Residual-moveout analysis in presence of strong lateral velocity anomalies

Introduction

Methods to perform wave-equation migration velocity analysis without requiring velocity-spectra picking are attractive and have been the focus of substantial effort at SEP (Biondi, 2010,2008; Zhang and Biondi, 2011). Robustness should be an important characteristic of these methods. Therefore, we have focused on algorithms that extract velocity information from migrated angle-domain common image gathers (ADCIG) using a one-parameter residual-moveout analysis. One-parameter moveout analysis has the advantage of being more robust to noise, imaging artifacts, and cycle skipping than alternative methods for measuring residual moveout. It has been extensively used for ray-based migration velocity analysis where it has proven to provide useful information for velocity updating when the moveout parameter is picked from stack-power, or semblance, scans. However, it has not been extensively tested when the velocity information is extracted by computing the derivative of the focusing measure (e.g. semblance or stack power) around the origin of the moveout-parameter axis, as is required by the automatic methods we have been developing.

A one-parameter moveout cannot accurately describe the actual moveout of the migrated gathers in some important cases, such as in the presence of strong lateral velocity anomalies and anisotropy. When the velocity errors are large and the migrated gathers display a significant (i.e. larger than the dominant wavelength in the image) moveout at wide angles, numerical differentiation of stack-power scans can be prone to errors caused by cycle skipping.

To test the robustness of one-parameter moveout analysis for automatic wave-equation migration velocity analysis, I performed numerical experiments on a simple synthetic data set. I started from the angle-domain image generated from data that were modeled assuming a strong velocity anomaly, but migrated with a constant background velocity. In the migrated image there are areas that illustrate both the challenges described above (complex residual moveout and cycle skipping caused by large velocity errors). At the original frequency band of the data (25 Hz dominant frequency) the straightforward computation of the gradients would likely result in poor convergence. After I applied a low-pass filter to the data (high cut at 8 Hz) the gradient becomes better behaved. However, seismic data are not currently recorded with sufficient signal-to-noise ratio at arbitrarily low frequencies. Smoothing the velocity spectra along the moveout parameter axis is a simple remedy that does not require low-frequency data. This smoothing is sufficient to overcome the problems identified from the test data set. Numerical differentiation of smoothed stack-power scans provides useful information to be used for a tomographic update even when the data are migrated at full bandwidth.

Residual-moveout analysis in presence of strong lateral velocity anomalies