Recognition and reconstruction of coherent energy with application to deepseismic reflection data

Reflections in deep seismic reflection data tend to be visible on only a li mited number of traces in a common midpoint gather. To prevent stack degene ration, any noncoherent reflection energy has to be removed. In this paper, a standard classification technique in remote sensing is pre sented to enhance data quality. It consists of a recognition technique to d etect and extract coherent energy in both common shot gathers and final sta cks. This technique uses the statistics of a picked seismic phase to obtain the likelihood distribution of its presence. Multiplication of this likeli hood distribution with the original data results in a "cleaned up" section. Application of the technique to data from a deep seismic reflection experi ment enhanced the visibility of all reflectors considerably. Because the recognition technique cannot produce an estimate of "missing" d ata, it is extended with a reconstruction method. Two methods are proposed: application of semblance weighted local slant stacks after recognition, an d direct recognition in the linear tau-p domain. In both cases, the power o f the stacking process to increase the signal-to-noise ratio is combined wi th the direct selection of only specific seismic phases. The joint applicat ion of recognition and reconstruction resulted in data images which showed reflectors more clearly than application of a single technique.