Th ELI1 8 Efficient Land Seismic Acquisition Sampling Using Rotational Data P. Edme* (Schlumberger Gould Research), E. Muyzert (Sclumberger Gould Research) & E. Kragh (Schlumberger Gould Research) SUMMARY We further investigate the benefit of a novel ground-roll attenuation method using newly available measurements of the rotational components of the wavefield. This method is based on adaptive subtraction using the rotational components as noise models. Field tests demonstrate that these noise models are significantly better than those provided by the horizontal components of conventional 3C geophone data. Because the technique is applied locally at an individual receiver station, the coherent noise attenuation performance is independent of the spatial sampling, in contrast to multi-trace velocity filtering. This approach can allow a relaxation of the field effort associated with spatial sampling to attenuate ground-roll and therefore impact overall acquisition costs. 76 th EAGE Conference & Exhibition 214 Amsterdam RAI, The Netherlands, 16-19 June 214
Introduction Survey design for land seismic programs typically deals with two spatial sampling regimes, one for the desired reflection signal and the other for coherent noise propagating with relatively slow apparent velocities (ground roll). The requirements for noise sampling are generally more onerous than for signal. The standard methods to attenuate ground-roll are based on the combination of spatial filtering in the field, through multi-element source and receiver arrays, and multi-channel velocity filtering in data processing, which allows discrimination between the apparent velocities of signal (fast) and coherent noise (slow), under the condition that the wavetrains are recorded spatially unaliased, requiring additional field effort. As an alternative, the acquisition of multi-component (3C) data has been proposed where the horizontal data components are used as noise models for ground-roll attenuation based on adaptive subtraction or polarization filtering (e.g. Özbek, 2; Franco and Musacchio, 21). This 3C noise attenuation strategy has the advantage of being local to the sensor. With no further spatial sampling criteria related to noise attenuation to be satisfied, spatial sampling in the field can be for signalonly, allowing a reduction in overall field effort. In this paper, we demonstrate the benefit of using rotational data instead of horizontal geophone data as noise models for adaptive noise subtraction and we compare this novel approach to conventional multi-trace velocity filtering. Rotational data Conventional geophone sensors measure the translational components of the seismic wavefield. The rotational components, or equivalently the curl of the wavefield, have so far been largely neglected in the seismic industry, probably due to the fact that robust dedicated rotational sensors are not yet commercially available (with the required fidelity for seismic applications). However, as demonstrated by Edme and Muyzert (213), the rotational components at the surface can be estimated by differencing the outputs of closely spaced conventional vertical geophones. With the appropriate distance between sensors for the differencing (typically less than a third of the smallest wavelength of interest), the estimated rotational data can be locally expressed as slowness-scaled versions of the vertical wavefield, i.e. for a given plane wave: (1) where Z is the vertical component of the acceleration, R XZ and R YZ are the rotational components around the two perpendicular horizontal axes, in the X-Z and Y-Z plane respectively. The local scaling factors p X and p Y are the inline and crossline horizontal slowness (inverse of apparent velocity, also commonly called ray parameter). The slowness scaling of equation 1 implies that: - Slowly propagating events (typically ground-roll) are amplified, as opposed to fast propagating events (typically refraction/reflection signal) that are attenuated (all relative to Z). - Rotational data contain mostly noise (inline and crossline propagating for R XZ and R YZ respectively) and, therefore, provide noise models. - Rotational data polarization (versus Z) is linear for both signal and noise, in contrast to horizontal geophone data whose noise polarization can be very complex. - Rotational data cannot be used for polarization filtering but are a better representation of the noise (i.e. exhibit a better coherency with Z) than horizontal geophone data because they are less contaminated by S waves or Love waves. Most of these expectations can be observed in the field data example of Figure 1. This small dataset was acquired in quite homogeneous near-surface conditions. However, even in these conditions, it can be seen that both the X and Y components contain events that do not correlate with Z (i.e. S waves and Love waves for X and Y, respectively). In comparison, the rotational data are a better representation of the noise, as demonstrated by the significantly better noise attenuation results after adaptive subtraction. Using both the R XZ and R YZ components allows further improvement in the results, by also attenuating side scattering noise (the latter possibility was presented by Edme et al. 213). Added synthetic reflections (at 55ms and 9ms) allowed us to check for potential signal damage. 76 th EAGE Conference & Exhibition 214 Amsterdam RAI, The Netherlands, 16-19 June 214
After adaptive subtraction Z using X using R XZ using R XZ & R YZ 5 1 5 1 5 1 5 1 15 15 15 15 Y X R XZ R YZ Love wave S-wave 5 1 ( ) 5 1 5 1 5 1 15 2 4 6 15 2 4 6 15 2 4 6 15 2 4 6 Figure 1 Common-receiver gather. Top panels: Z-component data before and after adaptive subtraction using different noise models. Bottom panel: Noise models used for the adaptive subtraction. The same parameterization was used for the subtraction (window length of 5 ms and filter length of 12 ms). The same t-x gain was applied on all data. Field data test UAE A field test experiment was conducted in the desert near Al-Ain, United Arab Emirates. This part of the Middle East is notorious for its complex near-surface conditions (due to large sand dunes for example), resulting in complex behavior of the multiply scattered ground-roll. The acquisition geometry was a crooked 2D line of 34 source points at 12 m spacing and 1 receiver stations also at 12 m spacing, where each station consists of a 3C geophone plus two additional vertical geophones separated by 62.5 cm (see Figure 2). The small, triangular-shaped single-sensor, multi-receiver groups are termed Composite-Point-Receivers (CPR). The R XZ and R YZ components are obtained by differencing the vertical geophone data in the inline and crossline direction, respectively. Data QC showed reasonably good quality of the estimated rotational components with less than 3% bad traces due to sensor coupling and/or orientation perturbations. This is the first realistic seismic dataset where both multi-component and rotational component are recorded. These data allow us to compare the effectiveness of ground-roll attenuation using either the rotational data or horizontal geophone data as noise model(s), as well as to compare these results to 76 th EAGE Conference & Exhibition 214 Amsterdam RAI, The Netherlands, 16-19 June 214
conventional multi-channel velocity filtering. Figure 3 shows the vertical component (commonreceiver gather) before and after noise attenuation using the different techniques. 3C 6m 6m 3C 62.5cm CPR #1 CPR #2 Figure 2 Layout design for the CPR acquisition. The receiver line is composed of 1 CPRs. Comparison of panels b and c further demonstrates that rotational data provide better noise models than horizontal geophone data. With the same parameterization for the adaptive subtraction (1 ms sliding window, 2 ms filter length), the result is significantly better with the rotational data (~12 db signal-to-noise ratio S/N improvement) than with the horizontal geophone data (~6 db S/N improvement). The reflection signal (which consists of added synthetics) remains masked below the noise cone in the t-x domain, but now clearly appears in the frequency-wavenumber domain (bottom panels). Note that the added synthetic reflections are also used to check for eventual signal damage (minor with the chosen parameterization). Figure 3 also compares the adaptive subtraction approach with conventional multi-trace velocity filtering. While the latter technique is more effective when the data are not spatially aliased (i.e. a trace interval of dx=12 m), comparison of panels c and d shows that the local adaptive subtraction offers better performance when the wavefield is more sparsely sampled (here dx=24 m). Spatial aliasing prevents the velocity filter to be effective (~5 db S/N improvement), despite a multi-pass filtering applied both in the shot and receiver domains. The effectiveness of the local subtraction technique is further confirmed by the stacked images (not shown here due to space limitations). In addition to providing a new approach to ground-roll attenuation and potentially allowing a relaxation of the sampling effort, the proposed technique does not suffer from other perturbation issues associated with large areal geophones arrays (e.g. due to statics and coupling variability). The use of multiple measurements per CPR station, with both inline and crossline rotational components as noise models allows attenuating inline as well as side-scattered noises. Conclusions We present a novel ground-roll attenuation method that takes advantage of a new type of seismic data: the rotational component of the wavefield. The technique is based on adaptive subtraction using estimated rotational components as noise models. Field tests demonstrate that these new noise models are significantly better than those provided by the horizontal components of conventional 3C geophone data. The technique is local to the sensor station, in contrast to methods based on array spatial filtering and like multi-trace velocity filtering. Field tests demonstrate that significantly better results are obtained when the data are spatially aliased. The method can allow a relaxation of the field effort associated with spatial sampling to attenuate ground-roll and therefore impact overall acquisition costs. Acknowledgements We thank numerous colleagues in WesternGeco and Schlumberger, in particular Nicolas Goujon, Tristan Hollande, Peter Nyhuus and the WesternGeco UAE field crew who assisted in the field trial. 76 th EAGE Conference & Exhibition 214 Amsterdam RAI, The Netherlands, 16-19 June 214
References Edme, P., Daly, M., Muyzert, E. and Kragh, E. [213] Sparse acquisition using rotational data. 75 th EAGE Conference & Exhibition, Extended Abstract. Franco, R. and Musacchio, G. [21] Polarization filter with singular value decomposition. Geophysics, 66, 932-938. Edme, P. and Muyzert, E. [213] Rotational data measurements. 75 th Exhibition, Extended Abstract. EAGE Conference & Özbek, A. [2] Multichannel adaptive interference cancelling. 7 th Annual International Meeting, SEG, Expanded Abstract, 288-291. a) b) c) d) Z (receiver gather) After AsubX After AsubR After efx 5 5 5 5 1 1 1 1 15 15 15 15 2 2 2 2 25 25 25 25 3 3 5 1 15 3 5 1 15 3 5 1 15 5 1 15 1 1 1 1 2 3 4 2 3 4 2 3 4 2 3 4 5 5 5 5 6 -.2.2 6 -.2.2 6 -.2.2 6 -.2.2 Figure 3 Common receiver Z gather before noise attenuation (a) and after noise attenuation using b) X and Y as noise models, c) R XZ and R YZ as noise models, d) velocity filtering (considering a receiver and source sampling of 24m). Bottom panels show the corresponding fk spectra (with colour map ranging from to2 db). The same t-x gain is applied to all data. 76 th EAGE Conference & Exhibition 214 Amsterdam RAI, The Netherlands, 16-19 June 214