Seismic multiple attenuation based on pre-stack reflectivity modelling

Abstract

A particular method of pre-stack multiple attenuation, based on generalised linear inversion (GLI) and the Haskell-matrix formulation, is investigated. The method uses GLI to obtain a 2-D earth-reflectivity function which produces a synthetic seismic record as close as possible to the observed seismic data. The forward model employs the Haskell-matrix method to compute the entire elastic response, including primaries and multiples, corresponding to an input reflectivity. The final reflectivity is used to generate a multiples-only signal, which is subtracted from the original. Initial trials have been carried out on noisy synthetic pre-stack gathers. Despite erroneous starting earth models, the inversion iterates robustly to provide an output record exhibiting excellent agreement with the observed record. This leads to effective multiple attenuation. The Haskell matrix method is naturally formulated in terms of wave slowness, and hence the inversion algorithm is most conveniently carried out in either the t-p domain, or the f-p domain. Our experimentation suggests that, in the presence of noise, the f-p domain is more robust than t-p. This method is computationally more expensive than conventional multiple removal strategies, such as those based on differential moveout or predictive deconvolution. Hence it is likely to have most potential where these approaches fail.