Suppression of water bottom multiples in shallow seismic data by predictive deconvolution

Abstract

Predictive deconvolution is a digital signal processing technique widely used to compress the seismic wavelet and remove multiples in seismic reflection data. In this paper, predictive deconvolution is used to suppress water bottom multiples in shallow seismic data collected with the IKB Seistec TM single channel seismic system. These data were acquired in the St John River at Fredericton during the summers of 2001 and 2002. The convolutional model is a method of explaining the origin of reflection seismograms, which are recordings of seismic pulses reflected back to surface from subsurface boundaries between different layers of sediment or rock. Deconvolution is a method of attempting to reverse convolution process and estimate the subsurface reflectivity based on an analysis of the recorded seismogram. In other words, convolutional model is estimate the seismic wave response with the earth reflectivity, but deconvolution is the reverse, finding the reflectivity. Initial tests of predictive deconvolution have proven that partially successful in removing multiples from the marine seismic data set. However, the results have not been as good as we had hoped. We examine field data, synthetic data and the underlying assumptions of the predictive deconvolution process to identity a number of possible reasons for our sub-standard result including (i) the large noise pulse at time zero in our seismic data, and (ii) the fact that our source wavelet does not appear to be minimum phase.