TY - GEN
T1 - LEARNING TO UNFLOOD THE SALT IN FULL-WAVEFORM INVERSION, APPLICATION ON VINTAGE GOM DATA
AU - Alali, A.
AU - Alkhalifah, T.
N1 - Funding Information:
We thank Mahesh Kalita for his valuable insights and SWAG group for their support. We are also grateful for the Supercomputing Laboratory at KAUST for providing the computational resources.
Publisher Copyright:
Copyright© (2022) by the European Association of Geoscientists & Engineers (EAGE). All rights reserved.
PY - 2022
Y1 - 2022
N2 - Building a velocity model in salt provinces is a challenging task. Traditionally the salt boundaries is manually interpreted by an iterative process of imaging, picking the top of the salt and flooding, re-imaging and picking the bottom of the salt for unflooding. This workflow is time-consuming and prone to errors. Full-waveform inversion (FWI) can be used to correct the erroneous picks of the salt boundaries. However, it requires low frequency and long offsets data to build an accurate salt body. We apply an FWI-based automatic unflooding process on vintage field data that do not meet these requirement by training a neural network using data with similar characteristics. The network is trained to unflood the salt and estimate the subsalt velocity from an inverted flooded model in a regression manner. The network shows good potential to unflood the vintage data and produce results comparable with the legacy model.
AB - Building a velocity model in salt provinces is a challenging task. Traditionally the salt boundaries is manually interpreted by an iterative process of imaging, picking the top of the salt and flooding, re-imaging and picking the bottom of the salt for unflooding. This workflow is time-consuming and prone to errors. Full-waveform inversion (FWI) can be used to correct the erroneous picks of the salt boundaries. However, it requires low frequency and long offsets data to build an accurate salt body. We apply an FWI-based automatic unflooding process on vintage field data that do not meet these requirement by training a neural network using data with similar characteristics. The network is trained to unflood the salt and estimate the subsalt velocity from an inverted flooded model in a regression manner. The network shows good potential to unflood the vintage data and produce results comparable with the legacy model.
UR - http://www.scopus.com/inward/record.url?scp=85142685476&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85142685476
T3 - 83rd EAGE Conference and Exhibition 2022
SP - 2974
EP - 2978
BT - 83rd EAGE Conference and Exhibition 2022
PB - European Association of Geoscientists and Engineers, EAGE
T2 - 83rd EAGE Conference and Exhibition 2022
Y2 - 6 June 2022 through 9 June 2022
ER -