Short period demultiple using iterative second order multi-dimensional predictive deconvolution

2D deconvolution is an attractive approach for short period multiple prediction as it does not require direct recording of the multiple generator and can model multiples relating to more than one multiple generator at a time. One drawback, however, relates to an inherent over prediction of mixed side multiples leading to an inconsistency in the amplitude of multiple predictions from one multiple order to the next. We present a modified form of predictive deconvolution that iteratively refines the prediction operator and prevents the over prediction of higher order multiples. The resulting multiple prediction is consistent with the amplitude of multiples in the recorded data, reducing the necessity for adaptive subtraction. The algorithm may be applied in 2D or 3D, or alternatively with a receiver only side 3D implementation suited to towed streamer geometries. The effectiveness of the algorithm is demonstrated on synthetic data as well as on two towed streamer 3D seismic data examples acquired in the North Sea.