INTRODUCTIONIn a previous paper, we presented a fully data driven method, Shallow Water Demultiple (SWD), based on the concept of multichannel prediction for attenuating surface related short period multiples in shallow water (Hung et al., 2010b). The described method is robust, but it may have problems with the first few orders of multiples generated by the seafloor (and any events just beneath the seafloor) when the seafloor is very shallow and hence its reflections are indistinct or in the postcritical angle range. In this case, the model of Water Layer Related Multiples (WLRMs) obtained from the convolution of multichannel prediction operators with seismic data is dominated by low frequencies, primarily due to the near-offset extrapolation required for the convolution process. This affects the multiple removal. Solutions derived from modelbased method (Pica et al., 2005) have recently been proposed for handling reverberations due to the shallow water layer. For instance, Wang et al. (2011) designed a model-based waterlayer demultiple (MWD) method that predicts multiples from a shallow seafloor, overcoming the above-mentioned issue faced by SWD; but at the expense of requiring accurate information of the multiple generators. For practical reasons, we normally only pick the seafloor, among all the shallow multiple generators, for the application of MWD. For very shallow water environment, however, even the seafloor may not be readily available.In this paper, we present a hybrid approach combining the strengths of SWD and model-based methods. We show that the multichannel prediction operator estimated by SWD can be used as an accurate kinematic representation of the seafloor reflection with a higher signal-to-noise (S/N) ratio, allowing the travel time of the seafloor to be automatically estimated. With this information the Green's functions of water-layer primary reflections can be modelled and used for multiple prediction and removal. We first present the theory of the method and then demonstrate its effectiveness on synthetics and field data acquired offshore Australia.
METHOD AND RESULTSThe DELPHI feedback model (Verschuur et al., 2005) forms the basis of surface related multiple elimination (SRME). In practice, the prediction is usually the first step of SRME, and in the form of (1) where M is multiple model, ΔP is primary response, P is acquired seismic data, and w is source wavelet. By formulating this in another way, Biersteker (2001) describes it as a multichannel prediction problem:where F is the time and offset truncated multichannel prediction operator, which can be estimated by minimizing the prediction error between seismic data P and multiple M.Comparing with the SRME formulation in Equation (1), the operator F is equivalent to a scaled version of the primaries (Hargreaves, 2006), we can make use of this property in performing MWD, but instead of requiring a known waterlayer model, we propose to derive the Green's function of the water-layer reflections by utilizing the travel time information of...