Wave-Equation Waveform Inversion for Crosswell Data M. Zhou and Yue Wang Geology and Geophysics Department University of Utah
Outline Theory: Wave Equation Inversion Theory: Wave Equation Inversion 2 Problems: Local Minima & Visco 2 Problems: Local Minima & Visco McElroy Xwell data McElroy Xwell data Conclusions Conclusions
Waveform Inversion = || P(x,t) – P(x,t) || 2 pred obs v(x,z) v(x,z) - v(x,z) P(x,t) pred [P(x,t) – P(x,t) ] pred obs Migration { Waveform Residual { Acooustic, Elastic or Viscoelastic
Why Waveform Inversion? 0 km Depth 0.2 km km Better Resolution & Focusing Blurring Faults LithologyDistortion Gas
Outline Theory: Wave Equation Inversion Theory: Wave Equation Inversion 2 Problems: Local Minima & Visco 2 Problems: Local Minima & Visco McElroy Xwell data McElroy Xwell data Conclusions Conclusions
Why Waveform Inversion is Not Used? Poor Convergence, Local Minima, Cost Localminima traveltime) v v 1 2 traveltime + 1 st Arrival) v v 1 2 v v 1 2 1 st -Arrival Waveforminversion Waveforminversion Global Minima
km/s Depth (m) 0 90 X (m) Model (Zhou et al., 1995) Model 1.5m X 1.5m grid 18 shots / 36 geophones 60 Hz Ricker wavelet
0 90 X (m) WIF km/s Depth (m) 0 90 X (m) Model 0 90 WT10 Model : Tomograms 0 90 X (m) WIF20 + WI10
X (m) Model WIF20 Time (sec) X (m) WT10 Model 2: Synthetic CSG
Outline Theory: Wave Equation Inversion Theory: Wave Equation Inversion 2 Problems: Local Minima & Visco 2 Problems: Local Minima & Visco McElroy Xwell data McElroy Xwell data Conclusions Conclusions
Problem: Invert Vp & Vs by elastic wave inver. Vp accurate by elastic waveform inversion Vp accurate by elastic waveform inversion Vs inaccurate by elastic waveform inversion Vs inaccurate by elastic waveform inversion Why? Insufficient physics in forward modeler Anisotropy? Attenuation? Biot? Coupling? Source effects? 3-D scattering? Source effects? 3-D scattering? Conjecture: Add attenuation
Waveform Inversion = || P(x,t) – P(x,t) || 2 pred obs v(x,z) v(x,z) - v(x,z) P(x,t) pred [P(x,t) – P(x,t) ] pred obs Migration { Waveform Residual { Viscoelastic Viscoelastic
0 90 Offset (m) Depth (m) P 0 90 Offset (m) S m/s m/s Fault Model
0 Time (s) Depth (m) Elastic Seismogram
0 Time (s) Depth (m) =40 Qp=40Qs=25 Viscoelastic Seismogram
0 90 Offset (m) Depth (m) P 0 90 Offset (m) S m/s m/s A B C Ray Tracing Tomograms
0 90 Offset (m) Depth (m) P 0 90 Offset (m) S m/s m/s A B C A B C D Viscoelastic Tomograms
0 90 Offset (m) Depth (m) P 0 90 Offset (m) S m/s m/s Elastic Tomograms
Outline Theory: Wave Equation Inversion Theory: Wave Equation Inversion 2 Problems: Local Minima & Visco 2 Problems: Local Minima & Visco McElroy Xwell data McElroy Xwell data Conclusions Conclusions
200 Sources * 200 Receivers 180 ft 2700 ft McElroy Geometry 500 ft
0 500 Depth (ft) Time (s) Shot Gather 101
Bandpass Filter Median Filter 3D ==> 2D Borehole Filter, wavelet and radiation pattern Processing
0 184 Offset (ft) Depth (ft) P Offset (ft) S B A C D ft/s ft/s Elastic Tomograms
2700 V Depth (ft) 3150 Receiver Well P-Velocity Profile: Tomo vs Sonic Log
2850 Depth (ft) 3050 V Source Well S-Velocity Profile: Tomo vs Sonic Log
Bandpass Filter Median Filter 3D ==> 2D Borehole filter, Wavelet and Src Rad. Qp and Qs by Harris’ Redshift Method Processing
Freq. ~ Time Covariance of Source Spectrum Appr. Qp Extract All Direct P Waves Estimate Q from 1 st Arrivals
Freq. vs Time Freq. vs Time Time (s) Freq. (hz)
B A C D Offset (ft) Depth (ft) P Offset (ft) S ft/s ft/s Viscoelastic Tomograms
2850 Depth (ft) 3050 V Source Well S-Velocity Profile: ViscoTomo vs Sonic Log
2850 Depth (ft) 3050 V Source Well S-Velocity Profile: ElasticTomo vs Sonic Log
0 Offset (ft) Depth (ft) A B Viscoelastic S-Tomogram
0 Offset (ft) Depth (ft) A B Elastic S-Tomogram
0 Offset (ft) Depth (ft) Visco. 184 C D Poisson Ratio
0 Offset (ft) Depth (ft) Elastic 184 C D Poisson Ratio
Conclusions 4 Iteration, Pressure Data, Visco.WTW4 Iteration, Pressure Data, Visco.WTW Visco better than ElasticVisco better than Elastic Issues:Issues: Q(z), multicomponent, anisotropy Q(z), multicomponent, anisotropy CPU Time CPU Time Intrinsic atten. vs Scattering atten.? Intrinsic atten. vs Scattering atten.? Scale of Q vs Velocity? Scale of Q vs Velocity? Need more Physics such as Anisotropy? Need more Physics such as Anisotropy?
B A C D Visco. WTW Tomograms Offset (ft) Depth (ft) P Offset (ft) S ft/s ft/s
S Tomogram Comparison 0 Offset (ft) Depth (ft) Visco. 184 C D
S Tomogram Comparison 0 Offset (ft) Depth (ft) Elastic 184 C D
Conclusions Visco.WTW High-resolution P- and S-velocity ! YES
Conclusions Visco.WTW Visco.WTW Porosity, Lithology, AVO YES
Acknowledgments Acknowledgments We are grateful for the support of : the 1997 members of University of Utah Tomography and Modeling/Migration Consortium
Viscoelastic CSP Gather Depth (ft) Time (s)
Evaluate Q Values Appr. Qp + Synthetic Modeling Test Qp Qs
CSP Gather Depth (ft) Time (s)
Viscoelastic CSP Gather Depth (ft) Time (s)
Elastic CSP Gather Depth (ft) Time (s)
Viscoelastic Waveform Inversion Depth (ft) Time (s) Depth (ft) Time (s) Depth (ft) Time (s) Elastic Viscolastic CSG101
P-Velocity Profiles : P-sonic Logs : Elastic P-velocity Profiles : Visco. P-velocity Profiles
S-Velocity Profiles 2700 V Depth (ft) 3150 V Receiver Well Source Well
S-Velocity Profiles : S-sonic Logs : Elastic S-velocity Profiles : Visco. S-velocity Profiles
Elastic CSP Gather Depth (ft) Time (s)
Viscoelastic CSP Gather Depth (ft) Time (s) Qp=80Qs=50
Outline Motivation+Theory WI Motivation+Theory WI Examples Examples Xwell Synthetic Xwell Synthetic CDP Synthetic : Chirp CDP Synthetic : Chirp CDP Gulf of Mexico CDP Gulf of Mexico Conclusions Conclusions
Synthetic CDP Model (FD Acoustic) Depth (m) Distance (m) (m/s) 40 m Synthetic CDP Model (100 shot gathers, 100 receivers/gather)
Traveltime Tomogram Depth (m) Distance (m) (m/s)
Multigrid for half comp. time Depth (m) Distance (m) (m/s)
Chevron Gulf of Mexico Seismic Line Courtesy of Alan Leeds Shots: 990 Channel: 180 Shot spacing: 25 m Receiver: 25 m Sample: 4 ms Length: 8 sec. Offset 173 – 4648 m
(m/s) CDP NUMBER Depth (m) Traveltime Tomogram Waveform
Waveform vs. Traveltime (m/s) CDP NUMBER Depth (m) WaveformTraveltime Gas?
Amplitude Vs. Offset Log10 Amplitude Log10 Offset (m) Waveform Data Traveltime
600 (msec.) 1200 Stacked Section (Waveform vs. Traveltime)
Conclusions Robust to initial model for Xwell. Robust to initial model for Xwell. High resolution tomograms. High resolution tomograms. WI vs. Traveltime Inversion: WIF + WI vs. WI: Stable for Models Tested. Stable for Models Tested. Sensitive to initial model. Sensitive to initial model. Will this approach work for Reflections? Will this approach work for Reflections?
Acknowledgements I am grateful for the financial I am grateful for the financial support from the members of support from the members of the UTAM consortium. the UTAM consortium.
Waveform Tomogram Depth (m) Distance (m) (m/s)
2D Synthetic Data (Blind test) Courtesy of Konstantin Osypov Shots: 401 Channel: 241 Shot spacing: 50 m Receiver: 25 m Sample: 4 ms Length: 2 sec. Offset: – 3000 m
Traveltime Tomogram m/s Horizontal distance (km) Depth (km)
Amplitude 0.1.WIF20 Amplitude Time (s) WT10 Model 2: One Trace Examples ObjectiveTheoryConclusionsMotivation
Waveform Tomogram m/s Horizontal distance (km) Depth (km)
600 (msec.) 1200 Stacked Section (Waveform vs. Traveltime) WaveformTraveltime
600 (msec.) 1200
Summary Traveltime Inversion Traveltime Inversion Waveform Inversion Waveform Inversion slow, sensitive to initial model high resolution fast, insensitive to initial model low resolution
Model 1: Model km/s 050 X (m) Depth (m) Model 1m X 1m grid 41 shots/geophones 200 Hz Ricker wavelet
Model 1: Tomograms 050 X (m) WI km/s Depth (m) Model 050 X (m) Tomo50
050 WIF30 + WI km/s 050 X (m) Depth (m) Model 050 X (m) WIF30 Model 1: Tomograms
Time (sec) X (m) Tomo50 Model WIF30 + WI Model 1: Synthetic CSG
RMS Waveform Residuals Number of Iterations Model 1: Residuals
Outline Motivation+Theory WI Motivation+Theory WI Examples Examples Xwell Synthetic Xwell Synthetic CDP Synthetic : Chirp CDP Synthetic : Chirp CDP Gulf of Mexico CDP Gulf of Mexico Conclusions Conclusions
Visco. Wave Equation Initial Velocity Model Vp, Vs, Qp, Qs Synthetic Seismograms Residuals = ||Syn. - Obs|| 2
u Perturbation of Lame parameters: Gradient Optimization f : from forward wavefield b: from adjoint wavefield
Spectrum Covariance Frequency (hz) Amplitude 1 1/e
Problem & Methodology Synthetic Data Example Field Data Example Conclusion and Discussion Outline
Full 2-D viscoelastic wave equationFull 2-D viscoelastic wave equation with memory variables. Standard with memory variables. Standard model spring-dashpot Qp and Qs model spring-dashpot Qp and Qs 2-D to 3-D conversion2-D to 3-D conversion Apply borehole transfer functionApply borehole transfer function Invert src waveletInvert src wavelet Invert src radiation patternInvert src radiation patternMethodology