In Figure , the straightforward 3-D interpolation scheme of SEP-73 is applied. For each output location, we select the N (here N=15) nearest neighbors. Each pair of traces is cross-correlated, and the results are combined to give an estimate of event coherency as a function of dip. The dip which gives the maximum coherency is chosen, and the traces are summed along this trajectory to give a new trace at the desired output location. Although the stack is over fifteen traces, the noise has not been attenuated significantly.
In Figure , the stack is weighted by the coherency, computed as a function of time, measured along the best dip direction. (Actually semblance is being used here instead of the generalized coherency of SEP-73.) This weighting supplements the noise suppression power of the stack and makes a cleaner result. In Figure , the median is used instead of the mean, and in Figure , both the median and semblance weight are used. This combination gives the best overall noise suppression.
An example of velocity filtering is shown in Figure . Here we have limited the algorithm to scan over only a certain range of dips; the more steeply dipping event is filtered out. Note that this method does not attempt to deal with the problem of aliased energy.