Estimating primaries by sparse inversion, a generalized approach

摘要

For an accurate interpretation of seismic data, multiple-free data are of great value. Removing surface multiples and interbed multiples proves to be challenging in many cases. The nowadays widely used method of Surface-Related Multiple Elimination (SRME) has lately been redefined as a full-waveform inversion process, resulting in the method of Estimation of Primaries by Sparse Inversion (EPSI). The new method is shown to be more accurate than the former method in several situations, because it estimates primaries such that they, together with their multiples, explain the input data. Its main advantage is that the minimum energy assumption in traditional multiple subtraction is avoided. The SRME methodology has been extended to the case of internal multiples by several authors, however, the involved subtraction of predicted multiples is probably even more challenging than for the surface-multiple case. Therefore, in this paper the EPSI method is generalized to remove both surface and interbed multiples. As in previous implementations of internal multiple removal based on data-driven convolution, the newly proposed scheme requires some knowledge about the subsurface: the data should be divided into (macro) layers and appropriate time windows must be selected. The method is tested on two 2D synthetic datasets to prove its viability. Furthermore, application to a 2D field dataset showed improved accuracy compared to conventional prediction and subtraction.