Multisource Least-Squares Migration of Marine Streamer Data with Frequency-Division Encoding Yunsong Huang and Gerard Schuster KAUST Outline • Multisource LSM • Problem with Marine Data • Multisource LSM with Frequency Division • Numerical results • Conclusions Multisource vs Benefit: Reduction in computation and memory Liability: Crosstalk noise … Multisource (2) vs T T mmig =[L +L ](d + d ) d2 1 d1 d1+d2 = [L1+L2]m ~ d migrate ~ L blended blended forward data modeling operator d1 +d2 T T m ~ [L1+L2](d1+d2) = L1standard d1+Lmig. 2d2+ L1d2+L2d1 crosstalk T T T T Multisource LSM Inverse problem: arg min J = m 1 2 ~ ~ || d – L m ||2 misfit d Iterative update: ~ T (k+1) (k) m =m +aLd K=1 K=10 Outline • Multisource LSM • Problem with Marine Data • Multisource LSM with Frequency Division • Numerical results • Conclusions Problem with Marine Data misfit = observed data erroneous misfit simulated data Outline • Multisource LSM • Problem with Marine Data • Multisource LSM with Frequency Division • Numerical results • Conclusions Solution - Every source sends out a unique identifier that survives LTI operations - Every receiver acknowledge the contribution from the ‘correct’ sources. observed simulated Frequency Division R(w) Nw frequency bands of source spectrum: Nw = 5 ttrav fpeak w 152 sources/group Group 1 2.2 km Outline • Multisource LSM • Problem with Marine Data • Multisource LSM with Frequency Division • Numerical results (2D) • Conclusions Migration images (input SNR = 10dB) an example shot and its aperture 0 304 shots in total b) Standard Migration c) Standard Migration with 1/8 subsampled shots d) 304 shots/gather 0 1.48 Z (km) a) Original 1.48 Z (km) 26 iterations 0 X (km) 6.75 0 X (km) 6.75 Convergence curves. Input SNR = 10dB 1 Normalized data misfit Conjugate gradient Encoding anew and resetting search direction 0.5 0.4 0.3 0.2 0.1 0 3 6 9 15 21 Iteration number 30 39 Sensitivity to input noise level 9.4 8.0 Computational gain 6.6 5.4 3.8 Conventional migration: 1 38 76 152 Shots per supergather 304 I/O considerations • Ns: # shots subsumed in a supergather • Nit: # of iterations that call for new encoding (i.e., new frequency division scheme) i) If data is stored on hard disk – The I/O cost of our proposed method is Nit/Ns times that of standard migration. ii) If data is stored on tape – The I/O cost of our proposed method is 1+ e times that of standard migration. I/O cost i) Data on hard disk ii) Data on tape Conventional migration Proposed method Stacked migration vs successive least-squares 3 1 2 stacked migration: 0 di Li m successive least-squares: 1 1 2 3 m (1) 2 m (2) 3 m = L†1d1 + L†2d 2 + L†1d3 L†d m (3) Outline • Multisource LSM • Problem with Marine Data • Multisource LSM with Frequency Division • Numerical results (3D) • Conclusions SEG/EAGE Model+Marine Data 100 m 256 sources 40 m 4096 sources in total 6 km 20 m 3.7 km 16 cables 13.4 km Numerical Results 6.7 km 3.7 km 13.4 km 8 x gain in computational efficiency What have we empirically learned? Stnd. Mig Multsrc. LSM IO 1 ~1/36 Cost 1 ~0.1 Migration SNR Resolution dx 1 ~1 1 ~double Cost vs Quality: Can I<<S? Yes.
© Copyright 2026 Paperzz