Summary. Introduction

Transcription

1 Model-based coherent noise attenuation for complex dispersive waves Claudio Strobbia, Alexander Zarkhidze, Roger May, John Quigley and Phil Bilsby, WesternGeco Summary The attenuation of source-generated coherent noise energy can be a challenging problem for land data where surface waves often exhibit complex behavior with multiple propagation modes, high lateral variability and relatively short wavelengths. The traditional acquisition and processing strategy for mitigation of coherent noise has combined analog spatial filtering through source and receiver arrays in the field, with multi-channel digital filtering in data processing. The field arrays act as complementary spatial anti-alias filters for data processing algorithms which have difficulty in dealing with aliased events. Limitations of the available processing procedures place constraints on the acquisition design which can potentially both limit flexibility and increase the cost of the acquisition. A new model-based approach to sourcegenerated coherent noise attenuation is presented, where the local properties of the multi-mode surface waves are estimated from the seismic data and used to generate a detailed model of surface-wave noise, spatially-variable over the survey area. The method has significant advantages with respect to the handling of aliased coherent noise energy, and robustness to spatial irregularities. The availability of effective processing tools for aliased noise attenuation can have a significant impact on required survey geometry, and on the cost of land exploration. Introduction Land field records can be heavily contaminated by different noise modes, in particular source-generated interface waves propagating in the near-surface. The attenuation of such coherent noise is a data processing step of primary importance to obtain clean high-resolution images of the subsurface and to extract reliable reservoir attributes. Despite progress in data acquisition and processing, many cases are still very challenging. Such interface waves are often of very large amplitude and may be highly dispersive with wave trains of long time duration. They appear as linear events only in laterally homogeneous media. The presence of large lateral variations in near-surface properties, together with the existence of multiple propagation modes, can introduce significant complexity to the kinematics of this part of the wavefield. Acquisition spatial sampling can be a critical factor as often very short noise wavelengths are involved, and aliasing can occur. When using receiver arrays in acquisition, one objective is to attenuate wavelength components of the propagating coherent noise equal or less than the dimension of the array itself. With the Stack Array concept, the field arrays provide a spatial anti-alias filter for acquired trace interval. In practice, coherent noise wavelengths longer than the array dimensions plus partially attenuated shorter wavelength noise are passed to data processing where, depending on the acquired trace interval, issues with aliasing can occur. Further, where source and receiver arrays do not have uniformly isotropic spatial response characteristics, and this is generally the case, azimuthallydependent distortions are introduced into the residual coherent noise present on the acquired traces. This complex distortion of the detailed characteristics of coherent noise propagation, due to the anisotropic array response, can be difficult to characterize, and ultimately proves to be a drawback with the traditional techniques. As practical land survey designs increasingly move to the full-azimuth ideal and we wish to exploit azimuthal processing and attribute extraction, azimuthal variation in the acquired datasets due to field practice are becoming a more significant problem for the processing geophysicist. Conventional data processing methods for coherent noise attenuation primarily consist of dip filters, such as FK and linear Radon filters, FX fan filters, and others, and are usually applied in a cascaded way to get the optimum response. Multichannel filtering, widely used with different geometries for both exploration and appraisal surveys, have some drawbacks. The filter impulse response can introduce artifacts (footprints) in the data and smear local anomalies affecting the noise and discontinuities. Where the multichannel algorithm has a strict assumption of regular input trace spacing, geometrical irregularities in the input data pose further challenges and may require data regularization. The most important and significant limitation is related to aliased components of the coherent noise, having apparent velocities and frequency content which do not allow effective discrimination and removal in a local data-driven filtering scheme. An alternative is the use of model-based techniques. With surface-wave analysis, modeling, and inversion (SWAMI), the surface-wave propagation properties are extracted, inverted for the near-surface characterization, and used to generate a surface-wave model to be subtracted (Strobbia et al., 2010). Significant advantages of this approach are the ability to deal with large spatial variability of the noise properties, spatial irregularity of the acquisition geometry, and a relaxation of the effective spatial sampling required in the field to handle coherent noise aliasing in a robust way. SEG San Antonio 2011 Annual Meeting 3571

2 The procedure comprises two steps. Firstly surface-consistent surface-wave propagation properties are extracted from the seismic field data with the highest possible spatial resolution; local, irregularly sampled trace super-gathers are created, which allow estimation of the surface-wave properties at wavenumbers beyond the nominal Nyquist wavenumber; an appropriate azimuth selection allows minimization of the impact of field arrays, if needed. For this initial analysis stage, an adaptive 3D aperture is designed, and all available data are used to exploit statistical data redundancy and generate the best possible areal volume of the surface wave propagation properties. Then, after the analysis, surface-wave noise model traces are generated integrating through the propagation property volume, location-by-location, using the actual acquisition geometry and local propagation properties. The modeling stage respects the true coordinates and source-detector distance of a source-receiver pair and can model noise which would appear aliased on the acquisition trace interval. The SWAMI method has been applied to several datasets from the Middle East, North Africa and Australia. The development of robust aliased noise attenuation has particular relevance for survey design and acquisition economics since it allows some relaxation of the sampling criteria for coherent noise in the field and supports the use of array-less point-source and point-receiver acquisition, with elimination of anisotropic array effects in the acquired data. Surface-wave analysis, modeling, and inversion In the past few years, methods have been developed for the processing and analysis of surface wave from high-volume production seismic surveys, to unlock the additional information they contain. Surface wave dispersion and attenuation characteristics can be estimated with a very high lateral resolution even in large 3D surveys. These property estimates are used for characterization of the near-surface using either a single-domain surface wave inversion, or in a joint inversion with other data, such as refraction traveltimes (Re et al., 2010). The near-surface characterization techniques are particularly robust in the presence of complex near-surface conditions, such as velocity inversions, and can be applied for statics estimation, for velocity modeling in the land depth imaging workflow. The availability of high-resolution areal volumes of the surface wave propagation properties also allows us to model surface waves and subtract them from the raw data. Such a simulation does not require a complete earth model, but can be based on a surface-consistent distribution of the propagation properties. An approximation of the surface wave equation in laterally varying media is used. The phase term is obtained by integrating along the horizontal rays in the wavenumber volume; the amplitude is predicted and matched to local amplitude spectra. An example of raw data and predicted noise, in a medium with lateral variations, is shown in figure 1. Figure 1: Example modeling and subtraction of one surface-wave mode. From left to right: real data, noise model, and straight subtraction. SEG San Antonio 2011 Annual Meeting 3572

3 Estimation of aliased coherent noise The method does not involve the application of multichannel filtering to the input trace data and can generate a noise model including aliased components. Local surface wave properties are estimated spatially, using all available data, the redundancy of 3D geometries and the properties of wavefield transforms for locally linear events; the properties can be extracted beyond Nyquist. The analysis in fact extracts the surface wave properties in supergathers, formed with frequency dependent aperture and offset selection. The radial surface wave phase, in these supergathers, is not evenly sampled, and therefore there is no fixed defined Nyquist wavenumber. This exploits the azimuthal response of a linear array for plane waves. The double integration over sources and receivers, varying offsets and azimuth, further enhances locally linear events. As an example, in figure 2, for a section of a receiver line with point receivers, we plot the simplified array response to a linear plane event with a wavenumber that at some frequencies exceeds three times the Nyquist wavenumber of the inline sampling. resolution for the longer wavelengths and maximum wavenumber for the short wavelengths. Regularity and smoothness criteria for the surface wave modal wavenumber (i.e. the relationship between wavenumber and frequency, for each of the surface wave modes) allows us to further increase the robustness of the estimation of the surface wave properties beyond the nominal Nyquist aliasing wavenumber. The spatial correlation of the surface wave property distributions allows an efficient QC of the analysis results. The generation of surface wave noise model traces including aliased noise components is performed for every input trace individually, without multi-channel filtering, based on its source-receiver coordinates. Although point-receiver and/or point-source acquisition is preferred, the presented method can also be effectively applied to data acquired with geophone and/or source arrays. However, where the use of arrays has distorted the information content of the surface waves in the short wavelength range this may lead to a decrease in the lateral resolution of the dispersion volume. Field arrays in 3D surveys, particularly linear inline arrays have a response which is azimuthally dependent. The analysis of such data sampling of the surface wave wavefield can be improved using a proper selection of data in azimuth and offset, from 3D acquisition geometry. The coarse inline sampling that can affect the inline response of the array can be mitigated by analyzing the linear events in the radial dimension for multiple shot locations. An example of modeling of aliased noise is shown in figure 3. Multiple propagation modes are present and the mode with lowest velocity is heavily aliased. Receiver spacing is 50 m. Benefits for data processing and acquisition A model-based method for the attenuation of complex dispersive noise, including aliased modes, presents a series of benefits for both the data processing and the acquisition. Figure 2: Array response of a supergather. A) The inline array response and B) the overall response for the 70-Hz frequency are shown together with C) the FK spectrum for a constant velocity event. With an appropriate data selection and with a frequency dependent selection of the azimuth boundaries for the integration, the properties can be extracted beyond the nominal Nyquist wavenumber, optimizing spectral Robustness is a key requirement: this predictive, modelbased noise attenuation method works with irregular input geometry, without the need of regularization or interpolation of data or of filter coefficients. Since the method is based on intrinsically using spatial property distributions extracted from the surface waves in the seismic data, space-variant parameterization of the process is in general not required and execution parameters are adapted automatically to the local properties. SEG San Antonio 2011 Annual Meeting 3573

4 From the data processing point of view, generating the noise model trace-by-trace, using the detailed volume of the noise propagation properties, avoids the negative effects of multichannel filtering, such as smearing and impulse responses left in the data. Sharp transitions in the nearsurface can create artifacts which do not stack out during processing, particularly affecting seismic attributes used to estimate reservoir properties. In fact the more detail we have in the recorded surface wave energy, the better the computed noise model, and the better and easier the match to the input data. Avoiding aggressive adaptive subtraction can help prevent signal damage during the subtraction process. The method is therefore signal friendly, also because intrinsically targets only a limited region of the data space where the noise is present, and is used only with very mild adaptive subtraction parameterization. Coupled with the ability to remove the aliased components of the surface waves this provides an improved, higher resolution dataset for the further controlled true amplitude processing, resulting in more reliable attributes for AVO and inversion. On the acquisition side, the opportunity to relax constraints on coherent noise sampling allows more flexibility in designing seismic surveys for the core objective of signal sampling. The introduction of acquisition systems with high source productivity, very high numbers of pointreceivers, in combination with processing techniques such as SWAMI provides enabling technology for practical high-density, full-azimuth, 3D surveys leveraging the latest illumination and imaging workflows. Conclusions We have presented a technology for coherent noise attenuation based on modeling dispersive modes in laterally varying media. The noise model can be created even for complex propagation and for irregular geometries. One of the key advantages is that it can be used to predict and remove aliased noise. It offers advantages over conventional approaches with finely spaced point-source and receiver data and it can be used for superior signal protection and preservation of seismic attributes, especially in the case of complex noise. With such data, it is part of an integrated workflow where the surface waves are analyzed, inverted for near-surface characterization, and modeled for noise attenuation. The technology has significant implications on land survey design, significantly changing the effort required in micro vs. macro geometry. It is predicted that a typical land survey will have live spreads that cover larger areas per shot, but fewer channels and sweeps per station, giving an overall improvement in resolution and illumination. With coarser conventional acquisition geometries, when the sampling is insufficient for the propagating wavelengths, the method can be used to predict and subtract aliased noise. Acknowledgements The authors thank WesternGeco for permission to publish this work. Figure 3: Example of modeling and subtraction of aliased surface waves. From left to right: real data, noise model, and straight subtraction SEG San Antonio 2011 Annual Meeting 3574

5 EDITED REFERENCES Note: This reference list is a copy-edited version of the reference list submitted by the author. Reference lists for the 2011 SEG Technical Program Expanded Abstracts have been copy edited so that references provided with the online metadata for each paper will achieve a high degree of linking to cited sources that appear on the Web. REFERENCES Re, S., C. Strobbia, M. De Stefano, and M. Virgilio, 2010, Simultaneous joint inversion of refracted and surface waves: Presented at the 80th Annual International Meeting, SEG. Strobbia C., P. L. Vermeer, A. Laake, A. Glushchenko, and S. Re, 2010, Surface waves: Processing, inversion, and attenuation: First Break, 28, no. 8, SEG San Antonio 2011 Annual Meeting 3575