SRT-Ai: identifying seismic reflection terminations using deep learning
File(s)
Published version
Author(s)
AlGharbi, Waleed M
Bell, Rebecca E
John, Cédric M
Type
Journal Article
Abstract
Seismic stratigraphy entails a regional scanning (reconnaissance) of seismic data to identify and annotate seismic reflection terminations. To identify these terminations in modern 3D seismic datasets, interpreters have to examine thousands of inlines and crosslines, which is a time-consuming process. Furthermore, accurate identification of these features relies heavily on human visual observation along with individual expertise.
A growing number of studies have shown promising results applying machine learning techniques to identify geological features from seismic data such as salt bodies and faults. However, the identification of seismic reflection terminations has not received the same level of interest and remains a manual process. One of the barriers to utilizing machine learning techniques in seismic interpretation is the lack of “labelled” data. In this study, we evaluate the ability of deep learning Convolutional Neural Networks (CNN) trained on synthetic seismic images to identify seismic reflection terminations.
A dataset comprising 160 000 synthetic seismic images that represent conformable and four types of seismic reflection terminations (truncation, toplap, onlap, and downlap) were created using geometric geological modelling and 1D convolution seismic modelling. The dataset was then split into two classes (“Contains Termination” and “No Termination”). A new CNN model architecture named “Seismic Reflection Terminations Attribute (SRT-Ai)” was trained on 80 % of the synthetic seismic dataset. SRT-Ai predicted the test set (remaining 20 %) with an accuracy and precision of 99.9 %. To test its generalization, SRT-Ai was also evaluated on real seismic images, achieving 91 % accuracy and 96 % precision against published interpretations used as reference labels. Qualitative analysis of predictions along seismic sections shows a strong correspondence between the model predictions and manual regional interpretations.
SRT-Ai is proposed as a screening tool that will assist seismic interpreters with the identification of major seismic terminations, minimise seismic interpretation uncertainties, reduce the time taken for seismic reconnaissance, and limit the reliance on human visual observation at the early stage of seismic interpretation process.
A growing number of studies have shown promising results applying machine learning techniques to identify geological features from seismic data such as salt bodies and faults. However, the identification of seismic reflection terminations has not received the same level of interest and remains a manual process. One of the barriers to utilizing machine learning techniques in seismic interpretation is the lack of “labelled” data. In this study, we evaluate the ability of deep learning Convolutional Neural Networks (CNN) trained on synthetic seismic images to identify seismic reflection terminations.
A dataset comprising 160 000 synthetic seismic images that represent conformable and four types of seismic reflection terminations (truncation, toplap, onlap, and downlap) were created using geometric geological modelling and 1D convolution seismic modelling. The dataset was then split into two classes (“Contains Termination” and “No Termination”). A new CNN model architecture named “Seismic Reflection Terminations Attribute (SRT-Ai)” was trained on 80 % of the synthetic seismic dataset. SRT-Ai predicted the test set (remaining 20 %) with an accuracy and precision of 99.9 %. To test its generalization, SRT-Ai was also evaluated on real seismic images, achieving 91 % accuracy and 96 % precision against published interpretations used as reference labels. Qualitative analysis of predictions along seismic sections shows a strong correspondence between the model predictions and manual regional interpretations.
SRT-Ai is proposed as a screening tool that will assist seismic interpreters with the identification of major seismic terminations, minimise seismic interpretation uncertainties, reduce the time taken for seismic reconnaissance, and limit the reliance on human visual observation at the early stage of seismic interpretation process.
Date Issued
2025-09-01
Date Acceptance
2025-07-24
Citation
Applied Computing and Geosciences, 2025, 27
ISSN
2590-1974
Publisher
Elsevier BV
Journal / Book Title
Applied Computing and Geosciences
Volume
27
Copyright Statement
© 2025 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).
License URL
Publication Status
Published
Article Number
100271
Date Publish Online
2025-07-26