Estimating CO2 saturation maps from seismic data using deep convolutional neural networks

Zi Xian Leong, Tieyuan Zhu, Alexander Y. Sun

Research output: Contribution to journalConference articlepeer-review

4 Scopus citations

Abstract

Estimating CO2 saturation from seismic data can be laborious due to the extensive processing steps and inversion workflows involved. We propose and develop a convolutional neural network (CNN) based inversion algorithm, in which the CNN takes in raw seismic data, and directly outputs the CO2 saturation map. This method is efficient and effective during inference. Our CO2 saturation dataset is crafted based on Cranfield CO2 pilot site geological model. We use viscoacoustic seismic modeling to include seismic velocity and attenuation dynamics caused by the replacement of brine by CO2 fluids in the generated seismic data. These steps encourage the CNN to learn from dataset which mimics field settings. Our CO2 saturation prediction results show high accuracies. In the future, we plan to test this method on a field dataset.

Original languageEnglish (US)
Pages (from-to)510-514
Number of pages5
JournalSEG Technical Program Expanded Abstracts
Volume2022-August
DOIs
StatePublished - Aug 15 2022
Event2nd International Meeting for Applied Geoscience and Energy, IMAGE 2022 - Houston, United States
Duration: Aug 28 2022Sep 1 2022

All Science Journal Classification (ASJC) codes

  • Geotechnical Engineering and Engineering Geology
  • Geophysics

Fingerprint

Dive into the research topics of 'Estimating CO2 saturation maps from seismic data using deep convolutional neural networks'. Together they form a unique fingerprint.

Cite this