This study used data from a three-month continuous

Transcription

1 SPECIAL Applications SECTION: and challenges Applications in and shear-wave challenges exploration in shear-wave exploration S-waves and the near surface: A time-lapse study of S-wave velocity KRISTOF DE MEERSMAN, CGGVeritas This study used data from a three-month continuous reservoir monitoring experiment in Peace River, Alberta, Canada to measure spatial and temporal variations in nearsurface S-wave velocity ( ) and attenuation ( ) in the weathering layer The permanently buried sources generate a strong refracted S-wave that was recorded on buried 3C receivers A method to perform receiver-side up-down separation and extract primary and ghost S-wavefields is presented These wavefields are then used to measure nearsurface for the near-surface layer above the buried receivers, which is the top 12 m in this case The measured values range between 180 and 220 m/s values between 8 and 22 Maps of show a robust correlation (low with low ) and a clear spatial variation that can be associated with soil type Both properties increase slowly over a three-month period in one section of the survey area while remaining constant in another The cumulative increase in is 10% Introduction Seismic exploration for shallow targets using S-waves or converted PS-waves should theoretically provide higher resolution images than P-waves The foundation for this statement is that lower S-wave velocities result in shorter wavelengths for the same temporal frequency, thereby offering greater resolution (Bale and Stewart, 2002) In practice, this perceived S-wave advantage hardly ever materializes and the blame is typically put on the near-surface weathering layer The near surface is believed to be highly attenuating (low ) as well as being characterized by extremely low and heterogeneous S-wave velocities Evidence to support these claims is mostly indirect and comes predominantly from processing PS data These data constitute the majority of exploration S-wave data sets and are often plagued by large S-wave statics, indicating a low, and a reduced bandwidth restricted to lower frequencies, further indicating a low Here, we like to address two questions regarding near-surface S-wave velocity ( ) and attenuation ( ) for a study area that is approximately 1 km 2 and located in Peace River, Alberta, Canada First, what is the spatial distribution and value of these parameters over an area that is expected to suffer from poor S-wave data quality? Second, do these properties change significantly over time and by how much? Answering these questions allows us to assess the impact of near-surface on our ability to conduct meaningful reservoir monitoring using (S-wave) time-lapse surveys Buried multicomponent receiver arrays: Land ghosts and how to use them to derive near-surface properties Recently, a number of Canadian oil sands operators have started experimenting with permanently buried receiver acquisition In some cases, the motivation to do this is purely economic: while the upfront costs are higher there may be a significant long-term cost saving In other cases, the motivation has been an attempt at improving the quality of PP and PS data by burying the receivers below the worst part of the weathering layer This may seem like a good idea at first, but the truth is that simply burying your receivers will replace one problem (the near surface) for another more difficult problem (receiver ghosts that are affected by the near surface) A receiver ghost is a free-surface multiple that arrives closely after its primary arrival (Figure 1) Seismic energy originating at a depth below a buried receiver array will be detected by this array as it propagates up toward the Earth s surface After reflecting at the Earth s surface, the wave will travel down into the Earth and pass the buried array a second time The first, upgoing, arrival is typically referred to as the primary wavefield whereas the second, downgoing arrival is often referred to as the ghost wavefield The upgoing arrival is not affected by the near-surface layer above the receivers while the downgoing is Ghosts are typically considered a Figure 1 Illustration of a plane SV-wave as it propagates up toward the Earth s surface where it is reflected down and back into the Earth Both the upgoing and downgoing wavefronts are recorded on a spread of buried 3C receivers The black arrows represent rays and propagation directions The wavefront is bold The blue and green arrows represent the polarity and direction of particle motion of the upgoing and downgoing signal SV energy does not propagate outside the plane of the drawing 40 The Leading Edge January 2013

2 simple inversion of this equation, yielding: (2) Note that this separation depends only on the ray angle and that it becomes singular for angles approaching = 0 and = 90 In practice, this singularity limits the applicability of the method to arrival angles that are sufficiently apart from horizontal and vertical, with a tolerance dependent on the signal-to-noise ratio of the data The ray angle is a space and time-dependent property and the generalization of Equation 2 to include these effects is straightforward We have also assumed that there is no S-to-P mode conversion at the free surface (SV reflection coefficient = 1) and no anisotropy of any symmetry Figure 2 Geometric relationship between the polarization vectors of the upgoing (blue) and downgoing (green) SV arrivals as they are recorded on the vertical and radial component of a buried 3C receiver The ray angle is defined as the angle between the vertical direction and the upgoing or downgoing rays noise problem, but here we will exploit them to derive nearsurface properties Arrival time differences between primary and ghost contain information on seismic velocity while amplitude differences contain information about attenuation for the layers above the receiver array S-wave up-down separation on buried 3C data Up-down separation of P-waves usually involves the combination of data recorded on hydrophones and vertically oriented geophones (PZ summation) or hydrophone data recorded at different depths (eg, slanted streamers) to remove the ghost wavefield Our method for S-wave deghosting uses data only from a single level of buried 3C accelerometers or geophones The separation method is for SV-type S-waves in the absence of azimuthal anisotropy Figure 1 shows a plane SV-wave and its receiver ghost as they propagate past a set of buried 3C receivers If the 3C data have been rotated to a radial-transverse-vertical coordinate system (Gaiser, 1999), no SV energy is expected on the transverse component so that it can be ignored The result is a two-component vertical-radial system Figure 2 shows the projections of the primary and ghost polarization vectors with respect to this system Primary (P) and ghost (G) energy will be recorded on both the vertical (V) and radial (R) component If defines the angle between vertical and the ray direction as per Figure 2, then the data recorded on the vertical and radial can be written as: (1) In compact matrix notation, d represents the matrix of recorded data, A is the ray-dependent projection matrix and m is the matrix with upgoing and downgoing wavefields Updown separation of the S-wave can be achieved through a Data: The 84-day continuous monitoring experiment The data set used in this experiment comes from a 4D monitoring experiment conducted by Shell and CGGVeritas at the Peace River heavy oil field in Alberta, Canada The survey consisted of 482 3C geophones buried at a depth of 12 m and 11 continuously emitting piezo-electric mini-vibrator SeisMovie sources cemented at a depth of 80 m (Forgues et al, 2011) The weathering layer in this area is an approximately 100-m thick glacial till The typical receiver and line intervals are 20 m and 180 m and the entire spread covers an area of approximately 1 km 2 The system recorded continuously for 84 days from 7 June until 29 August 2009, generating a daily data set of 11 3C shot gathers The principal goal of this acquisition was to monitor steam-assisted enhanced oil recovery through high-resolution P-wave time-lapse analysis However, as we will demonstrate later, the survey design and acquisition system is also perfectly suited to our S-wave deghosting method and subsequent time-lapse analysis of nearsurface The piezo-electric seismic source is critical to our analysis and merits a more detailed introduction It is effectively a dipole source with its long axis oriented in the vertical direction The theoretical far-field radiation pattern for such a source is rotationally symmetric about the vertical axis and contains both P-wave and SV-wave energy (Figure 3) Significantly more SV-wave energy is emitted than P-wave energy, 2 V with an S/P amplitude ratio that is proportional to P = 2 2 (Pujol and Herrmann, 1990) Most P-wave energy is emitted vertically, while no P-wave energy is emitted horizontally SVwave energy is preferentially emitted at an angle of 45 from the vertical and both up and down No SV-wave energy is emitted vertically or horizontally In short, while this source is well suited for P-wave reflection experiments, it will generate S-wave energy at take-off angles that are beyond the scope of a typical reflection survey Upward-emitted S-wave energy is expected to produce a strong direct arrival on the near offsets Downward-emitted S-wave energy is expected to refract at close proximity to the source as a result of the steep take-off angles (Figure 3) The velocity contrast at base of the glacial till is an excellent candidate to generate this refraction, which is recorded at medium-to-far offsets January 2013 The Leading Edge 41

3 Figure 3 A vertically oriented piezo-electric vibrating seismic dipole source is cemented in a well The source radiation pattern shown is rotationally symmetric about the vertical The source emits P-wave (blue) and SV-wave (red) energy SV-wave energy is emitted predominantly at angles close to 45 from vertical, both up and down Figure 4 shows the vertical and radial component of a typical shot gather The direct and refracted S-wave arrivals and their receiver ghosts are identified The prospect that most of the S-wave signal in our data is refracted energy is encouraging as it simplifies the up-down separation Refractions propagate to the surface with a constant ray parameter, meaning that they will pass the receiver array with a constant ray angle We can use Equation 2 without having to use a time- and offset-dependent ray angle or a multichannel transformation (eg, tau-p or f-k) In order to determine the optimal ray angle to perform the up-down separation of the refracted S-wave, a simple grid search was used The angle estimate was assessed by how well it minimized upgoing residual energy in the downgoing wavefield estimate This yielded a ray angle = 30 Figure 5 shows vertical and radial component data from a single receiver as well as the upgoing and downgoing wavefield estimates The data are shown at true relative amplitude and sorted by shot point and acquisition day This display is ideal to observe temporal changes in amplitude and arrival time at a specific receiver The ghost is delayed by approximately 100 ms relative to the primary While there are no significant time-lapse changes in the upgoing first arrival, we can observe a clear gradual 10-ms decrease in the arrival of its downgoing ghost There is also a significant amplitude difference between primary and ghost While divergence effects may contribute somewhat, there is no indication of a significant free-surface S-to-P conversion Near-surface time-lapse velocity analysis The time delay between primary and ghost arrivals can be used to estimate the average velocity in the near-surface at each receiver The velocity relates to the upper 12 m, or burial depth The S-wave refraction and its ghost are approximated by a local plane wave propagating along straight rays within the near surface Figure 6 shows an upgoing arrival recorded at D as well as the point B on the upgoing wavefront that will produce the downgoing arrival at D The time-delay Figure 4 Example of a typical (a) vertical- and (b) radial-component shot gather t between the upgoing and downgoing arrivals relates to a traveled distance of BC + CD The near-surface S-wave velocity is then: BC CD t 2zcos t Here, t is measured by crosscorrelating the up- and downgoing arrivals while the reflection angle = 30 is the same as used for the up-down separation and z is the receiver depth (12 m) For each receiver we obtain up to 11 (one for every shot) estimates for near-surface for each day of acquisition Nearoffset traces were removed from the analysis for lack of refracted energy A three-day running mean is calculated for each receiver and each day and mapped The resulting near-surface velocity contour maps for the first and last day of acquisition are shown in Figure 7a and Figure 7b A contour map of the difference is shown in Figure 7c Contour maps of subsequent days are extremely consistent and show only gradual changes The maps show that the near-surface S-wave velocity does not change with time in the northwest corner of the survey area and is between 180 and 220 m/s with a typical value of 200 m/s (3) 42 The Leading Edge January 2013

4 Figure 5 Example of a single-receiver gather sorted by source point and day Each gather consists of 11 panels (one for each source), each of which consists of 84 traces (one for each day) (a) Vertical component (b) Radial component (c) Upgoing wavefield (d) Downgoing wavefield In the southeast, we observe typical near-surface velocities that are only slightly greater than those in the northwest at the start of the experiment but increase gradually over time The cumulative velocity increase in the southeast varies spatially and is between 0 and 20 m/s This is a large (10%) timelapse signal, attributed to only a 12-m thick layer over an 84-day time span Near-surface time-lapse estimation The quality factor Q is a dimensionless quantity used to describe absorption (or attenuation) of seismic energy by the Earth Q is inversely proportional to the energy loss per wave cycle, Q = 2 E represents the energy of the wave and E the energy loss per wave cycle Within the seismic bandwidth, we assume Q is approximately invariant with frequency the so-called constant Q-model and can be estimated from amplitude spectra (Kjartansson, 1979) The Q-values reported in this study are computed using the log-spectral ratio method by fitting a line through the ratio of the log-amplitude spectra of a wave, measured at two points along its propagation path The slope m of this line is inversely proportional to Q so that The distance traveled is given by r while represents the propagation velocity Going back to Figure 6, one would ideally compute the slope m of the log spectral ratio between the upgoing S-wave energy at A and the downgoing energy at D These locations represent the same point on the wavefront, but at different times No receiver is located at A, but the distance between A and D is small enough ( 14 m) so that the upgoing amplitude at D can be used to substitute for that at A Taking 44 The Leading Edge January 2013 Figure 6 Geometric relationship between the rays of the upgoing and downgoing arrivals that is recorded by a single receiver Plane-wave propagation is assumed spectral ratios at the same receiver location produces values that are not biased by coupling variations between receivers The downside is that there will be some negligible spatial smearing While individual log-spectral ratio curves are noisy, shot averages show curves with a linear trend up to 45 Hz (Figure 8) Moreover, these average trend lines appear to intercept the origin, which indicates that the near-surface reflection coefficient is close to 1 and spreading effects within the topsoil are negligible Therefore, and to minimize the impact of noise on estimated values, it was decided to use an L1-norm straight line fit through the origin to estimate spectral slopes m These slopes relate to a total travel distance

5 Figure 8 Log spectral-ratio curves These were computed by taking the natural logarithm of the ratio between the upgoing and downgoing amplitude spectra (gray) Black lines represent shot-averaged spectralratio curves The dashed line represents the L1-norm gradient fitted between 5 and 45 Hz and intersecting the origin and, like, we observe clear spatial and temporal trends in The northwest-corner values are typically smaller than those in the SE corner ( = 10 versus = 15) While - values in the NE corner remain more or less constant with time; those in the southeast corner show a gradual increase (Figure 9c) The typical absolute increase is 15 In relative terms this is once again a dramatic time-lapse signal (10%) Figure 7 Near-surface S-wave velocity maps for the survey area Squares indicate source location and circles receiver locations The red line marks the border between Muskeg on the northwest and Parkland to the southeast (a) map for day 1 (b) map for day 8 (c) Difference When we combine this with the equation for and our definition of, we obtain: t mcos 2 Individual estimates are noisy For any given receiver, all 11 estimates (one for each shot) are median-filtered, after which a spatial mean filter with a seven-station window was applied to each receiver line The resulting near-surface maps for the first and last day of acquisition are shown in Figure 9a and Figure 9b A contour map of the difference is shown in Figure 9c Most values range between 8 and 22 (4) Discussion: Uncertainty We now consider the effects of errors in t and in ray angle on Considering the low values at this site, it is fair to question whether our t estimates could be biased by Q dispersion Lower is expected to produce more wavelet dispersion and thus larger t, resulting in lower Such a correlation between is observed at this site and, while some Q bias on the t estimates cannot be excluded, we don t expect it to be a major issue This is mainly because the crosscorrelations used to estimate t are dominated by a 25-Hz peak frequency and the range of values is probably too restricted to cause a significant bias at this frequency The total range of is 8 22 but most values lie between 10 and 17 We believe that errors associated with t estimates are small and mostly random Errors in estimates of the ray angle could also cause a systematic bias in both and The effect is similar to breaking the straight-ray assumption in the near surface An error of ±5 in will cause an approximate 5% shift to the estimated velocities and an approximate 10% shift to values Curved rays extend the traveled distance and will increase Even if a bias in were present in our results, the spatial and temporal variations are robust features Discussion: Results The maps of (Figures 7 and 9) contain a red line January 2013 The Leading Edge 45

6 soils) are more difficult to explain Inspection of the upgoing and downgoing first-arrival S-waves in Figure 5c and Figure 5d shows that the upgoing wavefield is more or less invariant with time while temporal changes occur in the downgoing wavefield This confirms that changes in are confined to the top 12 m of soil while no changes in occur between 12 m and the refracting layer which is probably 100 m deep Given the strong correlation between, it is fair to assume that changes in are also exclusively confined to the upper 12 m We speculate that these temporal changes are caused by a gradual drying of the soil Weather data from the vicinity of the survey area (Canadian NCDIA) shows a cumulative precipitation of 80 mm during acquisition with low precipitation for the last month Soil rigidity is expected to increase as water-saturated mud dries up This may well lead to the observed gradual increases in both Figure 9 Near-surface S-wave Q maps for the survey area Squares indicate source location and circles receiver locations The red line marks the border between Muskeg to the northwest and Parkland soil to the southeast (a) map for day 1 (b) map for day 84 (c) Difference that traces through the survey area This line marks the approximate boundary between Muskeg soils in the northwest and Parkland soils to the south The boundary was interpreted using a LIDAR satellite image and verified by the field operator (M Kiehn, personal communication) Muskeg soils are similar to bogs and consist of unconsolidated, water-saturated soils, rich in plant material Parkland soils are drier It is no surprise that Muskeg is not favorable to S-wave propagation and characterized by generally lower The temporal variations in to the southeast (Parkland Discussion: Where did my S-wave go? Low-velocity, low-q near-surface layers are detrimental to seismic bandwidth and signal-to-noise ratio Low Q will cause velocity dispersion and absorption Low velocities result in shorter wavelengths which increase scattering effects and generate large statics Low velocities also reinforce the impact of low Q values Indeed, as velocity decreases, the wavelength becomes shorter and more wave cycles are needed to propagate through a low-q zone To illustrate this we take a simple example based on the values reported here Given a 20-Hz signal that propagates through a near-surface layer of 10 m with V P = 2000 m/s and = 200 m/s, the P- wave will require 01 wave cycles while the S-wave requires a full wave cycle In other words, the S-wave attenuation is 10 times more than the P-wave for the same frequency Taking our earlier definition of Q = 2 and assuming Q = 10, we can compute that the S-wave will lose 62% of its energy to attenuation while the P-wave only 62% Dispersion effects also become important at these low Q-values For the same example and computed according to Aki and Richards (2002), velocity dispersion between a 5-Hz and 40-Hz signal will be 66% (eg, 200 m/s versus 213 m/s) If just 10 m of rock can have such a detrimental impact on S-wave propagation, it should be no surprise that in such situations S-wave (including converted-wave) images will be poor Conclusions We present a method to separate the upgoing and downgoing S-wavefields from raw vertical and radial components recorded by buried receivers The separated wavefields are then used to estimate in the upper 12 m of the weathering layer at the Peace River heavy oil field in Alberta, Canada, for 84 subsequent days S-wave velocity maps show a strong spatial correlation between lower (190 m/s) in areas with Muskeg soils (northwest corner) and higher (210m/s) in areas with drier Parkland soil types (southeast) Maps of near-surface correlate well with, both spatially and with time Typical values are between 8 and 22 They are slightly higher in the southeast corner than in the northwest ( = 15 versus = 10) Low near-surface 46 The Leading Edge January 2013

7 S-wave velocities combined with high rates of absorption (low ) are often quoted as the main reason for poor imaging results from S-waves and converted (PS) waves Especially for PS data, this is compounded by a typical and total lack of reliable near-surface S-wave velocity information This work confirms that in our survey area a low near-surface and are indeed the two important factors affecting S-wave quality Using buried 3C detectors, 12 m or less, and compensating for Q-dispersion may well prove to be the most effective tools for improving S-wave images in the area The near-surface time-lapse variations in velocity (and probably also ) are confined to upper 12 m of soil These changes are most likely linked to a gradual drying out of the soil As such we can expect smaller, though proportionally equivalent P-wave time-lapse changes This has serious ramifications with respect to time-lapse studies for reservoir monitoring on land The near-surface time-lapse variations reported here are significant and could easily interfere with or overprint time-lapse signals from a deeper reservoir Permanently buried sources and multicomponent (3C and 4C) receiver arrays, in combination with adequate deghosting strategies, hold the promise to provide high-quality timelapse land data free from a near-surface bias References Aki, K and P G Richards, 2002, Quantitative seismology, 2nd Edition: University Science Books Bale, R A and R R Stewart, 2002, The impact of attenuation on the resolution of multicomponent seismic data: 72nd Annual International Meeting, SEG, Expanded Abstracts, , dxdoiorg/101190/ Canadian National Climate Data and Information Archive (NC- DIA): Forgues, E, E Schisselé, and J Cotton, 2011, Simultaneous active and passive seismic monitoring of steam-assisted heavy oil production: The Leading Edge, 30, no 11, , org/101190/ Gaiser, J, 1999, Applications for vector coordinate systems of 3D converted-wave data: The Leading Edge, 18, no 11, , Kjartansson, E, 1979, Constant Q-wave propagation and attenuation: Journal of Geophysical Research, 84, B9, , dxdoiorg/101029/jb084ib09p04737 Pujol, J and R B Herrmann, 1990, A student s guide to point sources in homogeneous media: Seismological Research Letters, 61, Acknowledgments: The author thanks Shell and CGGVeritas for their kind permission to publish this work, and Sergio Grion and Sam Gray for their help with reviewing Comments on the original manuscript from Jim Gaiser and Rishi Bansal were much appreciated Many thanks to Richard Bale and the Onshore Reservoir Monitoring team of CGGVeritas, as well as Michael Kiehn and Peter Wills from Shell, for feedback on the subject Corresponding author: kristofdemeersman@cggveritascom 2013 Pacific South Honorary Lecturer Aeromagnetics A Driver for Discovery & Development of Earth Resources Dave Isles, Consultant, Perth, Australia Aeromagnetic surveys are very commonly under interpreted The potential value, captured during acquisition, is all too often unrealised at the interpretation and action stages of a project This presentation illustrates the fundamentals of robust aeromagnetic interpretation using telling case studies DATE LOCATION SECTION 7 Feb Tokyo, Japan SEG Japan 12 Apr Canberra, Australia ASEG, ACT Branch 16 Apr Adelaide, Australia ASEG, SA Branch 17 Apr Melbourne, Australia ASEG, VIC Branch 15 May Bandung, Indonesia Univ of Padjadjaran SEG/AAPG Student Ch 17 May Makassar, Indonesia Hasanuddin Univ SEG Student Chapter DATE LOCATION SECTION 20 May Semarang, Indonesia Diponegoro Univ SEG Student Chapter 22 May Malang, Indonesia Brawijaya University 25 May Yogyakarta, Indonesia Gadjah Mada Univ SEG Student Chapter 11 Jun North Ryde, Australia ASEG, NSW Branch 13 Jun Brisbane, Australia ASEG, QLD Branch For more information or to view previous HL presentations, visit: wwwsegorg/hl January 2013 The Leading Edge 47