GNGTS 2014 - Atti del 33° Convegno Nazionale

58 GNGTS 2014 S essione 3.1 surface seismic. SRME algorithm has some manifest advantages with respect to other multiple elimination techniques: in particular its accuracy and its fully data-driven nature that allows the application in the early stages of processing propose it as the benchmark for surface related multiple estimation and elimination. SRME approach is sketched in figure 1b: multiple reflection ray-paths can be divided into two (or more) reflected ray-paths (color-coded in Fig. 1b), each of which have been recorded by a different trace. Thus, for each input trace s-r , practical implementation of SRME involves the convolution of couples of traces s-n and n-r , where n is any possible Downward Reflection Point (DRP) belonging to a regularly sampled surface ϕ . The subsequent stack of the so-called Multiple Contribution Gather (MCG) allows to estimate the surface related multiples m(t) recorded by trace s-r : (1) The term w -1 represents the deconvolution by the source wavelet (which is often neglected in practical implementation). Then, an adaptive subtraction step allows to separate the interfering wavefields. 3D SRME limitations are mainly related to its computational cost, requiring appropriate strategies for data regridding (Dragoset et al. , 2010) and for Multiple Contribution Gather (MCG) aperture optimization (Bienati et al. , 2012). Moreover, it must be noted that the model independency of 3D SRME comes from a strict dependency on acquisition completeness, that cannot be satisfied by any practical acquisitions. SRME technique cannot be directly applied to WVSP acquisition geometries, as no data is recorded at sea-surface, therefore preventing the correct construction of theMultipleContribution Gather. Different approaches have been presented to overcome this limitation. Ma et al. (2011) proposed to use legacy surface data to complete WVSP acquisition geometries: however, a marine acquisition with the required coverage and orientation around the well cannot be always available. Moreover, such an acquisition, if it exists, have been recorded with different data resolution and in different environmental conditions. Conversely, Hokstad and Sollie (2005) proposed to model the missing surface data, by the mean of the information recorded by WVSP survey only. In particular their purpose is to simulate the ocean-bottom primary reflection only, using an approximation based on the DMO formula. Indeed, different multiple predictions based on Wave Equation Modeling (WEM) have been proposed to handle incomplete data acquisition for surface acquisitions (Wiggins, 1988; Pica et al. , 2005). Wang (2011) has shown how a Model-based Water-layer Demultiple (MWD) algorithm (based on the accurate knowledge of the bathymetry only) can outperform SRME data-driven approaches especially for shallow water environment, because of missing near-offset data and the poor quality of water layer primary reflections in the recorded data (as they are weak at large angles and often contaminated by other arrivals such as direct and refracted waves). In the present work, we similarly substitute the s-n trace recorded at sea-surface, that ismandatory for SRME convolutional formulation but missing inWVSP survey, with the Green’s function of the Water Layer Primary reflection only (dotted blue ray-paths in Fig. 1b): it can be easily and accurately computed when water velocity and bathymetry are known. Although only a subset of surface related multiples can be predicted, Water Layer Multiples w.l.m amplitudes may be significantly higher either than the reflection signals (which are attenuated during the propagation through the subsurface) or other surface-related multiples, lying in the same time window. Furthermore, hard seafloor and high structural attenuation in the subsurface exacerbate these effects. The surface- related multiple estimation procedure described in Eq. (1) is re-written as: (2)