Marine data and preprocessing

I tested the Monte Carlo automatic velocity picking algorithm on 300 CMPs of Gulf Coast data, donated to Stanford Exploration Project by Western Geophysical. Out of historical interest, this is the data set that Hale (1984) used in his classic paper on dip moveout analysis (DMO).

Each CMP gather is 24-fold and of 4 second record length. The midpoint spacing is 0.033 km, and the offset spacing is 0.134 km. I bandpassed the 4 ms data with a sixth-order Butterworth 10-40 Hz filter and decimated to 8 ms. I applied a mute at water velocity with a 250 ms ramp to complete the preprocessing prior to velocity analysis.

I ran my version of an NMO stacking velocity analysis on each of the 300 preprocessed CMP gathers (7200 traces total). I chose a stacking velocity range of 1.4-3.0 km/s in 0.040 km/s increments. I perform offset weighting, $\cos^2\theta$ weighting, and a time dependent divergence correction, during the computation of the velocity semblance spectra. Next, I do a little post-processing of the scans by doing a time-gated ``AGC'' using peak semblance, as a function of velocity, within a moving time window of 0.5 seconds, followed by a lateral moving average of 10 CMPs ( $\pm$ 165 m). This has the effect of normalizing and enhancing small semblance peaks as a function of time, and boosting semblance S/N by about 3:1. This careful preprocessing and velocity analysis has resulted in an excellent quality of the velocity semblance scans prior to picking, as is visible in Figure () as compared to the single fold CMP velocity scan in Figure ().