Least squares migration of elastic data Aaron Stanton and Mauricio Sacchi PIMS 2015
Outline Motivation Least squares migration of elastic data Adjoint (migration) operator – Wavefield de composition – Extrapolation – Imaging condition Forward (de-migration) operator – Adjoint of Imaging condition – Extrapolation – Wavefield re composition Preconditioning via Poynting vectors Example
Motivation To improve the imaging of converted wave data in the presence of noise, missing data, and poor illumination
Least squares migration with quadratic regularization L extrapolates P & S potentials and recompose into data components
The forward operator Extrapolation and wavefield recomposition: Split-Step Padé Fourier propagator Blending of wavefields into data components
What is H -1 ? H -1 blends wavefield potentials into data components
What is H -1 ? H -1 blends wavefield potentials into data components
What is H -1 ? H -1 blends wavefield potentials into data components
What is H -1 ? H -1 blends wavefield potentials into data components If we assume isotropy we can use Helmholtz decomposition
Helmholtz decomposition
Etgen, 1988
Helmholtz recomposition Etgen, 1988
The forward operator Extrapolation and wavefield recomposition:
The adjoint operator wavefield decomposition and extrapolation: implies that the adjoint operator could result in some crosstalk artifacts
Quadratic Regularization
Angle domain regularization By a change of variables z = Dm we write Where D -1 is smoothing across angles within each angle gather
Poynting vector method Imaging is done without spatial lags Vectors are calculated from source and receiver side wavefields independently From the source and receiver side Poynting vectors the angle can be defined in many different ways Higginbotham et al, 2010
Poynting vectors Typically implemented in RTM: This approximation leads to a technique to calculate Poynting vectors in WEM (Dickens and Winbow, 2011) (Yoon and Marfurt, 2006)
Poynting vectors in WEM 1.calculate gradient components for the source side wavefield: 2.inverse Fourier transform over the spatial axes (x and z) 3.obtain that corresponds to the time of reflection by calculating the zero- lag cross correlation with the receiver wavefield: 4.normalize the elements of 5.repeat steps 1 to 4 for the z-component of the source side wavefield to obtain 6.repeat steps 1 to 5 for the receiver side wavefield
Unit vectors
Angle with respect to vertical: Angle with respect to reflector normal: Sign of angle (Duan and Sava 2014): Conversion from unit vectors to angle
Three Interfaces
m ps for 1 shot gather
Angles for 1 shot gather
Corrected gather
MARMOUSI 2 Synthetic data example
vpvp
vsvs
Shot gathers
Migration: m pp 25˚ incidence angle
LS Migration: m pp 25˚ incidence angle
Migration: m ps 25˚ incidence angle
LS Migration: m ps 25˚ incidence angle
ANTICLINE MODEL Synthetic data example
Simulated OBC Acquisition ★ osx = 100m ; dsx = 100m ; nsx = 69 ; sz = 10m ; dgx = 8m ; gz = 550m ; recording aperture = 6608m
X-Component
Z-Component
Angles for 1 migrated shot
Adjoint Mpp Constant Incidence angle of 10˚
Least Squares Mpp Constant Incidence angle of 10˚
Adjoint Mpp x = 1600m
Least Squares Mpp x = 1600m
Adjoint Mps Constant Incidence angle of 10˚
Least Squares Mps Constant Incidence angle of 10˚
Adjoint Mps x = 1600m
Least Squares Mps x = 1600m
Misfit Iteration number Relative misfit
Conclusions We implemented elastic least squares migration using the one way wave equation The forward operator consists of scalar extrapolation of P and S potentials followed by wavefield recomposition The method has application in imaging, regularization and wavefield separation of multicomponent data
Acknowledgements We gratefully acknowledge the sponsors of the Signal Analysis and Imaging Group for their generous support