Automated velocity modeling with domain transformations

Kevin Gullikson; Arnab Dhara; Ram Tuvi; Mrinal K. Sen

doi:10.1190/image2024-4094451.1

Automated velocity modeling with domain transformations

Kevin Gullikson, Arnab Dhara, Ram Tuvi, Mrinal K. Sen

Research output: Contribution to journal › Conference article › peer-review

Abstract

We demonstrate a machine learning model that effectively learns to predict a velocity model of the subsurface directly from trace data. We do this by first transforming the data from the shot gather domain to the tau-p domain, then predicting the time-domain velocity model using a U-Net, and finally converting from time to depth domain by integrating the predicted velocity model. This pipeline is capable of predicting a velocity model suitable for input to FWI and is capable of scaling to both the volume and complexity in real marine acquisitions.

Original language	English
Pages (from-to)	60-64
Number of pages	5
Journal	SEG Technical Program Expanded Abstracts
Volume	2024-August
DOIs	https://doi.org/10.1190/image2024-4094451.1
State	Published - 1 Jan 2024
Externally published	Yes
Event	4th International Meeting for Applied Geoscience and Energy, IMAGE 2024 - Houston, United States Duration: 26 Aug 2024 → 29 Aug 2024

Keywords

acoustic
deep learning
inversion
Radon transform
velocity model building

ASJC Scopus subject areas

Geotechnical Engineering and Engineering Geology
Geophysics

Access to Document

10.1190/image2024-4094451.1

Cite this

@article{e140ad2f5e19420ba4929ddf4a74acd3,

title = "Automated velocity modeling with domain transformations",

abstract = "We demonstrate a machine learning model that effectively learns to predict a velocity model of the subsurface directly from trace data. We do this by first transforming the data from the shot gather domain to the tau-p domain, then predicting the time-domain velocity model using a U-Net, and finally converting from time to depth domain by integrating the predicted velocity model. This pipeline is capable of predicting a velocity model suitable for input to FWI and is capable of scaling to both the volume and complexity in real marine acquisitions.",

keywords = "acoustic, deep learning, inversion, Radon transform, velocity model building",

author = "Kevin Gullikson and Arnab Dhara and Ram Tuvi and Sen, {Mrinal K.}",

note = "Publisher Copyright: {\textcopyright} 2024 Society of Exploration Geophysicists and the American Association of Petroleum Geologists.; 4th International Meeting for Applied Geoscience and Energy, IMAGE 2024 ; Conference date: 26-08-2024 Through 29-08-2024",

year = "2024",

month = jan,

day = "1",

doi = "10.1190/image2024-4094451.1",

language = "English",

volume = "2024-August",

pages = "60--64",

journal = "SEG Technical Program Expanded Abstracts",

issn = "1052-3812",

publisher = "Society of Exploration Geophysicists",

}

TY - JOUR

T1 - Automated velocity modeling with domain transformations

AU - Gullikson, Kevin

AU - Dhara, Arnab

AU - Tuvi, Ram

AU - Sen, Mrinal K.

PY - 2024/1/1

Y1 - 2024/1/1

N2 - We demonstrate a machine learning model that effectively learns to predict a velocity model of the subsurface directly from trace data. We do this by first transforming the data from the shot gather domain to the tau-p domain, then predicting the time-domain velocity model using a U-Net, and finally converting from time to depth domain by integrating the predicted velocity model. This pipeline is capable of predicting a velocity model suitable for input to FWI and is capable of scaling to both the volume and complexity in real marine acquisitions.

AB - We demonstrate a machine learning model that effectively learns to predict a velocity model of the subsurface directly from trace data. We do this by first transforming the data from the shot gather domain to the tau-p domain, then predicting the time-domain velocity model using a U-Net, and finally converting from time to depth domain by integrating the predicted velocity model. This pipeline is capable of predicting a velocity model suitable for input to FWI and is capable of scaling to both the volume and complexity in real marine acquisitions.

KW - acoustic

KW - deep learning

KW - inversion

KW - Radon transform

KW - velocity model building

UR - http://www.scopus.com/inward/record.url?scp=105001133293&partnerID=8YFLogxK

U2 - 10.1190/image2024-4094451.1

DO - 10.1190/image2024-4094451.1

M3 - Conference article

AN - SCOPUS:105001133293

SN - 1052-3812

VL - 2024-August

SP - 60

EP - 64

JO - SEG Technical Program Expanded Abstracts

JF - SEG Technical Program Expanded Abstracts

T2 - 4th International Meeting for Applied Geoscience and Energy, IMAGE 2024

Y2 - 26 August 2024 through 29 August 2024

ER -

Automated velocity modeling with domain transformations

Abstract

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this