Suppressing nonphysical reflections in Green’s function estimates using source-receiver interferometry

Abstract

Seismic interferometry retrieves the Green’s function propagating between two receiver locations using their recordings from an enclosing boundary of sources. Theory requires that sources completely surround the two receivers, but constraints in exploration seismology restrict sources to locations near the surface of the earth. Seismic interferometry by crosscorrelation then introduces usually undesirable nonphysical reflections (spurious multiples) in the Green’s function estimates. We found that the dominant nonphysical reflections can be converted into physical reflections via convolution using source-receiver interferometry. The resultant Green’s functions display fewer nonphysical reflections and show significantly better agreement with the true Green’s functions than those obtained using crosscorrelational interferometry. Nonphysical reflections can be further suppressed by iterating the convolution step. By comparing the velocity spectra of the Green’s functions retrieved by crosscorrelational and source-receiver interferometry, we can retrospectively identify the dominant nonphysical reflections introduced by crosscorrelational interferometry. We found that the nonphysical reflections are particularly important for constructing the primary reflections and internal multiples in source-receiver interferometry. This is because the primary reflections and internal multiples cannot be created via the convolution of physical reflections. Instead, the primary reflections and internal multiples are retrieved by the appropriate convolution between a nonphysical and physical reflection. We compared crosscorrelational interferometry and source-receiver interferometry using synthetic towed streamer data for a 1D acoustic and 2.5D elastic model, respectively. We also found that the nonphysical reflections obtained using crosscorrelational interferometry allow for the direct estimation of interval velocities and layer thicknesses without the need to use Dix inversion in the 1D example.