Passive Seismic Monitoring of Carbon Dioxide Storage at Weyburn

Transcription

1 Passive Seismic Monitoring of Carbon Dioxide Storage at Weyburn James P. Verdon 1, Don J. White 2, J-Michael Kendall 1, Doug Angus 3, Quentin Fisher 3, Ted Urbancic 4 To be submitted to The Leading Edge 1 Dept. of Earth Sciences, Bristol University, Bristol, UK 2 Geological Survey of Canada, Ottawa, Canada 3 School of Earth and Environment, University of Leeds, Leeds, UK 4 Engineering Seismology Group Canada, Kingston, Canada Carbon capture and storage (CCS) is currently one of several candidate technologies for reducing the emission of industrial CO 2 to the atmosphere. As plans for large-scale geological storage of CO 2 are being considered, it is clear that monitoring programs will be required to demonstrate security of the CO 2 within the storage complex. Numerous geophysical monitoring techniques are currently being tested for this purpose including controlled-source time-lapse reflection seismology, satellite synthetic aperture radar interferometry, electromagnetic sounding, gravity, and others. Passive seismic monitoring is an additional technique under consideration that complements these other techniques, and has potential as a cost effective method of demonstrating storage security. This is particularly true over longer periods of time, as passive seismic arrays cost relatively little to maintain. Of the large-scale pilot CCS projects currently operational, thus far only the IEA GHG Weyburn-Midale CO 2 Monitoring and Storage Project has included passive seismic monitoring. Here we present the results from 5 years of passive seismic monitoring at Weyburn, and discuss the lessons learnt that can be applied when deploying passive seismics to monitor future CCS operations. Passive Seismic Monitoring Activities such as production of hydrocarbons or injection of CO 2 will alter the pore pressure, and therefore the effective stresses, both inside and around a reservoir. This can lead to the reactivation of pre-existing faults and fractures, or even the formation of new fault/fracture networks. Fracture formation and fault movement will emit seismic energy, which can be recorded on geophones installed in boreholes near the reservoir. Various methods exist to locate event hypocentres based on the energy recorded at the geophones. Many of the techniques used in passive seismics have their basis in global seismological research. Accurate location of events can identify active fault planes (e.g., De Meersman et al. 2009), and identification of focal mechanisms can reveal the style and orientation of deformation (e.g., Rutledge et al. 2004). Furthermore, as the seismic energy recorded has usually travelled exclusively through rock in or near the reservoir, wave propagation effects such as S-wave splitting can provide direct information about features in the reservoir such as the presence of aligned fractures and reservoir quality (e.g., Verdon et al. 2009). Passive seismic monitoring provides a different kind of information to controlled source techniques.

2 Recording is continuous, and information can be analysed in near real time. Yet, the technique can only image areas between where microseismic events are occurring and receivers are located. Furthermore, since the locations of microseismicity vary spatially, use of the temporal variations of microseismicity to monitor fluid related velocities changes is challenging. Whilst 4-D seismics is sensitive to changes in fluid saturation and stresses, passive seismic monitoring is an excellent technique for identifying geomechanical deformation induced by injection. The detection of active faults and fractures or new fractures generated within the caprock of a reservoir is important for CCS because these may provide pathways for CO 2 leakage, especially if they propagate far into the overburden. Along with fluid migration up well bores, leakage along faults and fractures represents one of biggest risks to secure storage. Furthermore, once geophones have been installed, the costs of maintenance and data processing are small in comparison with controlled-source seismic techniques. This is an important consideration for CCS where a site may need to be monitored long after injection has ceased and the field shut in. Overview of the IEA-GHG Weyburn-Midale Project The Weyburn-Midale field is located in the Williston Basin of southern Saskatchewan, Canada (Fig. 1). The reservoir is located in the Carboniferous Midale beds at depths of ~1430m. The reservoir consists of fractured lower carbonate and upper dolomite layers, overlain by an evaporite caprock. An important secondary seal is provided by the Lower Watrous member, which constitutes a thick layer of shale-rich Mesozoic sediments that lie just above the reservoir. Vertical stress at the reservoir depth is ~34 MPa and the minimum horizontal stress is estimated to be MPa NE-SW in this region (McLellan et al. 1992). The injection pressures were approximately 22 MPa. The field has been in production since 1954, where initial waterflooding commenced in the 1960s and horizontal infill wells were drilled in the 1990s. CO 2 injection was initiated in 2000 to enhance oil recovery, resulting in production levels seen in the mid-1970s. However, a research component was also included to test and develop techniques for monitoring large volumes of CO 2 in the subsurface. Current total injection rates are over 3 million tonnes of CO 2 /year in the Weyburn-Midale field the equivalent to CO 2 emissions from ~ cars. Injection rates for individual wells range from 50 to 500 tonnes per day. Controlled-source 4-D seismic monitoring has been largely successful in imaging the plumes of CO 2 migrating away from the injection wells (White 2009). Passive monitoring at Weyburn has focused on a single pattern within the field. In 2003, a passive recording array consisting of 8 triaxial 20-Hz geophones was cemented in an inactive vertical well that was located within 50 m of a planned new vertical CO 2 injection well (121/06-08). Geophones within the array were spaced at intervals of 25 m between 1181 m and 1356 m depth lying roughly 200 m above the top of the reservoir. Three surface calibration shots were used to determine the orientation of the horizontal sensors and to check the system. The system was operated in triggered mode using a trigger window length of 200 msec and requiring processed signal levels exceeding threshold on 5 of the 8 geophones for event triggering and data storage to be initiated. Passive monitoring commenced in August 2003, prior to the onset of water injection in December, 2004, and has been in semi-continuous operation to present with the exception of an

3 11-month down-time from December 2004 to October 2005 (Fig. 2). During this period approximately 100 locatable microseismic events have been recorded, documenting a low rate of low intensity (moment magnitudes of -3 to -1) microseismic activity (cf. Vacuum Field; Aneth Oil Field, Utah). The majority of the recorded microseisms are characterized by low frequency content (close to the resonant frequency of the geophones) and a dominant wavelength ranging between 165 m and 275 m (assuming P-wave velocities between 3300 m/s and 5500 m/s). Events have been located up to ~500 m from the geophones. Water injection in the nearby 121/06-08 vertical well began on December, 15, 2003 with a switch-over to CO 2 on January 22, Injection has continued under a WAG (water-alternating-gas) process. The injection history of this well is shown in Fig. 3. Event Locations Microseismic event hypocenters were determined by matching the observed P- and S-wave arrival times by ray tracing, and determining the propagation azimuth by hodogram analysis across the levels of the array. A 1-dimensional velocity model was adopted for the purposes of ray tracing based on a dipole sonic log from well Sensitivity of event locations to the velocity model was examined by varying velocities by +/- 250 m/s, with resultant location changes of 75 m north-south, 20 m east-west and 70 m vertically. In addition, errors in location are as high as several hundred metres in cases where the P- and S-phase onsets are emergent or poorly defined. Magnitudes were estimated using an automatic time-domain calculation (Urbancic et al. 1996). The complete set of microseismic events located from August, 2003 to June, 2009 is shown in Fig. 4, whereas a subset are shown in Fig. 5 with temporal clusters of events grouped by color. The seismicity can be correlated with injection/production events occurring in this part of the field. Pre-injection: Approximately 30 locatable events were recorded during the 4 month background period prior to the start of injection. Of these, the majority are related to well completion activities and perforation shots in the vertical injection well, which form a tight cluster of yellow dots centred on the injector in Fig. 5. The remaining events (yellow dots) form a diffuse distribution about the nearby horizontal production well (192/09-06). Similar events were recorded at various times during the overall monitoring period including a cluster of 15 events occurring on March 18-19, 2004 during well shut-in. They are characterized by good signal-to-noise levels (Fig. 6c) and relatively high frequencies (up to 150 Hz). The timing of these events correlates directly with periods when the production wells are shut-in, and thus are likely associated with local pressure recovery during shut-in that leads to shear failure. Start-up of injection: Water injection in well 121/06-08 began on December 15, The resultant increase in background noise levels caused the microseismic system to trigger continuously eventually leading to shut

4 down until January 12, Unfortunately, no useful data were acquired during this period. Water injection stopped in well 121/06-08 on January 22, at 8 am and CO 2 injection started the same day at 11 am. On January 21, 13h29, and January 22, 9h02, 13 events (purple dots in Fig. 5) were recorded, before the start of CO 2 injection. They form a spatial cluster that extends up to 300 m east of the injector toward the production well 191/11-08, which experienced a 40-fold increase in (CO 2 ) gas production within several days. In contrast to the pre-injection events near production well 191/09-08, these events are characterized by relatively low peak frequencies (20 to 30 Hz) and short separation between P- and S-waves making hypocentral location difficult and leading to large location uncertainties. The relatively low-frequencies of these events (see Fig. 6c) suggest that the inducing mechanism is gas or fluid movement. However, in that the events all occur prior to the onset of CO 2 injection, they clearly are associated with the movement of fluids other than CO 2. Another four events with waveforms similar to those of January 21 and 22 were recorded on January, 31 and February, 4 and 9, Period of high injection rate: The rate of CO 2 injection in 121/06-08 was increased by almost a factor of 2 (from approximately 70 Mscm to greater than 150 Mscm) for a period of 8 weeks from early May to July, 2004 (see Fig. 7). The seismic array was only operational during the latter stages of this period. Recorded microseismicity continued past the increased injection period for almost 4 weeks, over which a total of 34 locatable events were defined. The associated cluster of events (orange dots, Fig. 5) continues the trend, further northward that was initially defined by the injection-related events of Jan , 2004 (purple dots, Fig. 5). Furthermore, this overall spatial cluster correlates well with a negative lobe in the timelapse seismic amplitude difference map for this region. The last significant group of microseismic events occurred during Jan , 2006 when a total of 20 events occurring over a period of 4 hours were detected and located (light blue dots in Fig. 5). These events have characteristically low frequencies similar to the events recorded near the start of CO 2 injection. Since January of 2006, there have been fewer than 10 locatable microseisms. This paucity in microseismic activity may be real or could potentially be due to a reduced sensitivity of the recording array or increased noise levels over time. S-wave splitting The seismic energy recorded on the geophones has travelled only through rocks in and around the reservoir. As a result any wave propagation effects, such as attenuation or S-wave splitting, can be used to make inferences about the properties of these rocks without first having to disentangle the effects of propagation though the overburden. S-wave splitting is particularly useful, as it allows the direct measurement of any anisotropy present, which in reservoir settings may indicate the presence of sedimentary layering, aligned fractures or grain boundaries. Forward modelling using rock physics theory can be used to find the combinations of fracture geometries and sedimentary VTI fabrics that best fit the observed splitting measurements (see Verdon et al., 2009).

5 When a S-wave travels through an anisotropic region it is split into two orthogonally polarised S-waves, one of which travels faster than the other. We measure the difference in arrival times between the fast and slow phases, and the polarisation of the faster phase. By observing the time-lags and fast polarisations over multiple ray-paths it is possible to characterise the anisotropy of the reservoir. The passive seismic data from Weyburn were analysed for S-wave splitting using a semi-automated approach. This technique performs a grid search over time-lags and fast polarisations to find the combination of parameters that, when corrected for, do the best job of removing the effects of splitting. The effects of splitting are removed when the S-wave particle motion has been linearised the time-lag and fast polarisation that best linearises the particle motion are selected as the splitting parameters. The splitting produced by aligned fractures set in non-isotropic rocks is nonlinear and often not intuitive. ITo fully understand splitting observations it is necessary to construct forward models using rock physics theory. We use an approach that inverts for rock physics parameters based on minimising misfit between observed and modelled splitting observations. The free parameters that we use in our inversions are the Thomsen parameters γ and δ (giving the strength of a VTI sedimentary fabric), and the strike and density of a set of vertical, aligned fractures, α and ξ respectively. We plot the RMS misfit between model and observation as a function of these parameters. As we are here principally interested in identification of fractures, we only show the RMS misfit as a function of the fracture strike and density, at the best fit VTI strength. The inversion finds that the VTI strength is very low, which is to be expected for the intrinsic isotropic carbonate rocks that make up the Weyburn reservoir. The results for splitting during the first phase of passive seismic monitoring are shown in Fig. 9. The inversion identifies a NW-striking HTI fabric. This could represent either an open fracture set with this orientation or the maximum horizontal stress orientation (or, most likely, a combination of both). The NW strike matches a fracture set identified from core and image analysis (Wilson et al. 2004). However, the dominant fracture set is not the NW striking fractures but rather a NE striking fracture set. An important question to ask is why the splitting has imaged this secondary fracture set, and not the principal set? To answer this question we construct a simple geomechanical model to simulate the evolution of principal stresses during injection. Seismic observations and geomechanical modelling Passive seismic activity represents an observable manifestation of geomechanical deformation in and around the reservoir. Therefore, passive seismic observations can be combined with geomechanical models to further enhance the understanding of the subsurface. In this section we demonstrate how even simple geomechanical models can enhance seismic observations. We generated a simplified coupled stress-fluid flow model, representative model consisting of a flat,

6 rectangular reservoir with much larger breadth than thickness, set in a homogenous overburden. The simulations were performed using a code that explicitly couples the TEMPEST production simulation model (Roxar Ltd.) for the flow calculations and ELFEN finite element model (Rockfield Ltd.) for the geomechanical simulations. An MPI interface controls the transfer of fluid pressure data from TEMPEST to ELFEN, pore volume multipliers data from ELFEN to TEMPEST, and also at which time-steps it is necessary to make this information exchange. Three horizontal wells were modelled, with two production wells either side of the injection well (though using symmetry arguments to reduce computational time, only a quarter of the reservoir was modelled). The injection rates were set such that there was an increase in pore pressure at the injector from 15 to 20 MPa, and a decrease at the producer from 15 to 11 MPa after 8 years approximating the pore pressure changes seen at Weyburn. The Young's modulus for the reservoir was set to 12 GPa and the Young s modulus for the overburden was set to 10 GPa in the immediate overburden and decreasing in stiffness towards the surface. This model can be considered a highly simplified representation of the Weyburn reservoir, matching the general geometry and properties, but missing much detail. Changes in the magnitudes and orientations of the principal stresses will alter seismic properties. In particular, seismic anisotropy can be highly stress-sensitive. Verdon et al. (2008) develop a model to compute the magnitude and orientation of shear wave splitting induced by stress changes. For vertically propagating shear-waves, the fast shear wave will align with the principal stress direction. In Fig. 10a we plot the SWS predicted by the geomechanical model described above. The model at the beginning of injection is isotropic. Around the wells, the maximum horizontal stress becomes aligned perpendicular to the horizontal wells. Hence, the maximum horizontal stress is oriented NW and the NE-oriented fracture set would be closed even though this fracture set has the highest fracture density. Since the raypaths of the microseismic events travel predominantly through this region, the NW-aligned fracture set would have the dominant influence on the measured SWS resulting in the observed NW-striking HTI fabric. We can also consider the likelihood of failure by considering where in the model the effective stresses approached those required for failure. In Fig. 10b-e we plot the Mohr circle evolution from the initial state to the end of the injection period. The likelihood of failure (and therefore microseismicity) will be increased if the Mohr circle either translates to lower normal stresses, whilst maintaining its size (inducing shear failure), or if normal stresses increase such that pore collapse occurs. We note from Fig. 10 that above the production well the Mohr circle translates to lower normal stresses with no decrease in deviatoric stress, increasing the likelihood of shear failure, whilst around the production well the Mohr circle translates to significantly higher normal stresses, increasing the likelihood of pore collapse. In contrast, around the injection well there is a decrease in deviatoric stress, whilst above the injection well there is a smaller increase in normal stress. This suggests that in this scenario, areas around and above the production wells will be placed at much greater risk of failure than around the injector. Comparing this modelling prediction with event locations at Weyburn (Figures 4 & 5), we note that the majority of events are in fact located close to the production wells this is particularly evident in Fig. 5.

7 We must emphasise here that the geomechanical model discussed is simple in nature, and misses much detail about the reservoir in question. Nevertheless, the predictions it makes provide a good first-order match with both SWS observations and areas where the majority of events are located. This provides a useful demonstration of both how observations can be used to ground-truth geomechanical models, and how geomechanical models can provide added insight into microseismic interpretation. This link between geomechanical modelling and observation will be of great importance for CCS, not just for comparison with microseismicity, but with surface uplift and 4-D seismic as well. Discussion The temporal clustering of microseismic events is episodic, which raises the question of what causes these discrete episodes of localised deformation. Furthermore, if the low-frequency events are interpreted as fluid movement, why do we only see them occasionally if fluid movement is occurring continuously? A potential means of addressing the cause of the events is to study their focal mechanisms and moment tensors. Fluid movement would perhaps generate non-double-couple mechanisms. Lessons can be learned from studies of seismicity in other environments. For example, low-frequency tremor on volcanoes is usually associated with magma movement (e.g., Chouet 2003). Another benefit of focal mechanism analysis is that it provides an estimate of the orientation and magnitude of the stress tensor. This is important information for guiding injection strategies and groundtruthing geomechanical models. Unfortunately, such analysis cannot be done with a recording array in a single well. Previous studies have successfully calculated source mechanisms using longer arrays and two boreholes (e.g., Rutledge et al. Geophysics, 2003, for composite focal mechanisms). It is gratifying that the events observed until November 2004 show good correlation with the time-lapse seismic results. However, not all negative amplitude anomalies seen in the time-lapse data have associated events. This perhaps indicates that although microseismicity has been observed to track CO 2 movement, the spatial and temporal distribution of microseismicity appears not to be a consistent mapping tool. However, with our limited array and paucity of seismicity it is difficult to draw more firm conclusions about this. Depth uncertainty in the location of microseismic events is a significant issue, especially when trying to ascertain that CO 2 is secure in the reservoir. The depth resolution can be improved very easily. Past experience suggests that extended vertical arrays, or multi-well arrays, would constrain depths to within roughly 10m. Additionally, improved event location algorithms (e.g., multi-trace correlation methods - see de Meersman et al. Geophysics, 2009) would better constrain depths, despite being more time consuming procedures.

8 Another important point is whether or not microseismicity above the reservoir indicates top-seal failure and the migration of CO 2 into the overburden. Stress arching effects can lead to failure in the over- and sideburden, without any fluid leaving the reservoir. Also, fault movement does not always result in increased fault permeability (Fisher et al., 2003). To determine whether or not deformation results in increased fault permeability it is necessary to consider the rheology of the rock with respect to the stresses at the time of faulting. This underscores the importance of having a good understanding of the potential geomechanical behavior of the storage site. However, it is likely that fault reactivation and top-seal failure will be documented by a different spatial and temporal pattern in seismicity from those associated with stress arching effects. The magnitude versus offset plot (Fig. 8) indicates that surface arrays would have limited use for microseismic monitoring under conditions like Weyburn. The largest events recorded have a moment magnitude of less than 1.0, and most event are smaller than 2.0. Dense surface arrays would be required to detect such events, and their detectability would be strongly influenced by surface noise and nature of the event focal mechanisms (see Chambers et al., SEG, 2009). Furthermore, the data volumes associated with such large arrays would be considerably larger than those generated by a limited number of downhole monitoring arrays, which increases processing times and costs. A key questions is should CCS operations always/sometimes/never employ passive seismic monitoring, and how should this decision be made? Downhole monitoring is now a commonly used tool for monitoring the hydraulic stimulation of fractures. It presents a low-cost option for long term CCS monitoring. Ideally, such monitoring would record little induced seismicity. This would suggest that the CO 2 plume moves aseismically through the reservoir, inducing no-significant rock failure, as seems to be the case at Weyburn. An important first step in such a monitoring project would be establishing the pre-injection level of seismicity. It is conceivable that passive seismic monitoring of CCS would be of limited use in areas with high amounts of natural seismicity. As Weyburn is the only CCS project to employ such monitoring, it is difficult to form more definitive conclusions. Our work also suggests that another important pre-injection step is the development of a good geomechanical model of the reservoir. Forward modelling can be then used to predict the seismicity associated with various injection scenarios. Conclusions We have presented the results of 5 years of passive seismic monitoring at the IEA GHG Weyburn CCS/EOR project. At present this is the only large scale CCS project to have deployed passive seismic monitoring. We have found that microseimicity rates correlate with periods of elevated CO 2 injection rates, and also with changes in production activities in nearby wells. The distribution of event locations also appears to correlate with the regions of CO 2 saturation that have been identified using 4-D seismics. However, overall the rates of

9 seismicity are low. The low rates of microseismicity indicate that the reservoir is not undergoing significant geomechanical deformation, which is encouraging in regard to security of storage. We also demonstrate how shear wave splitting measured on microseismic events can be used to identify structures such as aligned fractures in the reservoir, and confirm the presence of one of the fracture sets identified in core samples. This is not the dominant fracture set. However, geomechanical modelling shows that the evolution of stress during injection is likely to preferentially open this set, making it dominate the splitting results. The geomechanical model can also be used to predict regions where microseismicity is likely to occur. These predictions also match with observed location clusters. At future CCS projects, the avoidance of geomechanical deformation is likely to be a key aim, and injection programs will be probably be tailored towards achieving this. The observations that can be made with passive seismic monitoring that there is a low rate of seismicity can be used to prove that geomechanical deformation is not occurring. When demonstrating security of storage it is equally as important to identify what is not happening (i.e., that there is no geomechanical activity), as it is to demonstrate what is happening. Passive seismic monitoring will provide a useful tool, and it is cost effective to run, requiring little in the way of maintenance and data processing. Suggested Reading IEA GHG Weyburn CO 2 monitoring and storage project summary report by Wilson et al. (2004) provides a summary of all monitoring activities at Weyburn. This book is available from the PTRC website ( Further information about geophysical monitoring at Weyburn can be found in Monitoring CO 2 storage during EOR at the Weyburn-Midale field by White (TLE, 2009) and Integrated Geophysical and Geochemical Monitoring Programs of the IEA GHG Weyburn- Midale CO 2 Monitoring and Storage Project by White and Johnson (GHGT9 Proceedings, 2008). Constraining fracture geometry from shear-wave splitting measurements of passive seismic data by Verdon et al (GJI, 2009) shows how shear wave splitting measured on passive seismic events can be used to constrain additional reservoir features. For details about the method used to model seismic properties induced by geomechanical deformation see Verdon et al (Geophysics, 2008) and Angus et al. (SEG, 2009), and for a more general discussion of these issues, the recent review by Herwanger and Horne (Geophysics, 2009) is recommended. Other references include: Urbancic et al. (1996) Automatic time-domain calculation of source parameters for the analysis of induced seismicity, Bulletin of the Seismological Society of America; McLellan et al. (1992) A multiple-zone acid treatment of a horizontal well, Midale Saskatchewan, Journal of Canadian Petroleum Technology; Chouet, B. (2003), Volcano seismology, Pure and Applied Geophysics; Rutledge, J. T., W. S. Phillips and M. J. Mayerhofer, Faulting induced by forced fluid injection and fluid flow forced by faulting: An interpretation of Hydraulic-fracture microseismicity, Carthage Cotton Valley field, Texas, Bull. Seism. Soc. Amer., 94, , 2004; de Meersman, K., J-M. Kendall and M. van der Baan, Relocation of the 1998 Valhall microseismicity: an integrated study of relocated sources, seismic multiplets and S-wave splitting, Geophysics, 2009; Chambers, K., O. Barkved and J-M. Kendall, Imaging induced seismicity with the LoFS permanent sensor surface array, Ann. Meeting of the SEG, Acknowledgements The authors would like to thank the PTRC and the Weyburn field operator, EnCana, for making the passive

10 seismic data available. James Verdon was funded by a UKERC Interdisciplinary Studentship. We are also grateful to the PTRC for funding. Shawn Maxwell and Marc Prince are acknowledged for their work on determination of microseismic hypocenters. We also thank the sponsors of the IPEGG consortium (BP, BG, StatoilHydro and ENI) and Rockfield Software (Martin Dutko) for support. Rockfield Software and Roxar Limited are thanked for providing copies of the ELFEN and TEMPEST software respectively. Finally, this work is part of the Bristol University Microseismicity Projects (BUMPS) and a contribution from the Bristol CO 2 Group (BCOG).

11 FIGURES (a) ) Figure 1. Location of the Weyburn field, set in the Williston Basin in Central Canada. all_located Microseismic Events Not Recording events time Figure 2. Histogram of located microseismic events from August 2003 to March 2009.

12 Figure 3. Monthly injection volumes for vertical injection well 121/06-08.

13 Figure 4. All located microseismic events (yellow dots) to June 2009 superposed on the 2004 time-lapse amplitude difference map (from 3D surface seismic) for the Midale Marly horizon. Green-to-orange and blue background colors represent negative and positive amplitude differences, respectively. 51 thousand tonnes of CO2 had been injected in the sub-vertical injection well (labelled) adjacent to the passive monitoring array at the time corresponding to the background time-lapse image.

14 Figure 5. Significant microseismic event location clusters for period of August, 2003 to January, 2006 superposed on the 2004 time-lapse amplitude difference map (from 3D surface seismic) for the Midale Marly horizon. Green-to-orange and blue background colors represent negative and positive amplitude differences, respectively. Event clusters are color-coded according to time intervals: pre-injection period (yellow); initial injection (purple); high-injectivity period (orange); low-frequency events from January, 2006 (light blue). Note that the light blue events occurred approximately 1 year following the time corresponding to the background time-lapse image.

15 Figure 6. Vertical-component waveforms for all levels of the monitoring array for a) Recording of surface calibration shot (0.125 kg), b) Event associated with a temporary production well shut-in, and c) Event associated with injection. An amplitude scale factor of 0.05 has been applied to a) relative to b) and c). Only seven of the eight channels were fully operational for recording of b) and c).

16 CO2 Injection Rate - Well 121/ Not Recording 8 9e+7 6 6e+7 4 Injection Rate (Mscm Figure H /1/03 2/1/04 4/1/04 6/1/04 8/1/04 10/1/04 12/1/04 Date Daily CO 2 injection volume (red), histogram of microseismic events (blue), and calculated cumulative seismic moment (maroon) for injection well 121/ Note period of high injection rate from May-July, e+7 Number of Events Cumulative Se Figure 8. Magnitude versus distance plot for events to March 31, 2004.

17 Figure 9. Results from the inversion of splitting measurements for fracture properties, showing the RMS misfit as a function of fracture strike and density. The 90% confidence interval is plotted in bold, and the best fit model is marked by the red lines. The inversion also considers the Thomsen parameters for a sedimentary fabric, finding the VTI strength is very low. The inversion finds a clear minimum for a fracture set striking to the NW. Figure 10. Results from a simple geomechanical model representing the Weyburn reservoir. (a) a map view of the SWS patterns that develop during injection and production, with the tick direction giving the fast direction (which matches the principal horizontal stress) and tick size giving the splitting magnitude. The reservoir limits are marked in blue and the injection and production wells in red. (b)-(d)the Mohr circle

18 evolution for cells (b) in the overburden above the injector, (c) in the reservoir by the injector, (d) in the overburden above the producer and (e) in the reservoir by the producer, with black circles showing the initial and red the final stress states