Abstract

Interbed multiples are frequently a problem at target levels in seismic data. These multiples are often generated in high velocity layers, and consequently have little differential moveout compared to the primaries with which they interfere. In addition, statistical methods such as predictive deconvolution frequently fail to remove these multiples (for example as a result of geological complexity) or else they attenuate primary energy beyond a level that is considered acceptable. In such cases, the only remaining approaches to attenuating the multiples are based on modelling and subtraction. This paper describes a modelling and subtraction based method that uses primary reflections in the data themselves to predict interbed multiples via a technique analogous to that used to predict surface multiples. The paper concentrates on the practical aspects of the method and includes synthetic and field examples. It is shown that reasonable predictions of interbed multiples can be obtained provided great care is taken at every stage of the data preparation. It is envisaged that, although presently in their infancy and computationally very expensive, methods similar to this will find their place in future processing flows when alternative algorithms are not successful.