Estimating VP/VS ratios using smooth dynamic image warping

Abstract

Multicomponent seismic images are composed of different combinations of downgoing and upgoing wavefields. Each wave mode has different propagation velocity and polarization direction and thus carries unique, direction-dependent information about the subsurface. Differences in propagation velocity cause events in converted-wave or PS images to appear at later times than the P-wave or PP image counterparts. Reflectivities are different for each wave mode, and therefore, multicomponent images are not related simply by time shifts. These complications historically required that the alignment, also called registration, of corresponding image features be done manually, which is a tedious process. To register PP and PS images automatically, we used a smooth dynamic image-warping algorithm that can be accurate with respect to problems unrelated to time shifts, such as differences in noise and reflection waveforms. Interval [Formula: see text] ratios can be estimated from derivatives of vertical shifts that align reflections in PP and PS images. To optimize accuracy of estimated time shifts and [Formula: see text] ratios, we automatically constructed a coarse lattice of points located on reflections with high amplitudes and then estimated time shifts at only those image samples. By adjusting the coarseness of the lattice, we trade off resolution of changes in [Formula: see text] with increased accuracy in [Formula: see text] estimates. By processing 3D PP and PS images, we learned that our estimates of [Formula: see text] cannot be obtained by smoothing time shifts estimated at every image sample.