Virtual Source: new method for imaging and 4D below complex overburden Andrey Bakulin* Rodney Calvert Shell International E & P Houston I will present you fundamentally new VSP imaging method base on time reversal that have been designed for two main purposes: Image below very complex near-surface Perform sensitive monitoring in cases of non-repeatable acquisition and near-surface changes The fundamental advantage of Virtual Source over conventional VSP is to be able to achieve that without any knowledge of overburden velocities or near-surface changes. Presented at SEG 2004, Denver

Virtual Source Sk Complex near surface Sk Virtual source (at R) R Surface array of sources that simulates virtual source Simpler “middle” overburden Virtual source (at R) R Complex near surface Target Sk Well Sk Skb Db This is a cartoon showing you the typical problem we try to address. We have really complex near surface or overburden (basalt layer, salt or just nasty near-surface layer). Then our target is at the bottom while in between we have simpler middle overburden. Then geophones should be placed in a well below the most complex near-surface part. This for example could be horizontal or slanted well of producer/injector or dedicated smart sidetrack. As soon as we have done that we are able to directly measure the Green’s function or propagation effects caused by near-surface. Using time reversal technique we can re-create new completely downhole dataset with Virtual Source placed at each geophone location. Since we use time reversal – we avoid building any velocity model whatsoever above the receivers where we have most complex near-surface part. In simple words, the bottom line is to get the red traces from blue traces. Once we obtain new virtual dataset with both downhole sources and receivers – we can perform conventional imaging of the lower part of the subsurface. To achieve that we only need bottom portion of the velocity model below the receivers which is usually simple to obtain. We claim this fundamental new tool to be able to image below extremely complex (realistic) overburden, the worse the better. This is the only source where we actually know the waveform, can control it to be perfectly minimum phase, has a downward radiation pattern, is constant for 4D even if the near surface changes, is constant for 4D even if the shooting geometry changes slightly, all without knowing the overburden model. If that sounds too god to be true - remember the penalty is to place geophones in the subsurface. It has been designed to address the cases where overburden complexity prevents us from imaging the deeper subsurface. It is not because the energy does not get through or attenuates, it because it does scatter in complex fashion and we are unable to construct the velocity model to image this data.

Virtual Source by time reversal Receivers Source There is no magic involved. Everything is based on sound principle of time reversal which is tuned for the problem at hand. Time reversal says that if we capture the wavefiled from a source at a surface, time reverse it, turn receivers into sources and send the time-reversed wavefield back – then we will observe the reverse propagation with wavefronts collapsing towards original source. All energy collapses back into the original source and because it is not taken out of the system then it radiates out again as if we were having Virtual Source at this location. Remember that time reversal holds for elastic waves in arbitrary heterogeneous media with any anisotropy. Receiver Sources

Virtual Source as time reversal on a computer reciprocity o t This is how these principles are applied in real life. linearity time reversal sum for a Virtual Source

VS data To make sure we understand how we generate offset VS data this is what we actually do when we want to generate trace for VS at Ralpha and receiver at Rbeta. We take the trace or portion of the trace recorded at Ralpha from source Sk. We time reverse it and convolve with the trace from source Sk to receiver Rbeta. This is nothing else but an approximation to a single-channel time-reversal. Out of the whole wavefield recorded at the surface we only back propagated the portion from one location only. Then we repeat this process for each of the surface location thus backpropagating the entire wavefield recorded at the surface. This is an approximation because in a strict time-reversal we should be convolving with the impulse response where influence of original source signature has been removed, whereas we convolve with the trace itself. This approximation works perfectly well but could be improved.

Synthetic model with horrible overburden (full elastic finite-difference modeling) m/s (m) Reservoir Well with receivers Sources Let me start with demonstrating you feasibility of VS approach on synthetic case study and then we will move on to real data example. We walk through all the steps starting from VS generation, then imaging obtained VS data, estimating 4D response with VS data and determining repeatability of VS data. At each step we will compare that with results coming from surface data. Displayed here is the P-wave velocity model that resembles subsurface of Peace River field in Canada. You can see that bottom part is really simple. It consists of few layered formations with reservoir at about 600 m depth. The main challenge comes from complex near-surface part. The surface itself is represented by challenging Muskeg swamp environment. In addition upper 200m consists of very heterogeneous glacial fluvial deposits originating due to glacial channels. We created this horrible synthetic model as worst case scenario to test the method. Velocity has huge variation from 1000 m to 2400 m/s. We have shot line at the surface and horizontal well with receivers. *generated synthetic dataset for 80 receivers in horizontal well sitting below this horrible zone (depth of about 400 m) and for shot line at the surface.

Receiver gather Here you can see representative receiver gather from this dataset. Since it was generated by finite-difference modeling code, it does contain all multiples, conversions, diffractions, everything – it is full wavefield. You may clearly see that you will not get away by just simple static corrections – little blow-up at the bottom shows you time window around first arrivals after hyperbolic kinematic corrections. Everybody I approached declined to recover any velocity from the data itself or image it anyhow. Let me know if any of you wish to try and I get you the data.

Black – virtual source gather, red – real downhole gather VS Black – virtual source gather, red – real downhole gather On the top of the slide you can see example of resulting virtual receiver gather with VS re-created in the middle geophone. You can now start to see some coherent hyperbolic events with this one being first interface below VS and somewhere here top reservoir and this strong event is bottom reservoir. To answer the question how well we have done, I plotted at the bottom comparison between VS data in black and newly computed data with real physical sources placed downhole in red. And you can see that we have done reasonably well.

Black – virtual receiver gather, red – real downhole gather VS Black – virtual receiver gather, red – real downhole gather VS  Looking at the VS gathers at few other locations along the line we can also see that we have done reasonably well reconstructing all possible offsets.

PSDM comparisons VS data: Surface data: migrated with exact velocity model of the overburden First interface (505m) How do we compare all that with the original VSP or surface-to-downhole data? Well, let us assume that we have built velocity model of the overburden and by virtue of magic it happens exactly to coincide with the true model.   Then we can indeed obtain really good depth image shown on the right. We have correctly imaged the plane interfaces including reservoir bottom and we the level of noise is reasonable. On the left you see the depth image of VS data. You can see that VS image is at least no worse than the one on the right. I can claim that certain things are slightly better on VS image but this is not the point. The point is that VS image was obtained without any knowledge of the near-surface velocity model while the one on the right is obtained with perfectly correct velocity model. To get VS image we only needed lower 1D portion of velocity model which is easily obtainable. This is the end of the imaging part with simple conclusion that if you can do it from the surface – you are in good shape. If you can’t build perfect velocity models for your overburden – then you might use VS approach to overcome that. Bottom reservoir (590m)

Downgoing waves Pdn Pdh Pdp Pds PSuw Profile Surficial stuff Harmon Fahler Pdn Pdh Pdp Pds PSuw Profile

PSDM Real downhole data VS data Spurious reflections from above Bottom reservoir (590m)

Radiation pattern of Virtual Source Plane array of sources above simulates VS with radiation pattern mainly radiating along downward hemisphere To excite the other half additional array is needed below VS Source array No energy upward VS

PERFECTLY REPEATING GEOMETRY Surface data: monitor/baseline (top) and difference (bottom) You may always say well - whatever horrible overburden is in 4D we can always subtract the data and get the difference thus canceling out effects of horrible (but constant) overburden. And you will absolutely be right: if we perfectly repeat geometry and if the overburden does not change then difference starts exactly at the top reservoir and the only problem is to image and calibrate that difference. Same scaling

Comparing baseline shifted by 5m B1: 160 surface shots dx=10m Z=15m B2: 160 surface shots dx=10m (shifted by 5m) In reality we never perfectly repeat our acquisition and even if we will – the near-surface always changes. So let us analyze what will happen for an imperfect acquisition. As we are bounded by 2D geometry I consider simplest case when we shot two lines over the same baseline earth that are simply shifted by 5m. Z=430m Target Z~600m

After “4D static” (time shifts introduced to match 1 and 2)  And you may see that difference gets smaller, but does not go away and it is again everywhere. Same scaling

VS gathers from dataset 1 and 2, and difference (bottom) Now let us repeat this exercise for VS data. Again we use two different shot lines shifted by 5m to generate two VS datasets. And here is the example of VS shot gather. You may see that we have much smaller difference everywhere and no static correction is necessary. Same scaling

Peace River 4D VSP Baseline - September 2002 (before steam injection) Monitor – December 2002 (after steam injection) Let us look at the real data example. 50 geophones were cemented in place and 4D VSP has been recorded at Peace River field with the objective of monitoring steam injection in the target below. Blue line shows locations of surface 100 shots, while green line is slanted observation well. Component used for initial study, along-the-well (450)

Raw data on the left compared with the synthetic on the right show that near-surface is much more complex than 1D model derived from seismic to represent it. Also notice that strongest event is the direct shear arrival converted at the surface.

Despite cementing geophones and dropping charges into the same hole – we are not able to achieve good repeatability because near-surface itself does change very dramatically. You can see that first P-arrivals travel on average 2 ms longer in December than in September. The shear wave travels 20 ms longer. The RMS level or energy of traces from the same shot may differ from 25 to 250 %. Therefore straight differencing of even time-shifted traces is likely to show us near-surface changes rather than reservoir anomalies.

Nevertheless straight generation of VS data does reveal even of interest. Shown is common VS gather with VS in the middle and receivers above and below. This is direct downgoing wave, while these two are reflections from the top and bottom reservoir. Comparing with the synthetic data –we find a good agreement.

Virtual Source Top base Bottom Top monitor Bottom

4D surface seismic Top base Bottom Top monitor Bottom

Is the fault a barrier?

Track changes in small compartments 1b 1a 2b 2a 3b 4b 4a 3a 1b 1a 2b 2a 3b 4b 4a 3a

Improved 4D in changing conditions Highly repeating overburden detail despite 4D statics (surface to well) despite imprints of different near-surface conditions despite difference in shot wavelets and frequency The reservoir shows considerable changes – which are real

Virtual Source imaging New method for imaging/monitoring below very complex near surface using downhole geophones Does not require velocity model between surface and geophones in a well Increased overburden complexity may indeed enhance the VS approach VS has advantageous downward radiation pattern Substantially higher frequency images compared to surface seismic – no mis-stacking We have demonstrated you new VS method that fills the gap between existing techniques and can confidently image and monitor below very complex near surface using array of downhole geophones. We have shown that on realistic synthetic and on field data. Since we directly measure the transmission response method does not require velocity between surface and well geophones and thus can be applicable for arbitrary complex near surface. It automatically handles 4D static and in large corrects for changes in the near surface which is what we need for monitoring And finally it may completely relax the requirement to repeat surface shot positions exactly that turn out to be very difficult in practice.

Virtual Source monitoring Automatically takes care of 4D static and other changes in the near surface May relax requirements to repeat surface shots positions exactly Much more repeatable data compared to surface seismic or conventional VSP We have demonstrated you new VS method that fills the gap between existing techniques and can confidently image and monitor below very complex near surface using array of downhole geophones. We have shown that on realistic synthetic and on field data. Since we directly measure the transmission response method does not require velocity between surface and well geophones and thus can be applicable for arbitrary complex near surface. It automatically handles 4D static and in large corrects for changes in the near surface which is what we need for monitoring And finally it may completely relax the requirement to repeat surface shot positions exactly that turn out to be very difficult in practice.

Acknowledgements Peter McGillivray and Shell Canada Charles Jones for help and support Paul Milcik for teaching us how to use various imaging tools Richard Cook, Barabara Yantis, Boudewijn Salomons and Uwe Kaestner for invaluable insights into seismic migration Finally I wish to thank all the people on these slide who assisted us throughout this study and made it possible.