Ensemble-based deconvolution: when and why

David C. Henley, Carlos A. Montaña, Gary F. Margrave

Techniques for deconvolving seismic data often use the statistical properties of the data themselves in designing operators to apply to the seismic traces. In the early stages of seismic processing, individual seismic traces are usually members of one or more ensembles like shot gathers or receiver gathers, and their statistical properties are related not only to their own intrinsic character, but also to that of neighbouring traces within the ensemble. We demonstrate that seismic traces contaminated primarily by bands of coherent noise are often best deconvolved singly by a non-stationary algorithm like Gabor deconvolution, but traces uniformly contaminated by varying levels of random noise are better deconvolved by estimating an average operator for all the traces in an ensemble.