If a large number of closely spaced parallel events buried in noise fall in a window, the calculated filter will consist of the central main spike plus small contributions from all parallel events. With a large number of parallel events, the contribution of any one event parallel to the event being predicted will be small. For example, data in the Gulf of Mexico generally has many closely spaced reflections. If a long prediction filter in time is used with this data, events far from the output will contribute to the output point, but since each contribution is small, the spurious events generated by the filter are unlikely to be a problem. In an area with only a few reflections that are widely spaced, parallel events may generate significant spurious events as seen in Figures and .
These demonstrations should not be construed as suggesting that lateral predictions produce unreliable answers. The spurious events are low amplitude and will generally be hidden by the remaining noise. Furthermore, if many parallel events exist within a design window, the spurious events generated are weakened and distributed over time so that the effect is reduced. Figure shows an f-x prediction over a dataset with many parallel events. While many of the expected spurious events are very weak, a spurious event close to the original events on the left is almost as strong as those seen in Figure . I suspect that these phenomena will cause problems only in cases where a few isolated reflections appear in a background of noise, and where weak reflections are being sought. In this case, a t-x prediction rather than an f-x prediction should be used to avoid generating events separated from the original events by significant times. Spurious events that appear as a change of wavelet are less likely to be interpreted as new events, but may interfere with the interpretation of subtle stratigraphic changes.