Conclusion

Two methods for subtracting surface-related multiples were presented. One method removes the multiples by adaptive subtraction, assuming that the primaries have minimum energy. One method removes the multiples with a pattern-recognition technique, assuming that the primaries and multiples have different multivariate spectra (patterns). Tests on a 2D synthetic dataset show that the pattern-based technique tends to separate primaries and multiples better than adaptive subtraction. In cases where the primaries and multiples are correlated, however, pattern-recognition can damage primaries. This effect is amplified by the Spitz approximation which prevents the noise and signal PEFs from spanning similar areas of the data space.