Summary We are interested in the problem of separating primary and multiple seismic signals based on their statistical properties. We present a novel adaptive subtraction method based on an information maximization principle. Compared with previous methods, our proposed method has the theoretical advantages of utilizing higher-order statistics of the data and incorporating the filtering nature of the adaptive subtraction problem into our algorithm formulation. Furthermore, our proposed method provides a flexible framework for simultaneously subtracting more than one set of multiple predictions from the input data. We use simulations to show that our proposed adaptive subtraction method outperforms the popular least-square adaptive subtraction and the independent component analysis methods both quantitatively, as measured by the mean-squared error, and qualitatively, as evaluated by the visual quality of the image reconstruction.

ABSTRACT No preview is available for this paper.

ABSTRACT Seismic reflection data obtained with a 486-channel streamer cable can be processed by array-forming techniques that include optimal weighting of individual channels and/or beam steering. Such processing can improve the resolution of horizons at early and intermediate times as well as enhance the continuity and clarity of later reflections. The originally recorded data may also be processed to simulate results obtainable with a wide variety of conventional streamers. To demonstrate data enhancement obtained by these techniques, a selected line was surveyed in the Gulf of Mexico, first with the 486-channel system and then with a conventional 48-channel streamer. Comparison of stacked sections shows that: long, steered arrays can enhance deeper events while retaining the high frequency content of shallow data, and short arrays at small group intervals allow finer resolution of shallow events. When the 486-channel data were processed to duplicate the 48-channel data, the larger system yielded somewhat better results; this improvement may be attributed in part to reduced cable noise. INTRODUCTION Since the beginning of seismic reflection exploration, there has been a growing trend toward more complex and specialized geophone spreads in the field. Examples of this trend include long arrays designed to suppress multiple reflections and diffraction noise, optimally weighted and non-uniformly spaced arrays designed to attenuate coherent noise such as ground roll, and ministreamers designed to obtain high-resolution shallow data. Although these techniques have provided improved sections in selected areas, the trend of specialization puts an undesirable constraint on the planning, processing, and interpretation of seismic surveys. Also "optimum" geophone arrays are often determined from past experience or by analyzing field tests. These practices can lead to resurveys when serious problems are discovered in the course of processing initial survey data or when interpretation targets change at later dates. Furthermore, no particular array design can provide optimized data at all depths over an entire survey. This constraint is especially limiting for marine surveys because a standard streamer cannot easily be reconfigured in the field. Innovations in electronics make possible a seismic streamer cable of 500 or more independent data channels. With such a large number of short receiver groups distributed within a standard length cable (say 3 km), the spatial density of observation points is comparable to that for individual hydrophones within the arrays of standard streamers. The individual groups in the SOO-channel streamer are, however, short enough to be considered as points. As a result, the recorded data are not constrained by restrictions imposed by a fixed choice of array design. Since routine processing of such a large number of channels could be prohibitively expensive, reduction of the data by array-forming is desirable at present. Now, however, because the array design occurs at the data processing stage, a new realm of options becomes available. In this paper implications of several of these processing options are investigated for a seismic line recorded with a 486-channel streamer in the Gulf of Mexico.

ABSTRACT Data-driven demultiple techniques such as the common surface-related multiple elimination method (SRME) rely on dense, regular data input, a requirement that is too strict to be met by towed-streamer acquisition. Full-azimuth, long offset acquisition is required to properly image complex structures in the Gulf of Mexico and other complex geologic settings. The move to these more complete geometries, such as circular dual-coil shooting, has increased the offset and azimuth scope of the input required by SRME. Generalized surface multiple prediction (GSMP) is an efficient 3D, true-azimuth variant of SRME that reduces the burden on input data regularization and interpolation by applying "on-the-fly" interpolation. This interpolation can be combined with increased data density prior to GSMP to achieve optimal predictions for demanding full-azimuth, long-offset geometries such as circular shooting. GSMP selects input traces for interpolation by minimizing the distance to desired traces, where distance is measured in four dimensions: midpoint and offset, and azimuth. This distance function can be analyzed to show potential uplift from various input geometries. In this paper, we show the improvement of circular-shooting multiple prediction through the addition of wide azimuth data, then analyze the input error trend to assess potential for further improvement. Presentation Date: Monday, October 17, 2016 Start Time: 1:50:00 PM Location: 142 Presentation Type: ORAL

ABSTRACT No preview is available for this paper.

ABSTRACT We describe a workflow to optimize surface related multiple elimination (SRME) aperture constrained by the travel path of the main sea-surface-generated multiples. An efficient ray-tracing algorithm captures the position of downward reflection points (secondary sources) at the sea surface for key multiple signatures, such as first and higher order salt-related peg legs. The secondary source’s distribution, relative to the midpoint of the target traces, is then adjusted by the extent of the Fresnel zone. Using spatially variant SRME aperture increases efficiency of modeling simple multiples while also improving the quality of complex multiples, which translates to both shorter turnaround time and better image quality, free of destructive multiple interference. Presentation Date: Wednesday, September 27, 2017 Start Time: 9:20 AM Location: 370A Presentation Type: ORAL

ABSTRACT Multivessel circular data acquisition (dual-coil shooting), with full-azimuth and long-offset coverage, has significantly improved our capability to image complex subsalt structures. While dual-coil data provides richer azimuth coverage and have longer offsets, the coverage for large offsets is generally less uniform than the dominant azimuths in traditional wide-azimuth and narrow-azimuth data. This irregularity in data distribution for long-offset coil data can lead to an inferior 3D surface-related multiple elimination (SRME) result. For this case study, we took acquired dual-coil data as input and used 5D interpolation via Matching Pursuit Fourier Interpolation (MPFI) (Schonewille et al., 2013) to output a supplemental data set on a regularized grid with evenly distributed sources and receivers. The advantage of this workflow, compared to differential moveout correction performed inside SRME, is that MPFI, with its antialiasing capability, offers much better events reconstruction, especially for traces in small to medium sized gaps. Dual coil acquisition is designed to have nonuniform spatial distribution of sources and receivers and this is a positive feature for any wavefield reconstruction algorithm. After the data reconstruction by MPFI onto a regular grid, both the supplemental dataset and the acquired data are included as input to perform 3D SRME for the large-offset coil data. This case study shows that the resulting multiple model matches the seismic measurements better and leads to a cleaner migrated image. Presentation Date: Wednesday, September 27, 2017 Start Time: 10:10 AM Location: 370A Presentation Type: ORAL

ABSTRACT No preview is available for this paper.

ABSTRACT Adaptive Noise Cancellation (ANC) was applied to three different noise-contamination problems encountered in seismic data acquisition. The solution to each problem required modifications to the generic ANC algorithm, as well as a unique means of obtaining a noise estimate. In all three cases ANC dramatically attenuated the noise without disturbing the underlying signal. INTRODUCTION The traditional means of dealing with noise that contaminates seismic data is to try and solve theproblem in the field. For example, a land crew experiencing 60-Hz power-line interference might insert a 60-Hz notch filter into their recording system. Sometimes, source and receiver array patterns are designed so as to attenuate bothersome noise. For some environmental noise, all too often a crew has no choice but to cease recording and wait for a quiter time. None of these solutions are satisfactory. Using special filters and arrays can degrade the desired signal; waiting for a quiter time is costly. Modern seismic recording systems, especially the new 24-bit systems, have a dynamic range wide enough to faithfully record the seismic signal, even in the presence of relatively strong noise. Under these renditions it becomes feasible to deal with noise during processing rather than during data acquisition. Generally, if the time and spatial properties of a particular noise allow it to be delineated from a seismic signal in some domain, then one can find a digital filter that attenuates the noise without harming the signal. For some types of noise though, a standard-type filter is not the best choice. For example, a digital 60-Hz notch filter corrupts the seismic signal, just as does its analog field counterpart. Adaptive Noise Cancellation is an alternative way of removing noise contamination from a signal ANC is, in essence, a scheme for optimally estimating the noise component of a measurement and then subtracting the estimated noise from the measurement. In this paper first explain how a generic ANC algorithm works. Then, I discuss attempts to use this technique to attenuate three kinds of obstinate noise sometimes found in seismic data. These were: single-frequency noise, such as power-line interference, coherent noise picked up by geophones in a dual-sensor oceanbottom cable, and environmental noise, such as that from nearby shipping lanes, found in streamer shot records. ADAPTIVE NOISE CANCELLATION The block diagram in Fig. 1 shows the main features of a simple, one-channel, digital adaptive noise canceller. Notice that the noise canceller has two inputs, one output, and a feedback loop. The primary input is a desired signal that has been corrupted by additive noise. The secondary input is an in exact "measurement" of the noise that contaminates the primary input. in practice (as will be seen below), the secondary input can be obtained in several ways: it may bean actual measurement, it maybe an educated guess at the noise, or it may even be derived from the primary input in some fashion. An ANC is based on the premise that the secondary input, when convolved with some digital filter.

ABSTRACT No preview is available for this paper.

Multiple contamination both surface related and interbeded are common and known to hinder the interpretation, reservoir characterization and inversion studies. Surface related multiple is always present and strong on marine data, but not often a problem on land data. Interbed multiple is rather dependent on the subsurface structures and not always obvious, however in presence of strong subsurface reflectors, interbed multiple contamination could be marked, this article mainly focus on IME – Interbed Multiple Elimination. Two methodologies including data driven Extended SRME IME in combination with radon transform based demultiplex and ISS (inversion scattering series) based IME are introduced. Good demultiple results are demonstrated by case study.

ABSTRACT Green’s theorem derived deghosting methods in the space-frequency domain using a horizontal measurement surface (M.S.) have been successfully applied to synthetic and field data. Based on Green’s theorem wavefield separation theory, this paper derives a 3D source and receiver deghosting formula for a depth-variable M.S. assuming its topography is known. In numerical tests, the model has a free surface and one horizontal reflector. We use the Cagniard-de Hoop method to generate synthetic data on horizontal, inclined, and undulated measurement surfaces, respectively. Numerical results show that the current Green’s theorem deghosting formula for a constant depth M.S. remains useful for a mildly depth-variable M.S.. When the actual M.S. deviates significantly from horizontal, the horizontal M.S. formula produces serious errors and artifacts whereas the new formula produces an e?ective and satisfactory result. While the analysis and tests in this paper are based on nonhorizontal towed streamers, the motivation (and future work) is for on-shore and ocean bottom acquisition. Under these circumstances, the deviation from horizontal acquisition can be significant and the ability to accommodate a variable topography can have a considerably positive impact on subsequence processing and interpretation objectives. Presentation Date: Monday, September 25, 2017 Start Time: 3:05 PM Location: 360A Presentation Type: ORAL

ABSTRACT Green’s theorem derived deghosting methods in the (-) domain using flat cables have been successfully applied to synthetic and field data. Based on Green’s theorem wavefield separation concepts, this paper derives a 2D acoustic receiver deghosting formula for a depth-variable cable assuming the shape of the cable is known. In numerical tests, the air-water boundary is assumed to be horizontal. We use the Cagniardde Hoop method to generate synthetic data on parabolic cables and on periodically semicircular cables, respectively. The normal derivative of the total field on the cable is assumed to be known and is estimated by finite difference. Numerical results show that current Green’s theorem up/down separation formula for a constant depth cable remains useful for a mildly depth-variable cable. When the actual cable deviates significantly from horizontal, the horizontal cable formula produces serious errors and artifacts whereas the new formula produces an effective and satisfactory result. While the analysis and tests in this paper are based on nonhorizontal towed streamers, the motivation (and future work) is for on-shore and ocean bottom acquisition. Under these circumstances, the deviation from horizontal acquisition can be significant and the ability to accommodate a variable topography can have a considerably positive impact on subsequence processing and interpretation objectives. Presentation Date: Wednesday, October 19, 2016 Start Time: 8:25:00 AM Location: 140 Presentation Type: ORAL

Summary In the deep-water regions of the Gulf of Mexico, our ability to image subsalt structures has improved significantly with wide-azimuth data, reverse time migration, anisotropic imaging and tomography. However, subsalt velocity updating remains difficult in areas with complex salt geometry. Features like steeply dipping salt flanks and subsalt three-way closures create scenarios where limited incident angles are available for residual curvature analysis and velocity updating. Long-offset and full-azimuth acquisition configurations are the latest acquisition technology designed to address these subsalt challenges. This paper presents improved tomography update and subsalt imaging from a staggered acquisition, which provides both ultra-long offset and full-azimuth data coverage.

ABSTRACT No preview is available for this paper.

ABSTRACT In seismic exploration, acquiring full surface coverage with point receiver data is generally consistent with wave theoretic processing requirements. It is important to characterize the receiver array effect on any wave theory based seismic processing method. Deghosting is a prerequisite for free surface multiple removal (FSMR), internal multiple attenuation/elimination, and the imaging and inversion of primaries. The effectiveness of deghosting directly affects the performance of these methods. We propose a new method for deghosting towed streamer data that modifies and extends the Weglein et al. (2002) algorithm and make use of the wavelet method presented by Guo et al. (2005). We test the method for a simple 1D acoustic model and conclude that very accurate deghosting results can be obtained when using point receiver data. Useful but less accurate results can also be obtained when using receiver array data.

ABSTRACT Data-driven SRME techniques do not require any knowledge of the subsurface (reflectivity, structures and velocities). However, these methods require a shot location at each receiver location, wherein lies the main difficulty for their 3D implementation. Today solutions involve reconstruction of the missing data or reconstruction of the missing multiple contributions. In the following, we present a model-based surface-related multiple modeling technique (SRMM) free from any constraint relating to the shot position (including OBC) and distribution.

Summary When ocean bottom cables (OBC) are used for seismic surveying it has become standard practice to remove water layer reverberations by combining the geophone and hydrophone data. One of the most important processing concerns in such dual-sensor summation has been high levels of geophone noise. While many techniques have been tried to address strong geophone noise (such as match filters and coherent noise rejection) in many cases the only workable solution is not to use the majority of the geophone data for a given receiver in the dual sensor summation (often by allocating it very small scalar multipliers). This paper describes a new technique, Dual Wavefield Noise Attenuation (DWNA), which attempts to attenuate high magnitude geophone noise on dual sensor data. DWNA uses dual sensor theory to identify signal and noise and separate the two elements of the seismic recording in an adaptive manner.

Abstract The "data-driven" feedback method for modeling and attenuation of free-surface multiples was applied in two case studies for sub-basalt and for sub-salt imaging, where multiple attenuation is an important and challenging processing step. Despite limitations due to cable feathering in one case and to 3-D wide-tow acquisition geometry in the other case, the feedback method performed better than conventional multiple attenuation methods. Two features of the new method appeared as particularly attractive during the case studies: velocity picking or interpretation of horizons were not required prior to the application of the method, and, primary events, including locally converted PSP events with slow apparent velocities, were well preserved. Further improvements in the performance of the feedback modeling and attenuation method could be expected as a result of optimized acquisition geometries and of recently introduced acquisition hardware for streamer steering and control. Introduction Several recent papers 1, 2, 3, 4 have documented benefits from "data-driven" algorithms which model and attenuate surface related multiples without requiring knowledge of sub-surface velocities, interpretation of main multiple generators or periodicity of the multiples. It is also well established that the performance of such methods is degraded when conditions of "dataset integrity" are not met 3 . In this paper we discuss two case study applications of the feedback approach 5 to modeling and subtraction of freesurface multiples. In each case study, strong multiples are a key processing issue not well solved by conventional velocity filtering or deconvolution methods. On the other hand, the feedback and related multiple attenuation methods have to overcome problems due to the sub-optimal wavefield sampling that can be currently achieved in practice. The first case study discusses multiple attenuation in the context of sub-basalt imaging using long offset (11.4 km) streamer data. The feedback multiple attenuation method improves upon the results of conventional multiple attenuation methods and in addition demonstrates preservation of locally converted PSP primaries. The performance of the multiple attenuation algorithms is related to the feathering observed in the data, which is variable and occasionally very strong. The second case study compares multiple attenuation results obtained for a typical wide-tow 3-D survey acquired in the Gulf of Mexico for deep-water sub-salt exploration. A pragmatic extension 6 of the feedback method to 3-D geometries performs better than the conventional parabolic Radon transform, even though the performance of the method degrades significantly as the cross-line distance increases. Sub-basalt imaging using long offset 2-D streamer data The challenges of seismic imaging below basalt 7 have led to innovative acquisition and processing approaches. Improvements in data quality have been reported as a result of acquisition and processing using long or ultra-long offsets 8 and imaging either with locally converted PSP waves 9, 10 or with P-waves 11, 12, 13. The rock properties of basalts and their seismic responses, in particular the relative strengths of PSP and P waves reflected beneath basalt, vary considerably 14, 15, 16.

ABSTRACT No preview is available for this paper.