A comparison of different methods of velocity analysis based on cross-correlation in reflection seismic data processing

This paper outlines the velocity analysis introduction and its structure briefly and then a comparison of calculating velocity spectra using six velocity analysis methods. Conventional methods of velocity analysis are equivalent to modeling prestack seismic data with events that have a hyperbolic moveout. Examples are included which demonstrate the depth, overlapping events and the details of the closed layers.
   Various types of measured coherency that can be used as attributes in computing velocity spectra are described here and some of them which are discussed in this paper are: Stacked amplitude (S), Normalized stacked amplitude (NS), Unnormalized cross-correlation (CC), Energy-normalized cross-correlation (EC), Semblance (NE) and AB Semblance. It is should be noted that all these methods are based on the correlation between the traces. The equations of different methods are described and coded based on the corresponding literature and then applied to synthetic data. The results of these methods are compared in velocity and time with each other and their accuracy are examined. Four factors are introduced to compare the results of the velocity analysis methods. The first factor is the contrast which means the ratio of the coherency in the exact velocity to the coherency of the near velocities. The second factor is the smearing on data that measures the accuracy of the method used in detecting the velocity and the time of the events; it equals to the smearing in half of the difference between the coherency of the pick and the average coherency of the background. The third factor is the ability of the velocity distinction, it means how much a method can make difference between two near velocities. And the last factor is the ability of time distinction which indicates how much a velocity analysis method is able to detect two near layers with higher resolution. Fitness plots compare the performances of the six methods when the velocity analysis is done on the same events in both time and velocity aspects. The sharpness of the fitness curves is in the relation with the velocity and time resolution. Then we introduced more noise to data and discussed the effect of noise on the quality of the velocity analysis. Also the effect of the noise contamination is clearly explained and can be seen in another fitness plot. Additionally synthetic data contains various multiples and overlapping events with different changes. Finally, CC, Semblance, and AB Semblance led to the best results, however; the AB Semblance proves its accuracy by maximizing a coherent measure in correct velocities and times and also by minimizing a coherent measure in incorrect velocities and times. Compared to the AB Semblance the vulnerability of the other methods to the coherent noise is better understood.
   In the next step, these methods were applied to real data belonging to one of the southern oil fields in Iran and again the AB Semblance led to the best results. This method, in contrast to others, does not stretch velocities and displays shallow events as clear as deep events. We concluded that the AB Semblance method is able calculate the velocity and distinguish between the closed layers clearly better than the other methods.