TY - GEN
T1 - Image-domain full waveform inversion
AU - Zhang, Sanzong
AU - Schuster, Gerard T.
N1 - KAUST Repository Item: Exported on 2020-10-01
Acknowledgements: We thank the 2013 sponsors of Center for Subsurface Imaging and Fluid Modeling (CSIM) at KAUST. We also appreciate the support from IBM Deep Research Center and KAUST High Performance Computing. The intensive computation in this works is conducted in Blue Gene P/Q.
PY - 2013/8/19
Y1 - 2013/8/19
N2 - The main difficulty with the data-domain full waveform inversion (FWI) is that it tends to get stuck in the local minima associated with the waveform misfit function. This is because the waveform misfit function is highly nonlinear with respect to changes in velocity model. To reduce this nonlinearity, we define the image-domain objective function to minimize the difference of the suboffset-domain common image gathers (CIGs) obtained by migrating the observed data and the calculated data. The derivation shows that the gradient of this new objective function is the combination of the gradient of the conventional FWI and the image-domain differential semblance optimization (DSO). Compared to the conventional FWI, the imagedomain FWI is immune to cycle skipping problems by smearing the nonzero suboffset images along wavepath. It also can avoid the edge effects and the gradient artifacts that are inherent in DSO due to the falsely over-penalized focused images. This is achieved by subtracting the focused image associated with the calculated data from the unfocused image associated with the observed data in the image-domain misfit function. The numerical results of the Marmousi model show that image-domain FWI is less sensitive the initial model than the conventional FWI. © 2013 SEG.
AB - The main difficulty with the data-domain full waveform inversion (FWI) is that it tends to get stuck in the local minima associated with the waveform misfit function. This is because the waveform misfit function is highly nonlinear with respect to changes in velocity model. To reduce this nonlinearity, we define the image-domain objective function to minimize the difference of the suboffset-domain common image gathers (CIGs) obtained by migrating the observed data and the calculated data. The derivation shows that the gradient of this new objective function is the combination of the gradient of the conventional FWI and the image-domain differential semblance optimization (DSO). Compared to the conventional FWI, the imagedomain FWI is immune to cycle skipping problems by smearing the nonzero suboffset images along wavepath. It also can avoid the edge effects and the gradient artifacts that are inherent in DSO due to the falsely over-penalized focused images. This is achieved by subtracting the focused image associated with the calculated data from the unfocused image associated with the observed data in the image-domain misfit function. The numerical results of the Marmousi model show that image-domain FWI is less sensitive the initial model than the conventional FWI. © 2013 SEG.
UR - http://hdl.handle.net/10754/627672
UR - http://library.seg.org/doi/abs/10.1190/segam2013-1238.1
UR - http://www.scopus.com/inward/record.url?scp=85058084149&partnerID=8YFLogxK
U2 - 10.1190/segam2013-1238.1
DO - 10.1190/segam2013-1238.1
M3 - Conference contribution
SN - 9781629931883
SP - 861
EP - 865
BT - SEG Technical Program Expanded Abstracts 2013
PB - Society of Exploration Geophysicists
ER -