Multiples in reflection seismic data ...use 'em versus lose 'em
Multiples are seismic events that have undergone more than one upward reflection during propagation. These events are information-bearing, but they have a complex relationship with the Earth. Researchers have to decide at the outset if they are going to use multiples to extract information, or treat them as noise and suppress them. Art Weglein and collaborators developed a set of methods ([1],[2]) for the removal of these events from multidimensional data without assuming anything about the Earth's variability or knowing, e.g., the velocity structure of the Earth below the experiment. They are widely used in industry, and remain state-of-the-art technology 15 years on. Something nice that we're learning is that these processing methods are of real value regardless of our "use 'em" or "lose 'em" stance. We do a bit of both.
LOSE 'EM: IMPLEMENTATION
Sam Kaplan has written transparent, fully-documented code implementing multidimensional versions of the Internal Multiple Attenuation (IMA) and Free-Surface Multiple Elimination (FSME) algorithms. Those codes have been released to sponsors, and are also now being used in a series of research projects and collaborations. Below you can see a quick illustration of data shot records (left) and corresponding IMA output (right).
LOSE 'EM: ADAPTIVE SEPARATION
Adaptive subtraction is the name given to a final step in de-multiple algorithms, in which the small imperfections of the multiple predictions are cleaned up so that they can be subtracted directly from the data. This is called adaptive subtraction (e.g., [3]). Sam has posed subtraction as a source separation problem using Independent Component Analysis ([4]). The data and predictions of the multiples at several orders (seen below) are input into the algorithm, which extracts the primaries (the signal) as one independent component.
USE 'EM: CORRECTING PRIMARIES FOR TRANSMISSION THROUGH AN UNKNOWN OVERBURDEN
Jose Lira has been exploring a couple of ways of using multiples to process primaries. The main thing he's doing is taking advantage of a known property of the inverse scattering series internal multiple attenuation algorithm, which is that the amplitude of the predicted multiple is in error by exactly the transmission down to, and across, the interface where it experiences its downward reflection. Using the picture below, the multiple abcdijkl, when predicted by the events abcdef, ghijkl, and ghef, in comparison with the actuall multiple provides a means to correct one or other of the events to the right. Correction of these primaries for transmission through an overburden, is a key goal of current imaging. Jose has submitted a paper on this just recently.
