GRC Transactions, Vol. 39, 2015

279

Analysis of Seismic Moment Tensor, Finite-Source Scaling and Fluid Imaging During EGS Resource Development at The Geysers, CA

Douglas S. Dreger1, Roland Gritto2, and O. Sierra Boyd1

1Berkeley Seismological Laboratory2Array Information Technology

KeywordsThe Geysers, EGS development, fracture characterization, subsurface fluid distribution, induced seismicity, seismic analysis

1. ABSTRACT

In this paper, we investigate seismicity in the vicinity of the EGS development at The Geysers Pratti-32 injection well. For this investigation we have incorporated the capability of simultaneously inverting full waveform data and body wave amplitude ratios to improve estimates of the source mechanism of small events. We have analyzed the full moment tensor and finite-source models of five Geysers events ranging in magnitude from M 3.7 to 4.5. The scaling relationship between rupture area and moment magnitude, found for the investigated events at The Geysers, resembles that of a published empirical relationship derived for field and aftershock data for events from M 4.5 to 8.3. In order to perform spatio-temporal analysis of the subsurface water saturation, we compiled and analyzed four seismicity catalogs for the area around the injection well over the time period from October 2011 to February 2015. The catalogs consist of automatically processed events, analyst reviewed events, double-difference located events, and a subset of events with high precision locations. The catalogs contain varying numbers of events with varying precision of hypocenter locations. The result of our analysis is that the detected events need to be relocated using a 3D joint inversion for double difference hypocenter locations and P- and S-wave velocity structure. The advantage of this approach is that it combines the individual strengths of the separate catalogs.

2. Background

Enhanced Geothermal Systems (EGS) resource development requires explicit knowledge of subsurface physical parameters to address the challenges associated with quantifying the development of the fracture network, to address reservoir complexity, and to optimize geothermal well siting. The current paper reports progress on our study to assess in-situ stress magnitude and orientation, kinematic fracture parameters including rupture size, and propagation velocity and volumetric distribution of the injected fluid during EGS resource development. The successful generation of a fracture network during EGS development requires a-priori knowledge of in-situ stress, natural fracture orientation and spacing, among other parameters. However, because the orientation and magnitude of in-situ stress may not be reliably available and injecting fluids at high rates and volume may disturb the natural stress state, it is advantageous to monitor in-situ stress during the injection process. Knowing the stress evolution during the injection process is imperative to deduce the orientation and to control the formation of the fracture network.

3. Introduction3.1 Seismic Moment Tensors

Seismic moment tensor analysis is now routinely performed by global and regional monitoring observatories around the world. The seismic moment tensor has been used to characterize the mechanisms of earthquakes occurring in a variety of

280

Dreger, et al.

environments including tectonic faulting, geothermal and volcanic systems, glacial systems, and resulting from human activities such as nuclear explosion monitoring, mining operations and hazard forensics, and during operations involving the introduction of fluids underground. In our previous work we have investigated M> 3.5 earthquakes at The Geysers geothermal field using regional broadband data, finding evidence of volumetric elements to the moment tensor indicating fluid involvement during the source process (e.g., Boyd et al., 2015). A challenge to moment tensor analysis of events during EGS activities is the low magnitudes of the events requiring the simultaneous use of multiple data types in order to constrain the seismic moment tensor. We have developed a method that enables the simultaneous inversion of seismic waveforms and P-wave first-motion polari-ties in the Hudson et al. (1989) or Tape and Tape (2014) source-type space (e.g., Nayak and Dreger, 2015) that finds not only the best fitting mechanism, but also characterizes the nature of possible non-double-couple seismic radiation and its uncertainty.

Robust catalogs of focal mechanisms are needed to track the change in in-situ stress during EGS activities, which can be used to characterize the seismically active fracture network. Given the focal mechanisms, it is then possible to use empirical Green’s func-tion (EGF) methods to estimate the seismic moment rate function, which can be used to develop catalogs of event stress drops, and more importantly, the finite extent of earthquake slip using kine-matic finite-source inversion (e.g., Mori and Hartzell, 1990; Dreger, 1997; Dreger et al., 2007). In addition, focal mechanism catalogs themselves can be inverted for the in situ principal stresses (e.g., Michael, 1987; Hardebeck and Michael, 2006).

In this work, we investigate seismicity in the vicinity of The Geysers Pratti-32 injection well for the period from 10/06/2011 to 02/28/2015 during EGS activities (Figure 1). For this project we have incorporated the capability of simultaneously inverting full waveform data and body wave amplitude ratios. The basic idea is that improved estimates of the source mechanism of small events including the strike, rake and dip of activated faults, along with the scalar seismic moment are achievable by combining sparse observations of waveforms, first-motions and amplitude ratios.

3.2 Subsurface Fluid ImagingTomographic imaging of seismic Vp/Vs-ratios is a common technique used to estimate fluid saturation in geothermal

reservoirs (Gunasekera et al., 2003; Gritto et al, 2013). However, these methods typically generate smoothed estimates of subsurface fluid saturation on a larger scale, which often is insufficient to resolve the spatially confined volume of EGS stimulation. Lately, double-difference travel times of seismic P- and S-waves have been used to image fluid saturation in the subsurface based on Vp/Vs-ratio. The advantage of using double-differences of P- and S-wave travel times between neighboring events is that medium effects along the common propagation paths from the events to a common station are eliminated, which results in more precise relative hypocenter locations (Waldhauser and Ellsworth, 2000). Applications include imaging the temporal changes of fluid distribution in fault zones (Lin and Shearer, 2007, 2009) and fluid migra-tion associated with volcanic activities (Dahm and Fischer, 2013). Even though the technique shows promise to resolve smaller-scale fluid distributions in the subsurface, it has not yet been applied to engineered reservoirs. In this study, we apply the double-difference Wadati technique based on high-accuracy P- and S-wave differential travel times derived from subsample waveform cross correlation to image fluid saturation in the subsurface based on Vp/Vs-ratio. The results will allow us to estimate both, the spatial and temporal distribution of injected fluids and thus the fluid migration velocity and spatial fluid distribution during reservoir stimulation. We will use high-frequency and broadband seismic data recorded within the vicinity of Prati 32 as shown in Figure 1.

4 Results4.1 Source Analysis

4.1.1 Seismic Moment Tensors

Thus far we have investigated the source mechanisms of several earthquakes associated with the EGS stimulation project using the moment tensor method developed at Berkeley (Minson and Dreger, 2008). Previous work (Guilhem et

Figure 1. Map of the region surrounding the Prati 32 injection well at The Geysers geothermal reservoir. The red box marks area for which pre- and post-injection seismicity is analyzed. The triangles represent the locations of the three-component geophone network operated by the Lawrence Berkeley National Laboratory (LBNL).

281

Dreger, et al.

al., 2014; Boyd et al., 2015) has shown that it is possible to use both the regional broadband- and local geophone data to estimate the seismic mo-ment tensor. Furthermore, for some of the events in The Geysers evidence of non-double-couple solutions was found indicating a volume increase component. This volume increase component could be due to the flashing of pore fluid, or to tensile stress induced by the introduction of cold water into the system. Figure 2 shows the full moment tensor solution for a Mw 3.7 event that occurred on January 21, 2014. The seismic waveform data were processed by removing the instrument response to ground velocity, followed by bandpass filtering of the waveforms with a causal 4th-order Butterworth filter with corners at 0.2 and 1.0 Hz. The mechanism has a small statis-tically insignificant volume increase and consists of a mixture of a compensated linear vector dipole (CLVD) and double-couple. A network sensitivity solution (NSS; Ford et al., 2010) obtained by simultaneously inverting the waveforms shown in Figure 2 and first-motion observations from 15 geophone stations indicates that the mechanism is deviatoric and that acceptable solutions include a pure-double-couple.

We are presently working through moment tensor solutions of M 2 events utilizing combinations of seismic wave-forms, P-wave first-motions and body wave amplitude ratios. Figure 3 presents the results of a moment tensor inversion using waveform data from 11 stations for a Mw 2.87 event. The stations are located at distances between 3 and 6 km from the source. The velocity data were instrument-corrected and filtered between 0.2 to 1.0 Hz. There is a good level of fit to the data and the azimuthal coverage results in a well-constrained predominantly normal-faulting mechanism.

4.1.2 Finite-Sources & ScalingGiven a moment tensor solution, it is

possible to test which of the two possible nodal (fault) planes is the causative plane. For small events related to the EGS stimulation project it is necessary to consider higher frequencies and therefore the use of theoretical Green’s functions based on simple plane layered veloc-ity models is not adequate. Therefore to derive empirical Green’s functions, the waveforms of nearly collocated smaller events are utilized. Mori (1993) demonstrated that the waveform of a smaller earthquake could be spectrally deconvolved from the larger target earthquake yielding an estimate of the seismic moment rate function (MRF). Several MRF can subsequently be inverted for fault slip, as well as to identify the causative rupture plane. Dreger et al. (2007) applied this approach at Parkfield, California and found that it is possible to obtain finite-source parameters for micro-earthquakes exhibiting rupture complexity, and kinematic parameters that are comparable to their larger counterparts.

Figure 4 shows the slip distribution of the Mw 3.7 event (Figure 2) obtained by deconvolv-ing the records of a nearby Mw2.5 earthquake. The slip distribution is complex, with 3 or 4 as-

Figure 2. Comparison of filtered (0.2 to 1.0 Hz) data (solid) and synthetic (dashed) for the January 21, 2014 Mw 3.7 earthquake.

Figure 3. Moment tensor inversion of 0.2 to 1.0 Hz velocity waveforms (black traces). Red dashed traces are synthetic seismograms.

282

Dreger, et al.

perities being well resolved. The moment rate functions show multiple pulses of mo-ment release that give rise to the observed slip patches. The average and peak slips are 4.3 and 26.6 cm, respec-tively. Using the method of Ripperger and Mai (2004) the slip distribution can be mapped into stress change. The average stress drop over the fault rupture is 11 MPa, with a peak stress drop of 68 MPa. These high stress drops result from the com-pact nature of the individual slip patches in the rupture model, and demonstrate that the field is under high stress.

Thus far we have analyzed the finite-source models of five Geysers events ranging in magnitude from 3.7 to 4.5. One objective of our research is to estimate the coseismic fracture density of the EGS seismicity cloud. This is achieved by first determining moment tensor solutions to obtain focal mechanisms and robust magnitude estimates. Second, suitable target EGF event pairs are found to obtain MRF, which are subsequently inverted for fault slip and the orientation of the causative fault plane. Stress change maps using the method of Ripperger and Mai (2004) may then be determined. Once slip models are available for a number of events over a range of magnitude, the scaling relationship between rupture area and magnitude can be determined.

Figure 5 shows the scaling of rupture area as a function of magnitude for the five events that we have studied. The data are compared to the empirical area-magnitude scaling relationship of Wells and Coppersmith (1994), which is based on field and aftershock data for events from M 4.5 to 8.3, where the majority of events are in the M 6-7 range. Thus the comparison is very good, considering the substantial extrapolation to lower magnitudes. Nevertheless, the extrapolated relationship is consistent with the rupture area estimates obtained for the five Geysers earthquakes. It appears that four of the five events are collinear. The average stress drops of these events are between 6 and 17 MPa. The largest event (Mw 4.5), which occurred on January 12, 2014, had an average stress drop of 1.6 MPa. Judging from Figure 4, the relatively low stress drop Mw 4.5 event may be more consistent with the Wells and Coppersmith (1994) scaling relationship, although there are presently too few points to be able to reach reasonable conclusions. Our ongoing work will add additional area-magnitude data for smaller events, followed by corner frequency measurements for each event. We will compare the several other scaling relationships that have been published, as well as regress The Geysers data using both magnitude and corner frequency as independent variables. Our thinking is that The Geysers may require an independent scaling relationship, owing to the pos-sibly high state of stress, as well as to the fact that the seismicity is occurring in a system in which pore fluids may have a strong influence on the effective stress. An advantage of finding a relationship between rupture area and both magnitude and corner frequency is that we can account for variations in stress drop for the numerous smaller events where we cannot obtain finite-source models. Using the calibrated scaling relationship and the locations of the events, we will then be able to map the coseismic fracture density of the seismicity cloud associated with the EGS injection.

Figure 4. a) Slip distribution for the January 21, 2014 Mw 3.7 event. The hypocenter is indicated by the white ellipse. b) Observed moment rate functions (black) are compared to synthetics at the four stations, HOPS, MNRC, CVS and BKS. The amplitudes are proportional to 7.5e19 dyne cm/s.

Figure 5. Relationship between finite-source rupture area and Mw. The diamonds show estimates of rupture area from finite-source inversion using an empirical Green’s function approach. The red line is an extrapolation of the Wells and Coppersmith (1994) empirical relationship based on field observations of large earthquakes.

a) b)

283

Dreger, et al.

4.2 Subsurface Fluid ImagingFour catalogs of seismicity recorded during

the injection activities at Prati 32 were compiled from the Northern California Earthquake Data Center (NCEDC). NCEDC houses all seismic data acquired in northern California, including seismicity observed at The Geysers. A compila-tion of seismicity recorded from October 6th, 2011 through February 28th, 2015 in the vicinity of the Prati 32 EGS injection well at The Geysers is presented in Figure 6. The seismicity recorded by the LBNL geophone network is presented in Figure 6a. It contains the most comprehensive catalog including seismicity with a magnitude of completeness of M = 1.1. The events have been detected with an event detection algorithm and located using 3D seismic tomography. Figure 6b represents the seismicity reported by the USGS, which is reviewed by data analysts with a magni-tude threshold of approximately M = 1.2. It can be seen that the number of events are greatly reduced and concentrated around the injection interval of Prati 32. The P- and S-wave phase arrivals of these events are more precise than those in 6a, because they have been reviewed and, if necessary, cor-rected by analysts. However, the location of these events is based on a 1D velocity model. The seis-micity derived from the double-difference catalog for northern California is presented in Figure 6c. It can be seen that the seismicity is located in clusters around the injection interval, as can be expected from double-difference location techniques. Similar to the events in 6b, the latter events were located using a 1D velocity model. Finally, Figure 6d represents a subset of the seismicity show in Figure 6a, where only events with a minimum of 16 phase picks were included. The result is a reduced number of events with higher precision hypocenter locations. Comparing the catalogs in Figure 6, it becomes apparent that the hypocenter locations in Figure 6b and 6d are comparable due to the review by USGS data analysts (Figure 6b) and the inclusion of only high-accuracy events (Figure 6d). In contrast, the double-difference catalog (Figure 6c) reveals clustering of event locations that appear reasonable, as they are located around the injection well. However, the depth error of these events appears larger, since most earthquakes are located above the injection interval. While this is not unreasonable, it is more likely that the events are located below the injection interval, as is typically observed for water injection at The Geysers (Gritto et al. 2013). The differences in the locations of earthquakes in Figure 6 are due to the use of different 1D and 3D velocity models and location techniques. Therefore, we will relocate the events to create a catalog that can be used for future subsurface fluid imaging. The location technique is based on 3D joint inversion for double difference hypocenter locations and P- and S-wave velocity structure. This approach has been successfully used in past studies at The Geysers (Gritto et al., 2013). The advantage of this approach is that it combines the individual strengths of the location strategies displayed in Figure 6.

b)a)

c) d)Figure 6. Seismicity recorded from October 2011 to February 2015 in the area sur-rounding Prati 32, as shown in Figure 1. The projection of the injection well Prati 32 is indicated by the solid curved line, with the blue section representing the water injection interval. a) Seismicity recorded and located by LBNL. b) Seismic-ity recorded by LBNL and located by the USGS (analyst reviewed). c) Seismicity recorded by LBNL and located using the double-difference location method. d) High precision subset of seismicity shown in a).

284

Dreger, et al.

5. Acknowledgement

This work was supported by the Assistant Secretary for Energy Efficiency and Renewable Energy, Office of Geo-thermal Technologies, of the U.S. Department of Energy under Contract No. DE-EE0006766.

6. ReferencesBoyd, O. S., D. S. Dreger, V. H. Lai, and R. Gritto (2015), A systematic analysis of seismic moment tensor at The Geysers Geothermal Field, California,

Bull. Seism. Soc. Am., Under revision.

Dahm, T., and Fischer, T. (2013), Velocity ratio variations in the source region of earthquake swarms in NW Bohemia obtained from arrival time double-differences. Geoph.. J. Int., doi: 10.1093/gji/ggt410.

Dreger, D. (1997), The Large Aftershocks of the Northridge Earthquake and their Relationship to Mainshock Slip and Fault Zone Complexity, Bull. Seism. Soc. Am., 87, 1259-1266.

Dreger, D., R. Nadeau and A. Chung (2007), Repeating earthquake finite-source models: Strong asperities revealed on the San Andreas Fault, Geophys. Res. Lett., 34, L23302, doi:10.1029/2007GL031353.

Gritto, R., Yoo, S.H., Jarpe, S., (2013), Three-dimensional seismic tomography at The Geysers geothermal field, CA, USA. 38th Workshop on Geothermal Reservoir Engineering, February 11–13, SGP-TR-198,Stanford University, Stanford, CA, pp. 1–12.

Guilhem, A., L. Hutchings, D.S. Dreger and L.R. Johnson (2014), Moment tensor inversions of M ~ 3 earthquakes in the Geysers geothermal fields, California, J. Geophys. Res., 10.1002/2013JB010271.

Gunasekera, R.C., Foulger, G.R., Julian, B.R., (2003), Reservoir depletion at The Geysers geothermal area, California, shown by four-dimensional seismic tomography. J. Geophys. Res., 108, 2134, http://dx.doi.org/10.1029/2001JB000638.

Hardebeck, J, A. Michael 2006, Damped regional-scale stress inversions: Methodology and examples for southern California and the Coalinga aftershock sequence, J. Geop. Res., doi:10.1029/2005JB004144.

Hudson, J.A., R.G. Pearce and R.M. Rogers (1989), Source type plot for inversion of the moment tensor, J. Geophys. Res. 94, no. B1, 765-774.

Lin, G., Shearer, P., (2007), Estimating local Vp/Vs ratios with similar earthquake 397 cluster, Bull. Seism. Am., 97 (2), 379–388, 398, http://dx.doi.org/10.1785/0120060115.

Lin, G., and Shearer, P.M., (2009), Evidence for water-filled cracks in earthquake source regions: Geophys. Res. Lett., v. 36, L17315, doi:10.1029/2009GL039098.

Michael, A, 1987, Use of focal mechanisms to determine stress: A control study, J. Geoph. Res., 357-368.

Minson, S.E. and D.S. Dreger (2008), Stable inversions for complete moment tensor, Geophys. J. Int. 174, 585-592, doi: 10.1111/j.1365-246X.2008.03797.x.

Mori, J. (1993), Fault plane determination for three small earthquakes along the San Jacinto fault, California: Search for cross faults, J. Geophys. Res., 98, 17,711– 17,722.

Nayak, A. and D. S. Dreger (2015), Source-type specific inversion of moment tensors, Bull. Seism. Soc. Am., Under revision.

Ripperger, J., and P. M. Mai (2004), Fast computation of static stress changes on 2D faults from final slip distributions, Geophys. Res. Lett., 31(18), L18610, doi:10.1029/2004GL020594.

Tape, W., and C. Tape (2012), A geometric setting for moment tensors, Geophys. J. Int. 190, 476–498, doi: 10.1111/j.1365-246X.2012.05491.x

Waldhauser, F. and W.L. Ellsworth, (2000), A double-difference earthquake location algorithm: Method and application to the Northern Hayward Fault, California: Bull. Seism. Am., 90, 1353-1368.

Wells, D. L. and K. J. Coppersmith (1994), New empirical relationships among magnitude, rupture length, rupture width, rupture area and surface displacement, Bull. Seism. Soc. Am., 84, No. 4, 974-1002.