATA C O U R T E SY O F S H E L L
ELASTIC
MP-FWI
DUG Elastic MP-FWI Imaging solves for reflectivity, Vp, Vs, P-impedance, IMAGING
S-impedance and density. It delivers not only another step change in imaging IS HERE
quality, but also elastic rock properties for quantitative interpretation and pre-
stack amplitude analysis — directly from field-data input. A complete replacement for
traditional processing and imaging workflows — we talk the talk and walk the walk!

[email protected] | dug.com/fwi
 The Leading Edge
Table of Contents
Special Section: Near-surface geophysics in the energy transition Departments
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

162 ��������������Introduction to this special section: Near-surface geophysics in the energy transition 160................ Editorial Calendar
C. J. Weiss 161................ President’s Page
163 ��������������Characterizing hydraulic fracture formation during enhanced geothermal system experiments using coda waves 230................ Board Report
K. Pradhan, P. Sprinkle, J. Chaput, H. Knox, and EGS Collab Team 231................ Announcements
170 ��������������Harnessing 3D ultra-high-resolution seismic technology for offshore wind farm development: Advancements, challenges, 232................ Reviews
and future prospects 233................ Membership
B. Caselitz, A. McKay, M. Widmaier, J. Oukili, D. Davies, and N. Pernin
235................ Meetings Calendar
178 ��������������Quantum-annealing-assisted surface-consistent phase and amplitude refraction residuals estimation 236................ Seismic Soundoff
D. Rovetta, M. Dukalski, and A. Kontakis

187 ��������������Ground-truth-calibrated onshore and offshore subsurface infrastructure image from deep-learning-based 3D inversion of On the cover: Photo by StudioFI/
magnetic data Shutterstock.
S. Mukherjee, J. Y. Guigne, G. N. Young, S. Adavani, K. Kennelley, D. Hoffmann, H. Shukla, R. S. Bell, and W. N. Barkhouse

206 ��������������Overcoming complex near-surface conditions for improved seismic imaging in Tarim Basin, China
DOI:10.1190/leedff.2025.44.issue-3

R. Zhao, D. Zheng, S. Yang, A. Zuo, Y. Chen, X. Zhu, K. Xu, and S. Gray

214 ��������������Exploring seismic data in the flowline domain: Automated extraction of unconformities, sequence boundaries, and
conformable reflections
D. Adelved, J. E. Lie, A. J. Bugge, and P. Bormann

224 ��������������A graphical approach to determine the relationship between intercept, gradient, and the common seismic rock properties:
Global model and application
D. Went

Episode 250:
Bridging science and society – The impact of
Geoscientists without Borders
with Craig Beasley

Episode 249:
Machine learning methods in geoscience
with Gerard Schuster

Episode 248:
Sponsor a podcast !
AI, FWI, and the future of subsurface imaging Contact [email protected] for
with Satinder Chopra and Heather Bedle more information.

158 The Leading Edge March 2025

 The Leading Edge
SEG Board of Directors
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

PRESIDENT TREASURER DIRECTOR AT LARGE

John Eastwood Xiaojun Huang Lillian G. Flakes
Calgary, AB, Canada ExxonMobil Prospect Dimensions
Houston, TX, USA Richardson, TX, USA

PRESIDENT-ELECT PAST PRESIDENT DIRECTOR AT LARGE

Joseph Reilly Arthur C. H. Cheng Catherine Truffert
Industry, TX, USA The Chinese University of Hong Kong IRIS Instruments
Houston, TX, USA Orléans, France

FIRST VICE PRESIDENT DIRECTOR AT LARGE DIRECTOR AT LARGE

Marianne Rauch Sandeep Chandola Constantine Tsingas
TGS Petronas Saudi Aramco
Houston, TX, USA Kuala Lumpur, Malaysia Dhahran, Kingdom of Saudi Arabia

DIRECTOR AT LARGE
DOI:10.1190/leedff.2025.44.issue-3

SECOND VICE PRESIDENT CHAIR OF THE COUNCIL

Aria Abubakar Sergio Chávez-Pérez Ge Jin
SLB Mexican Petroleum Institute Colorado School of Mines
Houston, TX, USA Ciudad de México, Mexico Golden, CO, USA

VICE PRESIDENT, PUBLICATIONS DIRECTOR AT LARGE

Kyle Spikes Joseph Ebeniro
The University of Texas at Austin Unversity of Port Harcourt
Austin, TX, USA Port Harcourt, Nigeria

The Leading Edge® (Print ISSN 1070-485X; Online ISSN 1938-3789) is published monthly by the Society of Exploration Geophysicists, 125 W. 15th St., JIM WHITE, Executive Director
Suite 100, Tulsa, Oklahoma 74119 USA; phone 1-918-497-5500. Periodicals postage paid at Tulsa, OK and at additional mailing offices. SCOTT SUTHERLAND, Managing Director, Business and
Geoscience Technology
Print subscriptions for professional members of the Society in good standing are included in membership dues paid at World Bank IV rates. JENNIFER COBB, Managing Director, Publications and Membership
Dues for Active and Associate Members for 2025 vary depending on the three-tiered dues structure based on World Bank classification of the JENO MAVZER, Director, Journals
member’s country of citizenship or primary work residence. Students may purchase print subscriptions by paying a discounted fee during dues STEVE BROWN, Managing Editor
renewal or by contacting [email protected]. Nonmembers may obtain online subscriptions by contacting [email protected]. Copies of single issues KELSY TAYLOR, Associate Editor
of The Leading Edge® may be purchased by inquiry to [email protected]. MARIA GEE, Graphic Designer
Corporations, universities, and other institutions can purchase access to The Leading Edge® via standalone subscriptions or one of several SARAH WEATHERS, Publishing Platform Manager
available subscription package options. SEG’s subscription pricing features three academic and three corporate tiers, providing affordable STACY BAKER, Publishing Platform Analyst
access to applied-geophysics content for institutions of any size. SEG subscription packages offer significant savings on the price of single-
publication subscriptions. Special pricing is available for organizations with multiple sites and for institutional consortia. For more information on Editorial information: 1-918-497-5503, [email protected]
package pricing, including single-site and multisite institutional subscriptions, please contact Patrick Riley at [email protected] or 1-918-497-5531. Advertising information and rates:
Institutional subscriptions for print delivery are available only with the purchase of a digital subscription. Rates are subject to change without 1-918-497-5524
notice. For more information, please visit https://seg.org/subscriptions. Kathy Gamble, [email protected]
Advertising rates for The Leading Edge® will be furnished upon request. No advertisement will be accepted for products or services that cannot Subscription information:
be demonstrated to be based on accepted principles of the physical sciences. For more information, contact [email protected]. Members, [email protected]
Statements of fact and opinion are made on the responsibility of the authors and advertisers alone and do not imply an opinion on the part of Nonmembers, [email protected]
the officers or members of SEG. Copyright 2025 by the Society of Exploration Geophysicists. The Leading Edge®, SEG®, and the SEG logo are Institutions, Patrick Riley, 1-918-497-5531,
registered marks of the Society of Exploration Geophysicists. All rights reserved. Material may not be reproduced without written permission. [email protected]
Printed in the USA. POSTMASTER: Send changes of address to
The Leading Edge
125 W. 15th St., Suite 100
Tulsa, OK 74119 USA

March 2025 The Leading Edge 159

 The Leading Edge THE LEADING EDGE
Editorial Calendar EDITORIAL BOARD
CHAIR
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Issue Special section theme Due date Guest editors Heather Bedle
University of Oklahoma
All Technical standalone articles ongoing TLE Editorial Board Norman, OK, USA
[email protected]

April 2025 Geothermal past due William Cumming

Michael Wilt1
Niels Grobbe
May 2025 Elastic FWI past due Kristopher A. Innanen University of Hawai‘i at Mānoa
René-Édouard Plessix Honolulu, HI, USA
[email protected]
Andrew Ratcliffe
Partha Routh
Ramesh Neelamani1
June 2025 Rock physics: Honoring Amos Nur’s past due Colin Sayers
Joël Le Calvez
legacy Madhumita Sengupta SLB
Chicheng Xu Sugar Land, TX, USA
Laurent Louis1 [email protected]

July 2025 4D FWI past due Kristopher A. Innanen

Francis Rollins
Madhav Vyas1
Madhav Vyas
August 2025 Geophysics for urban space and 15 March 2025 Derrick Dasenbrock BP America
infrastructure Yunyue Elita Li Houston, TX, USA
DOI:10.1190/leedff.2025.44.issue-3

[email protected]
Heather Bedle1
September 2025 Data-driven geophysics 15 April 2025 Ge Jin
Ivan Lim Chen Ning
Harpreet Sethi
Vladimir Kazei
Qie Zhang Aramco Americas
Vladimir Kazei1 Houston, TX, USA
[email protected]
October 2025 Geophysics for new energies 15 May 2025 Erkan Ay
Joël Le Calvez1
November 2025 Geophysics for mineral exploration 15 June 2025 Sarah Devriese
Yongyi Li Laurent Louis
Jiajia Sun Aramco Americas
Houston, TX, USA
G. Didem Beskardes1 [email protected]
December 2025 Advances in seismic acquisition 15 July 2025 Sandeep Chandola
Faizan Akasyah Ghazali
Shivaji Maitra
Ramesh Neelamani1 Ramesh (Neelsh) Neelamani
January 2026 High-performance computing in 15 August 2025 Madhav Vyas1 ExxonMobil
Houston, TX, USA
geophysics [email protected]
1
TLE Editorial Board coordinator

Michael Wilt
Lawrence Berkeley National Laboratory
Berkeley, CA, USA
[email protected]

TLE publishes special sections and standalone articles covering all aspects of applied
geophysics and related disciplines. Submission of articles is open to all. Please submit articles via
G. Didem Beskardes
the online manuscript submission system at https://mc.manuscriptcentral.com/tle. Submission Sandia National Laboratories
instructions are available at https://library.seg.org/TLE-authors. For full descriptions of special Albuquerque, NM, USA
[email protected]
section themes, see https://library.seg.org/TLE-sections. TLE Editorial Board coordinators work
with guest editors to coordinate and support the review process and also may serve as guest editors.
For additional assistance, contact [email protected].

160 The Leading Edge March 2025

 President’s Page
Appreciating the depth
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

of near-surface geophysics
This month’s author:
Catherine Truffert, SEG Director at Large

I f you look for a general definition of

“near-surface geophysics,” you will
discover a lot of them, which, one after the
At first glance, the near-surface sector
appears more heterogeneous than the oil
and gas sector. Its ecosystem is made up of
of artificial intelligence, in our lives
depends on knowledge of the subsurface
and its resources? How can we continue
other, will confuse you even more. We all smaller practitioners, with more limited to roll out information and communication
know that the term “geophysics” covers a human and financial resources. To some, technology without finding new mineral
range of physical, noninvasive methods for this sector may seem peculiar. Yet it offers deposits? Metal recycling can only start
exploring and understanding the subsur- geoscientists a wide range of subjects to when any type of infrastructure has
face. But what does near-surface geophysics study. It deals with many subjects of societal reached the end of its life. Before having
mean in this context? Can we identify a importance, such as land management, access to the secondary source (recycled
DOI:10.1190/leedff.2025.44.issue-3

depth limit that separates near-surface preservation of cultural heritage, and cli- metal), we need enough mineral resources
geophysics from deep? Is this limit around mate change mitigation. Geoscientists often for building new infrastructure.
200 m, 500 m, or 1 km? Do the rare excep- interact with the social sciences, as dem- Other societal challenges call for prog-
tions of hydrocarbon exploration methods onstrated by the emblematic Geoscientists ress in geosciences, particularly in the
at a depth of less than 200 m — conven- without Borders program. near-surface sector — drinking water
tional shallow deposits, bituminous oil, or The point here is not to pit the major resources, for example. This resource
coalbed methane — fall into the category oil and gas or mining industries against involves the subsoil, and knowledge of this
of near-surface geophysics? the more plural near-surface industries. is critical to ensure its preservation in terms
I invite you to start a debate with some We are all geoscientists seeking to push of quantity and quality.
geophysicists if you have a long evening back the frontiers of our knowledge of the So, what exactly is the field of “near-
ahead of you. geologic and geodynamic dimensions of surface geophysics”? It is a set of disciplines
In this issue’s special section, you will the subsurface. What does it hold? What designed to respond to a wide variety of
discover many facets of the geosciences is its history? What is its behavior under applications, not only for energy produc-
that fall within the near-surface domain. anthropic and climate change pressure? tion, but also for the basic needs of society.
The search for natural resources, the What are the processes it undergoes? How In this, it reconciles demanding work with
assessment of natural hazards, and the might it behave in the future? These are a passion that gives meaning to our actions.
monitoring or remediation of soil or fundamental, essential questions that lead Furthermore, the challenges to be met
groundwater pollution require a combina- us to manage, extract, or store resources require a foothold in research and develop-
tion of geological observations, geophysical and predict underground behavior — ment on the part of academia and research
imaging, and an understanding of bio- fundamental questions that are the subject centers, followed by practical applications
physical-chemical interactions. of intense research. or even innovation on the part of practi-
As geoscientists, the issues we face go Our lives and those of future genera- tioners. Thus, near-surface geophysics offers
far beyond the “what does near-surface tions depend on knowledge of the subsoil. a range of activities that spans the entire
geophysics mean” debate. So, after a few Our consumption of electricity is con- value scale from research and innovation
decades of experience in geosciences, I stantly growing, even though more than to applied studies.
agree that there are no depth limits to the 50% of its production is still based on SEG seeks to collaborate with other
near-surface domain. I would tend to coal, gas, and oil. The energy transition organizations that complement geophysics
delineate the near-surface field not by depth to a less carbon-intensive resource is as and their members or vice versa from geo-
of investigation but by type of resource. dependent on knowledge of the subsoil technical engineers, hydrologists, and other
For me, everything to do with hydrocarbon as are hydrocarbons. geoscientists to bring the best of all related
exploration remains a field outside the so- How many are aware that the use of disciplines together to solve some of the
called near surface. digital technology, and a fortiori the advent most pressing problems today.

March 2025 The Leading Edge 161

 Introduction to this special section:
Near-surface geophysics in the energy transition
Chester J. Weiss1,2
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

https://doi.org/10.1190/tle44030162.1

I n its most practical terms, the energy transition is a diversifica-

tion of and systems-level approach to the natural resources
and infrastructure therein required to meet the increasing global
interpretation, and acquisition efficiency. Such improvements aid
in estimating soil properties and building ground models necessary
for wind farm design. The authors offer example results from
demand for accessible, affordable, and sustainable energy. Ten Noorden van de Waddeneilanden Wind Farm Zone off the
Specifically, it manifests as an augmentation to traditional coast of the Netherlands.
hydrocarbon-based energy solutions with additional energy Rovetta et al. offer a compelling look into the future of
sources such as nuclear, wind, ocean-wave, hydro, biomass, computational geophysics through application of quantum
solar, and geothermal power. As such, the deep reservoir of computing to the challenge of deep-target seismic imaging in
knowledge matured through decades of research in oil and gas the presence of near-surface complexities. The resultant phase
has opportunity to both broaden its impact and inspire new and amplitude residuals from a near-surface structure are ana-
research in the emergent and rapidly evolving energy-technology lyzed in the context of a specialized quantum annealing process
landscape. The special section in this edition of The Leading Edge that exploits quantum superposition and tunneling to avoid
is dedicated to highlighting high-impact research in near-surface well-known inefficiencies in traditional optimization-based
geophysics as it borrows from the past successes of exploration approaches to stack-power maximization of seismic gathers. In
geophysics to shape the future of this landscape for the better- their method, residuals are instead quantified by a discrete
ment of people everywhere. quadratic model with a hybrid solver that utilizes both classical
DOI:10.1190/leedff.2025.44.issue-3

Near-surface geophysics is often defined as the application and quantum optimization.

of geophysical methods to investigate shallow (upper few tens Addressing the challenge that anthropogenic clutter mapping
to hundreds of meters) subsurface properties and processes. In poses in derisking exploration and characterization of previously
the energy sector, where exploration depths extend to about developed sites, Mukherjee et al. demonstrate a novel application
5 km, 1 to 2 km depths of investigation are relatively shallow, of artificial intelligence (AI) technology to magnetic field data.
and less than 1 km is very shallow. This special section considers In contrast to target reconstruction via traditional inversion or
geophysical methods and data analysis approaches to better trial-and-error curve fitting, the authors show how an AI workflow
understand properties and processes, at these relatively shallow not only is capable of recovering the location and distribution of
depths, as they relate to renewable energy resource characteriza- subsurface clutter but also offers a scalable framework for cases
tion, monitoring, and production. where the number of targets is large.
The range of contributions in this special section highlights Lastly, Zhao et al. present a case study from the Tarim
the breadth of exploration needs in the renewable energy sector Basin, China, where near-surface complexities — similar to
that near-surface geophysics can meet. For example, Pradhan those found in the Arabian Peninsula and the Andes of South
et al. call attention to the importance of understanding induced America — define a need for advanced seismic data acquisition
fracture patterns in enhanced geothermal systems (EGS), in which and analysis to minimize uncertainty in deep reservoir charac-
water heated by latent crustal heat is extracted via opened or newly terization. To that end, they describe progress in four key areas:
created fractures. EGS requires careful site characterization before constrained near-surface tomography, long-offset acquisition
and high-resolution imaging and data analysis after inducing for improved turning-ray tomography, finite-frequency wavefront
fractures (a.k.a. enhancement). The authors present coda wave tomography as an alternative to full-waveform inversion, and
interferometry as an approach to monitoring changes in fracture development of a comprehensive velocity model combining both
patterns, which are incredibly difficult to predict and image, by diving and reflected seismic waves. They demonstrate how these
utilizing late-arriving seismic scattering waves to detect small concepts, applied in the Tarim Basin, can be utilized elsewhere
changes in the fracture network at the EGS Collab Experiment in the world.
in South Dakota, USA.
A detailed understanding of subsurface structures is very Editors’ note: The editors of The Leading Edge wish to extend their
important for offshore wind development. Caselitz et al. present gratitude to Christine Downs of Sandia National Laboratories, whose
a fast, high-resolution, 3D seismic survey approach that improves expertise and dedication as guest editor were integral to assembling
upon 2D seismic surveys in terms of data resolution, quantitative this special section.

Sandia National Laboratories, Geophysics Department, Albuquerque, New Mexico, USA. E-mail: [email protected].
1

University of New Mexico, Department of Earth and Planetary Sciences, Albuquerque, New Mexico, USA.
2

162 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 Characterizing hydraulic fracture formation during
enhanced geothermal system experiments using coda waves
Kaushik Pradhan1, Parker Sprinkle2, Julien Chaput1, Hunter Knox2, and EGS Collab Team
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

https://doi.org/10.1190/tle44030163.1

Abstract better understand fracture creation, fluid flow, and heat transfer
Seismic waves are valuable for detecting structural variations processes relevant to EGS. The experiment was conducted in
in the subsurface and can be used to investigate enhanced geo- a drift 4850 ft below ground in a testbed consisting of a suite
thermal systems (EGS). Classic methods, like seismic reflection, of geophysical and geochemical instrumentation at the Sanford
struggle to resolve these effects when the perturbations are Underground Research Facility (SURF) in South Dakota.
confined to small volumes of rock, thus requiring other methods. Extensive site characterization was performed, including bore-
Multiply scattered waves are better suited to resolving small hole logging, core analysis, and stress measurements, to under-
structural changes due to their cumulative sensitivity acquired stand the geologic and geomechanical conditions of the rock
by their longer propagation times within the medium. With the mass (Kneafsey et al., 2018). The testbed included an injection
growing focus on renewable energy production, an improved borehole, a production borehole, and several monitoring bore-
characterization of fracture network geometry created during holes. This configuration enabled comprehensive monitoring
EGS stimulations is crucial. In this study, we leverage coda from of the stimulation process and subsequent fluid flow tests. A
waveforms generated by a continuous active seismic source to hydraulic fracture was created (Figure 1a) by injecting fluid
investigate fracture development during the EGS Collab into the rock mass under controlled conditions. The stimulation
Experiment 1 at the Sanford Underground Research Facility was carefully monitored using a variety of geophysical, hydro-
(SURF). By measuring waveform decorrelation, we use scattering logical, and geomechanical techniques to capture the fracture
DOI:10.1190/leedff.2025.44.issue-3

cross-section density as a proxy for the induced hydraulic frac- creation and propagation processes. High-resolution seismic
tures. Our approach implements a genetic algorithm to invert data were collected before, during, and after the stimulation
for scattering distributions, posing the problem as a nonlinear using the multilevel continuous active source seismic monitoring
optimization. We also constrain the scattering perturbations to (ML-CASSM) system (Ajo‐Franklin et al., 2011).
plane structures, enforcing realistic sparsity in fracture patterns
that is otherwise poorly resolved in linearized approaches. Results Scattered waves in enhanced geothermal systems
from synthetic examples demonstrate the effectiveness of this While EGS technology holds vast potential for sustainable
approach in recovering scattering density, highlighting its poten- energy production, it is important to understand and monitor
tial to complement methods that utilize induced seismicity for the induced fractures, as they control the efficiency of heat
improving characterization of fracture networks in enhanced recovery (Sun et al., 2017; Moska et al., 2021). The elastic moduli
geothermal reservoirs. and bulk density of rock formations control seismic wave veloci-
ties, and perturbations to these parameters alter wave propagation
Introduction in complex ways. Previously, induced microseismic events
Enhanced geothermal systems (EGS) harness the latent (Figure 1b) have been used to monitor fracture formation,
heat within the earth’s crust by injecting cold water into either location, and orientation, but they come with limitations
new or existing fractures in hot, dry rock at depth and circulat- (Schoenball et al., 2020; Fu et al., 2021). These include event
ing water through these fractures to extract hot water that can location errors and difficulty in detecting high-frequency, weaker
be used for heating homes and generating electricity (Tester tensile events, which can skew fracture plane characterization.
and Albright, 1979). However, the complexity of fracture Our research proposes using decorrelation of late-arriving seismic
networks and their evolution during stimulation creates chal- waves that result from the multiple scattering of energy in
lenges for monitoring and characterization. Over the past four heterogeneous media, that is coda waves, (Aki, 1969; Planès
decades, significant advances in EGS technology have shown et al., 2014; Xie et al., 2018) in conjunction with parametric
it to be technically feasible in terms of reservoir stimulation plane inversion to better understand the dynamic changes in
and production to generate net thermal energy (Sanyal and the fractured medium (Figure 1c).
Butler, 2005; Olasolo et al., 2016; Lu, 2018; Kumari and Recent work on coda wave interferometry (CWI) (Zhao et al.,
Ranjith, 2019; Sharmin et al., 2023). The EGS Collab 2017) and radiative transfer (Sens-Schönfelder and Wegler, 2006)
Experiment 1 focused on providing data for validating thermal, in heterogeneous media have significantly advanced the under-
hydraulic, mechanical, and chemical (THMC) models and to standing of scattering effects and the monitoring of material

Manuscript received 18 October 2024; revision received 16 January 2025; accepted 20 January 2025.
1
University of Texas at El Paso, Department of Earth, Environmental, and Resource Sciences, El Paso, Texas, USA. E-mail:
[email protected]; [email protected].
2
Pacific Northwest National Laboratory, Energy and Environment Directorate, Richland, Washington, USA. E-mail: [email protected];
[email protected].

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 163
 changes using diffuse waves. These works emphasize the sensitivity Limitations of microseismicity and application
of coda waves to subtle variations in the properties of the medium. of coda wave decorrelation
This approach has proven effective in fields such as seismology Coda waves are sensitive to small perturbations in the
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

(Mayor et al., 2014), civil engineering (Schurr et al., 2011; Stähler medium, making them ideal for detecting subtle changes in
et al., 2011), and materials science (Patra and Banerjee 2017), fracture networks. They have been used extensively to monitor
where traditional imaging techniques often struggle due to the velocity and scattering changes for a variety of applications (Grêt
high heterogeneity of the media. et al., 2006; Obermann et al., 2013; Niederleithinger et al., 2014).
The experimental rock physics community also uses coda We present examples of synthetic inversions using our approach,
waves recorded in laboratory experiments to characterize tailored to data sets recorded during the EGS Collab experiment
sandstone and granitic rocks, which are representative of those (Schoenball et al., 2020; Fu et al., 2021; Qin et al., 2024). In this
found in geothermal reservoirs (Singh et al., 2019). The method work, we employ CWI to monitor changes in the stimulated
utilizes sensitivity kernels derived from radiative transfer and geothermal reservoir, calculating decorrelation over time to infer
diffusion equations to model the interaction of scattered waves fracture development. In a geothermal reservoir characterized
with perturbations in the medium (Duran et al., 2020). The by high scattering, conventional imaging techniques relying on
advancement of these techniques has facilitated the precise direct waves fail due to pronounced heterogeneities and the small
identification and quantification of defects or changes in het- scale of the fractures themselves. Unlike conventional methods
erogeneous media, with applications in structural health (Martins et al., 2020) that utilize ballistic waves, our approach
monitoring and seismic hazard assessment. Where coda decor- exploits scattered wave sensitivity to subtle changes in the medium.
relation sensitivities tend to be spatially smooth under the Coda wave methods, validated through numerical experiments
assumption of diffusion or isotropic scattering and linear inverse (Duran et al., 2020), seek to identify multiple defects concurrently,
formulations, adding geometric constraints on allowable scat- tackling a persistent challenge in monitoring of diffuse media.
tering geometries reposes the problem in a nonlinear fashion The metric we employ, the temporal decorrelation of coda waves,
with a more limited model space. serves to quantify medium heterogeneity, specifically reflecting
DOI:10.1190/leedff.2025.44.issue-3

Figure 1. Experimental setup for the EGS Collab Experiment 1 (Sanford Underground Research Facility). (a) Experimental setup for a given injection event. Black lines are monitoring
wells, green line indicates injection well, and red line denotes production well. The yellow ellipse is the natural fracture. (b) Induced microseismicity in the volume (red dots). (c) Injection
pressures (top), decorrelation measurements (middle), and injected volumes (bottom) for different injection events. Shaded box is an example event studied in Figure 2.

164 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 the fractures formed during hydraulic stimulation. For our case, may offer valuable complementary information to traditional
decorrelation is a broad measure of how much the waveform of microseismic analysis.
a coda wave changes over time due to changes in the scattering The way energy decays in the coda envelope can be approxi-
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

properties of the medium, and this metric can be related to energy mated as an energy transport problem, which can be effectively
propagation via diffusion or radiative transfer (Planès et al., 2014; modeled using radiative transfer or diffusion models. These models
Margerin et al., 2016). are highly sensitive to scattering environments, making them a
Microseismicity, while valuable for providing initial insights natural choice for pinpointing localized changes. Given the linear
into fracture locations (Eisner et al., 2010), has several inherent relationship (Larose et al., 2010) between the measured decor-
limitations that can hinder a comprehensive understanding of relations (DC) and the scattering cross-section densities (Σ)
fracture dynamics. One significant drawback is the dominance through the kernel (K), the equation can be written in the form
of strong shear-mode failures in microseismic catalogs, which
often overshadow weaker tensile events. This bias can lead to an DC = K Σ, (1)
incomplete representation of hydraulic fracture behavior, as
tensile events play a crucial role in understanding the overall where DC represents the observed decorrelation vector. K is the
fracture system (Zhang et al., 2022). The challenges associated sensitivity kernel matrix (Figure 2a), which represents the expected
with microseismic event location further compound these limita- decorrelations for a given source-receiver configuration to scat-
tions. The swarm-like, overlapping nature of microseismic events tering cross-section density at each grid location in the model Σ,
frequently precludes the use of clear ballistic waves, making which is the parameter we aim to estimate (scattering cross-section
precise location determination difficult. Additionally, velocity densities in each voxel). The least-squares solution for Σ can be
model uncertainty can contribute to error in event location. As written as
a result, the located events provide only a partial picture of the
fracture system, leaving gaps in our understanding of the com- Σ` = (K T K)–1 K T DC, (2)
plete fracture network. To address these shortcomings and gain
DOI:10.1190/leedff.2025.44.issue-3

a more comprehensive view of fracture geometry, we are exploring where Σ` is the estimated scattering cross-section density. In cases
alternative methods, such as integrating coda wave data, which where we have prior knowledge of the model, denoted as Σprior, we

Figure 2. Example linear inversion for the injection event shown in Figure 1c (shaded box). (a) G-matrix of decorrelations recorded at the receiver array in the volume. (b) Inverted
distribution of scatterers for the period preceding the injection. (c) Inverted scatterers for the maximum decorrelation recorded during the injection. (d) Example inverted model for a
decorrelation synthetic for the same geometry, showing the impact of kernel and inverse smoothness with a failure to recover sparse scattering structures.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 165
 can incorporate this information to regularize and improve the implemented. In our example, we opted for a nonlinear genetic
stability of the inversion. The regularized solution is then given as algorithm due to its robustness in exploring complex, multimodal
search spaces and its flexibility in handling various constraints.
Σ` = (K T K + ƛ I )-1 (K T DC + ƛ Σprior).
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

(3) Genetic algorithms have been extensively used in geophysical

models, particularly for velocity model building (Sambridge and
Here, ƛ is the regularization parameter, I is the identity matrix, Drijkoningen, 1992; Jin and Madariaga, 1993; Lawrence, 2004),
and Σprior is the prior model information. The solution accounts because of their ability to efficiently navigate large parameter
for smoothness and prior knowledge and enhances the result of spaces and avoid local minima. The success of genetic algorithms
the inversion. in handling multiple solutions simultaneously and converging on
While a linear least-squares solution provides a straightforward a global optimum makes them particularly well-suited for solving
and computationally efficient way to estimate the scattering nonlinear inversion problems where subsurface structures exhibit
cross-section density, it assumes a degree of smoothness that may significant heterogeneity. However, the specific choice of optimiza-
not fully capture the complexity of subsurface heterogeneity. The tion technique is not critical to the overall approach. Alternative
inversion tends to produce overly smooth solutions because the nonlinear optimization methods, such as simulated annealing,
kernel K itself is derived from, for instance, a diffusion approxima- particle swarm optimization, or gradient-based techniques, could
tion that imposes smoothness on the estimated model Σ, and the also be employed depending on characteristics of the problem
pseudo-inverse itself is a form of minimum length regularization, and computational requirements. A synthetic example of a para-
which further exacerbates that effect. As a result, the sensitivity metric plane inversion for decorrelation measurements in the
of the decorrelation measurements to small-scale features is presence of noise is shown in Figure 4. The recovered plane model
diminished, causing fine-scale sparse variations in the scattering under the presence of strong measurement noise in Figure 4c
cross-section density to be unresolved. An example linear inversion shows that plane tilt and orientation are well recovered in the
for an injection stimulation and a synthetic test for the EGS inversion and that plane rotation and truncation are somewhat
Collab 1 experiment is shown in Figure 2. less well resolved. A statistical type of inversion (e.g., Bayesian)
DOI:10.1190/leedff.2025.44.issue-3

may help with the study of these various tradeoffs in future work.
Parametric inverse approach
Given that hydraulic fractures in rocks tend to propagate in Advantages of using scattered waves to
planar fashion according to pre-existing weaknesses, smoothly complement microseismic result
distributed scattering solutions are unlikely to be realistic. As such, CWI offers distinct advantages over traditional seismic
to address the challenge of accurately constraining fracture planes methods by capturing subtle variations in the medium, which
via decorrelation measurements, we propose a parametric plane is often beyond the detection capabilities of conventional tech-
inversion scheme that considers various parameters (such as plane niques (Zhang et al., 2020). The sensitivity is particularly ben-
position, orientation, width, and thickness) in a form of nonlinear eficial in environments like EGS where accurate fracture
geometric optimization. The inverse process first perturbs the plane characterization is essential for optimizing reservoir performance
parameters, which then map to the 3D grid of scattering values, (Magnusdottir and Horne, 2012). EGS requires a deep under-
thus limiting access of inversion-related steps to a subset of the standing of fracture dynamics to manage reservoirs effectively,
grid. A summary workflow is displayed in Figure 3. and our method excels in this context by providing a more precise
An efficient forward model mapping a truncated plane of measure of how fractures evolve under thermal, hydraulic, and
arbitrary geometry to a grid of 3D points was designed and mechanical stresses.
A key aspect of CWI in this application is the use of radiative
transfer (RT) theory, which offers a robust framework for modeling
the coda envelope in highly scattering media (Maeda et al., 2008).
When the mean free path — the average distance a wave travels
between scattering events — is very short, the RT equation simpli-
fies to the diffusion equation. This is particularly advantageous
because, in many practical scenarios, such as high-frequency
seismic experiments, the diffusion equation has an analytical
solution (Turner, 1998). By employing this approximation, we
can model wave propagation in a computationally efficient manner
without sacrificing accuracy.
One of the challenges in applying RT theory is the potential
smoothness issues of the diffusion or radiative transfer kernels.
These kernels describe how seismic energy diffuses or scatters
through a medium, and inaccuracies in their smoothness can lead
to erroneous interpretations of the subsurface structure. However,
this issue can be mitigated by using a parametric plane model to
Figure 3. Workflow for a parametric plane inversion using a genetic algorithm. constrain the heterogeneity in the medium. Given that fractures

166 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 4. Example of a parametric plane nonlinear inverse approach. (a) Depiction of an efficient parametric forward model that maps a truncated plane to a nearest grid point distribution
in 3D. (b) Example synthetic inversion for a known true plane model with added Gaussian noise to the decorrelation measurements. (c) Best fit genetic inverse model (red) versus true
model (blue) for this example.
DOI:10.1190/leedff.2025.44.issue-3

are generally planar in nature, modeling them as a parametric built to emulate EGS experiments, such as the EGS Collab
plane is a reasonable and efficient approach. This helps simplify Experiment 1 and show that, for noise-polluted decorrelation
the complex geometry of the problem and makes the inversion measurements produced by an arbitrary plane fracture, plane
process more stable, as the inherent nonlinearity of the system is orientation and offset can accurately be recovered despite the
better constrained. failure of classic linearized inversions. Ongoing work includes
Our future work will include implementing a multiple fracture applications to real-world EGS data sets, joint validation with
plane setup that further enhances the accuracy of CWI in complex microseismic monitoring, and discrimination of successful versus
geologic settings, especially where fractures are not confined to unsuccessful fracture propagation. As geothermal energy plays
a single orientation or plane. This approach will allow for the a more prominent role in the transition to renewable energy,
simultaneous modeling of multiple fracture systems, each char- advances in fracture characterization will be essential for opti-
acterized by its own parametric plane. Because the algorithm we mizing energy production and ensuring that fluid paths are
are using is inherently nonlinear, integrating multiple planes into adequately controlled.
the model is relatively straightforward. Each additional plane
adds another layer of detail, enabling more nuanced characteriza- Acknowledgments
tions of fracture networks. This setup will be particularly useful K. P. acknowledges support from the NSF grant NSF-EAR
in scenarios where fractures intersect or exhibit significant varia- #2113367, which was supplemented to fund the Geothermal
tions in orientation, as is often the case in EGS and other complex Internship at Pacific Northwest National Lab (PNNL). We extend
subsurface environments. It is crucial in EGS applications where our gratitude to PNNL for hosting K. P. as an intern and to the
managing the spatial and temporal evolution of fractures can have EGS Collab project and its entire team for their invaluable con-
a direct impact on the efficiency and sustainability of the reservoir tributions and support. The EGS Collab Team includes J. Ajo-
(Miranda et al., 2018). Future work will also focus on comparing Franklin, S. J. Bauer, T. Baumgartner, K. Beckers, A. Bonneville,
coda wave data with microseismic events in more detail to further L. Boyd, S. Brown, S. T. Brown, J. A. Burghardt, T. Chen,
refine our understanding of the stimulated zone. Y. Chen, K. Condon, P. J. Cook, D. Crandall, P. F. Dobson, T.
Doe, C. A. Doughty, D. Elsworth, J. Feldman, A. Foris, L. P.
Conclusion Frash, Z. Frone, P. Fu, K. Gao, A. Ghassemi, H. Gudmundsdottir,
Our study demonstrates the potential of CWI in monitoring Y. Guglielmi, G. Guthrie, B. Haimson, A. Hawkins, J. Heise,
and characterizing hydraulic fractures for EGS technology. M. Horn, R. N. Horne, J. Horner, M. Hu, H. Huang, L. Huang,
Given that hydraulic fractures mostly manifest as plane failures, K. J. Im, M. Ingraham, R. S. Jayne, T. C. Johnson, B. Johnston,
we present a strategy to significantly reduce nonuniqueness in S. Karra, K. Kim, D. K. King, H. Knox, J. Knox, D. Kumar, K.
the model space by imposing a parametric plane geometry to Kutun, M. Lee, K. Li, R. Lopez, M. Maceira, P. Mackey,
the inversion process, normally dominated by the smooth sen- N. Makedonska, C. J. Marone, E. Mattson, M. W. McClure, J.
sitivity of the scattering kernels. We present synthetic examples McLennan, T. McLing, C. Medler, R. J. Mellors, E. Metcalfe,

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 167
 J. Miskimins, J. Moore, J. P. Morris, S. Nakagawa, G. Neupane, review: Earth-Science Reviews, 199, 102955, https://doi.org/10.1016/j.
G. Newman, A. Nieto, C. M. Oldenburg, W. Pan, T. Paronish, earscirev.2019.102955.
R. Pawar, P. Petrov, B. Pietzyk, R. Podgorney, Y. Polsky, J. Popejoy, Larose, E., T. Planes, V. Rossetto, and L. Margerin, 2010, Locating a
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

small change in a multiple scattering environment: Applied Physics

S. Porse, S. Richard, B. Q. Roberts, M. Robertson, W. Roggenthen,
Letters, 96, no. 20, 204101, https://doi.org/10.1063/1.3431269.
J. Rutqvist, D. Rynders, H. Santos-Villalobos, M. Schoenball, Lawrence, J. F., 2004, Combined receiver-function and surface wave
P. Schwering, V. Sesetty, C. S. Sherman, A. Singh, M. M. Smith, phase-velocity inversion using a niching genetic algorithm: Application
H. Sone, F. A. Soom, P. Sprinkle, C. E. Strickland, J. Su, D. to Patagonia: Bulletin of the Seismological Society of America, 94,
Templeton, J. N. Thomle, C. Ulrich, N. Uzunlar, A. Vachaparampil, no. 3, 977–987, https://doi.org/10.1785/0120030172.
C. A. Valladao, W. Vandermeer, G. Vandine, D. Vardiman, V. Lu, S.-M., 2018, A global review of enhanced geothermal system (EGS):
R. Vermeul, J. L. Wagoner, H. F. Wang, J. Weers, J. White, M. Renewable and Sustainable Energy Reviews, 81, 2902–2921, https://
doi.org/10.1016/j.rser.2017.06.097.
D. White, P. Winterfeld, T. Wood, S. Workman, H. Wu, Y. S.
Magnusdottir, L., and R. Horne, 2012, Characterization of fractures in
Wu, Y. Wu, E. C. Yildirim, Y. Zhang, Y. Q. Zhang, J. Zhou, Q. geothermal reservoirs using resistivity: Proceedings of the 37th
Zhou, and M. D. Zoback, led by T. Kneafsey and D. Blankenship. Workshop on Geothermal Reservoir Engineering, SGP-TR-194.
Margerin, L., T. Planès, J. Mayor, and M. Calvet, 2016, Sensitivity
Data and materials availability kernels for coda-wave interferometry and scattering tomography:
Data associated with this research are available and can be Theory and numerical evaluation in two-dimensional anisotropically
accessed at https://gdr.openei.org/submissions/1229. scattering media: Geophysical Journal International, 204, no. 1,
650–666, https://doi.org/10.1093/gji/ggv470.
Martins, J. E., C. Weemstra, E. Ruigrok, A. Verdel, P. Jousset, and G.
Corresponding author: [email protected] P. Hersir, 2020, 3D S-wave velocity imaging of Reykjanes Peninsula
high-enthalpy geothermal fields with ambient-noise tomography:
References Journal of Volcanology and Geothermal Research, 391, 106685,
Ajo‐Franklin, J., T. Daley, B. Butler‐Veytia, J. Peterson, Y. Wu, B. https://doi.org/10.1016/j.jvolgeores.2019.106685.
Kelly, and S. Hubbard, 2011, Multi‐level continuous active source Mayor, J., L. Margerin, and M. Calvet, 2014, Sensitivity of coda waves
DOI:10.1190/leedff.2025.44.issue-3

seismic monitoring (ML‐CASSM): Mapping shallow hydrofracture to spatial variations of absorption and scattering: Radiative transfer
evolution at a TCE contaminated site: 81st Annual International theory and 2-D examples: Geophysical Journal International, 197,
Meeting, SEG, Expanded Abstracts, 3727–3731, https://doi. no. 2, 1117–1137, https://doi.org/10.1093/gji/ggu046.
org/10.1190/1.3627980. Miranda, M. M., C. Dezayes, N. Giordano, I. Kanzari, J. Raymond,
Aki, K., 1969, Analysis of the seismic coda of local earthquakes as and J. Carvalho, 2018, Fracture network characterization as input
scattered waves: Journal of Geophysical Research, 74, no. 2, 615–631, for geothermal energy research: Preliminary data from Kuujjuaq,
https://doi.org/10.1029/JB074i002p00615. Northern Québec, Canada: Proceedings of the 43rd Workshop on
Duran, A., T. Planès, and A. Obermann, 2020, Coda-wave decorrelation Geothermal Reservoir Engineering, SGP-TR-213.
sensitivity kernels in 2-D elastic media: A numerical approach: Moska, R., K. Labus, and P. Kasza, 2021, Hydraulic fracturing in
Geophysical Journal International, 223, no. 2, 934–943, https://doi. enhanced geothermal systems — Field, tectonic and rock mechanics
org/10.1093/gji/ggaa357. conditions — A review: Energies, 14, no. 18, 5725, https://doi.
Eisner, L., S. Williams-Stroud, A. Hill, P. Duncan, and M. Thornton, org/10.3390/en14185725.
2010, Beyond the dots in the box: Microseismicity-constrained Niederleithinger, E., C. Sens-Schönfelder, S. Grothe, and H.
fracture models for reservoir simulation: The Leading Edge, 29, no. Wiggenhauser, 2014, Coda wave interferometry used to localize
3, 326–333, https://doi.org/10.1190/1.3353730. compressional load effects in a concrete specimen: 7th European
Fu, P., M. Schoenball, J. B. Ajo‐Franklin, C. Chai, M. Maceira, J. P. Workshop on Structural Health Monitoring, IFFSTTAR.
Morris, H. Wu, et al., 2021, Close observation of hydraulic fracturing Obermann, A., T. Planès, E. Larose, and M. Campillo, 2013, Imaging
at EGS Collab Experiment 1: Fracture trajectory, microseismic preeruptive and coeruptive structural and mechanical changes of
interpretations, and the role of natural fractures: Journal of Geophysical a volcano with ambient seismic noise: Journal of Geophysical
Research: Solid Earth, 126, no. 7, e2020JB020840, https://doi. Research: Solid Earth, 118, no. 12, 6285–6294, https://doi.
org/10.1029/2020JB020840. org/10.1002/2013JB010399.
Grêt, A., R. Snieder, and J. Scales, 2006, Time‐lapse monitoring of rock Olasolo, P., M. C. Juárez, M. P. Morales, S. D´Amico, and I. A. Liarte,
properties with coda wave interferometry: Journal of Geophysical 2016, Enhanced geothermal systems (EGS): A review: Renewable
Research: Solid Earth, 111, no. B3, 2004JB003354, https://doi. and Sustainable Energy Reviews, 56, 133–44, https://doi.org/10.1016/j.
org/10.1029/2004JB003354. rser.2015.11.031.
Jin, S., and R. Madariaga, 1993, Background velocity inversion with a Patra, S., and S. Banerjee, 2017, Material state awareness for composites
genetic algorithm: Geophysical Research Letters, 20, no. 2, 93–96, part I: Precursor damage analysis using ultrasonic guided coda wave
https://doi.org/10.1029/92GL02781. interferometry (CWI): Materials, 10, no. 12, 1436, https://doi.
Kneafsey, T. J., P. Dobson, D. Blankenship, J. Morris, H. Knox, P. org/10.3390/ma10121436.
Schwering, M. White, et al., 2018, An overview of the EGS Collab Planès, T., E. Larose, L. Margerin, V. Rossetto, and C. Sens-Schönfelder,
Project: Field validation of coupled process modeling of fracturing 2014, Decorrelation and phase-shift of coda waves induced by local
and fluid flow at the Sanford Underground Research Facility, Lead, changes: Multiple scattering approach and numerical validation:
SD: Presented at the 43rd Workshop on Geothermal Reservoir Waves in Random and Complex Media, 24, no. 2, 99–125, https://
Engineering. doi.org/10.1080/17455030.2014.880821.
Kumari, W. G. P., and P. G. Ranjith, 2019, Sustainable development of Qin, Y., J. Li, L. Huang, M. Schoenball, J. Ajo-Franklin, D. Blankenship,
enhanced geothermal systems based on geotechnical research — A and T. J. Kneafsey, 2024, Source mechanism of kHz microseismic

168 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 events recorded in multiple boreholes at the first EGS Collab testbed: Geothermics, 120,
102994, https://doi.org/10.1016/j.geothermics.2024.102994. ModelVision
Sambridge, M., and G. Drijkoningen, 1992, Genetic algorithms in seismic waveform inversion:
Magnetic & Gravity
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Geophysical Journal International, 109, no. 2, 323–342, https://doi.org/10.1111/j.1365-

246X.1992.tb00100.x. Interpretation System
Sanyal, S. K., and S. J. Butler, 2005, An analysis of power generation prospects from enhanced
geothermal systems: Proceedings of the World Geothermal Congress. Simulate your target
Schoenball, M., J. B. Ajo‐Franklin, D. Blankenship, C. Chai, A. Chakravarty, P. Dobson, C.
before you fly
Hopp, et al., 2020, Creation of a mixed‐mode fracture network at mesoscale through
hydraulic fracturing and shear stimulation: Journal of Geophysical Research: Solid Earth,
125, no. 12, e2020JB019807, https://doi.org/10.1029/2020JB019807. SRTM + satellite + flight plan
Schurr, D. P., J.-Y. Kim, K. G. Sabra, and L. J. Jacobs, 2011, Damage detection in concrete
using coda wave interferometry: NDT&E International, 44, no. 8. 728–735, https://doi.
org/10.1016/j.ndteint.2011.07.009.
Sens-Schönfelder, C., and U. Wegler, 2006, Radiative transfer theory for estimation of the
seismic moment: Geophysical Journal International, 167, no. 3, 1363–1372, https://doi.
org/10.1111/j.1365-246X.2006.03139.x.
Sharmin, T., N. R. Khan, M. S. Akram, and M. M. Ehsan, 2023, A state-of-the-art review
on geothermal energy extraction, utilization, and improvement strategies: Conventional,
hybridized, and enhanced geothermal systems: International Journal of Thermofluids, 18,
100323, https://doi.org/10.1016/j.ijft.2023.100323.
Singh, J., A. Curtis, Y. Zhao, A. Cartwright‐Taylor, and I. Main, 2019, Coda wave interferometry Your target model and DEM
for accurate simultaneous monitoring of velocity and acoustic source locations in experimental
rock physics: Journal of Geophysical Research: Solid Earth, 124, no. 6, 5629–5655, https://
doi.org/10.1029/2019JB017577.
Stähler, S. C., C. Sens-Schönfelder, and E. Niederleithinger, 2011, Monitoring stress changes
DOI:10.1190/leedff.2025.44.issue-3

in a concrete bridge with coda wave interferometry: The Journal of the Acoustical Society
of America, 129, no. 4, 1945–1952, https://doi.org/10.1121/1.3553226.
Sun, Z., X. Zhang, Y. Xu, J. Yao, H. Wang, S. Lv, Z. Sun, Y. Huang, M. Cai, and X. Huang,
2017, Numerical simulation of the heat extraction in EGS with thermal-hydraulic-mechanical
coupling method based on discrete fractures model: Energy, 120, 20–33, https://doi.
org/10.1016/j.energy.2016.10.046.
Tester, J. W., and J. N. Albright, 1979, Hot dry rock energy extraction field test: 75 days of
operation of a prototype reservoir at Fenton Hill, segment 2 of phase I: Los Alamos Scientific TMI simulation
Laboratory Informal Report, LA-7771-MS, https://doi.org/10.2172/6070276.
Turner, J. A., 1998, Scattering and diffusion of seismic waves: Bulletin of the Seismological
Society of America, 88, no. 1, 276–283, https://doi.org/10.1785/BSSA0880010276.
Xie, F., E. Larose, L. Moreau, Y. Zhang, and T. Planes, 2018, Characterizing extended changes
in multiple scattering media using coda wave decorrelation: Numerical simulations: Waves
in Random and Complex Media, 28, no. 1, 1–14, https://doi.org/10.1080/17455030.2017
.1308042.
Zhang, H., Z. Shen, L. Xu, L. Gan, Z. Ma, Q. Wu, and D. Liu, 2022, Experimental investiga-
tion on hydraulic fracturing in cement mortar with tensile stress: Engineering Fracture
Mechanics, 259, 108058, https://doi.org/10.1016/j.engfracmech.2021.108058.
Zhang, S., L. Feng, and M. H. Ritzwoller, 2020, Three-station interferometry and tomography: Tensor simulation (Bzz)
Coda versus direct waves: Geophysical Journal International, 221, no. 1, 521–541, https://
doi.org/10.1093/gji/ggaa046.
Zhao, Y., A. Curtis, and B. Baptie, 2017, Locating microseismic sources with a single seismometer
channel using coda wave interferometry: Geophysics, 82, no. 3, A19–24, https://doi.
org/10.1190/geo2016-0404.1.

Tensor Research

[email protected]
www.tensor-research.com.au

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 169
 Harnessing 3D ultra-high-resolution seismic technology
for offshore wind farm development: Advancements,
challenges, and future prospects
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Bertrand Caselitz1, Allan McKay 2, Martin Widmaier 2, Julien Oukili2, Daniel Davies3, and Noemie Pernin1
https://doi.org/10.1190/tle44030170.1

Abstract
The global transition toward renew-
able energy has intensified the demand
for more offshore wind power genera-
tion. Advanced geophysical techniques
to enhance near-surface site character-
ization and ensure the rapid, safe, and
efficient installation of wind turbines
are of considerable value. Traditional
2D ultra-high-resolution seismic
(UHRS) methods, while useful, gener-
ally lack the resolution and spatial cover-
age required for modern offshore proj-
ects. Depending on the requirements,
DOI:10.1190/leedff.2025.44.issue-3

2D surveys are often acquired over Figure 1. Three-dimensional migrated stack: two time slices on the top and a section at the bottom.
multiple years. In contrast, 3D UHRS
technology, acquired in a single campaign, provides detailed and areal extent is also increasing. Offshore wind developments require
comprehensive subsurface data, significantly improving the reli- detailed knowledge of the subsurface, known as a ground model,
ability of ground models. This article highlights the value of 3D to ensure the structural integrity and longevity of wind turbines
UHRS technology in offshore wind farm development, demon- and their foundations. The complexity of the near-surface environ-
strating the importance of an integrated approach to survey design, ment, coupled with the need for rapid and accurate assessments,
acquisition, and data processing and how this enhances subsurface has driven the development and application of 3D ultra-high-
imaging, reduces uncertainties, and supports more informed resolution seismic (UHRS) technology.
decision making. The technology’s ability to capture volumetric Traditional 2D seismic methods, while useful, often lack the
data across an entire survey area or along corridors allows for the resolution and spatial coverage required for modern offshore projects.
accurate mapping of critical geological features, such as key soil As wind developments grow in size and number, the limitations
units and hazardous objects (e.g., shallow gas and boulders). The of 2D surveys, such as incomplete subsurface images and lower
expanded 3D UHRS volume offers flexible options for adjusting resolution, become increasingly apparent. In contrast, 3D UHRS
turbine locations if any anomalies are detected at the initial sites. surveys offer the ability to comprehensively characterize subsurface
Furthermore, 3D UHRS subsurface data provide the reliable structures, properties, and potential hazards, enabling more reliable
framework required for quantitative interpretation, particularly and robust site assessments (Figure 1). The continuous 3D data set
in predicting soil properties, thereby optimizing foundation designs offers more accurate interpretation of the soil units compared to a
and reducing the need for extensive geotechnical investigations. 2D data set. Additionally, it facilitates the detection of small-scale
Most importantly, an integrated 3D UHRS solution can reduce features like boulders or shallow gas pockets that could pose sig-
the time required for new wind farm developments to become nificant risks during turbine foundation work and any geotechnical
fully operational and deliver clean energy to consumers. sampling. A 3D UHRS volume is expected to deliver high-quality
input data that enable more detailed quantitative interpretation
Introduction (QI), aiding in the estimation of soil properties and the development
The global push toward renewable energy sources, especially of a comprehensive ground model (Figure 2). By providing the
offshore wind, has necessitated the use of advanced geophysical ability to assess soil properties and their variability across a site in
techniques for site characterization. Not only are the number of 3D, this approach could potentially optimize geotechnical sampling
offshore wind farm projects significantly increasing, but their and investigations and may even help reduce the scale and duration

Manuscript submitted 15 October 2024; revision received 9 December 2024; accepted 2 January 2025.
1
TGS UK, Weybridge, UK. E-mail: [email protected]; [email protected].
2
TGS Norway, Oslo, Norway. E-mail: [email protected]; [email protected]; [email protected].
3
TGS USA, Houston, Texas, USA. E-mail: [email protected].

170 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 UHRS data set enables faster and more confident decision making,
reducing the likelihood of delays during the construction phase.
This article focuses on 3D UHRS technology, from survey design
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

to QI, including acquisition and imaging. We will discuss the achieve-

ments and challenges in setting up this solution and explore the
opportunities that 3D UHRS technology presents for improving
and accelerating offshore wind farm development projects.

Survey design and acquisition

The goal of survey design is to find the optimal acquisition
parameters that meet the geophysical and geological objectives
while reducing survey turnaround time. Site characterization of
the near surface generally requires 3D UHRS data with a hori-
zontal bin size of approximately 1–2 m and a vertical resolution
less than 50 cm. Offshore wind surveys are mostly located in
shallow water, making near-zero-offset data important to achieve
the desired resolution. In the recent past, 3D UHRS survey designs
did not allow for the acquisition of large 3D surveys in a time- or
Figure 2. Three-dimensional UHRS applications: objects detection, geological cost-efficient manner. Davies and Rietveld (2020) explained that
interpretation, and soil properties estimation. it took six days to acquire 1 km 2 of UHRS data with an acquisition
bin size of 0.5 × 2 m because 118 sail lines were needed to cover
the area. It is clear that this kind of design is not viable for area
sizes commonly used in offshore wind projects, where a typical
DOI:10.1190/leedff.2025.44.issue-3

3 GW wind farm license may cover more than 500 km 2, as it

would take a year to acquire 60 km 2.
Building on experiences gained in seismic survey design for
oil and gas and carbon capture and storage, the use of wide-towed
multisources, in combination with a large number of streamers
(Widmaier et al., 2023), enables the acquisition of 3D UHRS
data in a time-efficient manner. Increasing the number of sublines
per sail line and speeding up the vessel should enable larger areas
to be acquired in less time. Thus, the strategy should be to use
the largest number of streamers and sources possible. Operational
and geophysical limitations must be addressed to ensure that
near-offset requirements, clean record, and desired spatial sampling
are met. Figure 3 shows an acquisition layout with eight streamers
separated by 12.5 m and three multilevel sparker sources, achieving
a bin size of 1.56 × 2.08 m. This layout has been used to acquire
approximately 600 km 2 in a a single campaign (Caselitz et al.,
Figure 3. Acquisition layout (unit in meters). 2024). The sources were fired sequentially at intervals of 250 ms.
The nominal source and receiver depths were 30–60 cm and 3 m,
of traditional geotechnical surveys, which are often extensive, time respectively. It is the largest full 3D UHRS survey ever conducted,
consuming, and costly. taking less than 130 production days to complete.
While 3D UHRS surveys generally require a higher initial To acquire uniformly distributed near-offset data, the three
investment compared to 2D surveys, they can be more cost effective sparkers were towed in a wide-tow source configuration above
in the long run, particularly in complex areas. The comprehensive the front end of the streamer spread. Figure 4 highlights the
data provided by 3D UHRS technology reduce the need for sampling in offset coverage in the crossline direction. The figure
additional surveys, potentially lowering overall project costs. shows that this acquisition setup is expected to provide sufficient
Additionally, the ability to acquire complete coverage in a single input data sampling to allow for reconstruction using 4D regu-
survey reduces the time required for data acquisition and inter- larization (inline × crossline × offset × time), even with small gaps
pretation, helping to keep projects on schedule. In offshore wind in the near-offset sampling. Streamer lengths of 100–150 m are
farm development, where timelines are often tight and any delays generally used in conjunction with fast-firing sources (typically
can have significant financial implications, the time efficiency 200–250 ms pop intervals) to build up fold coverage and improve
offered by 3D UHRS technology is a considerable advantage. By the signal-to-noise ratio of the final processing products. Despite
providing a more complete and accurate subsurface image, a 3D the relatively short streamers and therefore offset ranges, these

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 171
 acquisition configurations often provide sufficient incidence angles Deghosting. Receiver-side deghosting is particularly chal-
to support prestack QI work. lenging due to the high-frequency content of 3D UHRS data and
Acquiring 3D UHRS data presents several challenges, par- variations in receiver depth caused by sea state variations. The
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

ticularly in terms of maintaining data quality in variable sea deghosting process involves estimating the receiver depths in local
conditions. During UHRS surveys, streamers are towed close to time-slowness windows using an inversion-based methodology.
the sea surface, usually at a depth of 1 m or less. By towing the This approach accounts for the variations in receiver depth and
streamers deeper, down to 3 m, we were able to collect data even optimizes the deghosting process in the frequency domain, result-
in relatively rough sea conditions, significantly reducing weather- ing in a clearer and sharper seismic signal (Bekara et al., 2024).
related downtime. Another challenge associated with 3D UHRS Source-side deghosting, while less challenging, also requires
acquisition is the need for accurately positioning data, both spatially careful attention. The sparker or boomer units are suspended from
and vertically. Given the fine spatial resolution of the seismic data, floating devices, which naturally follow the sea surface, maintain-
even small positioning errors can lead to significant misalignments ing a consistent depth. This setup ensures that the source ghost
in the final seismic images. To address this, the survey is equipped notch frequencies remained stable across the survey, allowing for
with advanced GPS systems, and streamer positions are continu- a deterministic approach to be used effectively. It is important to
ously monitored using acoustic positioning devices. note that for multilevel sources, deghosting is addressed simultane-
To achieve an extremely high vertical resolution, electric ously with the designature, as the effects of source ghosts are
sources such as sparkers and boomers are used. These sources emit minimized during the shot firing.
signals at very high frequencies but lack low frequencies below Given the computational intensity of inversion-based deghost-
100–150 Hz, which limits signal penetration. This is not generally ing, a machine learning (ML) alternative has recently been
a concern as the zone of interest for offshore wind farm develop- introduced that delivers comparable quality results much more
ment lies within the first 50–100 m below the seabed. quickly. This solution employs a convolutional neural network
Depending on the wind farm development stage and/or budget akin to the one developed by Farmani et al. (2023) to denoise raw
constraints, a 3D UHRS survey may cover the entire area or focus shot gathers. Figure 5 shows a common-midpoint (CMP) stack
DOI:10.1190/leedff.2025.44.issue-3

on narrow 3D corridors that are aligned with the preliminary comparison between input, inversion-based receiver deghosting,
turbine locations. Given the importance of time-efficient solutions and ML receiver deghosting. In this example, the nominal source
in accelerating offshore wind farm development, recent surveys depth is 0.3/0.6 m and the receiver depth is 3 m. This deghosting
have incorporated additional geophysical measurements such as
multibeam echosounders, subbottom profilers, magnetometers,
and side-scan sonar. These supplementary data can be collected
simultaneously with the seismic campaign.
To enable the processing team to begin work immediately
after acquiring the first sail line, raw seismic shot gathers are
continuously streamed to the cloud via low-earth-orbit satellites.
It is important to note that the 3D UHRS data volume is substantial
due to the small spatial and temporal sampling intervals, neces-
sitating the use of high-performance computing (HPC) for process-
ing. For a given area size, 3D UHRS surveys are typically 5 to 10
times larger than standard oil and gas exploration surveys.

Data processing
The typical processing workflow for 3D UHRS data closely
resembles the one used for oil and gas seismic data, incorporating
steps like denoise, signal processing, demultiple, regularization, Figure 4. Offset coverage for three sail lines. The center of the sail lines corresponds to the
and migration. This enables the processing flow to take advantage red dashed lines.
of the technological advancements made in recent years including
broadband processing. However, due to the fine sampling interval
(e.g., 0.125 ms), variations in sea height are captured in the data
with high precision, presenting challenges during processing.
Wave heights as little as a few decimeters, significantly affect key
steps such as deghosting, sea surface statics, and demultiple. One
advantage of sparker and boomer sources is that they do not emit
frequencies below approximately 100 Hz. This means that swell
noise attenuation is not a concern, as this type of noise does not
interfere with the signal. Access to HPC resources is essential
for processing high-density and small sampling rate 3D UHRS
data in a timely manner. Figure 5. Premigration stack comparison: deghosting.

172 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 result shows that streamers can be towed at greater depths than Sea surface statics. Statics arise from sea state variations. In
previously expected, leading to reduced weather downtime and 3D UHRS surveys, statics correction is essential, as even minor
improved signal-to-noise ratio data. changes in source and receiver elevation relative to the seabed can
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Designature. Designature involves shaping each source sig- cause major time misalignments. The source tends to follow the
nature to a common broadband zero-phase wavelet to eliminate sea surface, while the receiver, usually towed at greater depth, is
variations between shots and sail lines. This step is essential for affected by a combination of sea heights, currents, and vessel
enhancing the resolution and consistency of the seismic data, speed. Statics correction, derived from water-bottom reflections
providing a robust platform for future interpretation work. Source compared with simultaneous multibeam echosounder data (MBES;
signatures are estimated from the seismic data on a shot-by-shot bathymetry data from the same seismic vessel) and receiver depths
basis using high-order statistics, as described by Bekara (2021). estimated during the deghosting step, compensate for sea surface
This method offers the advantage of not requiring alignment with variations. After applying statics correction, the data are redatumed
the water-bottom reflection, which is particularly challenging in and aligned with the bathymetric reference datum, taking tidal
the presence of sea surface statics. Additionally, source directivity time variations into account. This ensures that the final 3D volume
compensation is applied using modeled notional signatures, result- aligns with the measured bathymetry and delivers continuous
ing in enhanced resolution by harmonizing the wavelet across seismic events throughout the data set. Figure 7 illustrates on
emission angles and azimuths. Figure 6 shows on CMP stack the CMP stack how the water bottom appears more continuous and
successive changes associated with applying receiver deghosting, flat following the application of sea surface statics.
source deghosting, and designature, with a clear enhancement of Demultiple. Demultiple processing removes unwanted reflec-
the data in the shallow section. tions that can mask primary seismic events. The workflow
combines 3D surface-related multiple
elimination and 3D wave-equation
multiple modeling (Barnes et al., 2015)
to create accurate multiple models.
DOI:10.1190/leedff.2025.44.issue-3

These models are then subtracted from

the data, enhancing clarity.
Challenges arise during subtraction
due to sea surface statics. Wave height
variations affect multiples nonlinearly
Figure 6. Premigration stack comparison: deghosting and designature.
compared to primary events. To address
this, the adaptation process allows for
larger time-shift corrections in longer
(time) and narrower (space) windows.
This premigration demultiple step effec-
tively eliminates water-bottom rever-
berations and peg-leg multiples, even
in noisy areas with weak primary sig-
nals, as illustrated in Figure 8.
Imaging. Following the demultiple
process, the data undergo 4D regular-
ization, which involves reconstructing
well-populated offset classes from the
acquired data. This step is particularly
Figure 7. Premigration stack comparison: statics. important in 3D UHRS surveys, where

Figure 8. Migration stack comparison: demultiple.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 173
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 9. Migration velocity model rendered on migrated stack. The dashed line shows the location of the depth by time slice and the section.
DOI:10.1190/leedff.2025.44.issue-3

dense spatial sampling can lead to data gaps if not properly regular- with multiples (Lu et al., 2014) and least-squares migration, could
ized. The regularization process utilizes an antialias antileakage further enhance spatial and temporal resolution.
Fourier transform (Schonewille et al., 2009) to ensure high-fidelity
reconstruction of the seismic signal. Geological interpretation
The final stage of the processing workflow involves 3D Geological interpretation is essential for constructing a ground
Kirchhoff prestack time migration, which is critical for accurately model. Access to continuous 3D UHRS volumes offers a significant
positioning and focusing the seismic wavefield. The migration advantage over a grid of 2D UHRS data when it comes to inter-
process helps correct for the effects of dipping reflectors and other pretation. The primary goals of interpretation are to identify key
geological features, leading to improved interpretability of the soil units, map structural features, and detect potential geohazards
seismic data. Velocity model building is an integral part of the such as gas pockets, subsurface voids, or unstable layers. In cases
imaging process. The initial velocity model is generally derived of shallow gas, this information is often required at the start of a
from geotechnical P-wave velocity logs, when available, and refined geotechnical campaign. Therefore, beginning interpretation as
through iterative residual moveout (RMO) picking. The automated soon as UHRS acquisition starts is crucial. An ultra-fast-track
RMO picking is performed on a dense grid, ensuring that the migrated 3D UHRS volume can be progressively built while data
velocity model accurately represents the geological structure of are still being acquired.
the survey area. The final velocity model is then used to guide the With 2D UHRS data, interpretation is a labor-intensive
migration process, resulting in a high-resolution 3D seismic manual process. Applying this same approach to 3D UHRS data
volume. Figure 9 shows a migrated stack (time section and time would be inefficient and could take months to complete. To ensure
slice) with the velocity field overlaid. The velocity field correlates timely interpretation, semiautomated and fully automated methods
well with the geological structures. should be employed. Pauget et al. (2009) describe a technology
Limonta et al. (2024) demonstrate that a 3D broadband pro- that enables global geological modeling while offering users a
cessing workflow significantly improves vertical and spatial resolu- degree of control. While ML techniques are expected to facilitate
tion over a vintage 2D UHRS processing data set, where reflections this process on 3D UHRS data, we are still in the initial stages
are frequently mislocated and diffractions remain visible. of their development.
Although the current migrated product is of high quality,
future imaging enhancements could be made by migrating the Quantitative interpretation
data in the depth domain with a UHRS velocity model. This One of the most promising applications of 3D UHRS data is
would provide a more reliable product for direct interpretation in in the field of QI, particularly for predicting soil properties.
the depth domain. The velocity model could potentially be devel- Accurate soil property prediction is crucial for the design and
oped using techniques like full-waveform inversion (FWI) (Ryan installation of wind turbine foundations, as it directly impacts
et al., 2024). When available, legacy conventional oil and gas the stability and longevity of the structures. Traditional methods
seismic data can be utilized to build the background velocity of soil property estimation often rely on extensive borehole data
model. Additionally, other imaging methods, such as imaging and cone penetration tests (CPTs), which can be time consuming

174 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 10. Measured and predicted cone resistance profiles at seven CPT locations from the Ten Noorden van de Waddeneilanden Wind Farm Zone.

and expensive. However, integrating 3D UHRS data and geo- delivers more detailed and accurate results, particularly in capturing
technical data into QI workflows offers a more efficient and small-scale variations in soil properties.
potentially more accurate alternative. This could also lead to a Opportunities with QI. To derive soil properties such a Gmax
DOI:10.1190/leedff.2025.44.issue-3

reduced and more focused geotechnical campaign. The use of (small strain modulus), it is essential to move from acoustic to
certain attributes, such as RGB decomposition and coherency elastic inversion and estimate shear velocity and density. This
volumes, is generally valuable to QC the interpretation work. requires a sufficient range of incidence angles in the prestack
Data-driven soil property prediction. Our data-driven work- domain and consistent high quality across near-, mid-, and far-
flow for predicting soil properties using 3D UHRS data involves angle stacks. Hence, having “long offset” is important. This
several key steps. First, the seismic reflectivity and velocity data approach is currently in the proof-of-concept stage, and if suc-
obtained from the 3D UHRS survey are used to derive acoustic cessful, it could significantly impact how geotechnical surveys
impedance, which is then correlated with CPT measurements are executed during offshore wind development projects.
for key soil units. This approach allows for the prediction of soil
properties, such as cone resistance, across the entire survey area, Future directions and technological advancements
even in locations where CPT data are sparse or unavailable. The application of 3D UHRS technology in offshore wind
The value of this approach lies in its ability to provide detailed development is still evolving. As demand for renewable energy
variations in soil properties at the resolution of the 3D UHRS grows, further advancements in seismic acquisition, processing,
data volume. By leveraging the dense and high-quality data and interpretation techniques will be crucial to meeting the
provided by 3D UHRS technology, geophysicists can create industry’s needs. Several key areas of development are likely to
more quantitative and reliable ground models, reducing the need enhance the 3D UHRS capabilities in the future.
for extensive CPT campaigns and potentially lowering project Enhanced acquisition technologies. Advances in seismic source
costs. Furthermore, the ability to predict soil properties across and receiver technology will continue to drive improvements in
a large area allows for better planning and optimization of data quality and resolution. Having an acquisition geometry
turbine foundation designs, ensuring that they are tailored to where sources are located at the front end of each streamer would
the specific conditions at each site. provide regular near-zero-offset data. However, this would imply
The 3D UHRS poststack data from the Ten Noorden van de the use of deblending techniques that are able to clean overlapping
Waddeneilanden Wind Farm Zone has effectively demonstrated shot records.
the method’s capabilities (Polyaeva et al., 2024). The predicted Acquisition setup incorporating an air gun, could supply
cone resistance was validated against measured CPT data, and, the lower frequencies that are lacking in the UHRS data.
as shown in Figure 10, the two profiles align well, even in geologi- Furthermore, achieving longer offsets is possible by deploying
cally complex areas. Figure 11 illustrates the cone resistance ocean-bottom nodes on the seabed and/or using longer streamers.
attributes rendered onto the seismic section, interleaved with the This combination should provide suitable data quality to derive
CPT profile, further confirming the strong correlation between a reliable velocity model.
predicted and measured values. Additionally, the predicted cone Specialized acquisition methods can be employed to record
resistance provided a continuous 3D UHRS volume, enabling the shear waves or to focus on boulder detection. However, it is
development of a more quantitative ground model. Compared to essential to consider that these techniques may increase both the
traditional 2D UHRS predictions, the 3D UHRS approach cost and duration of development projects.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 175
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 11. Cone resistance property derived from 3D UHRS migrated stack rendered on the seismic cube interleaved with the measured cone resistance from six CPTs from the Ten Noorden
van de Waddeneilanden Wind Farm Zone.

The development of autonomous vessels equipped with Techniques such as controlled-source electromagnetics (CSEM),
seismic sensors presents opportunities for improving the efficiency marine magnetics, and high-resolution bathymetry can complement
and coverage of 3D UHRS surveys, especially in areas where 3D UHRS data, providing additional information on subsurface
traditional survey vessels are limited by obstacles and/or envi- properties that are not easily detected by seismic methods alone.
DOI:10.1190/leedff.2025.44.issue-3

ronmental constraints. For example, integrating CSEM data with 3D UHRS data
ML and AI in processing. The use of ML and artificial intel- can help identify variations in sediment conductivity, which is
ligence (AI) techniques in UHRS data processing is rapidly gaining useful for detecting gas hydrates, shallow gas, or fluid migration
traction. These methods have the potential to significantly acceler- pathways. Similarly, combining seismic data with high-resolution
ate the processing project turnaround time. By training ML bathymetric and multibeam echosounder data can provide a more
algorithms on a wide selection of 3D UHRS data sets, it is possible complete picture of the seafloor’s morphology, which is critical
to automate the key steps of the entire seismic processing sequence. for assessing potential seabed hazards and optimizing turbine
AI-based tools can also be used for feature recognition in foundation designs.
seismic data, such as detecting boulders, gas pockets, or faults,
which can be time consuming when performed manually. These Conclusion: The future of offshore wind farm
automated methods not only speed up the interpretation process development with 3D UHRS technology
but also reduce human bias and increase the consistency of As the offshore wind industry continues to grow, the role of
results. Finally, ML technology could play a significant role in 3D UHRS technology in supporting sustainable and efficient
generating reliable transfer functions between geophysical and wind farm development will become even more significant. The
geotechnical attributes. ability to provide detailed high-resolution 3D images of the
Advanced inversion techniques. The prestack inversion tech- shallow subsurface will remain critical for optimizing turbine
niques will focus on delivering even more reliable and accurate placement, minimizing environmental impacts, and reducing
models of subsurface geotechnical properties. Advances in elastic geotechnical risks.
inversion algorithms will enable more accurate predictions of Looking forward, advancements in acquisition technology,
parameters such as porosity, shear velocity, and density, which data processing, and QI will enhance the 3D UHRS capabilities
are essential for assessing soil stability and foundation integrity. and ensure that it remains a key tool in the development of offshore
These high-resolution models will provide developers with greater wind farms. The integration of AI, ML, and other geophysical
confidence in their site assessments and reduce the need for methods will allow for more efficient workflows and deeper insights
extensive ground truthing through boreholes or CPTs. into subsurface conditions, while ongoing efforts to reduce the
Moreover, improvements in FWI techniques, which utilize environmental footprint of seismic surveys will contribute to the
the entire seismic wavefield, will allow for the extraction of finer industry’s sustainability goals.
details from 3D UHRS data sets. FWI has already shown sig- By continuing to innovate and refine 3D UHRS technology,
nificant quality uplift in oil and gas exploration, and its application the offshore wind sector can accelerate the global transition to
in offshore wind farm development could lead to a step change renewable energy, driving down costs, improving project safety, and
in the accuracy of subsurface imaging. supporting the delivery of clean energy to millions of homes and
Integration with other geophysical methods. Integrating 3D businesses worldwide. With the combination of high-quality data
UHRS surveys with other geophysical methods will become and innovative technology, 3D UHRS technology is poised to play
increasingly important for comprehensive site characterization. an essential role in the future of renewable energy development.

176 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 Acknowledgments Second EAGE Marine Acquisition Workshop, https://doi.
We thank TGS management for their support in publishing org/10.3997/2214-4609.202034003.
and our TGS colleagues who have been involved in the 3D UHRS Farmani, B., M. Lesnes, and Y. Pal, 2023, Multisensor noise attenuation
with RIDNet: 84th Annual Conference and Exhibition, EAGE,
projects. We thank the Community Offshore Wind consortia,
Extended Abstracts, https://doi.org/10.3997/2214-4609.202310225.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

including RWE Renewables and National Grid Ventures, and Limonta, L., V. Butterworth, B. Caselitz, M. Lange, and J. Oukili,
Netherlands Enterprise Agency (RVO) for permission to show 2024, Elevating 3D ultra high resolution processing and imaging for
data examples. wind farm site characterisation: 85th Annual Conference and
Exhibition, EAGE, Extended Abstracts, https://doi.
Data and materials availability org/10.3997/2214-4609.2024101166.
Data associated with this research are confidential and cannot Lu, S., N. D. Whitmore, A. A. Valenciano, and N. Chemingui, 2014,
Enhanced subsurface illumination from separated wavefield imaging:
be released.
First Break, 32, no. 11, https://doi.org/10.3997/1365-
2397.32.11.78585.
Corresponding author: [email protected] Pauget, F., S. Lacaze, and T. Valding, 2009, A global approach in seismic
interpretation based on cost function minimization: 79th Annual
References International Meeting, SEG, Expanded Abstracts, 2592–2596,
Barnes, S. R., R. F. Hegge, H. Schumacher, and R. Brown, 2015, https://doi.org/10.1190/1.3255384.
Improved shallow water demultiple with 3D multi-model subtraction: Polyaeva, E., N. Pernin, B. Caselitz, Ruiz, R., and N. Lee, 2024, Soil
85th Annual International Meeting, SEG, Expanded Abstracts, properties prediction based on 3D ultra-high resolution seismic: A
4460–4464, https://doi.org/10.1190/segam2015-5856770.1. data driven inversion workflow: Fifth EAGE Global Energy Transition
Bekara, M., 2021, Mixed phase seismic wavelet estimation using the Conference and Exhibition, https://doi.org/10.3997/2214-
bispectrum: 82nd Annual Conference and Exhibition, EAGE, Extended 4609.202421164.
Abstracts, https://doi.org/10.3997/2214-4609.202010410. Ryan, C., K. Liao, H. Moore, H. Westerdahl, M. Thompson, Å. S.
Bekara, M., C. Davison, M. Lange, and L. Limonta, 2024, Deghosting Pedersen, J. Mispel, M. Wierzchowska, R. Dehghan-Niri, and Y.
of UHR seismic data via dynamic ghost tracking: 85th Annual Biryaltseva, 2024, Short streamer acquisition — The potential and
Conference and Exhibition, EAGE, Extended Abstracts, https:// the challenges: Fourth EAGE Workshop on Practical Reservoir
DOI:10.1190/leedff.2025.44.issue-3

doi.org/10.3997/2214-4609.202410379. Monitoring, https://doi.org/10.3997/2214-4609.202430024.

Caselitz, B., L. Limonta, J. Oukili, J. Tegnander, and V. Catterall, 2024, Schonewille, M., A. Klaedtke, and A. Vigner, 2009, Anti-alias anti-
3D ultra high resolution seismic processing — A case study from leakage Fourier transform: 79th Annual International Meeting, SEG,
offshore USA: Fourth International Meeting for Applied Geoscience Expanded Abstracts, https://doi.org/10.1190/1.3255533.
& Energy, SEG, Expanded Abstracts, 1577–1581, https://doi. Widmaier, M., C. Roalkvam, and O. Orji, 2023, Advanced 3D seismic
org/10.1190/image2024-4100740.1. crossover technologies between hydrocarbon exploration, CCS
Davies, D., and W. Rietveld, 2020, The journey to 1M resolution at Clair development, and offshore wind: First Break, 41, no. 11, 53–58,
— Lessons from acquisition and processing: Proceedings of the https://doi.org/10.3997/1365-2397.f b2023090.

ANNOUNCING

THE SEAM AI MODEL

Unlock the power of cutting-edge AI technology

using SEAM’s benchmark models.

Licensing now available!

LEARN MORE
seg.org/seam

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 177
 Quantum-annealing-assisted surface-consistent phase
and amplitude refraction residuals estimation
Diego Rovetta1, Marcin Dukalski2, and Apostolos Kontakis1
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

https://doi.org/10.1190/tle44030178.1

Abstract solving difficult optimization problems, leading to more precise

In an era of customized hardware designed to tackle specific resource exploration, environmental monitoring, and hazard
geophysical challenges, recent advancements in quantum computing assessment. There are complementary metal-oxide semiconductor
could offer promising solutions to complex global optimization (CMOS) technology-based devices, such as digital annealers or
problems using hybrid quantum-classical solvers. We introduce an vector engines, designed to solve very large quadratic unconstrained
application of seismic data processing where the stack power of the binary optimization problems (Afanasyev et al., 2021; Codognet
traces is maximized through quantum annealing for surface- et al., 2022). Among various QPUs, quantum-annealing (QA)
consistent phase and amplitude refraction residual calculations. machines use quantum mechanics to simultaneously and very
Refraction residual corrections are a necessary step in the processing efficiently explore potential solutions for problems, which in theory
workflow to obtain a detailed characterization of the subsurface. at large scale should be offering a significant advantage over
The problem of estimating accurate refraction residuals can be classical algorithms for certain optimization tasks (Brooke et al.,
solved in a framework where relative time shifts and amplitude 2001). However, only problems of a small number of dimensions
distortions of seismic traces are evaluated in multiple common- can be tackled by these machines currently. Hybrid solvers
midpoint-offset bins through computerized cross-correlation (McGeoch and Pau, 2020a) combine QA with classical computing
operations. The results are then used to build a system of linear techniques to capitalize on the strengths of both methodologies.
equations that is simultaneously inverted for surface-consistent This enables faster convergence to global optima, resulting in
DOI:10.1190/leedff.2025.44.issue-3

shot and receiver phase and amplitude variations. We rewrote the more accurate and reliable solutions compared to classical methods
problem by substituting the cross-correlation operations with global alone, and efficient management of large problems that would be
optimization performed by a “black-box” hybrid solver, which uses unattainable with QA alone. All of these approaches allow users
a physical quantum-annealing machine. With the projected hard- to run algorithms and hardware without needing in-depth quan-
ware and technology improvements of the near future, the new tum or classical combinatorial optimization expertise.
approach is expected to be more accurate than the current solution It is important to understand which geophysical applications
because it prevents the phenomenon of the so-called cycle skips in stand to benefit from the development of these technologies. Here,
the cross-correlation process. The method has been benchmarked we discuss a case study where the QA-assisted hybrid solver helps
with a classical global optimization algorithm and tested with characterize the shallow subsurface through improved estimation
promising results on synthetics of industrial size and relevance. A of phase and amplitude refraction residuals.
first attempt of application on field data has also been carried out. In land seismic imaging, accurately characterizing the near
Quantum-annealing-based global optimization can be applied to surface is crucial for data processing. The shallow subsurface often
other disciplines such as material science, logistics, and finance. presents abrupt physical property changes (i.e., seismic velocity
variations) due to complex geologic features, such as karst, sand
Introduction dunes, sabkhas, and wadis, in desert environments. Such subsurface
Applied geophysics is being revolutionized by new technologies variations can cause phase and amplitude distortions in the elastic
that are driven by advancements in data acquisition, processing, wavefield during seismic acquisition.
and computational power. The fast-paced developments in the Phase distortions can produce time delays in recorded arrivals,
latter enabled much of the progress observed in the past decade. which are referred to as statics. Long-wavelength statics can be
We are now in an era of increasingly specialized hardware that corrected using a smooth velocity model. However, shallow geology
is designed to tackle specific geophysical challenges. Graphics can result in relative time delays for different source and receiver
processing units (GPUs) do intensive parallel computations for locations that fall below the resolution limit of the velocity model
tasks such as seismic imaging. Field programmable gate arrays estimate, known as short-wavelength or residual statics.
(FPGAs) do highly efficient customizable processing for real-time Additionally, shallow anomalies can affect the amplitude of seismic
signal processing and data filtering. Finally, the quantum process- signals, causing relative variations for different source and receiver
ing units (QPUs) are expected to start the next revolution in what locations called amplitude residuals. Correcting phase and ampli-
we can compute within a realistic time. tude distortions in land seismic data is challenging due to the
There is also a class of combinatorial optimization processing presence of noise and unwanted signals that can compromise
units, which are purpose-built devices that promise to excel in reflection arrivals.

Manuscript received 9 October 2024; revision received 12 January 2025; accepted 16 January 2025.
1
Formerly, Aramco Delft Research Center, South Holland, Netherlands. E-mail: [email protected]; [email protected].
2
Aramco Delft Research Center, South Holland, Netherlands. E-mail: [email protected].

178 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 One approach to address this issue is to use refracted waveforms Method
to compensate for phase and amplitude residuals. The current Modification of the transmission-based surface-consistent
method is based on the cross correlation (XCORR) of each trace workflow. The approach used in this study is based on a trans-
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

of the seismic gather with a pilot trace obtained from all the traces mission-based surface-consistent solution (Colombo et al., 2020b),
of that gather, for example, by stacking them together (Colombo which involves sorting seismic data into CMP-X, CMP-Y, and
et al., 2016, 2020a, and 2020b). However, this process can suffer offset (XYO) bins. First, statistics are performed on the first-break
from cycle skips, which is when similar peaks within the signals picks of each XYO gather to identify and reject outliers and
to cross correlate are wrongly associated to the same geologic derive mean traveltimes for each bin (first block of Figure 1a).
features. This matching of noncorresponding peaks may lead to Then, a robust time-offset distribution is derived for each common
inaccurate results. midpoint (CMP) that is fitted with a constrained spline function
Alternatively, phase and amplitude residuals can be estimated and converted into a layered velocity profile. Alternatively, the
after maximizing the stack power of the seismic gathers. This time-offset distribution can be directly inverted for the layered
approach is challenging due to the multimodality of the objective velocity profile. All of the CMP profiles are interpolated to obtain
function, and current classical global optimization techniques are a 3D velocity model used for evaluating long-wavelength phase
often inefficient to find an acceptable solution in a reasonable corrections (second block of Figure 1a). The velocity model is
amount of time (Pierini et al., 2019). typically sufficient for providing long-wavelength statics, but it
The QA process, a form of quantum computing carried out can also be used as a starting model for other velocity model
on a specific hardware (quantum annealer) that leverages quantum building tools such as tomography, full-waveform inversion, or
superposition and tunneling, can be used to efficiently find global in a joint inversion scheme.
optima in optimization problems. QA can be applied to estimate Next, a set of trace-related amplitude (∆A( f )) and phase
residuals through stack-power optimization without suffering (∆Θ( f )) residuals are calculated in the XYO gather by means
from cycle skips or local optima traps. The problem can be for- of a cross-correlation operation of the early-arrival waveforms
mulated as a discrete quadratic model (DQM), and a hybrid solver with a pilot trace derived from the data (third block of Figure 1a).
DOI:10.1190/leedff.2025.44.issue-3

combining classical and quantum optimization can be used to This pilot trace can be, for example, the stack of all the traces
find an estimate of the residuals. in the gather. The convolutional wave propagation model has
Dukalski et al. (2023) demonstrated the application of this been adapted to the specific case of transmitted (rather than
method for phase residuals and for small-size synthetic data sets. reflected) wavefields, where seismic traces are decomposed into
Rovetta et al. (2023) tackled the same problem with a “black-box” the convolutional effects of the source and receiver functions,
solver combining classical and quantum computations and were and of the Earth’s Green’s function between the source and the
able to apply it to synthetic data sets of bigger size. receiver (Cambois and Stoffa, 1992).
This work further extends the approach of Rovetta et al. (2023) While operating in frequency domain and decomposing the
to amplitude residuals and applies it to both industrial-size synthetics propagation model in residual terms and in an average response
and small-size field data sets. Additionally, a simulated (or thermal) term from the XYO sorted gather, we obtain
annealing classical algorithm (ThAnn), which is to our knowledge
the closest analogue of the QA global optimizer among classical R( f ) = si( f )rj ( f )q( f ), (1)
algorithms (Kirkpatrick et al., 1983; Rothman, 1984), is used to
benchmark the quantum solution through global optimization. where R( f ) is the ratio between the input trace after propagation
All of the results shown in this paper have been obtained by compensations and an average gather response (pilot trace);
performing global optimization operations on a physical QA si ( f ) and rj ( f ) are the source and receiver surface-consistent
machine through a black-box hybrid DQM solver provided by residual terms, respectively; and q( f ) is a subsurface-related
the same vendor. We believe that this contribution could help residual term.
pave the way to the application of QA in geophysical global The phase and amplitude residuals associated with each trace
optimization problems of industrial size and relevance and eventu- are inverted for surface-consistent source and receiver residuals
ally to further fields of interest in the energy sector and beyond. (fourth block of Figure 1a) by solving the system of equations:

Figure 1. (a) Workflow of a conventional surface-consistent solution for refraction residuals estimation. (b) Modified workflow where the XCORR block is substituted with a stack-power
optimization performed by global optimization through thermal annealing (ThAnn) or QA.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 179
 ∆ A​( f )​ = |​ ​si​(​ f )|​ ​|​rj​(​ f )​||​ q​( ​f )|​ However, as shown by Dukalski et al. (2023), current quantum

{ ∆ Θ​( f )​ = ​Θ​s​​​(​ f )​+ ​Θ​r​​​(​ f )​+ ​Θq​​( f ) ​

​ ​  ​ ​. (2) machines can only reliably solve problems with a relatively small
i j number of variables, specifically where M × K is smaller than about
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

170. To overcome this limitation, we employed a hybrid DQM

The phase residuals Θsi(f ) and Θrj (f ), also called short- solver (McGeoch and Pau, 2020; McGeoch et al., 2020) that
wavelength statics, are added to the long-wavelength statics previ- combines classical and quantum components. The classical part of
ously calculated to give the total refraction statics. Total refraction the algorithm explores different solution subspaces, and for each
statics are then applied to the seismic traces. Similarly, after subspace, it calls the quantum part of the algorithm to provide a
spherical divergence compensation, the amplitude residuals |si(f )| result. Finally, the best outcome from all of the results is selected
and |rj(f )| are corrected through amplitude balancing or frequency- as the final solution. This hybrid approach allows us to solve larger
dependent deconvolution operations (Colombo et al., 2020a). problems than would be possible with a quantum machine alone.
In some cases, the XCORR method may encounter cycle- Because we make use of a black-box solution provided by the
skipping issues, where noncorresponding peaks are matched, vendor that also gives us access to the QA machine, we do not need
leading to errors in the calculation of relative time delays. An to deal with the implementation of the hybrid DQM solver com-
alternative approach for estimating residuals is to maximize the ponents, encoding to BQM included. However, we are currently
stack power, which is a global optimization problem (Ronen and designing our own hybrid solver to make use of the quantum annealer
Claerbout, 1985). The objective function is often highly complex at a lower level and to further optimize our processing workflow.
and multimodal, making it challenging to find the global maxi- This activity will be considered for another future publication.
mum using conventional global optimization techniques (Pierini
et al., 2019) such as ThAnn or genetic algorithms. However, QA Examples
can be successful in solving this problem. The accuracy and robustness of the method have been tested by
To address this challenge, the transmission-based surface- using synthetic and field data sets. As explained in the previous
consistent workflow has been modified by replacing the XCORR section, the problem of refraction residuals calculation is solved in
DOI:10.1190/leedff.2025.44.issue-3

operation with global optimization (Figure 1b) performed by a two steps. First, for each XYO gather, we calculate the phase and
classical (e.g., through ThAnn) or by a quantum (e.g., through QA) amplitude variation of each trace with respect to the gather pilot (or
machine. The novel approach can output identical stack-power stack) trace. We can perform this operation through XCORR or
solutions that differ by an overall shift, and hence different XYO ThAnn classical algorithms (CPUs only) or through a hybrid solver
bins need to be aligned relative to each other using their pilot traces. making use of a physical QA machine (CPUs and QPUs). Second,
Stack-power maximization on a QA machine. The stack-power the phase and amplitude trace residuals (outcome of the first step)
maximization of M traces di (with i = 1, …, M), shifted in time of are then used to compute the source and receiver refraction residuals
τi, in an XYO gather corresponds to the solution of the problem: by solving the inverse problem of equation 2. The outcomes of the
second step are merged so residual maps can be generated and show
2 the phase and amplitude residuals for each source and receiver
​ ​​​‖​∑ i=1​di​(​ t − ​τi​​)​‖​​2​,
M
​argmax​
τ
(3) location. These maps may then be used to analyze the shallow geology
of the investigated subsurface through a qualitative interpretation
or in a more quantitative approach to correct the seismic traces from
which corresponds to the following DQM (Dukalski et al., 2023): the amplitude and phase distortions due to the shallow geology and
to produce a better stack or migrated image of the subsurface
(Colombo et al., 2020a, 2020b). The use of the residuals maps goes
​argmax​​∑ i=1
M
​∑ j=i+1
M
​∑ a=1
K
​∑ b=1
K
​δ​y​​​a​​δ​y​​​b​​Iijab
​ ​. (4) beyond the scope of this paper, and it will not be discussed.
i j
y∈​{​ 1,…, K }​M​
Synthetic data example. The synthetic data were extracted
from a portion of the SEAM Arid model data set, which accurately
Here, δyia and δyjb are the Kronecker delta functions of the represents the complex subsurface features in desert environments
discrete variables yi used to control the application of the time (Oristaglio, 2015). Figure 2a shows the map of the first 50 m
shifts τa and τb to the i th and j th traces from one of the K possible depth average compressional wave velocities for the full model
shifts in the solution space SK = {τ1, …, τK}M . Here, Iijab represents (10 × 10 km) and for the area of interest (AOI), which is marked
the frequency domain inner product between spectra of the i th by a red square (3 × 3 km) and expanded in Figure 2b.
and j th traces shifted by τa and τb, respectively. About 144,000 traces have been recorded and 2200 XYO
Quantum annealers are specialized machines that utilize gathers have been processed in total, giving an average of about
qubits, which can exist in a state of analogue superposition of two M = 65 traces per bin. The number of allowed discrete time
bits, to solve complex optimization problems. These problems are shifts per trace has been set to K = 6 (time shifts from 0 to 30 ms,
first mapped onto a binary quadratic model (BQM), which with dt = 6 ms), corresponding to 665 possible configurations per
describes the energies and interactions between the qubits. To average bin.
solve the DQM equation 4 using a quantum machine, we need Figure 3 compares the result of the QA hybrid solver global
to convert it into a BQM by encoding the discrete variables yi in optimization of the stack power of one XYO gather (Figure 3d)
terms of binary variables xi. with (1) the original gather (Figure 3a), (2) the solution obtained

180 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 through XCORR (Figure 3b), and (3) the solution obtained through superiority of the first over the latter. ThAnn performed similarly
ThAnn (Figure 3c). The processing window used for XCORR, to QA, with improvement by a factor of about 20.
ThAnn, and QA was 120 ms around the first-break arrival. In this In addition to visually inspecting the lateral continuity of the
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

particular XYO gather, we saw one of the largest improvements seismic traces in the gather shown in Figure 3 (a well-established
in the stack power (by a factor of about 20) using QA compared qualitative method in the industry for evaluating trace alignment
to about 4 using the conventional XCORR approach, showing a solutions), we also employed a more quantitative approach. This
involved comparing the stack power of
the gather of realigned traces using dif-
ferent techniques. The best solution is
the one that gave the highest stack
power. Figure 4a shows the stack-power
distributions of the original gathers and
of the ones corrected by XCORR,
ThAnn, and QA for the whole AOI (all
of the processed XYO gathers). Figure 4b
reports the performances of the three
solutions: with stack-power improve-
ments of about 41%, QA and ThAnn
are better performing than XCORR,
which shows an improvement of 39%.
Figure 2. (a) Average compressional velocity model for the first 50 m in depth of the SEAM Arid model and (b) zoom of the AOI However, QA and ThAnn do not show
used in this work. significant differences, as confirmed by
the distributions of their stack-power
DOI:10.1190/leedff.2025.44.issue-3

improvements over XCORR (Figure 4c).

The differences in performance of the
various solvers vary dramatically
between gathers, and the source of this
variation needs to be investigated more
closely in follow-up publications.
However, the high quality of synthetic
data (noise free) makes the average
performance very similar, also showing
small differences in the shallow subsur-
face reconstruction.
The refraction residual (RR) maps
after inversion are shown in Figure 5,
where the left and right columns are for
phase and amplitude residuals, respec-
Figure 3. SEAM Arid model AOI example of alignment for one gather: (a) original data, (b) XCORR, (c) ThAnn, and (d) QA. The tively. We compare the QA results
latter shows better improvement both quantitatively and qualitatively when compared with the conventional method. (Figure 5c) with the conventional

Figure 4. (a) Histograms of the stack power of original data and of data corrected using XCORR, ThAnn, and QA techniques for the SEAM Arid model AOI. (b) Overall stack-power
improvements over original data using XCORR, ThAnn, and QA techniques. (c) Histogram of the difference in the stack-power improvements of QA and ThAnn over XCORR.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 181
 method XCORR (Figure 5a) and with ThAnn (Figure 5b). In this selected one swath of the available data of about 20 × 6 km
case, the results of ThAnn and QA are generally comparable with (Figure 7) and an AOI (red rectangle) of about 6 × 2 km.
the conventional solution for both phase and amplitude. As shown Approximately 860,000 traces have been recorded and 8500
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

in Figure 6, this is also confirmed by analyzing the differences of XYO gathers have been processed in total, giving an average of
the residual estimations between XCORR and ThAnn and between about M = 100 traces per bin. The number of allowed trace discrete
XCORR and QA that are generally negligible. time shifts has been set to K = 8 (time shifts from 0 to 28 ms,
Field data example. We now discuss the application of the being dt = 4 ms), corresponding to 8100 possible configurations
algorithm to field data acquired in a desert environment. We per bin.
DOI:10.1190/leedff.2025.44.issue-3

Figure 5. Estimation of the SEAM Arid model AOI residual maps. The left and right columns are for phase and amplitude residuals, respectively. The rows show the results of (a) XCORR,
(b) ThAnn, and (c) QA. The results of ThAnn and QA are generally comparable with the conventional solution for both phase and amplitude.

182 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms
DOI:10.1190/leedff.2025.44.issue-3

Figure 6. Differences of the SEAM Arid model AOI residual estimations between XCORR and ThAnn (first row) and between XCORR and QA (second row) for phase (left column) and amplitude
(right column). The differences are generally negligible.
Figure 8 is the result of the QA hybrid solver global optimization
of the stack power of one XYO gather (Figure 8d), compared with
(1) the original gather (Figure 8a), (2) the solution obtained through
XCORR (Figure 8b), and (3) the solution obtained through ThAnn
(Figure 8c). As in the previous section for the synthetic data example,
a processing window length of 120 ms around the first-break arrival
has been chosen for each XCORR, ThAnn, and QA. In this
particular XYO gather, the stack power improved by a factor of
about 3 using QA and by about 1 using the conventional approach,
showing a superiority of the first over the latter. ThAnn performed
similarly to XCORR, with only a slight improvement of 10%.
Even in this case, the visual inspection of the lateral continuity
of the seismic traces in the gather of Figure 8 has been complemented
by comparing the stack power of the gather of the realigned traces
by the different techniques. Figure 9a shows the stack-power
distributions of the original gathers and of the ones corrected by
XCORR, ThAnn, and QA for the whole AOI (all of the processed
XYO gathers). Figure 9b reports the performances of the three
Figure 7. Phase and amplitude residuals solution using XCORR for a swath of seismic data solutions: with a stack-power improvement of about 44%, QA is
recorded in a desert environment. The red rectangle is highlighting the AOI where we better performing than ThAnn and XCORR, which show improve-
performed the calculation with the quantum annealer. ments of 43% and 38%, respectively. Similar improvements of QA
and ThAnn over XCORR can also be appreciated by analyzing
the distributions of Figure 9c. We observed that, compared to the
previous synthetic data example, the results of this field (noisy)
data test show smaller differences in the performance of the various

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 183
 solvers between gathers and a higher average improvement in the The main constraint is the limited number of variables (and
performance of QA over XCORR. corresponding qubits) we can deal with. If M is the number of
The phase and amplitude residual maps after inversion are seismic traces we want to process per gather and K is the number
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

shown in Figure 10, where the left and right columns are for phase of the allowed discrete time shifts per trace, when M and K are
and amplitude residuals, respectively. Also, in this case, we compare too big, we cannot fit the problem in the quantum computer. Even
the QA results with the XCORR and ThAnn ones, showing that if M and K are reasonably low enough to be managed, there is
all of the solutions are generally comparable for both phase and still an area where we cannot operate without working with hybrid
amplitude, with the QA solution being slightly smoother. This is solutions (red ellipse in Figure 12), which is combining QPUs
confirmed, especially for the amplitudes, by analyzing the differences with CPUs and GPUs. This application resides in this area, and
of the residual estimations between XCORR and ThAnn and its solution can be achieved with a black-box hybrid solver
between XCORR and QA, as shown in Figure 11. (McGeoch et al., 2020b).
We dealt with about 665 possible configurations in the synthetic
Discussion and way forward data example and with 8100 in the field data one. However, the
While an explanation of QA theory is beyond the scope of maximum data fold exceeded these values (86 traces for the
this publication, our goal is to highlight how QA machines can synthetic case and 845 traces for the field data case), giving for
already be used to solve simple geophysical problems using syn- many gathers more solution configurations. Because QA can deal
thetic and field data sets. This aims to draw the readers' attention with a limited number of traces/bin (and a maximum number of
to the current capabilities and scale of this technology, encouraging solution configurations), gathers with fold higher than this limit
them to view it as a practical tool rather than science fiction. were processed by splitting the data into different parts, with
Similarly, we decided not to give a description of the simulated smaller problems solved in parallel.
annealing algorithm, which is well known and sufficiently Our results show only slight improvements in the accuracy
described in the literature. However, it is important to remember of the solution of the residuals, and, at the moment, the
that QA is fundamentally different from
DOI:10.1190/leedff.2025.44.issue-3

simulated annealing. The latter com-

putes a global optimum through an
algorithm, which is highly inefficient
for large problems and cannot be com-
pleted in a reasonable time. In contrast,
QA finds the state of minimum energy
of a system, corresponding to the solu-
tion of the global optimization problem,
in a fraction of a second.
Figure 12 gives an idea of the
complexity of the problem we are try-
ing to solve. After the mathematical
formulation, a quantum formulation is
needed to run the optimization on a
quantum machine. This requires differ-
ent steps including Ising model building, Figure 8. AOI example of alignment for one field data gather: (a) original data, (b) XCORR, (c) ThAnn, and (d) QA. The latter shows
variable discretization, and encoding. better improvement when compared with the conventional method.

Figure 9. (a) Histograms of the stack power of original data and of data corrected using XCORR, ThAnn, and QA techniques for the field data AOI. (b) Overall stack-power improvements over
original data using XCORR, ThAnn, and QA techniques. (c) Histogram of the difference in the stack-power improvements of QA and ThAnn over XCORR.

184 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 calculation time cannot be compared with classical methods due quantum computers. Therefore, we believe that we will be able
to the limitation of available number of qubits and communication to obtain better and faster solutions than conventional techniques
delay between the classical and quantum machines. However, in the very near future. In parallel, we are also investigating
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

the progress of QPU technologies is speeding up, as well as the alternative solutions, where global optimization could be effi-
active research to find smart algorithms to fit the problem into ciently and accurately performed on customized machines.
DOI:10.1190/leedff.2025.44.issue-3

Figure 10. Estimation of the field data AOI residual maps. The left and right columns are for phase and amplitude residuals, respectively. The rows show the results of (a) XCORR, (b) ThAnn,
and (c) QA. The results of ThAnn and QA are generally comparable with the conventional solution for both phase and amplitude. The QA solution is slightly smoother than the others.

Figure 11. Differences of the field data AOI residual estimations between XCORR and ThAnn (first row) and between XCORR and QA (second row) for phase (left column) and amplitude (right
column). The differences are generally negligible, especially for the amplitudes.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 185
 Finally, having available quantum technologies to solve this
or similar optimization problems now opens new questions (as a
product of this research) such as: given multiple global optimizers,
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

how is it possible to rigorously evaluate their impact on the final

result? This is a research question that will keep coming back in
the context of benchmarking quantum versus classical methods.

Conclusion
Recent advancements in QA present a promising solution for
Figure 12. Conceptual workflow that is needed to solve a problem on a QA machine. On
complex global optimization problems by utilizing hybrid quan- the bottom left, a diagram shows when the problem can fit on a quantum computer. In
tum-classical solvers. In the field of seismic data processing, the this application, M corresponds to the number of seismic traces and K corresponds to the
estimation of phase and amplitude residuals in a transmission- number of time steps they are allowed to be shifted to correct for phase residuals.
based surface-consistent framework has been successfully addressed
using stack-power maximization with a physical QA machine. Codognet, P., D. Diaz, and S. Abreu, 2022, Quantum and digital
The method is expected to offer an advantage over conventional annealing for the quadratic assignment problem: Proceedings of the
techniques by avoiding the issue of cycle skips and providing a International Conference on Quantum Software, IEEE, https://doi.
org/10.1109/QSW55613.2022.00016.
more accurate solution. The approach has been validated on subsets
Colombo, D., F. Miorelli, E. Sandoval-Curiel, and D. Rovetta, 2016,
of the SEAM Arid model and on a field data set acquired in a pQC: A novel approach for robust automatic near-surface analysis
desert environment. in low-relief geology: The Leading Edge, 35, no. 11, 952–960, https://
The results show that the stack power is generally better maxi- doi.org/10.1190/tle35110952.1.
mized after aligning the traces with the quantum-assisted method Colombo, D., D. Rovetta, E. Sandoval-Curiel, and A. Kontakis, 2020a,
compared to the conventional technique. Additionally, the phase Transmission-based near-surface deconvolution: Geophysics, 85, no.
and amplitude residual maps obtained from both are consistent with 2, V169–V181, https://doi.org/10.1190/geo2019-0443.1.
DOI:10.1190/leedff.2025.44.issue-3

Colombo, D., D. Rovetta, E. Sandoval, and A. Kontakis, 2020b,

each other. A simulated annealing that performs global optimization
Transmission-based surface-consistent framework for residual statics,
classically has also been used to benchmark the quantum solution. deconvolution, and FWI: A new paradigm for near-surface analysis:
Overall, the use of the QA machine has proven to be effective The Leading Edge, 39, no. 6, 382–390, https://doi.org/10.1190/
in solving the problem of phase and amplitude residuals estimation tle39060382.1.
in a transmission-based surface-consistent framework, potentially Dukalski, M., D. Rovetta, S. van der Linde, M. Möller, N. Neumann,
offering soon a more accurate and efficient solution compared to and F. Phillipson, 2023, Quantum computer-assisted global optimiza-
conventional techniques. Similarly, the technique can be applied tion in geophysics illustrated with stack-power maximization for
refraction residual statics estimation: Geophysics, 88, no. 2, V75–V91,
to other global optimization applications in fields like material
https://doi.org/10.1190/geo2022-0253.1.
science, logistics, and finance. Kirkpatrick, S., C. D. Gelatt Jr., and M. P. Vecchi, 1983, Optimization
by simulated annealing: Science, 220, 671–680, https://doi.
Acknowledgments org/10.1126/science.220.4598.671.
The authors would like to thank Xinru Tang, Han Yao Choong, McGeoch, C., and F. Pau, 2020, Hybrid solver for discrete quadratic
and Sven Müller for valuable discussions and for reviewing the models: white paper.
manuscript. McGeoch, C., P. Farré, and W. Bernoudy, 2020, D-wave hybrid solver
service + advantage: Technology update: Technical Report
14-1048A-A.
Data and materials availability Oristaglio, M., 2015, SEAM update: The Arid model — Seismic explora-
Data associated with this research are confidential and cannot tion in desert terrains: The Leading Edge, 34, no. 4, 466–468, https://
be released. doi.org/10.1190/tle34040466.1.
Pierini, S., M. Aleardi, and A. Sajeva, 2019, Comparisons of recent global
Corresponding author: [email protected] optimization algorithms: Tests on analytic objective functions and
residual statics corrections: 81st Conference and Exhibition, EAGE,
Extended Abstracts, https://doi.org/10.3997/2214-4609.201901383.
References Ronen, J., and J. F. Claerbout, 1985, Surface-consistent residual statics
Afanasyev, I. V., V. V. Voevodin, K. Komatsu, and H. Kobayashi, 2021,
estimation by stack-power maximization: Geophysics, 50, no. 12,
VGL: A high-performance graph processing framework for the NEC
2759–2767, https://doi.org/10.1190/1.1441896.
SX-Aurora TSUBASA vector architecture: The Journal of
Rothman, D. H., 1984, Nonlinear inversion, simulated annealing, and
Supercomputing, 77, 8694–8715, https://doi.org/10.1007/
residual statics estimation: Stanford Exploration Project, SEP-41,
s11227-020-03564-9.
297–326.
Brooke, J., D. Bitko, T. F. Rosenbaum, and G. Aeppli, 1999, Quantum
Rovetta, D., M. Dukalski, and A. Kontakis, 2023, Quantum annealer-
annealing of a disordered magnet: Science, 284, 779–781, https://doi.
assisted residual refraction statics estimation on the SEAM Arid model
org/10.1126/science.284.5415.77.
dataset: 84th Annual Conference and Exhibition, EAGE, Extended
Cambois, G., and P. L. Stoffa, 1992, Surface-consistent deconvolution in
Abstracts, https://doi.org/10.3997/2214-4609.202310278.
the log/Fourier domain: Geophysics, 57, no. 6, 823–840, https://doi.
org/10.1190/1.1443296.

186 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 Ground-truth-calibrated onshore and offshore subsurface
infrastructure image from deep-learning-based 3D inversion
of magnetic data
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Souvik Mukherjee1, Jacques Y. Guigne2, Gary N. Young 3, Santi Adavani4, Kevin Kennelley5, Dillon Hoffmann5, Harshit Shukla3, Ronald S. Bell6, and
William N. Barkhouse6
https://doi.org/10.1190/tle44030187.1

Abstract overall geophysical signal so that contributions from other targets

In this study, we demonstrate the application of deep-learning- can be understood (e.g., Stalnaker et al., 2006; Chen et al., 2012;
based 3D inversion of magnetic data to image subsurface infra- Mukherjee et al., 2012; Hoversten and Schwarzbach, 2018).
structure. We highlight results from two case studies: an onshore Of late, a surge in interest in providing more accurate informa-
survey at Texas A&M University’s Rellis Campus site and an tion about near-surface geophysical targets such as buried utilities,
offshore survey in the northern Gulf of Mexico (GOM), off the oil and gas pipelines, abandoned well casings, and unexploded
coast of Louisiana. The onshore case utilized drone-acquired ordnance has led to an uptick in efforts to better characterize the
magnetic data to map buried utilities before construction. The metallic target instead of isolating its response from other geo-
offshore case employed a boat-towed magnetometer approximately physical features of interest. Most of this effort has been focused
3.5 m above the seafloor to locate oil well conductors disrupted on either a direct interpretation of the observed geophysical
by Hurricane Ivan in 2004 and presently buried under 35 to 45 m response (examples include diffraction hyperbola [Ali et al., 2021]
of sediment. The inversion results at the Rellis site were validated and localized anomalies in 2D magnetic and electromagnetic
against excavation data, revealing strong agreement in target maps [Bernstein et al., 2023]) or making limited 2D to 3D
DOI:10.1190/leedff.2025.44.issue-3

location and depth (within 17 cm). In the GOM survey, the detection and inferences such as depth to target for buried cables
artificial intelligence (AI)-driven inversion successfully extended and pipelines (e.g., Guigné and Gogacz, 2015; Li et al., 2020).
conductor imaging beyond the limits of acoustic methods, provid- The acquired geophysical data over cultural targets contain
ing critical information on conductor geometry near the well more detail than has been meaningfully extracted. Successful
conductor bay. This work highlights the effectiveness of AI-driven extraction of this information can significantly impact strategic
inversion techniques in enhancing subsurface imaging, offering commercial and safety-critical applications. For example, utility
cost-effective and scalable solutions for applications in utility companies consider four different levels of information when plan-
mapping, environmental monitoring, and hazard assessment. The ning new construction or relocating existing utility lines: quality
results demonstrate that AI-based workflows can be adapted to levels D, C, B, and A, in ascending order of importance and reli-
various geophysical settings, providing new opportunities for ability in making business decisions (e.g., Kraus et al., 2012).
high-resolution imaging of complex subsurface features in onshore Quality levels assigned to buried utilities are defined by the
and offshore environments. American Society of Civil Engineers (ASCE) in its Standard
Guideline 38-22 (ASCE, 2022). Quality levels are assigned to
Introduction utility lines on CAD drawings submitted to design engineers
Well casings, subsurface metallic pipes, underground storage based on the amount and type of investigation used to find and
tanks, etc., belong to a class of geophysical targets called subgrid map the lines. They give the engineer an idea of the risk associated
entities (e.g., Weiss, 2017). This is because the significant contribu- with the correct location of the lines, based on their quality level.
tion to the observed signal comes from a 3D material property Quality levels progress from low to high confidence, D to A,
distribution where one dimension (wall thickness) is orders of based on these general criteria: Level D is determined by examining
magnitude smaller than the others. This makes creating as-is records from previous work by utility owners, “Call Before You
geometries for modeling the response of such targets extremely Dig” centers, other investigators, or any other publicly available
difficult. Also, for many geophysical applications, the presence data sources, with no field investigation. Level C is determined
of such bodies is often deemed a hindrance to understanding the by surveying visible surface features associated with buried utilities
response of a geophysical investigation’s main target: a subsurface such as valves or terminal boxes. Level B is determined most often
geologic entity. As such, much effort has been concentrated on by using pipe and cable locators that put a signal on the utility,
modeling the response of such targets and isolating it from the either electromagnetically or directly, and then tracing the path

Manuscript received 22 October 2024; revision received 6 January 2025; accepted 16 January 2025.
1
EmPact-AI, Houston, Texas. E-mail: [email protected].
2
Kraken Robotics, St. John’s, Newfoundland and Labrador, Canada. E-mail: [email protected].
3
Texas A&M Transportation Institute, College Station, Texas, USA. E-mail: [email protected]; [email protected].
4
S2 Labs, Portland, Oregon, USA. E-mail: [email protected].
5
Couvillion Group, New Orleans, Louisiana, USA. E-mail: [email protected]; [email protected].
6
Drone Geosciences, Houston, Texas, USA. E-mail: [email protected]; [email protected].

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 187
 of the utility with the locator’s receiver on the ground surface. et al., 2023; Guigné and Laidley, 2024). To address this, a boat-
Level A is determined by hand or vacuum excavation to physically towed high-resolution magnetometer array, located approximately
locate the utility. Level A can only be ascribed to locations where 3.5 m above the present-day seafloor, was used to acquire data.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

excavations have been made. Improving the results of quality level Given the relative insensitivity of the recorded magnetic field to
B designations by improved geophysical methods is the goal of the near-surface hydrocarbon cloud and the high sensitivity to
the work discussed here. iron-rich material like the conductors, this was deemed the most
Quality level B information can be derived from the interpreta- promising approach to complete the delineation of the conductor
tion of geophysical data using present-day methods described geometry as constrained in the acoustic imaging beyond row 11.
earlier. There is an order of magnitude cost difference when The acquired data were processed and then inferred for
acquiring level A data over level B. Complexity in geometry, subsurface structures using a trained deep machine learning
material property distribution, and other factors can distort results (ML) model. The resulting image is compared for location,
from direct interpretation of geophysical data, leading to costly depth, and geometry where there is overlap with the image from
acquisition of more level A data involving excavation of buried the acoustic coring.
utilities. With improvements in the accuracy and resolution of The rest of this paper is organized as follows. We will introduce
level B geophysical data, dramatic reduction in costs for such the general artificial intelligence (AI) workflow, emphasizing
activities can be envisaged. how it is adopted for geophysical inversion. Next, we will describe
In the current work, we describe the application of deep- how the workflow is deployed to Texas A&M University’s Rellis
learning-based 3D inversion of geophysical data to resolve sub- Campus site. This will include an overview of how data were
surface cultural targets, mainly pipes. We present two separate acquired and processed and a parallel process of preparing models
case studies: one from onshore Texas, where drone-based magnetic and generating data for AI system training. Next, we offer a
data were acquired over Texas A&M University’s Rellis Campus summary of the AI architecture, the choice for current applications,
site, and the other from the Gulf of Mexico (GOM), using a and a description of the training performance. A description of
boat-towed magnetometer located 3.5 m above the seafloor. inversion results for synthetic and field data will follow this.
DOI:10.1190/leedff.2025.44.issue-3

The first survey aimed to flag subsurface metal artifacts, mainly Finally, the inversion results from field data will be compared
utilities, prior to excavation and construction at the Texas A&M against excavation results.
Rellis Campus site. The inversion results from this site have been After this, we will discuss the study of the GOM platform,
compared against excavations carried out in two locations. This highlighting the different steps in planning the magnetometer
provides a measure of ground truth calibration to 3D inversions survey, data acquisition, processing, and preparation for inversion.
developed using deep learning. It enables the assessment of suit- We will also discuss the benefits of continuing to incorporate
ability for deploying such methods for applications in utility and a priori knowledge at every process step provided for iterative
construction projects. verifications in managing the data flow.
The GOM survey was motivated by a need to locate missing We will end with key conclusions relevant to each case study
sections of conductors that were disturbed during Hurricane Ivan and try to illustrate the broader implications of the results for
in 2004 and are currently buried under 35 to 45 m of sediment deploying this technology on a more scalable basis.
(Hoffmann et al., 2023). Kraken Robotics had undertaken an
innovative acoustic imaging seabed investigation at the site in AI imaging workflow
2022 (Laidley et al., 2023; Guigné and Laidley, 2024; Hoffmann Data form the cornerstone of any AI-based process.
et al., 2024). The company used its proprietary acoustic imaging Geophysical data are usually acquired with an end objective,
method called acoustic coring to image the location of these specifically to illuminate the critical characteristics of a subsurface
conductors lying approximately 40 m below the seafloor. (For target. Knowledge of the maximum expected depth of the target,
details on the acoustic coring approach, refer to Guigné and the target geometry, and median material property contrasts
Blondel, 2017.) The resolution and location of the conductors to against the background are usually the key considerations behind
the overturned platform were successfully delineated, except for the design of a geophysical data acquisition and processing work-
a section where gas in the sediment masked the features. The need flow. Thus, the central tenets of target geometry, median material
to further delineate the conductor extent toward the original well property contrast, and the expected depth range inform the two
bay location required a different approach, which led to the critical components of the AI imaging workflow (Figure 1):
proposed survey application of highly sensitive magnetometers,
with the intent of using deep learning protocols to extract, within 1) Data preprocessing to enhance signal from the target.
the magnetometer response, unprecedented details of the legacy 2) The generation of subsurface material property distributions
debris field, pipes, and conductors. The objective was to define if and the forward simulation of their response to build output–
a potential pathway remained for hydrocarbons to flow from the input pairs for training the AI model for geophysical
wells to the displaced platform location where oil and gas accu- inversions.
mulations were noted to form.
An extensive hydrocarbon cloud in the sediments prevented Subsequent sections of this paper discuss the different steps
acoustic coring and other seismic methods from imaging the of the AI imaging workflow, first in the context of the onshore
conductors past row 11 of the acoustic coring survey (Laidley drone magnetometer data set over Texas A&M University’s Rellis

188 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 1. Workflow for AI-based inversion of geophysical field data. The workflow has been customized for handling relatively large data sets by identifying a representative subset domain
within the total inversion volume. The AI system is trained in this domain, and the results are inferred on segments of field data that are deemed sensitive to subsurface structures within
that segment. The system is designed to highlight strong contrasts in material properties rather than their absolute values. As such, the inversion results are typically displayed as a
threshold volume of susceptibilities relative to the peak susceptibility value within an individual inverted domain.
DOI:10.1190/leedff.2025.44.issue-3

Campus and second in the context of

the offshore GOM towed magnetom-
eter data set.

Application to Rellis data set

Data processing and training model
preparation. A drone-based magne-
tometer (Figure 2a) was used to
acquire data over the construction site
Rellis #2 at Texas A&M University’s
Rellis Campus site (Figure 2b). The
survey aimed to identify potential
utilities and subsurface metallic bodies
that may impede construction.
Acquired total magnetic intensity
(TMI) data were processed to remove
the International Geomagnetic
Reference Field (IGRF) and regional
trends to obtain residual magnetic
intensity (RMI) data (Figure 2c). The
identif ied imaging domain in
Figure 2d is defined by the red dashed
line in Figure 2c, showing the extent
of contributing data. The total inver-
sion depth is determined from a priori
expectations of how deep man-made
metallic targets could be present in
the region (approximately 4 m below
ground). The AI training domain is
thus chosen to be a 20 × 20 × 4 m Figure 2. (a) Drone-based magnetometer used to acquire data. (b) The raw magnetic field data collected in a 5 × 5 m line spacing
using a drone with an average flight height of 2 m above ground level. (c) Data are processed for removal of deeper trends and
subset of the total imaging domain. highlighting of the signal, mainly from shallow targets such as buried pipes. (d) The total imaging domain corresponds to the red
The choice, among other consider- dashed line shown in panel (c). The trained AI system infers data within a small 20 × 20 × 4 m training domain. Once trained, the
ations, is guided by the potential AI network infers data across the larger imaging domain in increments of 20 × 20 m with a predefined (in this case 50%) overlap.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 189
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 3. Cultural subsurface targets of various shapes, size, and intensity are used for training the AI system. (a) USTs are modeled as equivalent solid blocks with variable magnetic
susceptibility. (b) Horizontal pipes oriented along local y-axis of imaging domain, similarly, modeled as solid blocks of equivalent susceptibility. (c) Horizontal pipes aligned along the x -axis
with similarly modeled property distribution. (d) Dipping pipes modeled in the same manner. (e) Combination of storage tanks, pipes, and other targets. (f) All these targets are embedded
in a background whose susceptibility is also variable.
DOI:10.1190/leedff.2025.44.issue-3

Table 1. Relative abundance of various model types used for generating synthetic data for the Rellis 2 data set. susceptibility (Figures 3b–3d). The train-
ing models data set also includes complex
Minimum percentage Maximum percentage targets involving a combination of stor-
within the total number within the total number age tanks, pipes, and other elements
Model class of samples of samples
(Figure 3e). In addition to target com-
Single rectangular block 10 20 plexity and material property, it is crucial
Single horizontal pipe oriented in y direction 10 20 to incorporate backgrounds of varying
Single horizontal pipe oriented in x direction 10 20 susceptibility, as the response of geo-
physical targets is sensitive to both the
Single pipe with arbitrary dip and azimuth 10 20
target material property and its contrast
Two rectangular blocks 5 15 relative to the background (Figure 3f).
A rectangular block and a horizontal pipe oriented in y direction 5 15 The material property distributions
A rectangular block and a horizontal pipe oriented in x direction 5 15 of the target and background are chosen
based on available geologic knowledge
A rectangular block and a pipe with arbitrary dip and azimuth 5 15
regarding local soil type, general sus-
Two pipes with arbitrary dips and azimuths 5 15 ceptibility variation in soil (e.g.,
A rectangular block, horizontal pipes in x and y directions, and a Dekkers, 1978), and the general sus-
5 15
pipe with arbitrary dip and azimuth ceptibility variation in iron and steel
objects (e.g., Barrows and Rocchio,
lateral extent of the magnetic field response from targets buried 1990) such as pipes and USTs as discussed in literature. Accordingly,
no more than 4 m below the surface. a susceptibility range between 10 –3 and 10 –2 (SI units) for back-
A diverse training model set is prepared (Figure 3), including ground sediments, with a contrast of 0.5 × 101 to 108 for buried
common cultural subsurface targets of various shapes, sizes, and steel and iron targets is used in the training models. Individual
intensities. The thickness of cultural targets, such as underground background model susceptibilities are randomly perturbed within
storage tanks (USTs), drums, and pipes, is typically subgrid. The the modeling volume.
modeling of the subgrid dimension is handled by upscaling on The various classes of models are generated randomly with
the invariant product relationship between magnetic susceptibility counters set to prevent any single class from being over- or under-
( χ) and wall thickness (t) so that hollow structures, such as USTs, represented in the final data set. The classes of models used, and
pipes, and drums, are modeled as solid entities (Um et al., 2015). their maximum and minimum distributions within the total
USTs are modeled as equivalent solid rectangular bricks with training data set, are presented in Table 1. The model classes are
variable magnetic susceptibility (Figure 3a). Horizontal pipes named after the target embedded at variable locations in the
oriented along the local x- or y-axis and dipping pipes of arbitrary subsurface x-, y-, z- space within a heterogeneous background.
orientation are similarly modeled as solid objects with equivalent The background susceptibilities (schematically shown in Figure 3f)

190 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 vary randomly within the modeling volume around a randomly
chosen median susceptibility value between 10 –3 and 10 –2 (SI
units) with arbitrary magnetization (can be positive or negative
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

relative to the incident magnetic field) and a randomly sampled

standard deviation, not exceeding 15% of the median value.
Choice of ML architecture to train system. ML architectures
are defined by their ability to mathematically interrogate features
in texts, images, and speech and then recognize similar features
in other entities once “trained” to do so via “learning” over various
feature distributions. It has become increasingly common to
deploy such architectures for inversion of geophysical data (e.g.,
Kim and Nakata, 2018; Puzyrev, 2019; Colombo et al., 2020;
Botelho et al., 2022; Zhang et al., 2022; Dhara and Sen, 2023).
Much of the effort has been focused on using variants of convo-
lutional neural networks (CNNs) to train a deep learning AI
model that is capable of inferring a 3D subsurface material
property distribution from an input geophysical data set usually
acquired on a 2D surface.
The inherent challenge of the “curse of dimensionality” Figure 4. Schematic representation of the U-Net architecture used for inverting geophysical
involved in resolving 3D targets from a data set acquired on a data. The geophysical data are transformed to have representative values in the modeling
2D surface (e.g., Mukherjee et al., 2021) haunts ML-based cells whose material properties are used for simulating the geophysical data. The U-Net is
approaches just like conventional approaches based on con- used to map the values of the data in these cells to the material property values present in
them. The architecture height denotes the number of learning layers used by the network.
strained least-squares optimization. But, generating a large The architecture width provides the number of starting filters being used in the topmost
DOI:10.1190/leedff.2025.44.issue-3

number of training data samples corresponding to a diverse set layer. The encoder (downsampling/contracting) path progressively reduces spatial
of models can provide a step-change improvement in the resolu- resolution while increasing the number of feature channels, enabling the network to learn
tion and accuracy of model recovery compared to least-squares abstract representations. Meanwhile, the decoder (upsampling/expanding) path recovers
inversion (e.g., Colombo et al., 2020; Mukherjee et al., 2022). spatial information through upsampling operations, while skip connections preserve fine
details by providing direct pathways for low-level features. See text for more details.
However, the challenges of time, computational complexity,
and cost remain major bottlenecks in deploying such architectures
for 3D inversions of large-scale commercial data (Mukherjee to handle input images of varying sizes, provided that the spatial
et al., 2022). dimensions allow for consistent tensor shapes across all downs-
With key changes to the form of the simulated input data ampling and upsampling operations. This architecture has dem-
used for training (Mukherjee et al., 2021), the geophysical inverse onstrated remarkable success across various domains, from medical
problem is reformulated as an image-to-image translation task in image segmentation to remote sensing, making it particularly
which we aim to learn a mapping from the transformed observed well-suited for our geophysical inverse problem.
survey responses to the underlying material properties that char- It is also well known that magnetic data are more sensitive to
acterize subsurface structures. the relative contrasts in magnetic susceptibility distribution of
CNNs have emerged as the cornerstone architecture (Li subsurface targets instead of their absolute susceptibility (e.g.,
et al., 2022) for image-related tasks, owing to their ability to Blakely, 1995; Maus and Haak, 2003). As such, relative suscep-
learn hierarchical features and exploit spatial correlations inherent tibility contrasts between subsurface objects are generally more
in image data. Fully convolutional neural networks (FCNs) have attractive practical targets for geophysical inversion than absolute
proven especially effective for dense prediction tasks where spatial susceptibility values. Cultural targets such as pipes and other
information must be preserved throughout the network. Among man-made metallic objects usually have a markedly different
FCN architectures, U-Net (Ronneberger et al., 2015) has become direction of magnetization relative to the ambient geomagnetic
the de facto choice for image-to-image translation problems due field and the host geology. Often, the orientation of these directions
to its elegant architecture that combines multiscale feature extrac- can be highly varied, but the effective susceptibility contrast in
tion with precise localization (Krizhevsky et al., 2012). the direction of the incident field is much stronger than the
U-Net’s distinctive encoder-decoder structure with skip con- background geology. Such “remanent” magnetization can lead to
nections (Figure 4) allows it to capture both fine-grained details erroneous estimates of absolute susceptibility values, but the strong
and global context (Le Cun et al., 2015). The encoder path pro- relative contrasts can still be useful in identifying the presence or
gressively reduces spatial resolution while increasing the number continuity of large metallic targets such as pipes and fallen oil
of feature channels, enabling the network to learn abstract rep- field conductors. As such, it is not uncommon to let inversions
resentations. Meanwhile, the decoder path recovers spatial infor- recover the effective susceptibility of targets along one component
mation through upsampling operations, while skip connections of magnetization that is either aligned with the incident magnetic
preserve fine details by providing direct pathways for low-level field or in the opposite direction (e.g., Sun et al., 2020), resulting
features. A key advantage of the U-Net architecture is its flexibility in the recovery of a negative susceptibility value.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 191
 Accordingly, the susceptibility value in the training model a width of 64 starting filters, applied to a general training
set is scaled from –1 to +1 such that the algorithm learns to predict model mesh containing 50,000 to 250,000 cells, in order to
the relative susceptibility variation in the subsurface due to changes recover “simple shaped models” (rectangular blocks, single
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

in similarly scaled residual magnetic field and remanent magnetiza- horizontal pipe, etc.).
tion can be taken into account. (For more details, see Mukherjee • For recovering more complex models, or for improving the
et al., 2022.) This results in a simple scalar relative susceptibility resolution and accuracy of recovered models, the total number
parameterization in settings with complex magnetization but of data–model pairs needed for training should increase, and
high relative susceptibility contrasts. then, eventually, the training architecture height and/or width
In addition to the efficient image-to-image translation of should also be increased.
single valued magnetic data pixels being mapped to single valued • Additional models can be introduced based on how the training
magnetic susceptibility values, the ML architecture can leverage results generalize (whether the relative loss values between
the rapid fall-off in sensitivity away from the target to essentially training and validation models are diverging, if some subset
design a modular training unit that can be used to cost effectively of model class is performing much better, or much worse than
infer the subsurface material property distribution (in this case, expectations, etc).
effective susceptibility) within a larger volume of interest, using
the “moving footprint”-based approach for mobile data acquisition The tetrahedral-cells-based modeling mesh used had 76,800
systems (Cox et al., 2012). cells and uses the accurate analytic solutions from Okabe (1979),
Before discussing results, it will be helpful to introduce some implemented by Lelièvre and Jahandari (Lelièvre et al., 2012;
commonly used terms to discuss neural network behavior and Jahandari and Farquharson, 2013) for modeling the response of
performance: subsurface realizations. The system available for training (four
32 GB RAM GPU nodes on Iowa State University’s Bridges
• Architecture height: The number of “learning layers” used by Computing Network) could support a maximum of five layers
the training architecture. It is also sometimes called the depth and 128 starting filters for an input–output training data–training
DOI:10.1190/leedff.2025.44.issue-3

of the U-Net, defined by the number of downsampling/ model pair of this individual size (76,800 cells). The total number
upsampling block pairs. Each level processes features at a of samples used was 190,000 (following the model and data
different spatial resolution. generation scheme outlined in Table 1) on this training architec-
• Architecture width: The number of filters in the initial con- ture. The system was trained for approximately 100 hours and a
volutional layer, which doubles with each downsampling step. total of 289 epochs. The final loss on trained models was 0.0016,
This determines the network’s feature learning capacity at and the corresponding loss on the holdout (unseen by the trained
each scale. model) or validation set was 0.0018 root-mean-squared error.
• Loss: The difference between predicted model values and true Discussion of imaging results: Synthetic data. To test if the
model values used in training (and validation), analogous to ML architecture is fully trained, some synthetic models
“misfit” in geophysical terminology. Mean squared error is (Figures 5a, 5c, and 5e) that were not part of the training were
used as the metric here. generated and given to the trained architecture to infer. The
• Model weights: The learnable coefficients within the network inferred model results are provided in Figures 5b, 5d, and 5f.
that are adjusted during training to capture the mapping “Conventional” geophysical targets where none of the three
between input and output data. dimensions are subgrid, such as the upscaled USTs, are recovered
• Batch size: The smallest number of samples used to update within ±2% of the actual volume. The top of the structure is
model weights during training. If there are 1000 samples for located at less than 1% of true model depth in 147 epochs of
example, a batch size of 5 will mean 5 of these samples are training for a given ML architecture.
used at a time for updating model weights. The model weights Pipe-like structures, on the other hand, where two of the
for multiple batches can be updated in parallel. The model dimensions, even after upscaling, are considerably small (0.5–1 m
weights for the entire data set will be updated over 200 such in a tetrahedral mesh where average cell size is 0.5–1 m) in two
batch sizes in a single iteration. of the three dimensions, as seen in Figures 5c and 5e, require
• Epoch: An update of the model weights using a batch size of many more epochs (289 in this instance) to reach similar levels
samples at a time, for all the data samples present in the of model recovery (Figures 5d and 5f).
training data set, and the corresponding prediction of loss Discussion of imaging results: Field data. For imaging field
values for both the training and the validation data set at the data, the entire imaging domain (Figure 2d) is subdivided into
end of the update. overlapping (by 50%) inference volumes that are of the same
dimension (20 × 20 × 4 m) as the training domain volume (the
Based on prior experience (Mukherjee et al., 2021, 2022), dashed outline of the box in Figure 2c and the overlapping
discussions, and personal communications with other authors inference volumes in Figure 2d). The inversion results on indi-
(Puzyrev, 2019), general “rules of thumb” suggest: vidual imaging subdomains were obtained via inference on the
appropriate 20 × 20 m data subset. The inference takes less than
• The total number of models needs to exceed a minimum 1 s on a 16-core 128 CPU unit and can be performed simultane-
threshold of 100,000 for a network height of five layers and ously. Thus, once trained, an arbitrary number of such imaging

192 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 domains can be imaged in less than
1 s, implying large data sets can be
inverted in a reasonably short time.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

The high positive (greater than 0.05)

and low negative (less than –0.05)
threshold relative susceptibility values
for the entire inversion domain are
displayed in Figure 6a. For the signifi-
cant short- to medium-wavelength
anomalous volume imaged (highlighted
in orange ovals) in Figure 6a, there is a
corresponding expression of high posi-
tive and/or low negative anomaly in the
residual magnetic field data (Figure 6b),
which are highlighted by the same
orange ovals. Moreover, the strong
“low–high” alternating anomalies on
the left side of the map in Figure 6b and
subparallel to the y–axis are now appar-
ent as a pipe in the threshold volume in
Figure 6a. Two excavation locations,
marked in violet ovals in Figure 6c,
show characteristically high values of
DOI:10.1190/leedff.2025.44.issue-3

magnetic anomaly in Figure 6b and a

corresponding high relative susceptibil-
ity value distribution in Figure 6a, both
denoted by similar violet-colored ovals
in the respective locations in Figures 6a
and 6b. More details on the comparison
of the AI-based inversion with the
excavation results are provided in the
Figure 5. (a) Two storage tanks, heterogeneously modeled at varying depths. (b) AI imaging results using five layers, 128 starting
next section. filters, network with 150,000 samples, and 147 training epochs. (c) A buried pipe uniformly magnetized. (d) Recovered image
Comparison against excavation using five layers, 128 starting filters, 190,000 samples, and 289 training epochs. (e) A buried pipe heterogeneously magnetized.
results. The susceptibility contrasts (f) Recovered image using five layers, 128 starting filters, 190,000 samples, and 289 training epochs.

Figure 6. (a) Inferred relative susceptibility threshold

images on 49 different imaging volumes with 50%
overlap between the imaging domains. (b) Recovered
image anomalies show good correspondence between
the anomalous magnetic data (marked with orange
circles) with the orange circles around the anomalous
susceptibility volume in panel (a). (c) Digital elevation
model (DEM) of the region. The surveyed area is
highlighted by the dotted red rectangle shown in
panel (b). The coordinate system is transformed
to local coordinates following the one displayed in
panel (b). The overall relief in the surveyed region is
less than 1 m and can be considered flat for purposes
of the AI-based inversion. (d) Texas A&M University’s
Rellis Campus site. There are two excavation sites
circled in violet and the corresponding imaged volume,
the anomalous data, and the DEM in panels (a), (b), and
(c) are circled in the same color. See figure legend for
additional clarification.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 193
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 7. (a) Extracted positive and negative relative susceptibility volume thresholds from the inversion volume. Displayed at 10% opacity. The extracted volumes specifically for the two
excavation sites are overlain and displayed at 50% opacity for better visualization. The location of an x -z cross section at y = 5 m is also shown. The location of the excavation lines 1–2,
and 3–6 are marked. (b) The x -z cross section shown in panel (a) using wireframe visualization. The relative susceptibility values are color coded by the color of the mesh. The extent of the
excavated line as seen in the photograph in panel (d) is demarcated by the dotted red line. Most of the pixels within this zone are “red,” implying a much higher contrast relative to the other
values in the section. The extracted background volume thresholds displayed at 10% opacity in the background confirm the same. (c) Photograph of the excavated structure marked by 1–2
in panel (a). (d) Photograph of the excavated 8-inch line marked by 3–6 in panel (a). (e) Extracted threshold relative susceptibility volumes at both excavation sites. Comparing with the
photographed structures in panels (c) and (d), the difference in material property at the two locations as well as the difference in complexity of the structures in site 1–2 relative to 3–6
are apparent. The ability of the recovered inversion image to reflect this difference is remarkable given the underlying data were acquired at 5 m line separation spacings.
DOI:10.1190/leedff.2025.44.issue-3

along the excavated lines (Figure 6c) have been extracted and and 6. The excavated line in Figure 7d also shows zones of non-
overlain on the overall relative susceptibility variation within the uniformity in the middle, but not enough information is available
entire inverted volume (Figure 7a). Contrasting anomalous struc- at this time to calibrate and correlate such changes meaningfully
tures are observed at excavation locations 1–2 versus 3–6 (Figures 7c beyond first-order visual observations. Moreover, this inversion/
and 7d). At the site location marked by pins 3 and 6 in Figure 7a, inference was done on data acquired at an approximate 5 × 5 m
a single line 8 inches in diameter was excavated between 36 and line spacing, so very detailed analysis of inversion performance
42 inches below ground (Figure 7d). The relatively high susceptibil- relative to the excavated 8-inch line is not merited.
ity threshold volume (≥0.01) extracted in the neighborhood of The imaging cell size is approximately 0.5 m, or approximately
this location (pins 3–6) shows relatively high susceptibility values 20 inches. The rendered top of structure in the section is approxi-
along this trend (Figure 7a) until the extremity at pin 6. On the mately 0.5 m below ground. The “smoothing” applied to inversion
other extremity at pin 3, there is a bend noted in the high relative results usually means “high confidence” location of the top of
susceptibility trend (Figure 7a) toward the y-axis that seems to structure is 1 or 2 pixels below the resolved top of structure. This
correspond to a bend in the excavated line in Figure 7d. Not implies a mean uncertainty in the location of the top of the
enough information is available regarding the continuity of the structure to be somewhere between 0.75 and 1.25 m, or 29 to
line beyond this bend, so further substantiation of the bend in 49 inches. Given the actual excavated location of 36 to 42 inches,
the susceptibility trend and the corresponding bend in the line is this translates into a mean error of ±7 inches, or approximately
currently not possible. 17 cm for this instance.
A vertical x–z cross section at y = 5 m, which is subparallel The excavated section 1–2 in Figure 7c shows a more complex
to the overall high susceptibility trend, is also extracted from this metallic structure. Extremity point 1 in the section may mask
volume and displayed in Figure 7b using wireframe visualization. potential continuity of the structure under the overburden. The
The relative susceptibility values are color coded by the color of extracted relative susceptibility volume corresponding to this
the wireframe mesh in the displayed section. The relative suscep- section is shown in Figure 7b. It seems to be part of an extended
tibility values of the extracted volume are displayed in the back- moderately high susceptibility volume trend that diminishes in
ground at diminished opacity. The extent of the excavated line as value shortly beyond the extremity at location 1, where either
seen in Figure 7d is demarcated by the dotted red line. Most of the extension of this excavated structure or a separate similar
the pixels within this zone are “red,” implying a much higher structure may be present. The other extremity of the structure
contrast relative to the other values in the section. The extracted at location 2 seems to correspond closely with the near end of
background volume thresholds displayed in the background this susceptibility trend as extracted in the susceptibility volume
confirm the same. The sharp termination of the high red anomaly in Figures 7b and 7e.
corresponds reasonably well with the extremities pinned by points Figure 7e displays the extracted volume of high (greater than
3 and 6. The top of the red anomaly in this cross section (Figure 7b) 0.01) relative susceptibility values in the vicinity of the excavated
is a relatively flat line with some discontinuities in relative sus- lines only. Comparing the magnitude of the extracted values at
ceptibility values in the middle between the extremity points 3 the two locations with the pictures of the excavated material, it

194 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 seems clear that the overall iron content by volume in the structures distribution of weak particle/particle unconsolidated bounded
at these locations is very different. At excavation site 3–6, the very fine sediments dispersed in the upper seabed region. A
mostly “red hot” high relative susceptibility values correspond more densely consolidated geologic dominant basement (likely
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

well with the excavated extents of the 8-inch line, while at locations indicative of the paleo bathymetry during the hurricane) was
1–2, the excavated structures are complex, containing both vertical revealed at approximately 136–165 ft (41–50 m) below the current
chimney-like and horizontal pipe-like segments. The structures mudline (Figure 9d). Smooth, continuous linear conductors
are thinner and show more wear and tear compared to the observa- reposing onto this basement geology indicated that the pipes
tions at location 3–6. The recovered image in Figure 7e at this were not fractured (apart from three through-the-wall conductor
location 1–2 represents the vertical and horizontal components damage/leak locations).
with lower relative susceptibility overall than at site 3–6. With The endings of the row 11 pipes (Figure 9b) show an abrupt
5 m line spacing data, this is a remarkable achievement, difficult spread-out termination (Figure 9c) with an undulating humped,
to achieve with current non-AI imaging methods. With more slightly down-looking character. Several potential causes were
training and higher-resolution data, there is room to further identified:
improve image fidelity and resolution.
• The lines may have been kinked and going downward at a
Application to Gulf of Mexico data set sharp angle, so acoustics could not image the conductors.
Hurricane Ivan event. In September 2004, Hurricane Ivan • The conductors may have severed and no longer exist at row 11.
precipitated a low-probability, high-consequence event, resulting Instead, the remnant 10-inch liners or smaller bore tubing,
in a producing oil and gas platform toppled in approximately which feeds oil and gas into the conductors, is present but
140 m of water and dragged off station approximately 500 ft cannot be seen due to their smaller diameter.
(150 m) (Figures 8a–8c). Twenty-eight oil and gas well conductors • In the region around row 11, gas-laden soil inhibited the
were buried beneath the mudline under sediment and have been acoustic ability to get a return and collect data.
discharging oil and gas into the GOM since the incident occurred.
DOI:10.1190/leedff.2025.44.issue-3

In November 2018, the worksite was partially federalized, and Magnetic survey feasibility. The lack of knowledge about the
oil spill response proposals were solicited to design, build, install conductors from row 11 to row 21, where the producing well bay
and operate a system to capture the ongoing release of oil at the is located, precipitated the need to explore alternate methods
site (Hoffmann et al., 2023). beyond acoustic imaging to determine conductor extent and/or
Acoustic Corer survey. Plugging and abandoning these wells current status, i.e., still intact and continuous, or severed, detached,
using any top kill solution would require knowing the burial depth and scattered elsewhere. The significant amounts of iron/steel in
of conductors beneath the seabed. Standard geotechnical survey the conductors and the well-known sensitivity of magnetic sensors
methodologies like ground-penetrating
radar and physical probing could not
determine the depth of conductor burial.
In May/June 2022, Kraken Robotics
conducted a subbottom/below-mudline
survey of the GOM site using its pro-
prietary Acoustic Corer imager
(Figures 9a and 9b), (Laidley et al.,
2023; Guigné and Laidley, 2024).
Sixty-three Acoustic Corer surveys
were conducted for Couvillion Group
(Hoffmann et al., 2024) to determine
the extent, expanse, orientation, and
characteristics of the conductors and
other components from the platform’s
former well bay location to the ero-
sional pit adjacent to the downed
jacket. After processing and interpret-
ing the acquired data, the entire
expanse of the oil well conductors was
detected and imaged with exact loca-
tions of conductors delineated within
the sub-seabed (Figures 9c and 9d).
Figure 8. (a) Snapshot of the erect GOM platform just before Hurricane Ivan. (b) Platform being toppled by hurricane. (c)
The results exhibited the presence
Platform reaching its present-day resting place. (d) Schematic representation of the subsurface situation after the hurricane.
of chaotic mudslide clay blocks inter- A significant portion of the structure is buried below the present bathymetry. The displaced conductors are thought to connect
spersed within a depositional with the well heads along a relatively straight path. Geophysical imaging is required to verify this hypothesis.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 195
 to the presence of ferruginous targets spurred the need to explore detection and imagery, there are 25 known conductors ranging
the feasibility of a very dense grid magnetic survey to delineate in diameter from 3 to 30 inches or approximately 7.5 to 75 cm
conductor extent beyond row 11. to consider in the modeling. The conductors are tightly bundled
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

While attractive on paper, the inherent challenges of detecting together, with spacing varying from a few centimeters to a meter.
subgrid geophysical targets located 40 m below the mudline are Thus, it is possible to consider these conductors as a single
daunting. To assess the possibility and design an appropriate magnetic entity with a single effective susceptibility depicting
survey, a careful feasibility study was undertaken first. The results their aggregate presence within a geophysical forward-modeling
of this feasibility and survey design forward-modeling study are cell (Figures 10a–10c).
encapsulated in Figures 10–12. The known image extents and geometry of the conductor from
The most common method for modeling such subgrid targets the corer survey (Figure 11b) are used to define and discretize a
is to build an upscaled, effective susceptibility model. The conductor susceptibility model (Figure 11a) for testing the feasibil-
well-known principle of equivalence shows that, within limits, ity of the magnetic survey. The accurate magnetization directions
the response of a pipe-like target with a given material property of the conductor package can be arbitrary. However, the effective
and thickness will be equivalent to another target whose product magnetization component along the direction of the incident
of material property and thickness are the same as the first geomagnetic field will influence the observed magnetic data most.
target (e.g., Hoversten and Schwarzbach, 2018). Additionally, As such, the upscaling relationships previously discussed are used
the total volume of susceptible and nonsusceptible material to define an effective susceptibility model where the direction of
within a single model grid cell can be averaged to a single magnetization is along the incident geomagnetic field.
effective susceptibility. The orientation of the conductor package geometry relative
The known ground condition of the conductors lends itself to the incident geomagnetic field also strongly impacts the
well to this modeling approach. Based on a priori knowledge, distribution of observed magnetic anomalies from the conduc-
the precision of the Acoustic Corer, and existing conductor tors. As such, the feasibility study considered two end cases
DOI:10.1190/leedff.2025.44.issue-3

Figure 9. (a) Location of the site damaged by Hurricane Ivan (2024). (b) The region where Acoustic Corer data were acquired. There are 17 rows of acoustic coring data along columns A, B,
and C. The conductors imaged with acoustic data terminate around row 11. The imaging with magnetic data will show that the conductors continue past row 11 and thin out considerably
around row 14 before continuing farther. (c) Image of the conductors from acoustic coring. The conductors are partially imaged with the direct connection to the well head not clear. (d)
Summary image of the status of the conductors imaged with Acoustic Corer, their geologic location within the present day and the paleo sea floor at the time of the incident, the location of
the conductor well bay, and the uncertainty of the image in between.

196 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 for conductor geometry relative to the incident geomagnetic horizontal component (along east–west) of the magnetic field
field: (1) the long axis of the conductor package is aligned (Figure 11d).
along the magnetic field’s most substantial horizontal com- Based on currently available options, the nearest distance a
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

ponent (along north–south) (Figure 11c), and (2) the long axis magnetometer could be placed in proximity to the conductors is
of the conductor package is aligned along the weakest approximately 3.5 m above the current seafloor. This can be

Figure 10. (a) Magnetic susceptibility value ranges for pipes (NCEI, 2024). (b) The hollow cylinder pipes can be modeled as solid pipes and upscaled in diameter by reducing their
susceptibility. Under limited circumstances, the product of pipe thickness and susceptibility is invariant. (c) Two pipes separated by a small distance (approximately 1 m in this example)
can be aggregated and modeled as a single pipe 1 m in diameter.
DOI:10.1190/leedff.2025.44.issue-3

Figure 11. (a) Modeling the expected magnetic response from the conductors imaged from the acoustic core data. The data acquisition system is a magnetometer towed by a boat 3.5 m
above seafloor. (b) The imaged conductors in Acoustic Corer data as reference. (c) Response mapped at 5 × 5 m spacing for the larger component of the horizontally incident magnetic field
aligned in the direction of the conductor orientation. (d) Similar response for the shorter component of the horizontally incident magnetic field aligned in the direction of the conductor
orientation.
Figure 12. (a) Modeling the expected magnetic
response from the conductors 1000:1 contrast. (b)
At 100:1 contrast. (c) 10:1 contrast. (d) The modeling
takes into account the variations due to bathymetry
and resulting differences in distances from sensor to
target location. The recorded relief in bathymetry is
12 m, which can have a significant impact on recorded
response. (e) The susceptibility model of conductor
geometry derived from the acoustic core image used
for simulating the responses in panels (a), (b), and (c).

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 197
 shows a robust and measurable anomaly
stemming from targets with a contrast
of at least 100:1 (Figure 12b) relative to
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

the background. Contrasts lower than

that would become progressively chal-
lenging to discern, and it is doubtful that
contrasts of 10:1 (Figure 12c) would be
detectable for conductor packages at the
given depths below mudline, which has
a 12 m relief (Figure 12d).
Data acquisition and processing. The
magnetic data were acquired using a
remotely operated vehicle-based OFG
Hypermag system — Ocean Floor
Geophysics’ (OFG) proprietary subsea
Figure 13. (a) Acquired magnetic data. The imaged extent of conductors from Acoustic Corer data is shown by the blue dashed
magnetometer array. The high-resolu-
lines. The digitized points of the fallen platform are shown by the black dots. The location of the well bay is shown by an
aggregate of black dots. (b) The fallen platform as imaged from sonar data. (c) The interpretation suggests a much deeper tion area was surveyed with a 3.25 m
buried extent of the platform. line spacing. In addition, the following
specifications were observed:
achieved by towing the magnetometer underwater on a boat
(Figure 11a). The simulated survey response shows a consistently • The main heading was set to 40° from true north.
solid magnetic field in the range of 25–75 nT along the length of • Main grid lines cover the area of interest plus a 100 m buffer
the conductor and beyond and across the width of the conductor on each side.
DOI:10.1190/leedff.2025.44.issue-3

for at least 25–30 m. The distribution pattern of the anomaly • A 20 m safety buffer around the fallen GOM platform was
changes from symmetric (Figure 11c) to strongly asymmetric implemented.
(Figure 11d) with changing orientation of the incident magnetic • Crosslines were added to the line plan to facilitate QC.
field relative to the long axis of the conductor package. However, • Line spacing was optimized based on the distance between
both end member orientations note a conspicuous, reliably detect- sensors in the magnetometer array.
able, and measurable response (above usual instrument and geologic • The total number of grid main lines and spacing of background
noise levels). Moreover, up to a successive data point spacing of main lines was fixed based on the requirements of an adopted
5 m, along an acquisition line and also line spacing, the associated Zamboni survey pattern.
anomaly pattern presents no discernible aliasing effect from a
practical point of view (mean square difference between the OFG took the following steps to produce a final data set for
gridded maps is less than 5%). final processing: The depth was adjusted for the unaccounted
This confirmed two important issues for the go-forward lever arm in the real-time solution and corrected for tides. The
decision: (1) the survey needs to extend beyond 75 m on either altitude data were smoothed to remove spurious data before
side of the long axis of the imaged conductor geometry and a merging with the magnetic data. Lever arms were applied to
50 m buffer at least beyond the last “hypothesized” extent of the the 1 Hz real-time navigation solution data to produce accurate
conductor (connecting up to the well conductor bay); and (2) the positioning for each sensor.
line spacing and along-line acquisition points must be less than The corrected depth, smoothed altitude data, and individual
5 m to prevent major aliasing artifacts. sensor positions were used to compute each magnetometer’s
The other issue is establishing the contrast in material property position (x, y, z, altitude).
between the conductors and the background sediments and ensuring The TMI data were reduced to an RMI by making the fol-
a minimum threshold of contrast above which the conductor bundle lowing adjustments:
would be detectable by a magnetic survey. Based on the works of
Ellwood et al. (2006) and Xinjing et al. (2019), magnetic susceptibil- RMI = TMIcomp – FIGRF–13 – Fdiurnal – Fship, (1)
ity contrasts between GOM sediments and metallic iron pipes
(conductors) are expected to be of the order 1:104 –1:108. This where, F IGRF–13 is the IGRF computed at each sensor location;
assumption is based on the iron content within the conductors. Fdiurnal is diurnal correction; and Fship is correction due to the
If the total volume of conductors in a given section of the grid ship located above the magnetometer.
is small, or integrity and/or corrosion issues cause metal loss, this Figure 13a displays the final prepared data set in the local
contrast can be lower. As such, it is necessary to consider lower coordinate system. Three distinct major anomaly patterns are
contrast values to establish the limits on detectability for the given observed:
depth of conductors (Figures 12a–12c). For the given extent of the
conductor geometry (Figure 12e), and their location below a variable 1) Positive anomaly is primarily associated with the
depth bathymetry (Figure 12d), the observed magnetic response conductors.

198 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 2) Negative anomaly is associated with the toppled platform on large surficial toppled structure complicates the signal from the
the bottom right side. conductors and well heads. For example, the negative anomaly from
3) Positive anomaly is associated with the toppled platform the platform counteracts the positive anomaly from the conductors,
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

farther up on the right side. diminishing its potential presence nearer the toppled platform.

Furthermore, earlier analysis of the

digitized platform in Figure 13c sug-
gests a deeper buried segment of the
toppled platform where it rests at the
time of this writing.
Postacquisition modeling and inter-
pretation. The equivalent source method
(e.g., Dampney, 1969) allows for using
more superficial structures to mimic the
observed response of those contributors
to the observed field, which are not the
primary targets for geophysical inter-
pretation. The concept is extended to
the GOM data set for modeling the
response of the toppled platform. The
digitized points from the toppled plat-
Figure 14. (a) Acquired magnetic data. The imaged extent of conductors from Acoustic Corer data is shown by the blue dashed
form have been aggregated to represent
lines. The digitized points of the fallen and displaced GOM platform are shown by the black dots. The green dashed rectangles
partially submerged rectangular slabs, around these points represent rectangular slabs (see panel [c]) used to model the contribution from the platform to the
DOI:10.1190/leedff.2025.44.issue-3

and the responses were modeled along observed magnetic response. (b) The modeled response from the susceptibility distribution in the acoustic imaged conductors
with the imaged conductor (Figure 14c). (blue dashed lines) and the displaced GOM platform (green dashed rectangles) modeled as rectangular slabs of constant
The current acquisition survey specifica- susceptibility value (see panel [c]). (c) The GOM platform modeled as rectangular slabs of constant susceptibility value can
produce both positive and negative anomalies (see panel [b]) of comparable magnitude and shape as seen in the acquired data
tions generated the response
(panel [a]) near the platform extent. The modeled response from the conductor bundle (colored pink) is positive (as shown
(Figure 14b). Compared to the observed in panel [b]), but clearly not enough conductor has been imaged in the acoustic data to account for the entire extent of the
data in Figure 14a, it is easy to recognize positive anomaly in the observed data (as seen in panel [a]).
that (1) the conductors imaged in the
acoustic data provide a magnetic field
response that shows a positive anomaly
corresponding to the positive anomaly
in the acquired data positioned at the
exact location, and (2) the extent of the
conductors/pipes and the significant
associated debris field stemming from
the platform’s toppling have yet to be
imaged by the acoustics, as the anomaly
in the acquired data extends well beyond
the conductor dimensions imaged in
the Acoustic Corer data.
Scenario testing by extending the
existing conductors and adding detached
conductors on either side of the primary
conductor bundle provides some inter- Figure 15. (a) Acquired magnetic data. The imaged extent of conductors from Acoustic Corer data is shown by the blue
esting insights. The strong positive dashed lines. Conductor package extended in both length and width to model magnetic response. The extended conductor
anomaly in the acquired data can be segments are shown by dotted magenta lines. The digitized points of the fallen platform are shown by the black dots. The green
dashed rectangles around these points represent rectangular slabs used to model the contribution from the platform to the
explained (Figure 15c) by extending the observed magnetic response. The location of the well bay is shown by an aggregate of black dots. (b) The modeled response
conductor bundle 40 m to the northeast from the susceptibility distribution in the acoustic imaged conductor (blue dashed lines), its extended segments in magenta,
of the known location of the remnant the displaced GOM platform (green dashed rectangles), and the well bay (shown by an aggregate of black dots). All these
well bay (Figure 15b). elements are necessary for the first-order approximation of the observed magnetic anomaly. (c) The GOM platform modeled
Moreover, the anomalous response as rectangular slabs of constant susceptibility value can produce both positive and negative anomalies (see panel [b]) of
comparable magnitude and shape as seen in the acquired data (panel [a]) near the platform extent. The modeled response from
from the toppled platform contains nega- the conductor bundle (colored pink) is positive (as shown in panel [b]) and needs to be extended both along length and width
tive and positive anomalies that the large (dashed yellow lines) to account for the full extent of the anomaly observed in panel (a). The well bay (modeled as purple-colored
equivalent slab model can replicate. The vertical slabs) response is muted relative to this conductor bundle.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 199
 Also, the long-wavelength nature of the signal means that geometric shapes like rectangular blocks, uniformly dipping pipes,
some effects from the toppled platform will impede the potential etc. The models capture various combinations of conductors
signal for the wells. approaching the vertical well pad, moving away from the well
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Subtracting the response of the modeled slabs from the pro- pad, being in isolation (Figure 17), and metallic elements that are
cessed RMI data reveals the continuity of the high positive not conductors, such as rectangular slabs. Table 2 summarizes
anomaly toward the toppled platform (Figure 16b). the various model classes and the permitted bounds of their relative
Training models. As before, the entire imaging domain is abundance within the training data set.
subdivided into overlapping (by 50%) inference volumes of the In addition, random sections of the recorded bathymetry
same dimension (60 × 60 × 60 m) as the training domain volume. within the survey domain were selected to place the models
Incorporation of a priori knowledge forms an important aspect under it, and the receiver locations randomly perturbed above
of developing training models and data sets. In this instance, the it given the spread in the above-ground-level data for the sensor
existing image of the conductors from the Acoustic Corer survey, locations.
the original location of the well conductor bay and the total number Data generation was conducted on Amazon Web Services
of slots feeding into it (28 slots, implying 28 vertical cased wells (AWS) clusters. Training was performed on a Shadeform cloud
from the slot locations), and the various scenarios tested by the computing platform. The 60 × 60 × 60 m training domain was
post-data-acquisition forward modeling are among the key infor- discretized into 226,000 roughly uniform tetrahedral cells with
mation used to develop a diverse, randomly selected, yet pertinent a maximum cell size of 2 m. This individual model size allowed
training data and model set, in addition to the utilization of regular for a maximum size deep learning U-Net architecture comprising
six layers and 128 starting filters on the allocated 8 GPU cluster,
where individual units were 80 GB RAM. A feature-engineered
magnetic response and the corresponding model, normalized
between –1 and +1 (Mukherjee et al., 2022), were used as input–
output pairs for training this architecture.
DOI:10.1190/leedff.2025.44.issue-3

In summary:

1) A total of 202,000 3D susceptibility models and their magnetic

response at 2 m intervals (high resolution) were generated.
2) Approximately 100,000 CPU hours were utilized to produce
the models and create the training data. A single forward run
consisted of the response of 1800 data points for a 226,000-cell
susceptibility model.
3) One hundred ten training epochs were completed using eight
A100 GPU clusters, each with 80 GB RAM on a Shadeform
cloud computing platform.
4) Training loss: 0.0018; validation loss (on data not used in
training): 0.0019.

Figure 16. (a) Acquired magnetic data. The imaged extent of conductors from Acoustic
Corer data shown by the blue dashed lines. The digitized points of the fallen platform are
shown by the black dots. The green dashed rectangles around these points represent
rectangular slabs used to model the contribution from the platform to the observed
magnetic response. (b) The residual magnetic field after subtracting out the effects of the Figure 17. Based on known observations and previous experience, a set of models were
fallen and displaced GOM platform. The green dashed rectangles (in panel [a]) around the used for training the AI architecture. The models captured a variety of potential expected
digitized black dots of the platform are not present here, symbolizing the subtraction. The scenarios between the conductors and the well head. This includes situations like
high positive anomaly associated with the conductor package now extends farther toward conductor bundle reaching very close to the well head, far away from the well head, and
the fallen platform. also, response of metallic targets that are nonconductors, i.e., not pipe-shaped metal.

200 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 The trained model used the acquired magnetic data to infer around the primary conductor bundle, consistent with the forward-
the possible configuration and geometry of the conductors, well modeling results discussed previously (Figure 15).
conductor bay, jacket, and other subsurface ferromagnetic com- Figure 18c presents a cross-section view showing the depth
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

ponents in the GOM platform. of the conductor segments to be primarily within 35 to 45 m below
AI inversion results. The magnetic
Table 2. Relative abundance of various model types used for generating synthetic data for the GOM survey.
field data used to infer the possible con-
figuration and geometry of the conduc- Minimum percentage Maximum percentage
tors, well jacket, and other subsurface within the total within the total number
ferromagnetic components in the GOM Model class number of samples of samples
platform are confined to the dashed rect- Single rectangular block 5 15
angular outline shown in Figure 18a. The
Single pipe with arbitrary dip and azimuth 10 20
choice was governed by the overarching
project objective to “see” the extent of the Well conductor bay with randomly chosen wells between 1 and 28 5 15
conductors beyond row 11 (geographic Randomly selected sections of conductors imaged from Acoustic Corer 10 25
location of row 11 depicted in Figures 9b Two rectangular blocks 5 10
and 9c), toward the well pad, where gas
A rectangular block and a pipe with arbitrary dip and azimuth 5 10
in the sediment blocked the acoustic
imagery from mapping the presence of Two pipes with arbitrary dips and azimuths 5 15
the conductors or associated pipes. In Three pipes with arbitrary dips and azimuths 5 15
addition, unlike the “practically flat topog- Randomly chosen one, two, or three pipes with arbitrary dip and
raphy” scenario (Figure 6c) in the Rellis azimuth and well conductor bay with randomly chosen wells between 10 15
data set (recorded relief less than 1 m 1 and 28
within the surveyed domain), the field Randomly selected sections of conductors imaged from Acoustic Corer
DOI:10.1190/leedff.2025.44.issue-3

10 15
data to be inferred needed to be converted and well conductor bay with randomly chosen wells between 1 and 28
into a local coordinate system with special
attention to the local bathymetry
(Figure 12d), as model depth and sensor
heights are calculated from the trans-
formed bathymetry (approximately 12 m
relief in the surveyed region) in the local
coordinate system before inference.
As with the previous case study, the
inferred volume was scaled, and a cut-off
threshold (greater than 0.008) was used
for the high end of the relative suscep-
tibility distribution.
Figure 18b shows this threshold
value distribution in the plan view. We
see good correspondence between the
acoustic image (pink) and the high rela-
tive susceptibility threshold values from
the inferred AI solution, where both
solutions are present. The current image
shows the conductor body extending
beyond row 11, where the image from
the acoustic core terminated. There is
some thinning of the conductor width
around row 14 or 15, which continues
Figure 18. (a) Acquired data vis-à-vis the imaging domain within which the data will be used to infer the relative susceptibility
for about 40 m before taking a sharp volume. Visual interpretation of the raw magnetic data anomaly is complicated due to the extensive metallic debris surrounding
curved jog toward the well conductor the fallen platform and the conductors. AI imaging is needed. (b) Threshold relative susceptibility distribution from AI
bay. As the main conductor body inference after 110 epochs shown in plan view. The imaged volume shows good correspondence with existing conductor image
extends toward the well conductor bay, from Acoustic Corer data and then extends conductor continuity beyond extents imaged by Acoustic Corer. Detailed view of
there is also plenty of metallic debris conductor geometry curving toward the original well bay, potentially retaining connection with it. it is not possible to glean
this insight from visual interpretation of acquired magnetic data. Several isolated dead conductors imaged separate from the
and “dead conductor” material (which main conductor body. (c) Cross section view of the same image. Remarkable confinement of majority of the conductor geometry
are likely disconnected from the well and associated dead conductor and other large metallic structures within or just above the inferred paleo bathymetry surface,
conductor bay) scattered in bands which is a proxy for the sea floor at the time of Hurricane Ivan in 2004.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 201
 the mudline. Much of the primary conductor package is either using Acoustic Corer, indicating that the drill had encountered
just above or just beneath the acoustic basement imaged from the debris and not the conductor-related pipes. Magnetic data are more
Acoustic Corer data, lending credence to this reflector being a sensitive to cumulative scattered metallic debris than acoustics and
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

proxy for the old seafloor at the time of the hurricane. There is can thus pick up and be influenced by the debris in many places.
some unidentified debris sticking up above the paleo bathymetry It is sometimes difficult to separate the contributions of the debris
but under the present-day seafloor. from the conductor package as they respond as a field sum depend-
Overlaying the imaged response with the metal contact points ing on the presence, continuity, and concentration of steel. Hence,
made by Fugro’s drilling program highlights this issue there is a reliance on the alignment with the imaged section of the
(Figures 18b–20). The blue spheres in these figures show the Fugro conductors using the acoustic coring images to study and guide
drilling locations where contact with metallic structure was made. the AI-based enhancements and confirm the results.
The conductors imaged in acoustic data align with most of these A vertical cross section of the relative susceptibility value
penetrations. Still, many of these drilling penetrations seem to distribution through the main conductor body (see Figure 19a for
occur at shallower depths relative to the imaged conductor package location of the slice) and surrounding sediments is displayed in
Figure 19b. The top of the high relative
susceptibility section lies conformably
just below (less than 1 m) the acoustic
image of the conductor. The high sus-
ceptibility zone follows the acoustic
package upward before def lecting
downward moving toward the well bay.
Thus, where both magnetic and acoustic
core images of conductors exist, the
alignment is near perfect (the tops
DOI:10.1190/leedff.2025.44.issue-3

coincide well within 1 m at depth). The

conductor trends are confidently
inferred based on this alignment. To
better understand the trend, location,
and orientation of the main conductor
complex, the inferred relative suscepti-
bility volume was clipped to remove
most of the other dead conductors and
Figure 19. (a) Similar plan view image as Figure 18b. The location of a vertical cross section slice through the main conductor body
other metallic debris. The results are
through y = 235 m is shown. It is useful to see the variation of the imaged target relative to the background. (b) The x -z cross section
slice shows remarkable conformity of the imaged target with the Acoustic Corer image (within 1 m) and clear continuity beyond displayed in Figures 20a and 20b.
imaged extents in Acoustic Corer. The variations in susceptibility in the middle of the target hints at potential damage and possible Figure 20a presents a 90° rotated
partial loss of integrity within some of the conductor bundle. Others continue farther toward the well bay. (relative to Figure 18b) plan view

Figure 20. (a) A 90° rotated (relative to Figure 18b) plan view showing clipped image of the high relative susceptibility distribution around the core central region between the well bay and
the imaged conductor package from acoustic core data. It appears that two major linear structures were lain on the paleo seafloor at the time of the event: a “bifurcation” into a likely “dead
conductor” debris segment and another, “potentially live,” segment that bends toward the wellhead and connects to it. (b) A slightly oblique and significantly zoomed-in view of the clipped
and thresholded relative susceptibility volume. Accentuates the relationship between the main conductor body imaged with Acoustic Corer vis-à-vis with magnetic data. The change in the dip
orientation of the acoustic package along with its location is matched in close details in the inferred magnetic data. Beyond the region overlapping with the acoustic image, the main conductor
package dips gently downward and thins out substantially near row 14 but connects with the well conductor bay (depicted by the yellow star) through the jog in structure near the well bay debris.

202 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 showing this clipped image of the high relative susceptibility magnetometers. The imaged results are consistent with indepen-
distribution around the core central region between the well bay dent verification of these structures via excavation, drilling, and/
and the conductor package imaged from acoustic core data. It or acoustic imaging. The infrastructure targets have varied from
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

seems reasonable to interpret that when the platform toppled and shallow onshore (within 2 m below ground surface) to relatively
transported 500 feet away to its current resting place, it likely deep offshore (approximately 40 m below mudline).
pulled out much of the metal associated with the infrastructure The resolution is consistent with changing depths: individual
including the conductors. It appears that two major linear struc- pipes with high-resolution second-order features (“vertical chim-
tures were lain on the paleo seafloor at the time of the event: (1) a neys” in horizontal pipes) are resolved even in relatively sparse
“bifurcation” into a likely “dead conductor” debris segment and (approximately 5 m) line spacing for shallow onshore targets,
(2) another “potentially live” segment that bends toward the well while a conductor bundle offshore, containing a variable estimated
head and connects to it. 25 individual pipes of varying diameters and lying 40 m below
Figure 20b shows this clipped image with a slightly oblique mudline, is resolved as a linear conductor bundle package with
and significantly zoomed-in view. This accentuates the relationship continuity from the well bay to the oil and gas collection site.
between the main conductor body imaged with Acoustic Corer Accuracy of the results both onshore (within centimeters at
vis-à-vis with magnetic data. The change in the dip orientation 1.5 m below surface) and offshore (within a meter at 40 m below
of the acoustic package along with its location is matched in close mudline) using remotely sensed magnetometer data highlight the
details in the inferred magnetic data. After conforming in the possibility of utilizing geophysical techniques for new applications
region overlapping with the acoustic image, the main conductor in subsurface infrastructure reconnaissance. In addition to first-
package dips gently downward and thinning out substantially order applications of constraining depth and lateral location of
near row 14 (see Figure 9b for geographic location), but likely subsurface utilities, it may now be possible to resolve second-order
continues toward the well bay, albeit along a trend approximately features, the quality of which has room for improvement with
40 m northwest of the original well bay location. However, as it closer line-spaced data collection and finer modeling grid definition
approaches the well bay in strike direction, it bends (jogs) toward for complex use cases. As such, specific sizes of material discon-
DOI:10.1190/leedff.2025.44.issue-3

the metallic debris field around the well conductor bay. The main tinuity and/or significant metal loss for near-surface pipes can
conductor package connects with the well conductor bay through now be investigated in future applications.
the jog in structure near the well bay debris. In addition, the onshore and offshore imaging applications
The primary magnetic anomaly associated with the conductor do not require alterations in the ML architecture. The same
complex is represented by the red color in the anomaly map of architecture can be extended for multiple use cases by adding
the acquired magnetic data (Figure 13a). It shows a dominant, more layers or starting filters as necessary. Furthermore, as
strong response influenced by steel debris, located approximately evident from the results, the total number of samples used for
40 m away from the well conductor bay, implying the magnetic training the system is relatively consistent (190,000–202,000).
signal from the bay is diminished and cancelled in strength (due Thus, using a modular trained data set over a small volume and
to remanent magnetization in different directions). The trained then imaging via inference to arbitrarily large data sets without
AI model is able to differentiate between these various contribu- additional training allows for a highly scalable approach to
tions and reconstruct an easily interpretable 3D image of the multiple imaging applications with minimal accrual of compu-
conductor package and the steel debris field surrounding the tational expense.
original bay structure (Figures 18–20). This is impossible to discern Moreover, AI imaging algorithms are agnostic of the geophysi-
from the raw magnetic anomaly map. cal sensor. As such, it is possible to generate similar training
The deep learning application reveals an evident branching models for nonmagnetic data such as electromagnetic and seismic
of a portion of the conductor bundle that traverses off and stops and obtain still higher and sometimes complementary resolutions
approximately 40 m northeast of the original well conductor bay, for targets where necessary.
composed of fragmented conductors or pipe-related debris
(Figures 18b and 18c). A separated pipe trend continues past Acknowledgments
row 11, curving and leading directly toward and around the well The authors thank the interagency committee involved in
conductor bay (Figures 18b, 19b, 20a, and 20b). As presented funding the study at the felled producing platform in the GOM.
through the AI application, this curving pipe path, traversing Special mention goes to the Kraken Robotics geophysical team
directly to the well conductor bay, shows a flow path from the of Ryan Laidley and Stephanie Abbott, along with Oceaneering
well bay back to the downed jacket and the collector oil/gas dome International’s Devon Crosby. The Texas A&M Transportation
(Figures 20a and 20b). Without such enhancement, the interpreta- Institute (TTI) is acknowledged for supporting the field data
tion of the raw magnetic anomaly by itself would have inferred acquisition, processing, and application of AI for interpretation
an isolated termination 40 m north of the well bay. under the TxDOT (Texas Department of Transportation) Project
0-7127 Examine Reconnaissance Scanning of Underground
Conclusion Utilities in the ROW. We also thank Cesar Quiroga, utility
Using deep learning AI, we have presented two examples of engineering program manager, for facilitating the process and
3D imaging of metallic infrastructure from remotely located permitting the publication of results from the project.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 203
 In addition, we would like to thank Iowa State University’s Dhara, A., and M. K. Sen, 2023, Elastic full-waveform inversion using
Bridges-2 and Jetstream computing platforms and Principal a physics-guided deep convolutional encoder–decoder: IEEE
Investigator Aditya Balu of the CIS220048 Project for computing Transactions on Geoscience and Remote Sensing, 61, 5913118, https://
doi.org/10.1109/TGRS.2023.3294427.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

access. We also acknowledge the efforts of Peter Kowalczyk,

Dekkers, M. J., 1978, Magnetic properties of sediments, in R. W.
Karen Weitemeyer, Fernanda Viana da Conceição, Allison Proctor, Fairbridge, ed., Encyclopedia of sedimentology: Springer, 684–692,
and Matthew Kowalczyk of Ocean Floor Geophysics in participat- https://doi.org/10.1007/3-540-31079-7_130.
ing in the surveying and developing a high-quality magnetic data Ellwood, B. B., W. L. Balsam, and H. H. Roberts, 2006, Gulf of Mexico
set for inversion. sediment sources and sediment transport trends from magnetic sus-
Finally, we thank the wonderful editorial team of The Leading ceptibility measurements of surface samples: Marine Geology, 230,
Edge that facilitated the publication of this paper in as smooth a no. 3–4, 237–248, https://doi.org/10.1016/j.margeo.2006.05.008.
Guigné, J. Y., and P. Blondel, 2017, Acoustic interrogations of complex
process as possible. In particular, we extend thanks to both of our
seabeds: Springer.
anonymous reviewers who have provided many insightful com- Guigné, J. Y., and A. Gogacz, 2015, Mapping of sub-seabed anomalies
ments, suggestions, and also raised very important questions that in seismically heterogeneous marine soils: Institute of Acoustics,
have all contributed toward making this manuscript a much- Proceedings of Seabed and Sediment Acoustics, https://doi.
strengthened contribution. org/10.25144/16050.
Guigné, J. Y., and R. Laidley, 2024, Acoustic interrogation and imaging
Data and materials availability of complex subsurface seabed buried debris: Institute of Acoustics,
Proceedings of the International Conference on Underwater Acoustics,
Data associated with this work may be released partially and
https://doi.org/10.25144/22233.
on a case-by-case basis. The corresponding author will be the Hoffmann, D. R., T. M. Couvillion, K. J. Kennelley, and W. J. Couch,
initial point of contact in this regard. 2023, A subsea oil spill response solution for a toppled platform that
has been leaking oil into the Gulf of Mexico since 2004: Offshore
Corresponding author: [email protected] Technology Conference, https://doi.org/10.4043/32268-MS.
Hoffmann, D., J. Y. Guigné, R. Laidley, and K. J. Kennelley, 2024, A
DOI:10.1190/leedff.2025.44.issue-3

References geotechnical acoustic survey to address well abandonment options

Ali, H., N. S. M. Ideris, N. Syahirah, A. F. Ahmad Zaidi, M. S. Zanar for a toppled offshore platform: Offshore Technology Conference,
Azalan, T. S. Tengku Amran, M. R. Ahmad, N. Abdul Rahim, and https://doi.org/10.4043/35403-MS.
S. A. Abdul Shukor, 2021, Ground penetrating radar for buried utilities Hoversten, M. G., and C. Schwarzbach, 2018, Monitoring hydraulic
detection and mapping: A review: Journal of Physics: Conference fracture volume using surface-to-borehole EM and conductive prop-
Series, 2107, 012056, https://doi.org/10.1088/1742-6596/ pant: 88th Annual International Meeting, SEG, Expanded Abstracts,
2107/1/012056. 863–866, https://doi.org/10.1190/segam2018-2984673.1.
ASCE, 2022, Standard guideline for investigating and documenting Jahandari, H., and C. G. Farquharson, 2013, Forward modeling of
existing utilities (ASCE 38-22): American Society of Civil Engineers. gravity data using finite-volume and finite-element methods on
Barrows, L., and J. E. Rocchio, 1990, Magnetic surveying for buried unstructured grids: Geophysics, 78, no. 3, G69–G80, https://doi.
metallic objects: Groundwater Monitoring and Remediation, 10, no. org/10.1190/geo2012-0246.1.
3, 204–211, https://doi.org/10.1111/j.1745-6592.1990.tb00016.x. Kim, Y., and N. Nakata, 2018, Geophysical inversion versus machine
Bernstein, B., Y. Li, and R. Hammack, 2023, Automated detection of learning in inverse problems: The Leading Edge, 37, no. 12, 894–901,
ferromagnetic pipelines from magnetic total-field anomaly data using https://doi.org/10.1190/tle37120894.1.
convolutional neural networks: Third International Meeting for Kraus, E., K. Obeng-Boampong, and C. Quiroga, 2012, Utility investiga-
Applied Geoscience & Energy, SEG/AAPG, Expanded Abstracts, tion trends in Texas: Transportation Research Record, 2309, no. 1,
1201–1203, https://doi.org/10.1190/image2023-3912400.1. 209–217, https://doi.org/10.3141/2309-20.
Blakely, R. J., 1995, Potential theory in gravity and magnetic applica- Krizhevsky, A., I. Sutskever, and G. E. Hinton, 2012, ImageNet clas-
tions: Cambridge University Press, https://doi.org/10.1017/ sification with deep convolutional neural networks: Advances in
CBO9780511549816. Neural Information Processing Systems, 1097–1105.
Botelho, S., S. Mukherjee, V. Rao, and S. Adavani, 2022, Deep learning Laidley, R., J. Y. Guigné, M. Kotsi, S. Abbott, C. A., Kilic, D. Hoffmann,
software accelerators for full-waveform inversion: Second International and K. Kennelley, 2023, Imaging conductor pipes in the Gulf of
Meeting for Applied Geoscience & Energy, SEG/AAPG, Expanded Mexico using 3-D high-resolution seismic data: Containing one of
Abstracts, 3681–3685, https://doi.org/10.1190/image2022-w17-01.1. the largest oil spills in US history: Proceedings of OCEANS 2023,
Chen, J., M. Tompkins, P. Zhang, M. Wilt, and R. Mackie, 2012, https://doi.org/10.23919/oceans52994.2023.10337084.
Frequency-domain EM modeling of 3D anisotropic magnetic perme- LeCun, Y., Y. Bengio, and G. E. Hinton, 2015, Deep learning: Nature,
ability and analytical analysis: 82nd Annual International Meeting, SEG, 521, no. 7553, 436–444, https://doi.org/10.1038/nature14539.
Expanded Abstracts, https://doi.org/10.1190/segam2012-0308.1. Lelièvre, P. G., C. G. Farquharson, and C. A. Hurich, 2012, Joint
Colombo, D., W. Li, E. Sandoval-Curiel, and G. McNeice, 2020, inversion of seismic traveltimes and gravity data on unstructured
Deep-learning electromagnetic monitoring coupled to fluid flow grids with application to mineral exploration: Geophysics, 77, no. 1,
simulators: Geophysics, 85, no. 4, WA1–WA12, https://doi. K1–K15, https://doi.org/10.1190/geo2011-0154.1.
org/10.1190/geo2019-0428.1. Li, C., D. Liu, J. Meng, J. Liu, and Y. Zhang, 2020, The positioning of
Cox, L. H., G. A. Wilson, and M. S. Zhdanov, 2012, 3D inversion of buried pipelines from magnetic data: Geophysics, 85, no. 6, J111–J120,
airborne electromagnetic data: Geophysics, 77, no. 4, WB59–WB69, https://doi.org/10.1190/geo2020-0157.1.
https://doi.org/10.1190/geo2011-0370.1. Li, Z., F. Liu, W. Yang, S. Peng and J. Zhou, 2022, A survey of convo-
Dampney, C. N. G., 1969, The equivalent source technique: Geophysics, lutional neural networks: Analysis, applications, and prospects: IEEE
34, no. 1, 39–53, https://doi.org/10.1190/1.1439996. Transactions on Neural Networks and Learning Systems, 33, 12,
6999–7019, https://doi.org/10.1109/TNNLS.2021.3084827.

204 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 Maus, S., and V. Haak, 2003, Magnetic field annihilators: Invisible Ronneberger, O., P. Fischer, and T. Brox, 2015, U-Net: Convolutional
magnetization at the magnetic equator: Geophysical Journal networks for biomedical image segmentation: Proceedings of the
International, 155, no. 2, 509–513, https://doi.org/10.1046/​ International Conference on Medical Image Computing and
j.1365-246X.2003.02053.x. Computer-assisted Inter vention, 234 –241, https://doi.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Mukherjee, S., R. S. Bell, W. N. Barkhouse, S. Adavani, P. G. Lelièvre, org/10.1007/978-3-319-24574-4_28

and C. G. Farquharson, 2022, High-resolution imaging of subsurface Stalnaker, J. L., M. E. Everett, A. Benavides, and C. J. Pierce, 2006,
infrastructure using deep learning artificial intelligence on drone Mutual induction and the effect of host conductivity on the EM
magnetometry: The Leading Edge, 41, no. 7, 462–471, https://doi. induction response of buried plate targets using 3D finite-element
org/10.1190/tle41070462.1. analysis: IEEE Transactions on Geoscience and Remote Sensing,
Mukherjee, S., M. E. Everett, A. I. Benavides, and R. Fernandes, 2012, 44, no. 2, 251–259, https://doi.org/10.1109/TGRS.2005.860487.
Mutual coupling of conductive, permeable targets and its effects on Sun, J., A. T. Melo, J. D. Kim, and X. Wei, 2020, Unveiling the 3D
CSEM response: 82nd Annual International Meeting, SEG, Expanded undercover structure of a Precambrian intrusive complex by integrating
Abstracts, https://doi.org/10.1190/segam2012-1505.1. airborne magnetic and gravity gradient data into 3D quasi-geology
Mukherjee, S., P. G. Lelièvre, C. G. Farquharson, and S. Adavani, 2021, model building: Interpretation, 8, no. 4, SS15–SS29, https://doi.
Three-dimensional inversion of geophysical field data on an unstruc- org/10.1190/INT-2019-0273.1.
tured mesh using deep learning neural networks, applied to magnetic Um, E. S., M. Commer, G. A. Newman, and G. M. Hoversten, 2015,
data: First International Meeting for Applied Geoscience & Energy, 2015, Finite element modelling of transient electromagnetic fields
SEG/AAPG, Expanded Abstracts, 1465–1469, https://doi. near steel-cased wells: Geophysical Journal International, 202, no. 2,
org/10.1190/segam2021-3583565.1. 901–913, https://doi.org/10.1093/gji/ggv193.
NCEI, 2024, Magnetic field calculators, https://www.ngdc.noaa.gov/ Weiss, C. J., 2017, Finite-element analysis for model parameters distrib-
geomag/calculators/magcalc.shtml#igrfwmm, accessed 10 February uted on a hierarchy of geometric simplices: Geophysics, 82, no. 4,
2025. E155–E167, https://doi.org/10.1190/geo2017-0058.1.
Okabe, M., 1979, Analytical expressions for gravity anomalies due to Xinjing, H., L. Shan, Z. Zhoumo, and L. Jian, 2019, Magnetic perme-
homogeneous polyhedral bodies and translations into magnetic ability measurement for steel pipe immersed in geomagnetic field:
anomalies: Geophysics, 44, no. 4, 730–741, https://doi.org/10.1190/1. Journal of Magnetism and Magnetic Materials, 491, 165621, https://
1440973. doi.org/10.1016/j.jmmm.2019.165621.
Puzyrev, V., 2019, Deep learning electromagnetic inversion with con- Zhang, L., G. Zhang, Y. Liu, and Z. Fan, 2022, Deep learning for 3-D
DOI:10.1190/leedff.2025.44.issue-3

volutional neural networks: Geophysical Journal International, 218, inversion of gravity data: IEEE Transactions on Geoscience and Remote
no. 2, 817–832, https://doi.org/10.1093/gji/ggz204. Sensing, 60, 5905918, https://doi.org/10.1109/TGRS.2021.3110606.

WORKSHOP

New Geophysical
Aspects for Production
Enhancement
13–15 May 2025 • Muscat, Oman

Register by 12 April and save!

The workshop will focus on three key themes: advanced seismic

techniques for high-resolution imaging, integrated geophysical
approaches for reservoir characterization and monitoring, and
geophysics in integrated field development and optimization.
It aims to provide a comprehensive understanding of how
these techniques can be leveraged for both onshore and
offshore operations to enhance production, optimize reservoir
management, and drive sustainable field development.

go.seg.org/ngap25

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 205
 Overcoming complex near-surface conditions for improved
seismic imaging in Tarim Basin, China
Ruirui Zhao1,2,3, Duoming Zheng1,2,3, Shanshan Yang1,2,3, Anxin Zuo1,2,3, Yangyang Chen1,2,3, Xianhuai Zhu4, Ken Xu4, and Sam Gray5
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

https://doi.org/10.1190/tle44030206.1

Abstract
Seismic imaging in the Tarim Basin, China, presents signifi-
cant challenges due to complex near-surface conditions, including
desert environments, foothills, and loess mountains. These chal-
lenges are comparable to those found in regions such as the Arabian
Peninsula, North Africa, the Andes Mountains, and other complex
terrains worldwide. Recent advancements in seismic imaging
aimed at overcoming these obstacles include: (1) advanced con-
strained near-surface tomography, which has significantly
enhanced the robustness of near-surface velocity-depth model
estimation, leading to improved resolution of deep reservoir Figure 1. Near-surface challenges in the Tarim Basin. Seismic exploration complexity
images; (2) a recent ultra-long-offset (greater than 15 km) experi- escalates from (a) the arid near-surface desert to (b) the foothills and (c) loess regions in
ment in the foreland basin of the Tarim Oilfield, demonstrating the Tarim Basin.
that more accurate and deeper near-surface velocity models can
be generated on land using turning-ray tomography; (3) finite- this article, versions of the advancements can still be applied
frequency wavepath tomography, which has been proven through in those areas.
DOI:10.1190/leedff.2025.44.issue-3

field data examples to be a robust alternative to traditional full- In 2018, Tarim Oilfield Company established a task force to
waveform inversion for land seismic data; and (4) integrated address near-surface complexities and enhance reservoir imaging.
tomography, which combines diving waves and reflected seismic High-graded proposals for solutions included (1) generalized
data to develop a comprehensive velocity model from shallow to sand-velocity-curve-constrained tomography, with the help of
deep sections for anisotropic prestack depth migration from machine learning, to improve the accuracy and reliability of
topography. These technologies, developed for and applied to the estimated near-surface velocity models; (2) ultra-long-offset (greater
Tarim Basin, can be applied, with modifications, to other land than 15 km) surveys on land, aimed at imaging ultra-deep reservoirs
basins around the world. 8000 m below the surface; (3) finite-frequency wavepath tomog-
raphy to improve the resolution of estimated near-surface velocity
Introduction models; and (4) integrated tomography using both refraction and
The challenges of seismic exploration in the Tarim Basin, reflection data to improve the robustness of velocity model building
primarily stemming from its complex near-surface conditions, for imaging deep reservoirs. Implementing these four key solutions
are analogous (though not identical) to those faced in the has resulted in step-change improvement in structural imaging
Arabian Peninsula and the Andes Mountains of South America under complex near-surface conditions. We highlight finite-fre-
(Gray and Zhu, 2019). These include low-velocity arid desert, quency wavepath tomography (Zhang et al., 2021) as a major
high-velocity conglomerate rocks in foothills regions, and a component of this improvement. This technology serves as a useful
mixture of low-velocity loess materials on the earth’s surface land alternative to full-waveform inversion (FWI), which is widely
and high-velocity conglomerate rocks beneath the surface in applied to marine data (Ramirez et al., 2020) but often fails on
the southwestern Tarim Basin. The seismic exploration com- land because of large differences between observed and simulated
plexity increases from the desert to the foothills and loess wavefields due to varying wavelets and noise types such as P- and
mountains (Figure 1). In the Arabian Peninsula and Andes S-wave scattering, back scattering, and multiple refractions caused
Mountains, some of these problems are replaced by others by the challenging near-surface conditions (Stork, 2024).
(extreme, highly karsted, near-surface variations in Arabia; This manuscript builds on an abstract presented during the
highly vegetated surface in some Andes regions), leading to Recent Advances and the Road Ahead session at the Fourth
severe acquisition challenges. Although these challenges require International Meeting for Applied Geoscience & Energy (Zhu
different details in solution methods from what we describe in et al., 2024).

Manuscript received 7 October 2024; accepted 16 December 2024.

1
PetroChina, Tarim Oilfield Company, Research Institute of Exploration and Development, Korla, Xinjiang, China. E-mail: zhaoruirui-tlm@
petrochina.com.cn; [email protected]; [email protected]; [email protected]; [email protected].
2
CNPC, R&D Center for Ultra-Deep Complex Reservoir Exploration and Development, Korla, Xinjiang, China.
3
Engineering Research Center for Ultra-Deep Complex Reservoir Exploration and Development, Xinjiang, China.
4
Forland Geophysical Services, Houston, Texas, USA. E-mail: [email protected]; [email protected].
5
Viridien, Calgary, Alberta, Canada. E-mail: [email protected].

206 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 Figure 2. Sand-velocity-curve-constrained near-surface
tomography. (a) Uphole surveys where color indicates
uphole depth; (b) base elevation of low-velocity
volume of the initial velocity model constructed using
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

a multilayer perception neural network (c) based on

sand-velocity curves, uphole surveys, and rock physics.
(d) Near-surface velocity model after tomography
constrained by the uphole survey data (a) only. (e) Near-
surface velocity model after tomography constrained by
the constructed sand-velocity-curve volume (b).
DOI:10.1190/leedff.2025.44.issue-3

Figure 3. Brute stacks with (a) field statics and (b) sand-velocity-curve-constrained tomostatics.

(velocity as a function of sand dune thickness), rock physics

relationships, and explosive charge depths. This approach generates
a high-resolution ultra-shallow velocity volume (Figure 2b) that
extends coverage to areas lacking uphole data. In this study, a
multilayer perceptron neural network (Figure 2c) was employed
to model the shallow sand velocity variations. The network is
trained and validated using measured uphole velocity profiles
(depth-velocity functions) along with their (x,y) spatial locations.
Once trained, the model predicts shallow velocities at all remaining
(x,y) locations within the survey area.
The inclusion of these predicted velocities as constraints in
Figure 4. Development trend of near-surface velocity inversion on land. One direction is to tomography significantly enhances the accuracy of near-surface
increase inversion depth; another direction is to increase resolution of estimated velocity models. velocity estimates and static corrections, as demonstrated in
Figures 2d, 2e, and 3. This methodology, while tailored for desert
Constrained near-surface tomography conditions, can be adapted for use in other challenging environ-
Seismic inversion is highly sensitive to the choice of initial ments, such as loess mountains, where uphole surveys, if available,
velocity model. Accurate near-surface velocity estimates, which could similarly inform the neural network’s predictions for loess-
are essential for reliable deeper velocity model building, depend velocity variations.
heavily on a well-constructed initial near-surface model. In desert
regions, uphole surveys are valuable tools for velocity estimation Ultra-long-offset inversion on land
and static corrections. However, their deployment is typically One of the directions in near-surface velocity estimation is to
sparse, with survey points spaced 1–2 km apart (Figure 2a) due increase the maximum depth of estimated velocity models, from
to the high costs involved. conventional surveys with maximum offset less than 8 km to
To address this limitation, we utilize a deep learning approach ultra-long-offset surveys with maximum offset greater than 15 km
to complement uphole data, incorporating sand-velocity curves (Figure 4). Recently an ultra-long-offset seismic acquisition was

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 207
 conducted in the carbonate province of the Tarim Basin, character-
ized by diverse surface conditions including farmland, villages,
Gobi, and alluvial fans with varying conglomerate rock sizes
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

(Figure 5). Forty ultra-long-offset “land-streamer” lines (Figure 6)

were deployed in conjunction with a conventional 3D survey
(Figure 5), utilizing 1322 dynamite and vibroseis sources. A land
streamer survey, as illustrated in Figure 6, is designed for intensive
target-oriented seismic exploration in areas with complicated near-
surface and subsurface geologic conditions, containing densely
deployed shot lines and receiver lines along a specific direction (Li
et al., 2020). This survey has longer offsets but narrower azimuthal
coverage than the conventional survey. A total of 48,558 land nodes
were strategically deployed at receiver locations (Figure 6). Using Figure 5. Study area of an ultra-long-offset experiment in the Tarim Basin.
land nodes makes the ultra-long-offset acquisition relatively afford-
able (Zhao et al., 2024). During processing, a matching filter needs
to be applied to balance amplitude, phase, and bandwidth of the
seismic data acquired using dynamite and vibroseis sources.
Figure 7 shows a representative dynamite shot record with a
15 km offset. Turning-ray tomography was performed using
normal-offset (less than 8 km) and ultra-long-offset (greater than
15 km) first arrivals (Figure 8). When the ultra-long-offset first
arrivals are employed, turning-ray tomography achieves a reliable
depth of approximately 4.0 km, about one-quarter of the maximum
DOI:10.1190/leedff.2025.44.issue-3

offset (Figure 8).

Finite-frequency wavepath tomography

Another direction in estimating near-surface velocity models
is to improve resolution, progressing from turning-ray tomography
to finite-frequency wavepath tomography and, ultimately, to FWI
(Figure 4).
Land FWI can improve the resolution of estimated near-
Figure 6. Acquisition geometry for ultra-long-offset acquisition indicated in Figure 5. Nodes
surface velocity models. However, significant challenges lie in were deployed at all receiver locations. Node (receiver) spacing = 25 m; shot interval =
accurately modeling the recorded seismic wavefields during inver- 50 m; receiver line spacing = 300 m; shot line spacing = 300 m; bin size = 12.5 × 25 m;
sion, especially considering the complex nature of near-surface record length = 10 s; fold = 200.
conditions, elastic wavefield scattering,
and the variability of wavelets along the
surface (Yilmaz et al., 2022), and land
FWI successes remain rare (e.g., Guo
et al., 2025). Some of the challenges
facing land FWI can be overcome using
an alternative called finite-frequency
wavepath inversion (Figure 9). This
technology has yielded encouraging
results (Figures 10–15), setting the stage
for further investigation and assessment
of its usefulness in the future. Unlike
FWI, f inite-frequency wavepath
tomography does not always require
knowledge of the wavelet for inversion,
so near-surface velocities on land can
be estimated using dynamic ray tracing
with the observed first-arrival travel-
times (Zhang et al., 2021) to achieve an
efficient alternative to FWI that Figure 7. A representative ultra-long-offset land shot record from the Tarim Basin. BTN = “black triangle” noise; CR = continuous
improves velocity estimation accuracy refractions; D = direct arrival; H (FB) = head wave (first break); MR = multiple refractions; SR = surface wave; R = reflection;
over ray tomography. Thus, in areas WR = wide-angle reflection. “Aoffset” stands for absolute offset. Total recording time was 10 s; the first 6 s are shown.

208 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 8. Ray density plot from turning-ray tomography, showing that (1) inversion depth increases with increased maximum offset used in turning-ray tomography, and (2) turning-ray
tomography is sensitive to the low-velocity zone indicated on the VSP plot shown on the upper right, where the blue TomoVel curve tracks the VSP curve. TomoVel stands for velocities
estimated from turning-ray tomography.
DOI:10.1190/leedff.2025.44.issue-3

Figure 9. Illustration of (a) wavepath and (b) workflow of implementing finite-frequency wavepath tomography (Zhang et al., 2021). The wavepath, also called sensitivity kernel, is a
function of frequency, Fresnel zone, and acquisition geometry.

where traditional FWI fails, finite-frequency wavepath tomog- turning-ray tomography (Figure 12b); and (3) a near-surface
raphy is a robust alternative to FWI on land (Zhu et al., 2024). velocity model estimated using finite-frequency wavepath tomog-
Unlike a fat ray, a wavepath, also called sensitivity kernel or raphy (Figure 12c). For models (2) and (3), the deeper velocities
“banana donut” (Woodward, 1992; Liu et al., 2009; Fu and (below 6 km) were the same as those used in the legacy model
Hanafy, 2017), is a function of frequency, Fresnel zone, and (Figure 12a). In the production workflow to generate the legacy
acquisition geometry. Because traveltime sensitivity to velocity velocity model, a maximum offset of 8000 m was used, and
perturbations is greatest along the wavepath edges and least along reflection tomography was applied to update velocities in both
the central raypath, the weighting factors in inversion are primarily shallow and deep sections. By contrast, ultra-long offsets (up to
distributed along the edges of the kernel (Figure 9a). 22,000 m) were utilized to estimate the near-surface velocity
Given a narrow band of frequencies, wavepath tomography models in turning-ray tomography (Figure 12b) and finite-
produces high-resolution near-surface velocity models, appears frequency wavepath tomography (Figure 12c).
to have better penetration in both vertical and horizontal direc- The common-image gathers (CIGs) after PSDM are displayed
tions, and is less sensitive to low-velocity zones and acquisition in Figure 13. It is evident that the flatness of the events on the
footprints than ray-based tomography (Figures 10 and 11). CIGs using the near-surface velocity model from finite-frequency
To validate the accuracy of the near-surface velocity model wavepath tomography (Figure 13c) is superior, particularly within
estimated using finite-frequency wavepath tomography, prestack the circles highlighting depths above 9000 m. This observation
depth migration (PSDM) was performed with three different indicates that the finite-frequency wavepath tomography-derived
velocity models for comparison (Figure 12): (1) a legacy velocity near-surface velocity model (Figure 13c) is more accurate com-
model generated from a production processing workflow pared to the legacy model (Figure 13a) and the model derived
(Figure 12a); (2) a near-surface velocity model estimated using from turning-ray tomography (Figure 13b). However, events

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 209
 below 10,000 m show little or no
improvement using wavepath tomog-
raphy (Figure 13c), suggesting that
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

deeper velocities need to be reestimated

after wavepath tomography. The aniso-
tropic parameters (ε, δ, dip, and azi-
muth) were kept consistent across all
scenarios. Following the application of
a stretch mute (purple lines in Figure 13),
the effective maximum offset used in
PSDM stacking was 17,000 m.
Figure 14 presents the shallow sec-
tions of the PSDM stacks derived using
the near-surface velocities from the
production workflow (Figure 14a),
turning-ray tomography (Figure 14b), Figure 10. Illumination plot from finite-frequency wavepath tomography. Compare with the ray density plot of Figure 8.
and finite-frequency wavepath tomog-
raphy (Figure 14c). The continuity of
reflection events and fault definition is
significantly improved in Figure 14c (as
highlighted by the arrows), demonstrat-
ing that the finite-frequency wavepath
tomography-derived near-surface veloc-
DOI:10.1190/leedff.2025.44.issue-3

ity model is more accurate.

Similarly, Figure 15 shows the deep
sections of the PSDM stacks. The
improvement in seismic images in the
deeper section (Figure 15c) using the
finite-frequency wavepath tomography-
derived velocity model is particularly
evident between depths of 6000 and
8000 m, especially in the circled areas
marked by arrows. However, below
8000 m, the improvements are limited
because the velocities below 6000 m
were the same as those in the legacy
velocity model (Figure 15a). The
enhancements observed between 6000
and 8000 m are partially influenced by
the improved shallow velocity model.
To achieve further improvements in
deep seismic imaging, a simultaneous
update of velocities from shallow to deep
sections is needed, using an integrated
tomography workflow as discussed in
the following section.

Integrated tomography
Integrated refraction/ref lection
tomography, also known as joint tomog-
raphy, proves particularly effective in
foothills areas (Jun et al., 2018; Li et al.,
2020; Zhao et al., 2023; Zhu et al., 2024).
This approach begins with the estimation Figure 11. (a) Velocity-depth models and (b) illumination plots of turning-ray tomography and finite-frequency wavepath tomography using
of a near-surface velocity model using the maximum offset at 8000 m (top panel) and 22,000 m (middle and bottom panels). The wavepath tomographic velocity model shows
turning-ray tomography (Figure 16). much greater resolution than the turning-ray tomographic models. Depth slices are extracted at z = 2040 m below the highest topography.

210 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 Subsequently, reflection tomography is employed to derive a com- tomography, have significant impact in imaging deep reservoirs
prehensive velocity model spanning from the surface to deep forma- after tilted transverse isotropy (TTI) anisotropic prestack depth
tions. Reflection tomography can be used after turning-ray tomog- migration from topography (Figure 17).
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

raphy; alternatively, it can be used to refine the turning-ray

tomography model with anisotropy in an iterative loop. Because Summary and conclusions
this loop uses both refractions (first breaks) and reflections, we can Figure 18 shows a roadmap of seismic imaging in areas under
update shallow velocities constrained by deep velocities and vice- complex near-surface conditions. First, turning-ray tomography
versa. The advantages of using joint tomography are illustrated in is employed to generate a low-resolution velocity model using
Figure 17, using a data set from a foothills area in the Tarim Basin. sand-velocity curves to provide a starting model (Figure 18a).
The high-velocity anomalies, probably conglomerate rocks delineated Second, the low-resolution velocity model is used as an initial
in the shallow part of the foreland basin (Figure 16) by turning-ray model for finite-frequency wavepath tomography to obtain a
high-resolution velocity-depth model
(Figure 18b). Third, joint tomography
is implemented to build an entire veloc-
ity/anisotropy model from shallow to
deep sections for TTI anisotropic
prestack depth migration, resulting in
an improved depth image for reservoir
delineation and characterization
(Figure 18c).
In summary, this paper provides an
overview of recent advancements and
practical applications of seismic imaging
DOI:10.1190/leedff.2025.44.issue-3

technologies in the Tarim Basin,

addressing the unique challenges posed
by complex near-surface conditions.
These challenges are analogous to those
encountered in many land exploration
areas globally.

• The implementation of sand-velocity-

curve-constrained tomography has
Figure 12. Velocity models constructed from: (a) production processing workflow, where the maximum offset used was 8000 m,
utilizing both turning-ray tomography and reflection tomography to build the model; (b) turning-ray tomography with a maximum significantly enhanced the accuracy of
offset of 22,000 m; velocities below 6000 m are the same as in (a); and (c) finite-frequency wavepath tomography with a near-surface velocity models, resulting
maximum offset of 22,000 m; velocities below 6000 m are the same as in (a). VSP velocities at well-A are used for calibration. in improved static corrections and

Figure 13. CIGs after TTI anisotropic prestack depth migrations using velocity models from Figure 12: (a) production processing workflow; (b) turning-ray tomography (0–6000 m in depth);
(c) finite-frequency wavepath tomography (0–6000 m in depth). After applying a mute along the purple lines, the maximum effective offset used for PSDM stacking is 17,000 m.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 211
 imaging. This advancement is particularly valuable in desert mountains, finite-frequency wavepath tomography has proven
regions with sparse uphole survey data. to be a robust alternative. Prestack depth migration utilizing
• Ultra-long-offset surveys, facilitated by the deployment of velocity models generated by finite-frequency wavepath
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

land nodes, have opened new opportunities for enhancing tomography has delivered superior shallow imaging, under-
seismic imaging of deep targets, offering significant improve- scoring the accuracy of velocity models derived from finite-
ments in resolution and structural clarity. frequency wavepath tomography in the shallow subsurface.
• While traditional FWI remains challenging for land seismic • Joint tomography, integrating refraction and reflection data,
data, particularly in complex terrains like foothills and loess has demonstrated its effectiveness in building comprehensive
DOI:10.1190/leedff.2025.44.issue-3

Figure 14. Shallow sections of PSDM using velocity models constructed from: (a) production Figure 15. Deep sections of PSDM using velocity models constructed from: (a) production
processing workflow; (b) turning-ray tomography; and (c) finite-frequency wavepath processing workflow; (b) turning-ray tomography; and (c) finite-frequency wavepath
tomography. In (b) and (c), velocities below 6000 m are the same as in (a). Arrows highlight tomography. In (b) and (c), velocities below 6000 m are the same as in (a). Arrows indicate
more continuous reflection events and better-defined faults with velocities estimated from more continuous reflection events and improved fault definition with velocities from finite-
finite-frequency wavepath tomography (c). frequency wavepath tomography (c).

Figure 16. Representative near-surface velocity model estimated using turning-ray

tomography in a foothills area of the Tarim Basin. The maximum offset used for picking
first arrivals is 8000 m. The reliable depth of the estimated near-surface model extends
approximately 1500 m below the surface.

212 The Leading Edge March 2025 Special Section: Near-surface geophysics in the energy transition
 velocity models from shallow to
deep sections. The methodology is
especially impactful in foothills
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

regions, enabling enhanced imag-

ing of deep reservoirs and contrib-
uting to more reliable subsurface
interpretations.

Acknowledgments
We extend our sincere gratitude to
Bangliu Zhao, Yalin Li, Haijun Yang,
Dajun Li, Ganglin Lei, Wensheng
Duan, Gengxin Peng, and Xingjun
Wang for their invaluable support Figure 17. Comparison of legacy (left) and reprocessed (right) PSDM for a representative inline in a foothills region (modified
throughout the duration of this study. from Li et al., 2020). Joint tomography was applied during reprocessing, incorporating the near-surface velocity model
We also wish to express our appreciation estimated from turning-ray tomography (Figure 16), resulting in the improved image (right). K represents the Cretaceous
formation, the primary target, while J indicates the Jurassic formation, a potential deep reservoir. The décollement at 12 km
for the insightful discussions provided depth is simplified (right), making interpretation easier.
by Houzhu (James) Zhang and Yan
Iliescu. Constructive suggestions and
comments from Sergio Chávez-Pérez,
Chengbo Li, Tracy Stark, and two
anonymous reviewers are greatly appre-
ciated. Additionally, we extend our
DOI:10.1190/leedff.2025.44.issue-3

thanks to Tarim Oilfield Company and

Forland Geophysical Services for per-
mission to showcase our results.

Data and materials availability

Data associated with this research Figure 18. Schematic roadmap of seismic imaging on land under challenging near-surface conditions, starting from (a) turning-ray
are confidential and cannot be released. tomography velocity model (velocity-depth slice) to (b) finite-frequency wavepath tomography velocity model (velocity-depth slice),
and (c) joint tomography velocity model corendered with TTI anisotropic prestack depth migration images (cross section).
Corresponding author: [email protected]

References Stork, C., 2024, How does the thin near surface of the earth produce up to 100
Fu, L., and S. M. Hanafy, 2017, Ray-tracing traveltime tomography versus times more noise on land seismic data than on marine data: SEG
wave-equation traveltime inversion for near-surface seismic land data: Distinguished Lecture, https://www.youtube.com/watch?v=IM4TGLP5P5I,
Interpretation, 5, no. 3, SO11–SO19, https://doi.org/10.1190/ accessed 13 February 2025.
INT-2016-0210.1. Woodward, M. J., 1992, Wave-equation tomography: Geophysics, 57, no.
Gray, S., and X. Zhu, 2019, Foothills seismic imaging challenges: Past, 1, 15–26, https://doi.org/10.1190/1.1443179.
present and future: 2nd SEG Foothills Exploration Workshop, https:// Yilmaz, O., K. Gao, M. Delic, J. Xia, L. Huang, H. Jodeiri, and A. Pugin,
doi.org/10.1190/FEW2019-01.1. 2022, A reality check on full-wave inversion applied to land seismic data
Guo, Y., A. Aziz, A. Sedova, M. Reinier, D. Donno, G. Lambaré, and D. for near-surface modeling: The Leading Edge, 41, no. 1, 40–46, https://
Carotti, 2025, Unlocking onshore imaging challenges with FWI: Case doi.org/10.1190/tle41010040.1.
studies from the Sultanate of Oman: The Leading Edge, 44, no. 1, 22–31, Zhang, H., H. Liang, H. Baek, and Y. Zhao, 2021, Computational aspects
https://doi.org/10.1190/tle44010022.1. of finite-frequency traveltime inversion kernels: Geophysics, 86, no. 1,
Jun, T., P. Gengxin, J. Jiao, G. Yan, and X. Zhu, 2018, Integrated turning-ray R109–R128, https://doi.org/10.1190/geo2020-0063.1.
and reflection tomography for velocity model building in foothill areas: Zhao, S., Y. Iliescu, X. Shang, H. Teng, Y. Wang, and X. Zhu, 2023,
Interpretation, 6, no. 4, SM63–SM70, https://doi.org/10.1190/ Foothills seismic imaging for deep reservoir exploration in Junggar
INT-2018-0005.1. Basin — A case study: The Leading Edge, 42, no. 10, 694–701, https://
Li, Y., X. Zhu, G. Peng, L. Liu, and W. Duan, 2020, Novel strategies for doi.org/10.1190/tle42100694.1.
complex foothills seismic imaging — Part 1: Mega-near-surface velocity Zhao, R. R., D. Li, S. Yang, A. Zuo, X. Zhu, K. Xu, and Y. Iliescu, 2024,
estimation: Interpretation, 8, no. 3, T651–T665, https://doi.org/10.1190/ Super-long-offset experiment for velocity model building: SEG 1st Tarim
INT-2019-0204.1. Ultra-Deep Oil & Gas Exploration Technology Workshop, 43–46,
Liu, Y., L. Dong, Y. Wang, J. Zhu, and Z. Ma, 2009, Sensitivity kernel for https://doi.org/10.1190/Ultra-Deep2024-012.1.
seismic Fresnel volume tomography: Geophysics, 74, no. 5, U35–U46, Zhu, X., R. Zhao, S. Gray, W. Duan, D. Zheng, G. Lei, G. Peng, et al.,
https://doi.org/10.1190/1.3169600. 2024, Seismic imaging in Tarim Basin: Overcoming complex near-surface
Ramirez, A. C., S. Baldock, D. Mondal, J. Gromotka, and M. Hart, 2020, conditions: Fourth International Meeting for Applied Geoscience &
Long offset ocean bottom node full-waveform inversion and multi-azimuth Energy, SEG/AAPG, Expanded Abstracts, 2489–2493, https://doi.
tomography for high-resolution velocity model building: North Sea, Utsira org/10.1190/image2024-4087690.1.
High: 90th Annual International Meeting, SEG, Expanded Abstracts,
750–754, https://doi.org/10.1190/segam2020-3428806.1.

Special Section: Near-surface geophysics in the energy transition March 2025 The Leading Edge 213
 Exploring seismic data in the flowline domain: Automated
extraction of unconformities, sequence boundaries, and
conformable reflections
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Dennis Adelved1,2, Jan Erik Lie2, Aina Juell Bugge2, and Peter Bormann3
https://doi.org/10.1190/tle44030214.1

Abstract evolution of sedimentary basins (Vail and Mitchum, 1977).

Working with seismic data in the flowline domain offers a Understanding these stratigraphic units and their boundaries is
promising approach to improving geologic interpretation, par- fundamental to unraveling the evolution of sedimentary basins.
ticularly in identifying unconformities and sequence boundaries. Since the mid-1990s, autotracking tools for seismic interpreta-
This method utilizes the seismic differential dip field as a fluid tion have been readily available through standard industry software
velocity field to extract geometric information and generate flow- packages (Henderson et al., 2007; Pauget et al., 2009; Marroquín,
lines. These flowlines, treated as distinct objects, represent the 2014; Eckersley et al., 2018; Vevle et al., 2018; Williams, 2018).
paths of hypothetical particles moving along local velocity vectors However, despite these tools, seismic interpretation still heavily
within the seismic data. By applying a scoring system based on relies on the manual efforts of experienced interpreters. Several
overlapping paths, major unconformities and sequence boundaries notable computational and computer-assisted horizon extraction
can be efficiently identified and extracted. An additional advantage methods have emerged, including techniques using unwrapped
of the flowline representation is its ability to capture the lateral instantaneous phase volumes (Stark, 2003, 2005; Wu and Zhong,
regional context of seismic geometries, enabling straightforward 2012) and methods based on local reflection slopes (Bakker, 2002;
grouping into stratigraphic sequences through simple clustering Lomask et al., 2006; Wu and Hale, 2013; Wu and Fomel, 2018).
DOI:10.1190/leedff.2025.44.issue-3

techniques, which simplifies the categorization of first-order Although many data-driven methods for horizon extraction
stratigraphic units. Furthermore, refining the flowline paths based successfully track coherent horizons, they struggle to correlate
on relative amplitude changes improves their alignment with dislocated horizons across faults and along unconformities within
conformable reflections, supporting a flowline-based approach to complex seismic volumes. Addressing unconformities resulting
effectively track these reflections. The flowline approach can also from significant erosion poses particular challenges, as they often
be extended into three dimensions, providing a straightforward truncate seismic horizons. Unconformities are crucial for under-
tool for preliminary geologic analysis in 3D seismic data. The standing sedimentary basin evolution, typically indicating changes
flowline-based workflows are demonstrated through application in depositional environments due to erosional events or hiatuses.
to two seismic sections, one of which addresses challenges posed Their interpretation can be labor-intensive, especially using tra-
by faulted areas and showcases the strategies used to resolve them. ditional seismic interpretation methods. Over recent decades,
various methodologies leveraging computer science, including
Introduction image processing and machine learning, have been proposed to
Seismic reflection data provide images of the earth’s subsurface, automate this process (Bahorich and Farmer, 1995; Barnes, 2000;
allowing geologists to map geologic structures and interpret van Hoek et al., 2010; Wu and Hale, 2016).
processes such as sedimentation history and tectonic activities. In this paper, we explore various aspects of working with
Mapping these structures from seismic images requires significant seismic data in the flowline domain. This approach involves
time and expertise in seismic interpretation and geophysical converting seismic data into flowlines by tracing the trajectories
understanding. Each seismic image reveals unique stratigraphic of imaginary particles within the differential dip field of the
sequences, characterized by variations in reflection properties like seismic image. One of the key advantages of the flowline repre-
continuity, amplitude, and frequency spacing (Badley, 1985). sentation is that it seems to be highly suitable for identifying and
These seismic sequences represent stratigraphic units consisting tracking sequence boundaries and unconformities, which can be
of conformable seismic reflections, indicating periods of consistent hard to track using correlation-based approaches. In the flowline
sedimentation conditions influenced by factors such as sediment representation, unconformities and sequence boundaries are
supply and relative sea level. The boundaries of these sequences, identified as regions where flowline paths converge. This allows
defined as unconformities or correlative conformities, mark us to create simple workflows for automatic identification and
changes in sediment deposition or nondeposition (Mitchum et al., extraction of these otherwise hard-to-track seismic features.
1977). Furthermore, seismic sequences can be categorized into Our approach builds upon existing methodologies that delin-
depositional strata packages, such as low-stand, high-stand, and eate seismic horizons by tracking the local dip of seismic events,
transgressive system tracts, offering valuable insights into the as demonstrated by de Bruin et al. (2006) and van Hoek et al. (2010).

Manuscript received 1 August 2024; revision received 18 September 2024; accepted 20 September 2024.
1
University of Oslo, Oslo, Norway. E-mail: [email protected].
2
AkerBP, Lysaker, Norway. E-mail: [email protected]; [email protected].
3
ConocoPhillips, Tananger, Norway. E-mail: [email protected].

214 The Leading Edge March 2025

 However, rather than primarily deriving secondary attributes, we of a flowline, we define an infinitesimal arc-length vector along
focus on the direct analysis and interpretation of seismic data the flowline, denoted as
within the flowline domain.
​d →s​
​ = dx​ˆi​ + dy j​ˆ​​,
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

In the following sections, we first detail the process of convert- (3)

ing seismic data into the flowline domain, which forms the founda-
tion for the subsequent workflows. We then explore how this where dx and dy are the infinitesimal displacement in the x and
representation enhances geologic interpretation through several y direction, respectively. Because the flowline should be parallel
workflows. First, we describe the extraction and ranking of seismic to V, the following must be true:
sequence boundaries and unconformities using a sorting and filter-
ing algorithm. We also present a cluster-based approach that ​ →s​
d ˆ​​ = 0​.
​ × V = ​(v dx − u dy) k​ (4)
groups similar flowlines to assist in identifying sequence boundar-
ies and their associated subsequences. By introducing an amplitude Rearranging equation 4, we can formulate it as an ordinary
constraint, in the form of an amplitude likeness criterion, we differential equation (ODE):
demonstrate that the flowline representation can be used to track dy v​(​x, y​)​
v
​_​ = _
​u​ = ​_ ​. (5)
conformable reflections. dx u​(​x, y​)​
Additionally, we introduce a straightforward pseudo-3D
approach to extend the flowline methods, taking a step toward Using the components of our differential field u(x,y) and v(x,y)
making these workflows applicable to 3D seismic data. Finally, and a point (xo,yo) through which the flowline passes, we can
we apply the workflows to two case studies from the Barents Sea, formalize the integral for the flowline as follows:
including an oil field, discussing the challenges of the flowline x
v​(x, y)​
representation in tectonically complex areas and proposing strate- ​y​(x)​ = ​∫​ ​_ ​dx + ​y​o ​. (6)
​x​ ​ u​(x, y) ​
o
gies to address them.
In practice, we obtain the flowline by solving the ODE numeri-
DOI:10.1190/leedff.2025.44.issue-3

Methodology cally using the fourth-order Runge-Kutta method (RK4). This

To characterize the dip field, we adopt a similar approach to method reduces truncation errors, allowing for accurate depictions
van Hoek et al. (2010) where we estimate the dip field based on of flowlines with larger step sizes (Butcher, 2016). RK4 strikes a
the structure tensor. The structure tensor, denoted as S, is defined good balance between computational efficiency and accuracy. The
by its eigenvalues λ1 > λ2 > 0 and the corresponding eigenvectors process for creating the flowlines can be outlined as follows:
{e 1,e 2}, which can be written as
1) Calculate the structure tensor field, S, from the gradient vector
​S = ​λ1​​​e1​ ​​e​T1​ + ​λ2​​​e2​ ​​e​T2​​. (1) field of the seismic image.
2) Define our seismic flow velocity field V(x,y) as the eigenvector
The largest eigenvalue, λ1, and its eigenvector, e 1, are aligned e 2 corresponding to the smallest eigenvalue λ2 at each
perpendicularly to seismic reflections, indicating maximum gradi- location.
ent alignment. The smallest eigenvalue, λ2 , with its orthogonal 3) Initialize the starting coordinate (xo,yo ) for the flowlines by
eigenvector, e 2 , aligns parallel to seismic reflections, which is extracting traces with a fixed increment along the seismic section
central to our analysis. and picking the sample on a specific event type on the trace.
In our study, we treat the differential dip field aligned parallel 4) For each starting coordinate, integrate the flowline using the
with the seismic reflections (e 2) as a fluid velocity field V(x,y), RK4 method with step size h.
which describes the local flow velocity vector at each point (x,y).
This resulting velocity field is static and does not change with For each step, calculate
time, as it is derived from the fixed amplitude image, making it
analogous to a steady flow. Flowlines are the curves whose tangents ​k​1​ = hv​(​x​n​, ​y​n​)​, (7)
are everywhere parallel to the fluid velocity field at all points

( 2)
(Chung, 2010). The velocity vector field in ℝ2 can be represented ​k​1x​ ​k​1y​
using the unit vectors along the x and y axes, given by î and ĵ, ​k​2​ = hv​ ​​x​n​+ _​ ​, ​y​n​+ _​ ​ ​, (8)
2
respectively. Thus, the velocity field is given by

( 2)
​k​2x​ ​k​2x​
V(x,y) = uî + vĵ, (2) ​k​3​ = hv​ ​​x​n​+ _​ ​, ​y​n​+ _​ ​ ​, (9)
2

where the velocity components in the x and y directions are given

by u = u(x,y) and v = v(x,y), respectively (Anderson, 1995). By ​k​4​ = hv​(​​x​n​+ ​k​3x​, ​y​n​+ ​k​3y​)​. (10)
placing an imaginary particle in this velocity field, we can trace
the flowline by tracking the particle’s movement, which will follow Update positions
the direction of the velocity vector at each point along its path, ​k​1x​+ 2 ​k​2x​+ 2 ​k​3x​+ ​k​4x​
effectively mapping out the flowline. To characterize the geometry ​x​n+1​ = ​x​n​+ _________________
  
​   ​​, (11)
6

March 2025 The Leading Edge 215

 ​k​1y​+ 2 ​k​2y​+ 2 ​k​3y​+ ​k​4y​ excluding internal reflections that overlap based on a threshold.
​y​n+1​ = ​y​n​+ ________________
  ​   ​​. (12) This method ensures accurate identification of unconformities
6
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

across different systems in the seismic section, regardless of

5) Repeat step 4 until (xn+1,yn+1) is outside the boundary of the truncation levels and is outlined as follows:
seismic image.
6) Store the parametrized path of each flowline for further 1) Construct a heatmap matrix H(x,y) by accumulating the
analysis. number of times a flowline intersects the matrix at position
(xn,yn).
The main steps of the flowline extraction workflow are illus- 2) Let 𝒮 = {s1,s2 , …, sn } represent the list of surfaces.
trated in Figure 1, going from the seismic section (Figure 1a) to 3) Let O(si) represent the overlap score for surface si, which we
the flowline representation (Figure 1d). define as the mean of the accumulated heatmap values along
the surface:
Application and discussion 1
In the following sections, we will provide an overview of the ​O​(​s​i​)​= _​ ​​∑ H ​ ​(x, y)​, (13)
|​ ​s​i​|​​(x,y)​∈​s​​
developed workflows and address one of the key aspects of the i

flowlines that we believe contributes to their effectiveness with

respect to identifying unconformities and sequence boundaries. where |si| is the length of si.
We will also investigate how the number of seed points used to 4) Sort the list of surfaces, 𝒮, in descending order based on their
initialize the flowline integration impacts the accuracy and overall overlap score O(si).
outcomes of the results. 5) Define the set 𝒮selected to hold the selected surfaces, initialized
as 𝒮selected = Ø.
Unconformity and sequence boundary extraction 6) For each surface si in the sorted list, add si to 𝒮selected if the
DOI:10.1190/leedff.2025.44.issue-3

Each flowline initialization simulates the path of an imaginary overlap U(si,sj) between si and surface sj ∈ 𝒮selected does not exceed
particle within a steady fluid velocity field (Figure 1d). The opacity the overlap threshold Uth:
of the paths reflects the frequency of
overlaps, with higher opacity indicating
more overlaps. These overlapping paths
often correspond to major unconformi-
ties or sequence boundaries in a seismic
section, as unconformities tend to
truncate other surfaces, creating shared
paths. To quantify this, we introduce
the overlap score, a metric that mea-
sures how much a flowline overlaps
with others, helping us identify those
likely to represent unconformities or
sequence boundaries.
However, relying solely on the
overlap score poses challenges. While
major unconformities or sequence
boundaries typically score highest in
highly truncated systems, internal
reflections within these systems also
produce high overlap scores. This can
overshadow less truncated systems,
where critical boundaries may be over-
looked due to lower scores being eclipsed
by those from more complex systems.
To resolve this, we propose a filter-
ing workflow. First, we calculate the
mean overlap score to emphasize the
most significant sequence boundaries
Figure 1. Illustrating the main steps of the flowline extraction workflow. (a) The input seismic section from the F3 Netherlands
or unconformities. Next, we apply a block. (b) Estimated dip field based on the orientation of the minimum eigenvector, e , in the seismic section. (c) The sampling
2
sorting and filtering process to retain process of the starting position of the flowlines. (d) The integrated flowlines for the seismic section were sampled on both
only the highest-scoring flowlines while peaks and troughs.

216 The Leading Edge March 2025

 the method can be seen in Figure 2.
We have plotted the seven highest-
scoring surfaces in Figure 2a. These
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

seem to coincide well with some of the

expected sequence boundaries in the
system. The 20 highest-scoring surfaces
are shown in Figure 2b, superimposed
on the seismic section.
The behavior of the flowlines in
relation to the seismic amplitudes
reveals interesting implications for
geologic interpretation. While not
directly dependent on seismic ampli-
tudes, flowlines are indirectly influenced
by them through the differential dip
field from which they are generated.
This relationship manifests in a note-
worthy tendency: the flowline paths
seem to trend toward the zero crossings
of the seismic signal, regardless of where
they are initialized.
We believe this stems from the
Figure 2. (a) The top seven scoring surfaces with respect to overlap score ordered from 1 to 7 superimposed on the seismic. (b) underlying principles of the method. By
DOI:10.1190/leedff.2025.44.issue-3

The top 20 highest-scoring surfaces using an overlap threshold of Uth = 0.7. following the direction of the minimum
gradient of the structure tensor, which
summarizes local amplitude gradients, the flowlines naturally
gravitate toward areas of least amplitude variation. These areas
will typically be closer to the zero crossings in the seismic data.
Figure 3 demonstrates this, showing flowlines initialized on peak
amplitudes. It is evident that these flowlines frequently trace the
zero crossings or follow closely around them. The benefit of this
characteristic becomes clear when considering unconformities or
sequence boundaries, which are often characterized by lateral
amplitude changes due to abrupt shifts in deposition or erosion.
The inclination of flowlines to follow zero crossings and the
insensitivity to amplitude changes proves particularly advantageous
in this context, as it enables the capture of subtle and complex
changes associated with these features.
Another critical factor in the flowline generation process is
the number of seed points used to initiate the flowlines, as this
can significantly impact the results. To achieve an accurate flowline
Figure 3. Illustrating the tendency of flowlines to trace the zero crossing or follow closely representation of the seismic section, it is essential to generate a
around them even when initialized on peaks. The flowlines are plotted in red. sufficient number of flowlines, ensuring they are sampled from
across the entire image. For optimal accuracy, using all trace
​𝒮​selected​ = ​{​s​i​| U​(​s​i ​, ​s​j​)​ ≤ ​U​th ​∀ ​s​j​ ∈ ​𝒮​selected​}​, (14) locations as seed points would be ideal, but this is computationally
expensive, especially during downstream sorting and filtering
tasks. Instead, we employ a fixed-size increment for selecting seed
where U(si ,sj) is defined as the fraction of overlapping points points, balancing accuracy with efficiency. The impact of this
between the surfaces si and sj , ranging from 0 to 1. approach is shown in Figure 4, which demonstrates the results
7) Continue the selection process until all valid surfaces have of integrating flowlines sampled at every 20th, 50th, and 100th trace
been added to 𝒮selected . (representing 6%, 3%, and 2% trace sampling, respectively). Larger,
laterally extensive features like unconformities or sequence bound-
The sorting of the surfaces before filtering ensures that we aries are reliably captured at all increments, while smaller features,
retain the highest-scoring surface within a truncation complex such as internal sequence reflections, are more sensitive to the
of flowlines, which often corresponds to the major unconformities trace sampling frequency. These suggest that varying the seed
or the sequence boundaries within each system. The result of increment based on the specific can be beneficial. A smaller

March 2025 The Leading Edge 217

 increment provides better resolution for detailed facies mapping, defined by the range between the uppermost and lowermost
while a larger increment is sufficient for capturing broader, regional flowlines assigned to a specific cluster. All intermediate flowlines
unconformities, reducing computational time without compromis- within this range can then be extracted and reclustered using
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

ing accuracy for major structures. K-means clustering. For example, Figure 6, shows the reclustering
of the light blue sequence (fourth from the top) from Figure 5
Identifying stratigraphic sequences into four distinct clusters. This reclustering enables more detailed
One of the main advantages of working with seismic data in analysis of the geologic variations within each sequence.
the flowline domain is its ability to preserve geologic relationships
over long distances, capturing long-range geometric dependencies Tracking conformable surfaces
and regional context. This preservation helps in better understand- A key aspect of seismic interpretation is tracking conformable
ing the geometry of seismic features and makes it easier to group surfaces, represented by continuous reflections. Flowlines excel
flowlines into geologic units using
simple clustering methods. Each flow-
line is represented as a feature vector,
allowing for effective classification and
analysis of geologic structures. One way
to do this is by using the coordinates
along the flowline’s path, which capture
long-range geometric information in
the data. The feature vector can be based
on the x, y, or both the x and y coordi-
nates along each flowline. Because
flowlines can have different lengths, we
DOI:10.1190/leedff.2025.44.issue-3

extend (or pad) each flowline so that

they all match the length of the longest
one in the data set. Figure 5 shows how
the flowlines are divided into eight
clusters using K-means clustering, where
only the y coordinate along each flowline
is used as the feature in the feature vec-
tor. These clusters correspond closely
with the expected stratigraphic
sequences in the section. Once the
seismic sequences are identified, the
cluster boundaries can be used to extract
the sequence boundaries.
Additionally, each identified seis- Figure 4. A section of 100 traces showing the effect of the trace sampling frequency (every 20th, 50th, or 100th trace). The lateral
mic sequence can be further subdivided extensive seismic features are preserved in all realizations. However, the finer features such as the facies reflections require
into subsequences. Each sequence is higher sampling density to be preserved.

Figure 5. Visualization of the K-means clustered result of the extracted flowlines, segmented into eight distinct clusters, each represented by a different color. When coloring the flowlines
based on their assigned cluster, we obtain stratigraphic sequences that closely correspond to expected stratigraphic units in the seismic section.

218 The Leading Edge March 2025

 at identifying unconformities by focusing on geometric structure After extraction, we process the flowlines to create a single
and being relatively insensitive to amplitude variations. This flowline per reflector, similar to unconformities. A well-tracked peak
insensitivity helps flowlines gravitate toward zero crossings, reflection should follow the maximum peak value and remain con-
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

which effectively highlights unconformities regardless of lateral tinuous. Our selection is based on two criteria: reflection length and
amplitude changes. amplitude. Our process begins by removing shorter flowlines when
However, this same strength in tracking unconformities they overlap with longer ones, keeping the longest representation.
presents a challenge when attempting to follow conformable We then sort the remaining flowlines by descending mean amplitude.
surfaces. To address this, we introduce an amplitude likeness Finally, overlapping sections are removed from lower-amplitude
metric, α, which is the relative difference between the amplitude flowlines to preserve the highest-amplitude representations.
values between two consecutive points along a flowline: The result of this process is a set of flowlines that aim to
represent each reflector in the seismic image. These flowlines are
designed to follow peak values, maintain continuity along the
  
​
|​ ​(​A​i​− ​A​i−1​)|​ ​ ​.
​α​(i)​ = 1 − _____________
   (15) reflection, and balance both length and amplitude considerations.
​|​(​A​i​)​|​+ ​|​(​A​i−1​)​| ​
Figure 7 illustrates the seismic section with conformable reflections
tracked using our proposed method. The tracking was performed
This metric serves as a stop criterion for flowline integration, on the peak event with an amplitude likeness threshold of αth = 0.3.
where the flowline stops if the amplitude likeness, α, falls below This example suggests that the flowline methodology can be used
a certain threshold αth. This ensures that each point added to the to reasonably track continuous reflections in the seismic data as
flowline does not differ significantly from the event type we aim parameterized line segments in 2D.
to track. To improve flowline integration, we calculate a new
differential dip field based solely on the specific event type we Toward 3D
aim to track. For instance, if tracking peaks, we first mask out Transitioning from 2D to 3D flow analysis involves significant
the troughs from the seismic image before deriving the dip field. complexities due to the added dimension, introducing extra
DOI:10.1190/leedff.2025.44.issue-3

Figure 6. Visualization of further segmentation of the light blue (fourth from the top) in Figure 5 using K-means clustering, segmented into four distinct clusters.

Figure 7. Illustrating the tracking of conformable surfaces by including the amplitude threshold, α th = 0.3, as a constraint for the flowline integration process.

March 2025 The Leading Edge 219

 variables and more intricate interactions within the flow. This Figure 8 illustrates the surface obtained by extending an initial
complexity increases the computational burden and challenges in flowline using the 2.5D method. This surface was processed by
numerical integration. While our current workflow is not suffi- importing the surface points into the OpenMind interpretation
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

ciently robust to be directly applied to 3D seismic surface identifica- software, where basic postprocessing techniques, including
tion, we aim to demonstrate the feasibility of extending it through smoothing and gap interpolation, were applied. This demonstrates
a simplified pseudo-3D approach, which we term “2.5D.” This that the proposed pseudo-3D method can produce a reasonable
method retains the fundamental aspects of 2D analysis while approximation of the expected stratigraphic surface in 3D.
allowing us to extract unconformities in 3D. We regard a 3D The advantage of this method lies in its simplicity and effi-
surface as a collection of 2D flowlines that track the same structure ciency, providing a practical approach for achieving a 3D extension
in both the inline (IL) and crossline (XL) direction. We call this from our initial 2D framework. However, one of the complexities
collection 𝒮. First, we extract flowlines in 2D along an initial ILi, inherent in this method is related to the variability in flowline
using the previously outlined unconformity extraction workflow tracking across separate 2D inline slices. There is no guarantee
and select the 2D surface we wish to track in 3D to form 𝒮. that the tracked flowlines in different slices will belong to the
The points along surface, 𝒮i , are used as seed points for same surface even when using the seeding strategy proposed in
extracting flowlines in the crossline direction. We take a step our 2.5D method. The seeding strategy in our method only
∆ IL in the inline direction and use the positions where the recently increases the likelihood that they will align closely enough to
tracked crossline flowlines intersect ILi+∆IL . Each seed point in provide a coherent surface representation.
ILi+∆IL gives rise to a set of candidate flowlines, 𝒞i+∆IL . This step Despite this limitation, the pseudo-3D method offers valuable
is crucial because the highest-scoring flowline regarding overlap insights into the subsurface structure. While it does not capture
between two inline slices does not necessarily correspond to the the full complexity of the real 3D geology, the results are sufficient
same surface. By initializing the starting seed points from our to enhance our understanding of stratigraphic relationships and
previous surface, we maintain the context of the 3D structure, the overall subsurface structure.
maximizing continuity and coherence in the constructed surface.
DOI:10.1190/leedff.2025.44.issue-3

This approach helps to minimize misalignment and inaccuracies, Application on different data sets
especially in complex geologic formations. From the candidate In this section, we will demonstrate the applications of the
flowlines, we select the flowline that achieves the highest overlap flowline workflow on two different examples. Both examples are
score in the heatmap calculated in ILi+∆IL . This process is repeated taken from modern broadband 3D seismic data sets, covering the
for every inline, and the coordinates along the flowlines in 𝒮 are southern part of the Loppa High in the southwestern Barents
written to file. Sea (Vinje et al., 2017). First, we demonstrate the extraction of

Figure 8. The 3D surface extended from a selected 2D flowline in the F3 Netherlands block. The flowline is initialized on IL = 440 and we employ a step size of ∆ IL = 5. A vertical time shift of
up to 10 samples is allowed between neighboring points. The surface is smoothed, and gaps are filled using OpenMind’s interpretation workflows.

220 The Leading Edge March 2025

 first-order stratigraphic units in a relatively simple tectonic setting, of the tilted basement by initializing flow only from the seismic
followed by addressing the challenges that arise in more complex samples located above the basement and ensuring no flow extends
tectonic environments. into the basement.
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Example 1. The section in Figure 9a shows the part of the After generating the flowline representation of the seismic
Loppa High that crosses the Alta discovery, showing a tilted data, we resample each flowline to match the length of the longest
basement block overlain by eastward-thickening stratigraphic flowline in the data set, using the y-coordinates as features.
sequences of Upper Permian and Lower Triassic age (Lie et al., Figure 9b displays the stratigraphic units identified through
2018). To extract these stratigraphic sequences, we apply the K-means clustering, initialized with six clusters, which seem to
previously described clustering workflow. Because our aim is to correspond reasonably well with the expected first-order strati-
focus on the stratigraphic sequences, we minimize the influence graphic units in the section.
DOI:10.1190/leedff.2025.44.issue-3

Figure 9. (a) Seismic section over the Alta discovery in the southwestern Barents Sea. (b) The same section with K-means clustered flowlines superimposed, representing the first-order
stratigraphic units.

Figure 10. A seismic section, showing a more complex tectonic setting from the southwestern Barents Sea. (a) The flowline representation of the seismic section, where the flowlines
are terminated using fault likelihood. Five of the reconnected flowlines derived from our method are superimposed on (a) the flowline representation and on (b) the seismic section. The
lowermost unconformity above the faulted region is marked with a thick red line at the top.

March 2025 The Leading Edge 221

 Example 2. One of the primary challenges in working with that do not align with the seismic stratigraphy. To resolve this,
flowline representation is dealing with faulting. Typically, we we terminate flowlines at fault boundaries using fault likelihood
assume that the dip field aligns with the stratigraphy, allowing volumes. We then reconnect the terminated flowlines using
flowlines to follow continuous depositional patterns. However, dynamic time warping to restore regional context and lateral
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

when tectonic activity introduces faults, this assumption no longer continuity across faults. This method shows that the flowline
holds, leading to arbitrary and erratic flowlines that fail to represent workflow can be effectively adapted for more tectonically complex
meaningful stratigraphic information. To address this challenge, geologic settings.
a simple yet effective solution is to terminate flowlines at fault
boundaries. This can be achieved by utilizing a fault likelihood Acknowledgments
volume (Hale, 2013), which identifies regions where faults are The Netherlands F3 seismic data are provided by dGB Earth
most likely to occur. By stopping flowlines at these boundaries, Sciences B.V. through OpendTect. D. Adelved would like to thank
we ensure that they do not extend across fault planes, thus prevent- AkerBP Norway, the Norwegian Research Council, and
ing distortions in the geologic interpretation. ConocoPhillips Norway for funding and contributing to the
While this method yields a more accurate representation of ongoing PhD project. Furthermore, D. Adelved would like to
the seismic stratigraphy in the flowline domain, it has the drawback thank AkerBP Norway for providing the Barents Sea data sets.
of losing lateral continuity across faulted strata, which limits the Finally, we would like to thank GeoMind for providing the
ability to maintain the regional context of the flowlines. We licensed OpenMind software used in parts of this work.
address this issue by using dynamic time warping to generate a
correlation section, similar to the approach proposed by Bugge Data and materials availability
et al. (2019), allowing us to reestablish the lateral continuity of The code related to the open-source data set in this research
the flowlines across the faults. will be made publicly available under an open-source licence at
Figure 10 presents the results of our workflow applied to a https://github.com/adelved/seismic-flow.
more tectonically complex setting, located farther west of the
DOI:10.1190/leedff.2025.44.issue-3

previous example. Both the flowline domain (Figure 10a) and the Corresponding author: [email protected]
seismic section (Figure 10b) are shown. In both figures, the
lowermost continuous flowline above the faulted reflectors is References
highlighted in red, with a set of reconnected flowlines superim- Anderson, J. D., 1995, Computational fluid dynamics: The basics with
posed. These reconnected flowlines align well with the faulted applications: McGraw-Hill.
Badley, M. E., 1985, Practical seismic interpretation: International
stratigraphy in Figure 10b, suggesting that our approach can be
Human Resources Development Corporation.
used to maintain the effectiveness of flowline representation even Bahorich, M., and S. Farmer, 1995, 3-D seismic discontinuity for faults
in complex stratigraphic settings. and stratigraphic features: The coherence cube: The Leading Edge,
14, no. 10, 1053–1058, https://doi.org/10.1190/1.1437077.
Conclusions Bakker, P., 2002, Image structure analysis for seismic interpretation:
In this study, we explored the potential of working with PhD thesis, Delft University of Technology, https://repository.
seismic images in the flowline domain, which offers several tudelft.nl/islandora/object/uuid%3Af3f60507-5b08-43cb-a309-
c3446c972395, accessed 19 March 2024.
advantages for geologic interpretation by capturing long-range
Barnes, A. E., 2000, Attributes for automating seismic facies analysis:
geometric dependencies and building regional context. Flowlines 70 th Annual International Meeting, SEG, Expanded Abstracts,
naturally converge near sequence boundaries and unconformities, 553–556, https://doi.org/10.1190/1.1816121.
facilitating their identification, ranking, and extraction based on Bugge, A. J., J. E. Lie, A. K. Evensen, J. I. Faleide, and S. Clark, 2019,
overlap with other flowlines. This characteristic, combined with Automatic extraction of dislocated horizons from 3D seismic data
their insensitivity to amplitude variations, makes flowlines par- using nonlocal trace matching: Geophysics, 84, no. 6, IM77–IM86,
ticularly effective for tracking unconformities or sequence bound- https://doi.org/10.1190/geo2019-0029.1.
Butcher, J. C., 2016, Numerical methods for ordinary differential equa-
aries, which often exhibit lateral changes in amplitude due to
tions, third edition: Wiley.
shifts in deposition or erosion. Chung, T. J., 2010, Computational fluid dynamics, second edition:
The regional context provided by the flowline domain also Cambridge University Press.
lends itself well to machine learning applications, such as cluster- de Bruin, G., H. Ligtenberg, N. Hemstra, and K. Tingdahl, 2006,
ing, allowing for the efficient grouping of flowlines by similarity Synchronized sequence stratigraphic interpretation in the structural
and the extraction of seismic sequences and subsequences. We and chrono-stratigraphic (Wheeler transformed) domain: Research
also introduced a method for tracking conformable surfaces by Workshop — From Seismic Interpretation to Stratigraphic and
Basin Modelling, Present and Future, EAGE, https://doi.
using the relative amplitude change along the flowlines to steer
org/10.3997/2214-4609.201403022.
their paths to better align with the conformable reflections. Eckersley, A. J., J. Lowell, and P. Szafian, 2018, High-definition frequency
While the extension of flowlines into 3D is still ongoing, our decomposition: Geophysical Prospecting, 66, no. 6, 1138–1143,
pseudo-3D approach shows promise for preliminary geologic https://doi.org/10.1111/1365-2478.12642.
analysis of well-defined surfaces. Hale, D., 2013, Methods to compute fault images, extract fault surfaces,
Faulting remains a significant challenge in the flowline domain and estimate fault throws from 3D seismic images: Geophysics, 78,
because following the dip in faulted areas often results in flowlines no. 2, O33–O43, https://doi.org/10.1190/geo2012-0331.1.

222 The Leading Edge March 2025

 Henderson, J., S. J. Purves, and C. Leppard, 2007, Automated delineation Vail, P. R., and R. M. Mitchum Jr., 1977, Seismic stratigraphy and global
of geological elements from 3D seismic data through analysis of multi- changes of sea level, part 1: Overview, in C. E. Payton, ed., Seismic
channel, volumetric spectral decomposition data: First Break, 25, no. 3, stratigraphy — Applications to hydrocarbon exploration: AAPG,
87–93, https://doi.org/10.3997/1365-2397.25.1105.27383. https://doi.org/10.1306/M26490C3.
Lie, J. E., V. Danielsen, P. E. Dhelie, R. Sablon, R. Siliqi, C. Grubb, van Hoek, T., S. Gesbert, and J. Pickens, 2010, Geometric attributes for
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

V. Vinje, C. I. Nilsen, and R. Soubaras, 2018, A novel source-over- seismic stratigraphic interpretation: The Leading Edge, 29, no. 9,
cable solution to address the Barents Sea imaging challenges: Marine 1056–1065, https://doi.org/10.1190/1.3485766.
Acquisition Workshop, EAGE, https://doi.org/10.3997/​ Vevle, M. L., I. Aarnes, K. Ledsaak, R. Hauge, and A. Skorstad, 2018,
2214-4609.201802101. Facies modelling of a real-life fluvial system using a modern object-
Lomask, J., A. Guitton, S. Fomel, J. Claerbout, and A. A. Valenciano, based algorithm: 80th Conference and Exhibition, EAGE, Extended
2006, Flattening without picking: Geophysics, 71, no. 4, P13–P20, Abstracts, https://doi.org/10.3997/2214-4609.201800792.
https://doi.org/10.1190/1.2210848. Vinje, V., J. E. Lie, V. Danielsen, P. E. Dhelle, R. Silliqi, C.-I. Nilsen, E.
Marroquín, I. D., 2014, A knowledge-integration framework for inter- Hicks, and A. Camerer, 2017, Shooting over the seismic spread: First
preting seismic facies: Interpretation, 2, no. 1, SA1–SA9, https://doi. Break, 35, no. 6, https://doi.org/10.3997/1365-2397.35.6.89461.
org/10.1190/INT-2013-0057.1. Williams, R. M., 2018, Derisking the Thebe discovery through cognitive
Mitchum, R. M., Jr., P. R. Vail, and J. B. Sangree, 1977, Seismic stra- interpretation: First Break, 36, no. 3, 71–78, https://doi.org/​10.3997/1365-
tigraphy and global changes of sea level, part 6: Stratigraphic inter- 2397.n0079.
pretation of seismic reflection patterns in depositional sequences, in Wu, X., and S. Fomel, 2018, Least-squares horizons with local slopes
C. E. Payton, ed., Seismic stratigraphy — Applications to hydrocarbon and multigrid correlations: Geophysics, 83, no. 4, IM29–IM40,
exploration: AAPG, https://doi.org/10.1306/M26490C8. https://doi.org/10.1190/geo2017-0830.1.
Pauget, F., S. Lacaze, and T. Valding, 2009, A global approach in seismic Wu, X., and D. Hale, 2013, Extracting horizons and sequence boundaries
interpretation based on cost function minimization: 79th Annual from 3D seismic images: 83rd Annual International Meeting, SEG,
International Meeting, SEG, Expanded Abstracts, 2592–2596, Expanded Abstracts, 1440–1445; https://doi.org/10.1190/
https://doi.org/10.1190/1.3255384. segam2013-0296.1.
Stark, T. J., 2003, Unwrapping instantaneous phase to generate a Wu, X., and D. Hale, 2016, Automatically interpreting all faults, uncon-
relative geologic time volume: 73rd Annual International Meeting, formities, and horizons from 3D seismic images: Interpretation, 4,
SEG, E x pa nded Abst rac ts, 1707–1710, ht t ps: //doi. no. 2, T227–T237, https://doi.org/10.1190/INT-2015-0160.1.
DOI:10.1190/leedff.2025.44.issue-3

org/10.1190/1.1844072. Wu, X., and G. Zhong, 2012, A method for generating a seismic wheeler
Stark, T. J., 2005, Generation of a 3D seismic “Wheeler diagram” from a volume via a relative geologic time volume: 82nd Annual International
high resolution age volume: 75th Annual International Meeting, SEG, Meeting, SEG, Expanded Abstracts, https://doi.org/10.1190/
Expanded Abstracts, 782–785, https://doi.org/10.1190/1.2148275. segam2012-1177.1.

Koichi Ha
yashi

Phil Sirles
Xavier Comas

In one photo and one sentence …

share your best field or laboratory experience with us!
Showcase your photography skills for a chance to win a prize at the International Meeting for
Applied Geoscience and Energy (IMAGE), taking place 25–28 August in Houston, Texas, at
the George R. Brown Convention Center. Submit your photos by 31 March 2025. Winners will
be announced at the Near Surface Technical Section (NSTS) Reception. This is a unique
opportunity to have your work recognized by leading professionals in the field of geoscience
and energy.
go.seg.org/nsphoto

March 2025 The Leading Edge 223

 A graphical approach to determine the relationship between
intercept, gradient, and the common seismic rock properties:
Global model and application
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

David Went1
https://doi.org/10.1190/tle44030224.1

Abstract can be used to generate other rock properties (e.g., Young’s

Globally applicable forward amplitude variation with offset modulus) or to provide a more bespoke reservoir characterization,
models of sand-shale systems are displayed using the intercept- where suitable log data are available.
gradient crossplot, color coded by the magnitude of reflectivity
of the common seismic rock properties: bulk modulus, shear Rock properties and angle-dependent seismic reflectivity
modulus, Lamé’s parameter lambda, Poisson’s ratio, and VP /VS . Most geoscientists working in the subsurface with seismic
Lines of equal magnitude are threaded through the data to data will be aware that contrasts in impedance are the cause of
establish the angular relationship of each rock property to seismic reflections. When building a normal incidence synthetic
intercept and gradient. In many cases, these generic models will seismogram, the sonic and density logs are used to calculate
be sufficient to design a seismic inversion capable of distinguish- acoustic impedance. Changes in acoustic impedance at reflecting
ing lithology and fluid. An inverted seismic section over a interfaces result in either a positive or negative reflection coefficient
producing field illustrates how fluid-sensitive (rEEIχ27) and (spikes), the magnitude of which is defined by equation 1:
fluid-insensitive (shear-impedance) attributes conform closely
to the generic model. Where suitable data exist, the modeling I 2 − I1 /(I 1 + I 2), (1)
DOI:10.1190/leedff.2025.44.issue-3

process can be modified to provide a more tailored reservoir

characterization and inversion routine. where I 1 is the impedance of the incident rock, and I 2 is the
impedance of the reflecting rock.
Introduction When a reflection coefficient series is convolved with a wavelet,
Analysis of amplitude variation with offset (AVO) or amplitude it yields a synthetic seismic trace. The earlier acoustic-impedance-
variation with angle (AVA) is an integral part of the prospecting generated seismogram is only strictly valid at zero offset. With
process in many plays globally. It is also key in reservoir charac- increasing incidence angle, contrasting lithologies shear at different
terization as part of field development and for monitoring produc- rates, which necessitates contrasts in shear velocity to be added
tion. It typically involves measurement of the amplitude at zero into the equation. Hence, reflectivity becomes defined by contrasts
offset (termed the intercept [I ]) and the rate at which amplitude in elastic (angle-dependent) impedance rather than acoustic
changes with incidence angle (termed the gradient [G]), using impedance (Mukerji et al., 1998; Connolly, 1999; Whitcombe et al.,
either seismic gathers or angle stacks. The most common reason 2002). The rate at which P-wave reflection coefficient changes
for doing it is to determine fluid type (hydrocarbon versus brine) with offset is described by the Zoeppritz equation, or more practi-
and to discriminate the effects of fluid, lithology, and porosity. cally, one of the approximations to it (most commonly the Shuey
Certain rock properties are commonly noted to be key to establish- equation) (Zoeppritz, 1919; Shuey, 1985). To build a forward
ing these effects. For example, the shear modulus (µ), by definition model of angle-dependent reflectivity, VP, VS , and density are
in Gassmann’s law (Gassmann, 1951; Smith et al., 2003), is required for the incident and reflecting rocks. Generic global
insensitive to fluid type. Shear impedance is closely related to models of angle-dependent reflectivity for sandstones and shales
shear modulus and also displays a very low sensitivity to fluid using the Shuey equation were presented by Went (2021). These
type. On the other hand, acoustic impedance, bulk modulus, models illustrate a full range of possible outcomes for intercept
Lamé’s parameter lambda (λ), Poisson’s ratio (σ), and VP /VS are and gradient based on geologically feasible combinations of VP,
far more sensitive to fluid changes. The equivalence of the common VS , and density. Models are depth dependent and may be cor-
seismic rock properties to elastic or extended elastic impedance roborated by other global studies (e.g., Mur and Vernik, 2019).
at different angles has been made previously, though typically
more as an aside than as a focus of study (e.g., Mallick, 2001; Method
Whitcombe et al., 2002). The purpose of this paper is to describe The most common rock properties, their common abbrevia-
a simple graphical method to establish the relationship between tions, and formulae (different combinations of VP, VS , and density)
the common rock properties and intercept and gradient, and to are listed in Table 1. The VP, VS , and density data used by Went
show how they can be generated from seismic data. The method (2021) to generate globally applicable intercept and gradient results

Manuscript submitted 6 August 2024; accepted 3 October 2024.

1
TGS Geophysical ASA, Weybridge, UK. E-mail: [email protected].

224 The Leading Edge March 2025

 for a depth of 1750 m below sea bed in a normally compacting combinations of VP, VS , and density, they are, in essence, different
basin are shown in Table 2. The resulting model is displayed in forms of elastic impedance. The reflection coefficient for each of
Figure 1. The same data are used to generate the common seismic the main rock properties (e.g., shear modulus for shale 1 over
rock properties, acoustic impedance, shear impedance, bulk sand 1) can be calculated using equation 1, in the same way as for
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

modulus, Lamé’s parameter lambda, shear modulus, Poisson’s any other impedance contrast. The results for each of the seismic
ratio, and VP /VS . These are also displayed in Table 2. Because the rock properties in each of the cases (shale 1 over sand 1, shale 1
common rock properties are each described by different over sand 2, etc.) are shown in the lower part of Table 2. Gassmann
fluid substitution was used to generate modified VP, VS , and density
Table 1. Seismic rock properties, their common abbreviations, and equations used to values (not shown) in the presence of hydrocarbons, using the
generate them from well logs and seismic data.
following in-situ fluid properties: oil density (0.7), bulk modulus
Attribute Abbreviation Equation (0.85), water density (0.98), and bulk modulus (2.57). These
VP VP
modified values are used to generate each hydrocarbon case
Velocity
reflection coefficient.
Shear velocity VS VS The reflection coefficient data for each seismic rock property
Density ρ ρ were sorted by magnitude. The intercept-gradient crossplot data
Acoustic impedance AI ρVP were then color coded by the magnitude of each seismic rock
property to enable the angular relationships to be visualized,
Shear impedance SI ρVS
determined, and compared.
Bulk modulus K ρVP 2 − 4/3μ
Shear modulus μ ρVS 2 Results
Lamé’s lambda λ ρVP 2 − 2μ Displays of crossplots of intercept versus gradient, color coded
by the reflectivity of the evaluated seismic rock properties, are shown
Poisson’s ratio σ VP − 2VS /2( VP − VS )
2 2 2 2
in Figures 2 and 3. Each intercept-gradient crossplot is shown twice:
VP / VS VP / VS VP / VS (1) with a clear view of the data color coded for the attribute and (2)
DOI:10.1190/leedff.2025.44.issue-3

Table 2. Global model data for 1750 m TVDBSB (after Went, 2021). Rock properties are calculated from the model data. Rock property reflectivity is prefixed R_ and is for shale 1 over
shale 2, sand 1, sand 2, sand 3, and sand 4. HC denote a hydrocarbon case.

Input data Shale 0 Shale 1 Shale 2 Org. Sh Sand 1 Sand 2 Sand 3 Sand 4
VP (km/s) 2.700 3.048 3.260 2.700 3.672 3.300 3.048 2.800
VS (km/s) 1.212 1.480 1.643 1.420 2.097 1.798 1.595 1.396
Density (g/cc) 2.27 2.35 2.40 2.22 2.32 2.25 2.20 2.16
AI (GPa) 6.14 7.17 7.81 5.99 8.51 7.43 6.72 6.04
SI (GPa) 2.76 3.48 3.94 3.15 4.86 4.05 3.52 3.01
VP / VS 2.23 2.06 1.98 1.90 1.75 1.84 1.91 2.01
Poisson’s ratio σ 0.37 0.35 0.33 0.31 0.26 0.29 0.31 0.33
Bulk modulus K (GPa) 12.13 14.99 16.85 10.22 17.66 14.81 13.00 11.30
Lambda λ (GPa) 9.90 11.56 12.53 7.23 10.87 9.96 9.26 8.50
Shear modulus μ (GPa) 3.34 5.15 6.47 4.48 10.20 7.28 5.61 4.20
Porosity ø (%) 20% 24% 27% 30%
R_VP / VS –0.039 –0.019 –0.040 –0.081 –0.058 –0.038 –0.013
R_VP / VS HC –0.105 –0.093 –0.081 –0.067
R_σ –0.039 –0.024 –0.057 –0.146 –0.090 –0.053 –0.017
R_σ HC –0.227 –0.178 –0.144 –0.114
R_SI 0.116 0.062 –0.049 0.166 0.075 0.005 –0.073
R_SI HC 0.163 0.071 –0.002 –0.080
R_μ 0.213 0.113 –0.070 0.329 0.171 0.043 –0.101
R_K 0.105 0.058 –0.189 0.082 –0.074 –0.071 –0.140
R_K HC –0.009 –0.124 –0.212 –0.306
R_Lambda 0.077 0.041 –0.230 –0.031 –0.074 –0.110 –0.152
R_Lambda HC –0.187 –0.258 –0.316 –0.383

March 2025 The Leading Edge 225

 with dashed “contour” lines separating seismic rock properties of A color-coded display of acoustic-impedance reflectivity is
different magnitudes. These straight lines are threaded between the not really needed because it is the same as intercept reflectivity.
data points and clearly illustrate the angle of rotation (chi [χ]) on the In other words, the color coding would reveal vertical lines of
intercept-gradient crossplot the seismic rock property represents. equal magnitude, confirming acoustic impedance is the same as
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Figure 1. (a) Angle-dependent reflectivity plot of sin2θ versus amplitude or reflection coefficient showing intercept and gradient for the cases in Table 2 extrapolated to 90°. (b) The same
data plotted as an intercept versus gradient crossplot color coded for shale over rock type. Global model for 1750 m TVDBSB (after Went, 2021).
DOI:10.1190/leedff.2025.44.issue-3

Figure 2. Intercept-gradient crossplots color coded by moduli reflectivity: (a) and (b) Figure 3. Intercept-gradient crossplots color coded by moduli reflectivity: (a) and (b)
bulk modulus, (c) and (d) λ, and (e) and (f) VP /VS . Plots (b), (d), and (f) show dashed lines Poisson’s ratio (σ), (c) and (d) shear impedance, and (e) and (f) shear modulus (µ). Plots
separating moduli reflectivity of different magnitude. The lines define the χ angle. (b), (d), and (f) show dashed lines separating moduli reflectivity of different magnitude. The
lines define the χ angle.

226 The Leading Edge March 2025

 intercept reflectivity, with an increasingly negative reflection
coefficient displayed to the left on the x-axis of the crossplot.
The color-coded display of bulk modulus reflectivity allows
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

the threading of lines of equal reflection coefficient that are inclined

at a crossplot rotation angle (χ) of 10° from the vertical (Figures 2a
and 2b). A similar plot but with lines of equal reflection coefficient
defining a rotation angle of 19° defines Lamé’s parameter lambda
(Figures 2c and 2d).
The color-coded display of VP /VS reflectivity shows lines of
equal reflectivity substantially rotated, defining a crossplot rotation
angle of 45° (Figures 2e and 2f). That is, it has equal parts of
intercept and gradient defining it. Poisson’s ratio reflectivity is
also defined by lines of equal reflectivity, defining a rotation angle
of 45° (Figures 3a and 3b). This is not surprising because Poisson’s
ratio is a monotonic function of VP /VS (Table 1).
The color-coded display of shear-impedance reflectivity shows
lines of equal reflectivity that run orthogonal to Poisson’s ratio
reflectivity, defining an attribute oriented at a crossplot (χ) rotation
angle of –45° (Figures 3c and 3d). A closely related attribute,
shear modulus, shows color-coded data that define a rotation
Figure 4. Intercept-gradient crossplot color coded by lithology and fluid, showing angle (χ) of –51° (Figures 3e and 3f).
the relationship to the common rock property vectors as defined by the graphical The results for all the evaluated rock properties are summarized
determination. Global model for 1750 m TVDBSB. in Figure 4. The extended elastic impedance attribute EEIχ27 is
DOI:10.1190/leedff.2025.44.issue-3

also displayed as the attribute oriented

at the optimal angle to isolate fluid
effects. Lambda, Poisson’s ratio, and
VP /VS are oriented, broadly speaking, in
a similar direction and are therefore also
sensitive to fluid effects. Shear imped-
ance and shear modulus by contrast are
oriented in a near orthogonal direction
to the fluid-sensitive attributes and
should therefore not show a fluid effect
in models or seismic inversions.

Application
Although the results summarized
in Figure 4 were, strictly speaking,
determined for a burial depth of 1750 m,
they are practically valid over a greater
depth range because typical background
intercept and gradient relationships
change slowly from 1 to 3.5 km of burial
(Went, 2021). Figure 5 shows example
seismic inversions to test for fluid effects
over a proven producing oil field in the
North Sea at a depth of 1300 m. The
AVO inversions are displayed as relative
impedances and were generated follow-
ing the method described in Went et al.
(2023). The rEEIχ27 attribute (Figure 5a)
shows a fluid anomaly with low values
of rEEIχ27 corresponding to the proven
oil in the producing field (hydrocarbon
Figure 5. Seismic inversions for (a) rEEIχ27 showing the fluid effect of oil present in a producing North Sea field and (b) shear
impedance (rEEIχ−45) where no anomaly is present because shear impedance is insensitive to fluid type. The upper dashed effect). The shear-impedance (rEEIχ-45)
black line (arrowed) marks the top of the oil sands in the producing field, and the lower black dotted line marks the base. attribute (Figure 5b), on the other hand,

March 2025 The Leading Edge 227

 shows no such anomaly, consistent with
a rock property insensitive to fluid effects
and supporting the notion the rEEIχ27
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

anomaly is indeed a fluid effect caused

by an impedance-softening hydrocarbon
fluid. Seismic data crossplots of relative
acoustic impedance versus gradient
impedance, color coded by rEEIχ27
(f luid-sensitive impedance) and
rEEIχ-45 (shear impedance), further
confirm this conclusion (Figure 6) and
conform favorably with the crossplots
of the model data (Figures 1b, 2, and 3).
Additional field examples demonstrating
the applicability of this method are
presented in Went et al. (2025).

Conclusion
The common seismic rock properties
are each forms of angle-dependent
elastic impedance. Reflectivity resulting
from contrasts in the common rock
properties can be calculated in the same
DOI:10.1190/leedff.2025.44.issue-3

way as for any other impedance contrast.

Using previously justified globally appli-
cable model data, the angular relation-
ships of the common rock properties to
intercept and gradient are established
and displayed. The rock properties Figure 6. Seismic data crossplots of relative acoustic impedance versus gradient impedance: (a) background location
broadly group into two classes: (1) those color coded by rEEIχ27 (fluid-sensitive impedance), (b) background location color coded by rEEIχ−45 (shear impedance),
that are fluid sensitive (acoustic imped- (c) producing oil field color coded by rEEIχ27 (fluid-sensitive impedance), and (d) producing oil field color coded by rEEIχ−45
ance, bulk modulus, lambda, Poisson’s (shear impedance). The oil sands are represented by data in the southwest corner of plots (c) and (d) and are highlighted by
ratio, and VP /VS) and (2) those that are anomalously low values (yellow-green) of rEEIχ27 but nonanomalous mid- to high-range values (red-blue) of rEEIχ−45 (shear
impedance). Cubes of seismic data are approximately 6 km2 × 150 m/s.
insensitive to fluid effects (shear modu-
lus and shear impedance). Designs for
appropriate elastic inversions of seismic data can use the global Mallick, S., 2001, AVO and elastic impedance: The Leading Edge,
model presented earlier where log data are either absent or compli- 20, no. 10, 1094–1104, https://doi.org/10.1190/1.1487239.
ant with the model. Alternatively, the modeling process can be Mukerji, T., A. Jørstad, G. Mavko, and J. R. Granli, 1998, Near and
far offset impedances: Seismic attributes for identifying lithofacies
performed using local log data to provide a more tailored reservoir
and pore fluids: Geophysical Research Letters, 25, no. 24, 4557–
characterization and inversion routine. 4560, https://doi.org/10.1029/1998GL900187.
Mur, A., and L. Vernik, 2019, Testing popular rock-physics models:
Acknowledgments The Leading Edge, 38, no. 5, 350–357, https://doi.org/10.1190/
Thanks are due to the two anonymous reviewers for their tle38050350.1.
thoughtful comments and to TGS for permission to publish. Shuey, R. T., 1985, A simplification of the Zoeppritz equations:
Geophysics, 50, no. 4, 609–614, https://doi.org/10.1190/1.1441936.
Smith, T. M., C. H. Sondergeld, and C. S. Rai, 2003, Gassmann
Data and materials availability fluid substitutions: A tutorial: Geophysics, 68, no. 2, 430–440,
Data associated with this research are available and can be https://doi.org/10.1190/1.1567211.
obtained by contacting the corresponding author. Went, D., 2021, Practical application of global siliciclastic rock-
property trends to AVA interpretation in frontier basins: The
Corresponding author: [email protected] Leading Edge, 40, no. 6, 454–459, https://doi.org/10.1190/
tle40060454.1.
References Went, D., M. Bamford, J. Rogers, S. Brown, and G. Turner, 2025,
Connolly, P., 1999, Elastic impedance: The Leading Edge, 18, no. 4, Characterising hydrocarbon discoveries and prospects in the Tay
438–452, https://doi.org/10.1190/1.1438307. Sandstone using relative elastic inversion: Greater Pilot area,
Gassmann, F., 1951, Über die elastizität poröser medien: Vierteljahrsschrift Central North Sea: Proceedings of the First Energy Geoscience
der Naturforschenden Gesellschaft in Zürich, 96, 1–23. Conference, Geological Society, London.

228 The Leading Edge March 2025

 Went, D., R. Hedley, and J. Rogers, 2023, Screening for AVA
anomalies in siliciclastic basins: Testing a seismic inversion
method in the Mississippi Canyon, Gulf of Mexico: First Break,
41, no. 9, 75–81, https://doi.org/10.3997/1365-2397.f b2023076.
ON DEMAND
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Whitcombe, D. N., P. A. Connolly, R. L. Reagan, and T. C.

Redshaw, 2002, Extended elastic impedance for fluid and lithol-
ogy prediction: Geophysics, 67, no. 1, 63–67, https://doi.
org/10.1190/1.1451337.
Zoeppritz, K., 1919, Erdbebenwellen VII, VII b, Über Reflexion
und Durchgang seismischer Wellen durch Unstetigkeitsflächen:
UNLOCK THE POWER OF IMAGE ‘24
Nachrichten von der Königlichen Gesellschaft der Wissenschaften ON-DEMAND RECORDINGS
zu Göttingen.
Did you miss a session at IMAGE ‘24, or want
to revisit a favorite presentation? Now’s your
chance to dive deeper into the insights and
innovations that shaped the conference.

Your exclusive on-demand access to IMAGE ‘24

is here!

Watch and replay your favorite sessions from

the comfort of your home or office. This gated
content provides in-depth insights and valuable
resources that can help you advance your
career in geoscience.
DOI:10.1190/leedff.2025.44.issue-3

SCAN THE CODE AND

START WATCHING NOW!

Latin America Honorary Lecturer

MARIA ANGELA
CAPELLO
Rocking Sustainability: Geophysics’
Role in a Greener, Prosperous and
Socially-Conscious Latin America
8 April • 6 May • go.seg.org/hl_mac

supported by

March 2025 The Leading Edge 229

 The Leading Edge
Board Report
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

SEG Board of Directors actions from Approved the candidates for the 2025–2026 SEG Distinguished
September to December 2024 Instructor Short Course (DISC) instructor as recommended by
Approved minutes from Board meetings on 25 August 2024, the DISC Subcommittee.
25 September 2024, 24 October 2024, and 21 November 2024.
Approved the appointment of Antonis Giannopoulos to fill the
Accepted the July to October 2024 financial statements. vacant representative position in District 7.

Ratified amendments to the Constitution and Bylaws of the Approved the SEG 2025 operating budget as recommended by
American Geosciences Institute (AGI) approved on 1 June the SEG Finance Committee.
2023 via e-mail vote by the AGI Member Society Council.
Approved the SEG Foundation 2025 budget as recommended by
Approved updates to the Reviews Committee Policy and the SEG Foundation Board of Directors.
Procedures to reflect a new way to send review books to
reviewers.

Approved a request from the SEG Council to add the Council

Medal procedures to the SEG Policy and Procedures,
Section II.1.
DOI:10.1190/leedff.2025.44.issue-3

Approved the proposed updates to the Geoscientists without

Borders Program Committee procedures.

Approved the recommendation to formally decline the request

from AAPG and SPE to join as a cosponsoring organization
to the Water Resources Management System.

Approved appointment for Huyen Bui to serve on the SEG

Foundation Board of Directors for the 2025–2027 term.

Approved a reappointment for Nick Moldoveanu and new

appointments for Sissy Theisen and Matthias Imhof to serve
on the SEAM Board of Directors.

Approved the following student chapter petitions:

• Polytechnic University of Man Geophysical Society
• Federal University of Rio de Janeiro Geophysical Society

Approved implementation of Phase I and II of the Achievement

Tracking System as proposed.

Acknowledged that all members are in receipt of the SEG IRS

Forms 990 and 990-T for 2023.

230 The Leading Edge March 2025

 The Leading Edge
Announcements
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Nominees for Board of Directors

The following Active members have been selected by the SEG District 7
Committee on Nominations and have agreed to be candidates for Rob Hardy and Emin Sadikhov
the 2025 SEG Board of Directors election: District 8
Paolo Mazzucchelli and Veronica Pazzi
President-elect District 9
Rocco Detomo, Jr. and Susan Webb Elly Bogi Mulumbu and Diego Quiros
Second vice president District 10
Adriana Citlali Ramírez and Glenn Wilson Ritesh M. Joshi and Nimisha Vedanti
Vice President, Publications District 11
Alison Malcolm and Chester Weiss Yi Shen and Hua Wang
Director at large District 12
Patricia de Lugao and Paulo Johann Faizan Akasyah Ghazali and Takeshi Tsuji
Director at large
Aiman Bakhorji and Chao Wang Rules governing the election process are provided in the SEG
Election Campaign Policy: https://seg.org/About-SEG/Elections/
In addition to the Board of Directors election, all Districts Campaign-Policy.
are holding elections for one new representative. Nominees are Additional nominations by petition, signed by at least 40
as follows: Voting members and accompanied by the written consent of
DOI:10.1190/leedff.2025.44.issue-3

the candidate and a brief biography, ready for publication, may

District 1 be submitted prior to 15 April. The biographies of the Board
Robert Blanchard and Amit Padhi candidates and their position statements will be published on
District 2 the SEG website and in TLE. Ballots will be sent by 15 May
Phil Gallaway and Edith Miller to eligible voters who have a valid e-mail address on file. Voting
District 3 will end on 30 June with results announced mid-July. New
Stanislav Glubokovskikh and Dan Hollis Board members will begin their term at the end of IMAGE
District 4 2025. Elected District Representatives will serve a two-year
Irina Filina and Shannon Jeffries term beginning 1 August 2025. Active, Emeritus, Honorary,
District 5 Life, and Associate members are eligible to vote on all matters
Matthew van den Berghe submitted to the membership; however members who have not
District 6 paid 2025 dues by 1 May will not be eligible to vote in the
Teresa Santana and Juan Soldo 2025 election.

SYMPOSIUM This event unites industry leaders to explore Egypt’s

evolving energy landscape. With hydrocarbons remaining
Leveraging Technologies key, focus areas include unlocking frontier basins,
to Unlock Egypt’s optimizing production, and attracting investment.
Energy Potential Experts will address technical challenges in the Western
28–30 Apr ‘25 • New Cairo, Egypt Desert, Nile Delta, Gulf of Suez, and Red Sea, highlighting
new technologies, digital transformation, and lessons
Register by 27 Mar and save from analogous basins. Through collaboration and
knowledge sharing, the symposium aims to drive
sustainable exploration, efficiency, and support Egypt’s
2030 energy strategy.
go.seg.org/mega25

March 2025 The Leading Edge 231

 Reviews
C o o r d i n at e d by Julie Aitken
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Energy and Civilization: A History, by Vaclav Smil, ISBN 978-0-262- The book is crammed with facts such as the food energy in a
33831-8, 2017, The MIT Press, 568 p., US$39.95 (hardcover), whale (80,000–800,000 megajoules), the power output of buffaloes
$22.99 (e-book). (250–550 watts), how much land was required to support a tra-
ditional city of 500,000 people (150,000 hectares), and percussive

S everal decades ago, I became interested in how social evolution

might be related to energy use. It did not take long for me to
discover this was an enormous topic, the understanding of which
drilling was invented by the Chinese more than 2000 years ago.
The text is supported by excellent graphs, charts, and images.
Numerous boxes discuss topics of particular interest.
required research far beyond my casual level of interest. When I Perhaps I am just old or lazy, but I struggled sometimes to
saw a reference to this book, I offered to review it for the Society. fight my way through the detail and grasp the author’s main
According to the author’s website, Dr. Smil has published 47 points. Similar books I have reviewed made this easier by includ-
books in the fields of energy, environmental and population ing an introduction to each chapter, telling the reader what the
change, food production, history of technical innovation, risk main points are, and a summary at the end of each chapter
assessment, and public policy. This book is an update of his 1994 reminding the reader what those points were. I reviewed an
book, Energy in World History. e-book version, and this may have affected how I assimilated the
The book has seven chapters. The first chapter gives back- information. I find e-books excellent for reading novels on
ground to the book and defines terms relevant throughout the airplanes, but not so good for reading nonfiction books filled
subsequent chapters: energy, power, energy density, power density, with information.
etc. The next two chapters focus on the links between energy use The primary message I got from this book is that energy is
and the evolution of humans and the development of tools, in necessary but not sufficient for development; i.e., understanding
DOI:10.1190/leedff.2025.44.issue-3

particular advances in agriculture. The fourth chapter discusses a society’s access to and use of energy does not by itself allow the
the uses and limits of traditional energy sources — muscles, wind, prediction of that society’s development. Societies with similar
water, and biomass combustion. The fifth chapter describes the opportunities have developed differently due to factors such as
transition to fossil fuels — coal, oil, gas, and nuclear energy — and education, legal framework, commercial rules, availability of
the innovations that drove it. The sixth chapter goes into more capital, and investment in basic research.
depth about how energy use has increased in agriculture, industry, This book satisfied my interest in energy and society. I recom-
transportation, communication, and warfare. It includes a discus- mend it to anyone with similar interests.
sion of both the positive and negative aspects of this growth. The
final chapter discusses how energy use is related to human develop- — Ray Wood
ment and the other factors that control that development. Hastings, New Zealand

WORKSHOP

Role of Fiber
in Geophysics:
Now and Beyond
23–25 June 2025 • Houston, TX, USA go.seg.org/rfgb25

232 The Leading Edge March 2025

 The Leading Edge
Membership
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

A pplications for Active membership have been received from

the candidates listed below. This publication does not consti-
tute election but places the names before the membership at large
Requirements for membership
Active: Eight years of professional experience practicing
or teaching geophysics or a related scientific field. Mem-
in accordance with SEG’s Bylaws, Article III, Section 5. If any bership applications and details of other types of member-
member has information bearing on the qualifications of these ship, including Associate, Student, and Corporate, may
candidates, it should be sent to the SEG president within 30 days. be obtained at https://seg.org/membership.

For Active membership Benmahfoudh, Khalid (USA) Li, Jing (China)

Aguilera Alonso, Elvira (USA) Bin Zaqr, Salem Mohsen (USA) Li, Shenglin (China)
Al Jawad, Rami Majed (Saudi Arabia) Bishop, Lila (USA) Li, Yang (China)
Al Jindan, Mohammad (Saudi Arabia) Boulanger, Olivier (Canada) Liu, Shuyang (China)
Al Saleh, Ahmed (USA) Butterworth, Vanessa (France) Liu, Xu (Saudi Arabia)
Al Tawil, Aus (Saudi Arabia) Cailly, Frederic (France) Luo, Renze (China)
Al-Aboud, Fares (Saudi Arabia) Campero Cervantes, Monserrat (Mexico) Maman Koumba, Gilles (France)
Al-Otaibi, Ahmed (Saudi Arabia) Carrillo, Pedro (USA) Mandal, Partha (India)
Al-Qarni, Mohammed (USA) Caruso, Nicholas (USA) Mansoor, Khalid (France)
Alabkari, Mohammed (Saudi Arabia) Carvajal, Carla (USA) Martindale, James (Australia)
Alangari, Aljawharah (USA) Castellanos Guevara, Victor (Colombia) McGlinchey, Dom (USA)
DOI:10.1190/leedff.2025.44.issue-3

Alawi, Mohanned (USA) Chamlagain, Deepak (Nepal) Merida Rodriguez, Amanda (USA)
Albadi, Amenah (USA) Coates, John (USA) Milewski, Adam (USA)
Albarrak, Hussain (USA) Cogan, Michael (USA) Morgan, Elodie (France)
Aldhalea, Omar (Saudi Arabia) de la Torre Guzman, Jorge (France) Mueller, Sven (Netherlands)
Alfaraj, Hussain (USA) Dhamen, Khalid (Saudi Arabia) Muller, Ana (Brazil)
Alghamdi, Amer (USA) Duhailan, Mohammed (Saudi Arabia) Murti, D. V. R. (India)
Alghamdi, Lina (USA) Farris, Matthew (USA) Niccoli, Matteo (Canada)
Alghamdi, Mujahed (USA) Fayez, Fahad (USA) Nowak, Ethan (USA)
Alhajri, Saif (USA) Formby, James (UK) Obregon, Freddy (USA)
Alharthi, Abdulrahman (USA) Gao, Longfei (USA) Ogu, Elemele (USA)
Alhussaini, Fahad (USA) Garden, Mikael (Austria) Onyango, Evans (USA)
Alkhalaf, Sulaiman (Saudi Arabia) Ghamdi, Khalid (Saudi Arabia) Ostrikoff, James (USA)
Almulla, Abdullatif (USA) Godbey, Will (USA) Otiebi, Munahi (Saudi Arabia)
Almutairi, Mohammed (USA) Goswami, Anjali (India) Owah, Evans (France)
Alnabbat, Mohammed (USA) Gupta, Gaurav (India) Paños, Luis (France)
Alqahtani, Bader (USA) Halder, Subrata (India) Pistoun, Patricia (USA)
Alruwayi, Abdullah (USA) Hammami, Dieb (France) Puzyrev, Vladimir (USA)
Alsaad, Ali (Saudi Arabia) Harris, Peter (UK) Qadous, Nidal (USA)
Alsawidan, Abdullah (Saudi Arabia) Harrison, Antony (Austria) Qiu, Changkai (China)
Alshahrani, Saeed (USA) Haryanto, Haryono (Indonesia) Raghvani, Mahesh (Australia)
AlShareedah, Nouf (USA) He, Tao (China) Rufaii, Khalid (Saudi Arabia)
Alsleebi, Marwa (USA) Hillock, Peter (France) Rulff, Paula (Netherlands)
Alsulaiman, Hussain (USA) Ibrahim, Yazki (Saudi Arabia) Ruterbories, Daniel (USA)
Alzayaer, Ahmed Kamal (USA) Ipoola, Fisayo (Nigeria) Sabor, Faisal (USA)
AlZayer, Ramzy (Saudi Arabia) Iso, Shinichiro (Japan) Sain, Kalachand (India)
Anam, Sameer (France) Jawas, Abdullah (France) Seelam, Naresh Kumar (India)
Anderson, Walter (USA) Keefer, Donald (USA) Shadoan, Tanner (USA)
Andrew, Adeola (USA) Kothwaal Sheik, Parwez Pasha (USA) Shahraeeni, Mohammad (Denmark)
Araujo, Mario Neto (Brazil) Krupnova, Yana (UAE) Sharahilli, Mahmoud (Saudi Arabia)
Armelle, Laurent (France) Kumar, Rajeev (India) Shen, Yi (USA)
Asem, Pouyan (USA) Lee, Chung Shen (Netherlands) Shusaku, Harada (Japan)
Ayyad, Wadyan Osama (Saudi Arabia) Li, David (USA) Sindi, Abdulaziz (USA)
Bakhsh, Ahmed (Saudi Arabia) Li, Gang (China) Singh, Kushwant (Australia)

March 2025 The Leading Edge 233

 Stephens, James (USA)
Sungkorn, Radompon (USA)
Totani, Maaku (Japan) Explore & Advance
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

Totten, Matthew (Namibia)

Trekas, Daniel (USA)
Tuluca, Florina (Romania) in Latin America
Vasyutkin, Serge (Saudi Arabia)
Wang, Tsili (USA)
Wanjala, Cleophas (Kenya)
Wilson, Alex (France)
Yang, Zhiping (USA)
Yavuz, Sinem (UK)
Yuan, Wei (China)
Zhang, Letian (China)
Zhang, Yijie (China) 28–29 May 2025
Zhu, Yaping (USA) Cartagena, Colombia

CCUSevent.org/latinamerica
DOI:10.1190/leedff.2025.44.issue-3

SPONSORING ORGANIZATIONS:

9TH SEG EUROPE

VIRTUAL STUDENT
CONFERENCE
13 MAY 2025 • SUBMIT YOUR ABSTRACT

This virtual conference is organized by students, for

NEAR SURFACE
students, and aims to help students network and
share their work in a short-format setting.
RESEARCH
It’s open to all final year undergraduate, masters, AWARD GRANT
and PhD students across Europe. Students who have
not yet been able to present at a major conference
but would like to share their work are encouraged The SEG Near-Surface Geophysical Research
to attend (check event webpage for applicable Endowment helps potential key contributors in
countries). Abstracts are due 10 April. the field of near-surface geophysics. The award
Why participate? is funded by the Near-Surface Geophysics
• Share research ideas Section of SEG.
• Network with peers across Europe
• Develop science communication Submit your application by 31 March!
Questions?
[email protected] seg.org/programs/student-programs
go.seg.org/vsc9

234 The Leading Edge March 2025

 The Leading Edge
Meetings Calendar
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

See https://seg.org/events/events-calendar for the latest updates.

MARCH 2025 6–8 MAY

Asia Pacific Summit on Drone Geophysics
16–19 JUN
AGU-SEG Geothermal Workshop
3–5 MAR Website coming soon https://seg.org/calendar_events/agu-seg-
Carbon Capture, Utilization, and Storage (CCUS) Virtual geothermal-workshop
https://ccusevent.org/2025 Salt Lake City, Utah, USA
Houston, Texas, USA 13 MAY
9th Virtual Student Conference 18–20 JUN
6 MAR https://seg.org/calendar_events/9th-seg-virtual- Beneath the Surface: Innovations in Geoscience
Denver Geophysical Society 3D Seismic student-conference https://seg.org/calendar_events/advances-in-
Symposium Virtual subsurface-characterization-techniques
https://denvergeo.org/events/30th-3d-seismic- Hyderabad, India
symposium-reflections-on-the-range 13–14 MAY
Denver, Colorado, USA 3rd Geothermal Geophysics Workshop 23–25 JUN
Website coming soon Role of Fiber in Geophysics: Now and Beyond
21–23 MAR New Zealand https://seg.org/calendar_events/role-of-fiber-in-
10th International Symposium on Energy Science geophysics-now-and-beyond
and Chemical Engineering 13–15 MAY Houston, Texas, USA
https://www.isesce.org 2nd Seabed Seismic Workshop — Imaging
DOI:10.1190/leedff.2025.44.issue-3

Ningbo, Zhejiang, China Innovation for the Future

https://seg.org/calendar_events/2nd-seabed-
JULY 2025
24–26 MAR seismic-workshop-innovation-for-the-future 2–4 JUL
Realizing the Energy Potential of Angola Kuala Lumpur, Malaysia 13th International Workshop on Advanced
Symposium Ground Penetrating Radar — IWAGPR2025
https://seg.org/calendar_events/energy- 13–15 MAY https://www.iwagpr2025.gr
potential-angola-margin New Geophysical Aspects for Production Thessaloniki, Greece
Luanda, Angola Enhancement
https://seg.org/calendar_events/geophysical- 4–6 JUL
APRIL 2025 aspects-production-enhancement-2025
Muscat, Oman
Near-Surface Geophysical Exploration and Geo-
Disaster Prevention Technology Workshop
14–16 APR https://seg.org/calendar_events/seg-near-
Asia Pacific Symposium on Digital 19–21 MAY surface-geophysical-exploration-and-geo-
Transformation in Energy Net-Zero Emissions disaster-prevention
https://adtedigital.com https://seg.org/calendar_events/net-zero- Chengdu, China
Kuala Lumpur, Malaysia emissions-workshop-2025
Virtual 7–8 JUL
28–30 APR SEG/SPE 2nd Integrated Reservoir
Leveraging Technologies to Unlock Egypt’s 28–29 MAY Characterization, Surveillance and
Energy Potential CCUS Latin America Management: Advances and Challenges
https://seg.org/calendar_events/egypt-energy- https://ccusevent.org/latinamerica/2025 https://seg.org/calendar_events/seg-spe-
potential-2025 Cartagena, Colombia 2nd-integrated-reservoir-characterization-
New Cairo, Egypt surveillance-and-management-advances-and-

JUNE 2025 challenges

MAY 2025 9–11 JUN

Kuala Lumpur, Malaysia

5–8 MAY
Offshore Technology Conference
Unconventional Resources Technology
Conference
AUGUST 2025
https://2025.otcnet.org https://urtec.org/2025/save-the-date 25–28 AUG
Houston, Texas, USA Houston, Texas, USA International Meeting for Applied Geoscience &
Energy (IMAGE)
https://www.imageevent.org
Houston, Texas, USA

March 2025 The Leading Edge 235

 S e i s m i c S o u n d o f f — C o o r d i n at e d by Andrew Geary
Saving lives before disaster strikes
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

I n episode 243 of SEG’s Seismic Soundoff podcast, host Andrew

Geary talks with Ron Harris, geological sciences professor at
Brigham Young University, about his groundbreaking work in
The best way is to support them in
what they are doing. For example,
almost all the funds for my nonprofit
disaster preparedness. Harris shares insights from his Global (In Harm’s Way) go to transportation
Sustainability Lecture (supported by the Baker Hughes Founda- costs and support of earthquake and
tion) and the Geoscientists without Borders (GWB) Java WAVES tsunami evacuation drills. We send
project. Listen to the full episode at https://doi.org/10.1190/ most of it to Indonesia and the people
seismic-soundoff-episode243. in You Inspire, and some of their other
colleagues use it to go from village to
Andrew Geary: What insights from the 2004 Indian Ocean village delivering the message, which
tsunami changed your professional perspective? is a dream come true for me. After
that 2004 event, I was like, we failed. We absolutely failed. We
Ron Harris: Like most people in our science, we were focused nailed where the earthquake was going to be and how big it was
on assessing the risk. Communication of the risk and implementa- going to be, but we needed to go village to village. We had two
tion of risk-reduction strategies we just thought someone else was years. We could have saved tens of thousands of lives in those
going to do. The 2004 event changed the world because we realized two years between when we realized what was going to happen
that no one was really tasked with that important part of disaster and when it actually did, just by saying, when you feel the ground
mitigation. We wrote a paper a couple years before the 2004 event, shake, evacuate. They stood around waiting for some kind of
talking about how the seismic gap of Sumatra was very dangerous message from the government. But most of the cellphone towers
DOI:10.1190/leedff.2025.44.issue-3

and could have an 8-plus earthquake. The earthquake was a 9.2, were knocked out by the earthquake, and there was no com-
and 250,000 people died. munication. There was chaos and now the 20-20-20 principle,
Then, there was the 2011 earthquake in Japan that killed which many parts of Indonesia have adopted. They adopted it
10 times less people. The only difference ... was preparedness. in the year 2020 to try to help people understand the significance
That is something we can control, and that changed everything of it. Just that is going to save thousands of lives.
for me. I started walking across the courtyard to talk to my social-
scientist friends, and I said this is not just a geological or geophysi- Geary: How has support from organizations like GWB
cal issue. In communicating the risk to them and working side influenced your work with In Harm’s Way?
by side to implement risk-reduction strategies, we need help. We
are not trained in this, but it is the most important part of what Harris: I was at an American Geophysical Union meeting, and
I call full-spectrum disaster mitigation, which is assessment there were no sessions on disaster mitigation. I gave a talk on it
coupled with communication coupled with implementation. The and a poster session. So many people came up to me like, “Wow!
most important of those three is implementation. That is the one This is amazing. We need to do more of this.” One person came
that was not being done. I would give us a D on the communica- up to me and asked if I knew anything about GWB. I said, no.
tion, but I would give us an F on the actual action items that we He said, you should contact them because they are aligned with
had to do to protect people. what you are doing. I contacted them and was so amazed that
there was somebody who was interested in what we were doing.
Geary: Could you tell me about the WAVES team and They provided our first grant and the resources we did not have.
how you bridge scientific expertise with community-level
risk reduction? Geary: What encouragement would you offer a student
doing this work or a person like yourself in the midst of
Harris: The first thing is you have to work with people who know doing this work?
how to communicate with locals. The most important part of our
team is the Indonesian students we have trained over the years. Harris: It is great to have great outcomes, but it is the right
One of the most satisfying things about this is those teams went thing to do, no matter what the outcomes are. It truly is. I feel
back home to Indonesia and started a project called You Inspire. like I was called to this. My experience in Indonesia set me up.
You Inspire is dedicated to disaster risk management, and in order When that 2004 event happened, I went through a complete
to be part of You Inspire, you have to be less than 30 years old. paradigm shift about what was the most important part of disaster
Three of my former students run the organization, and it has mitigation. That is what I would say to encourage people who
captivated youth throughout the country. They are all focused on are interested in this — to do it from the heart. Your heart is
different disciplines and on reversing the trend of increasing losses telling you to do it. And my heart is telling me to do this. My
to nature through simple communication and implementation of heart is saying, no matter what people think, this is the right
risk-reduction strategies. thing to do.

236 The Leading Edge March 2025

 SUBMIT
Downloaded 07/23/25 to 122.161.72.155. Redistribution subject to SEG license or copyright; see Terms of Use at http://library.seg.org/page/policies/terms

YOUR
US
as,
ex
n ,T
sto
ou
•H
st
ugu
8A
–2
25

ABSTRACTS
Share the Future of Geoscience
and Energy at IMAGE ’25
DOI:10.1190/leedff.2025.44.issue-3

The call for

abstracts for
IMAGE ‘25 is now open!
Join the leading assembly of applied
geoscientists, thought leaders, and
technical experts from around the world.
Share your breakthroughs! We’re seeking
compelling case studies, cutting-
edge technological advancements, and Submission
groundbreaking research discoveries. Deadline:
First-time presenter? No problem! IMAGE ‘25 15 March
welcomes both seasoned speakers and new 5:00 p.m., CDT
voices. Visit IMAGEevent.org to submit your
abstract today!

IMAGEevent.org