Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
1
A comparison of seismicity rates and fluid injection operations in Oklahoma and California: Implications for crustal stresses
Thomas Goebel California Institute of Technology
_______________ Seismological Laboratory, California Institute of Technology 1200E. California Blvd., Pasadena California 91125, USA. [email protected]
2
1. Abstract 1 2 Fluid injection into deep wellbores can increase pore pressure, reduce effective stress, and 3 trigger earthquakes. The extent of the seismogenic response to injection provides insight 4 into how close faults are to failure within the injection-‐affected area. This study examines 5 the seismogenic response to injection operations in hydrocarbon basins in California and 6 Oklahoma. I test whether there are significant spatial and temporal seismicity rate 7 variations and determine the corresponding timing and location based on nonparametric 8 modeling of background seismicity rates. Oklahoma experienced a recent surge of seismic 9 events, which exceeded the 95% confidence limit of Poissonian background rates in ~2010. 10 Annual injection volumes in OK increased systematically between 1998 and 2013 and have 11 been connected to several earthquake sequences [e.g. Keranen, et al. 2014]. In CA, injection 12 volumes increased monotonically between 2001 and 2009; however, the seismogenic 13 response was very limited, and devoid of large-‐scale background rate increase. I perform a 14 detailed comparison of injection parameters in OK and CA, focusing on well density, 15 wellhead pressures, peak and cumulative rates as well as injection depths. I find no 16 detectable difference that could readily explain the observed seismicity rate changes in OK 17 and lack thereof in CA. A strongly different seismogenic response to similar pressure 18 perturbations indicates that the injection parameters considered are only of secondary 19 influence on the resulting earthquake activity. I suggest that the primary controls on 20 injection-‐induced earthquakes may be the specific geologic setting and the stress state on 21 nearby faults. 22
2. Introduction: Injection induced earthquakes in California and the central 23 and eastern U.S. 24 25
The injection of waste fluids into deep disposal wells and its environmental 26 consequences are a growing concern in the central and eastern U.S. Such injection activities 27 can increase pore pressures and poroelastic stresses, which may trigger earthquakes on 28 faults close to failure [e.g. Kim, 2013; Ellsworth, 2013]. Several regions within the central 29 and eastern U.S. exhibited a pronounced increase in seismic activity coincident with 30 injection operations in near-‐by wastewater disposal wells. The corresponding seismicity 31 sequences include, the 2011 Mw4.7 Guy, Arkansas [Horton, 2012]; the Mw3.9 Youngstown, 32 Ohio [Kim, 2013]; and the Mw5.7 Prague, Oklahoma (OK) earthquake sequence [Keranen et 33 al., 2013]. Many of these sequences were associated with nearby wastewater injection into 34 high-‐permeability aquifers overlying igneous basement. These basement layers, which are 35 hydraulically connected to the above reservoirs, host the majority of the induced 36 earthquakes including the largest magnitude events. 37 38 Previous studies examined isolated cases of likely injection induced seismicity; however, a 39 synoptic identification of induced seismicity and its underlying causes is still missing in OK 40 and the central U.S. In central CA, a comprehensive, regional study revealed that induced 41 seismicity is rare considering the extensive injection activity that has occurs in close 42 proximity to active faults [Goebel, et al. 2015]. The authors identified three induced 43
3
seismicity sequences with magnitudes up to Mw4.7, based on a rigorous statistical 44 assessment of correlations between injection and seismic activity. 45 46 This study focuses on a large-‐scale assessment of differences in injection operations and 47 possibly induced seismicity in CA and OK. These two regions exhibit strong differences in 48 tectonic deformation and seismic activity. While OK experienced generally low seismicity 49 rates until 2009 [Ellsworth, 2013], seismicity rates in CA are high, as a result of pervasive 50 tectonic deformation along many active faults. To isolate the possible influence of injection 51 activity from tectonic earthquakes in CA, I limit the analysis to seismicity that occurred in 52 the major hydrocarbon basins. 53 54 The article is structured as follows. First, I determine the cumulative distribution of 55 earthquake rates and annual injection rates in CA and OK between 1980 and 2014. I then 56 test for statistically significant increases in background seismicity rates, and determine 57 when and where they occur. Finally, I examine whether differences in seismicity rate can 58 be traced back to differences in injection parameters between CA and OK. 59
3. Seismicity and injection data in CA and OK 60 61 This work concentrates on the most widely available and homogeneous seismicity and 62 injection data-‐sets, including earthquake catalogs, fluid-‐injection volumes, well-‐head 63 pressures and injection depths. Injection data have been archived by the OK Corporation 64 Commission (OCC) since ~1975 and the CA Department of Conservation, Division of Oil, 65 Gas and Geothermal Resources (DOGGR) since ~1977. The seismicity record is archived by 66 the USGS Advanced National Seismic System in CA and is available from the OK Geological 67 Survey. 68 69 Much of the seismicity in CA is tectonically driven and localized along major fault traces. I 70 exclude earthquakes along major faults and solely select seismicity within large 71 hydrocarbon basins. For this purpose, I compute the largest convex hull of well location 72 vertices, using a Delaunay triangulation algorithm. The subset of seismicity within the hull 73 corresponds to 12% of the total seismicity in CA (Figure 1). For OK essentially the entire 74 (99%) earthquake record is used. 75
I evaluate changes in network recording quality as a function of time based on 76 variations in the magnitude of completeness (Mc). The latter is computed by minimizing the 77 misfit between the observed frequency-‐magnitude-‐distribution and the modeled fit 78 assuming power law distributed data above Mc [Clauset, et al. 2009]. In CA, Mc is generally 79 close to 2 after 1995 but also shows short-‐period fluctuations, e.g., connected to the 1994, 80 Mw6.7 Northridge earthquake (Figure 1b). In OK, Mc varies between 1.5 and 2.5. 81 Consequently, the seismicity record in both regions is cut below Mc = 2.5 to ensure catalog 82 completeness. The following analyses are based on the cut catalogs within the convex hulls. 83 84
[Figure 1 here] 85 86
4
In addition to the seismicity records, I examine large-‐scale injection data to identify 87 possible temporal and regional variations. The overall well density in OK is ~0.05 km-‐2 (i.e. 88 about one well every 20 km2) which is significantly lower than in CA basins with a density 89 of 0.7 km-‐2 (i.e. one well every 1.4 km2). In CA, injection data is recorded monthly for each 90 well and categorized according to the type of injection activity, i.e., wastewater disposal 91 (WD), and enhanced oil recovery (EOR) wells. The latter are further classified into water 92 flooding (WF), steam flooding, cyclic steam injection, and pressure maintenance. Injection 93 into WD and WF wells represent the largest contribution to the total injected volumes. In 94 OK, insufficient information about WD vs. EOR is available so that both well types are 95 treated jointly. The overall contribution of WD wells to annual injection volumes is ~50% 96 in OK [Murray, 2014], and only ~20 to 30% in CA. In CA, cumulative injection rates are 97 published for each year by the DOGGR, however, with a significant time-‐lag, so that 98 presently 2009 is the most recent year with a complete record. Injection rates peaked in 99 the mid 1980s, and in 2000 and have been continuously increasing after 2001 (Figure 2a). 100 Moreover, some regions in CA such as the major oilfields in Kern county1 experienced a 101 continuous increase in injection rates over even longer periods, e.g. since ~1995, 102 highlighting that injection rate variations are not homogeneous in space. This is explored in 103 more detail below through a spatial correlation of large-‐volume injection wells and 104 seismicity rate variations. 105
In OK, the cumulative annual injection volumes were determined by summing over 106 all individual well records from the OCC database and comparing them to published values 107 between 2009 and 2013 [Murray, 2014]. Murray, [2014] performed a detailed data quality 108 assessment and removed obvious outliers. Consequently, the annual injection rates are 109 lower compared to the OCC database, especially after 2010. Nevertheless, both datasets 110 indicate a systematic increase in annual injection rates between 2008 and 2013 (Figure 111 2b). The OCC injection data prior to 2006 is incomplete but suggest a largely monotonic 112 increase between 1998 and 2013, assuming that relative injection variations are reliably 113 depicted even in the incomplete dataset. The annual injection volumes in OK are only 114 comparable to CA in 2006 with 2.0 Gbbl/yr and 2.3 Gbbl/yr respectively, and are generally 115 lower than in CA during all other periods. Moreover, the injection in OK occurs over a wider 116 area so that the difference in injection volume per area is even higher in CA compared to 117 OK. In addition to differences in fluid injection, net-‐fluid production rates and volumes may 118 influence the seismogenic response in a region. However, the assessment of net-‐fluid 119 production is complicated by geologic and reservoir complexity as well as fluid density 120 variations and thus not considered here. 121 122
[Figure 2 here] 123 124
125 126
1 The data for Kern county oilfields was attained from the DOGGR online data base: www.conservation.ca.gov/dog/ last accessed: 12/2014
5
4. Spatial and temporal changes in background seismicity rates 127 128 Seismicity rate changes are commonly dominated by aftershock clustering (Figure 2), and 129 thus provide only limited insight into changes in external driving forces such as pressure 130 increases. Such additional forces are more readily assessed using the rate of independent 131 mainshocks, i.e., the background seismicity rate (λ0) [Hainzl, et al. 2006]. To determine λ0, I 132 employ two different methods: 133
1) I remove all events within a specific space/time window after a mainshock, here 134 the largest magnitude event within a sequence. This method takes advantage of the 135 strongly localized occurrence of aftershocks in space and time. The size of this window is a 136 function of mainshock magnitude [Gardner & Knophoff, 1974]. As expected, the aftershock 137 removal demotes abrupt seismicity rate changes and the large rate increases, for example, 138 connected to the Northridge earthquake sequence, disappear. 139
2) To avoid inherent biases of the aftershock window selection, I also use a 140 nonparametric method to determine mainshock fractions and rates [Hainzl, et al. 2006]. 141 The method uses a Gamma-‐distribution to fit the observed inter-‐event time distributions 142 (ITDs), i.e. the distribution of time intervals between consecutive earthquakes [e.g. Hainzl 143 et al. 2006]: 144
𝑝 𝜏 = 𝐶 ∙ 𝜏!!!𝑒!!! ,
(1)
where C is a scaling constant, 𝜏 is the inter-‐event time normalized by earthquake rate, and 𝛾 145 and 𝛽 are parameters describing the shape and scale of the underlying distribution. The 146 background seismicity rate can be computed from the parameter 𝛽 and the observed rate, 147 𝜆, of the clustered catalog: 148 𝜆! =
1𝛽 ∙ 𝜆.
(2)
The Gamma-‐distribution parameters were estimated using a maximum likelihood fit of 149 ITDs within sliding time windows between 1980 and 2014. 150 Applying the non-‐parametric method to the CA seismicity dataset, I find that λ0 151 decreased systematically from 12 to 6 events per year between 1983 and 2011, however 152 this variation falls well within the 95% confidence interval of Poissonian distributed data 153 with a mean of 9 events per year (Figure 3a). In OK, λ0 shows little variation between 1983 154 and 2009, but increased rapidly thereafter up to 23 events in 2011 and 58 events in 2013. 155 These values exceed the 95% confidence interval of Poissonian distributed data with a 156 mean of 4 in ~2010. I tested the sensitivity of these results to the selected time window by 157 varying its length from 4 to 15 years. The results are independent of particular time 158 window selections such that the systematic increase in λ0 is observed consistently for all 159 time windows as well as for the declustered catalog using method (1). 160 161
[Figure 3 here] 162 163 In addition to changes in λ0 over time, I examine spatial variations in numbers of 164 mainshocks and correlate them with locations of large-‐volume injection wells in OK and CA. 165 Spatial variations are computed within a grid of 0.1 degree spacing in CA and 0.4 spacing in 166 OK. The rates at each node are then smoothed using an isotropic, Gaussian kernel 167
6
preserving the initial rates and normalized by the corresponding time interval and grid 168 spacing. I compare the smoothed rates in CA before and after the most recent, pronounced 169 increase in fluid injection rates in 2001 (see Figure 2a), and highlight regions that 170 experienced an increase in λ0 above the 95% confidence limit of a Poissonian process 171 (Figure 4). CA exhibits only localized areas with significant rate increase, connected to the 172 2013, Mw4.8 Isla Vista event [-‐119.9°N, 34.4°W] and the 2008, Mw5.4 Chino Hills 173 earthquakes [-‐117.8°N, 34.0°W]. 174 175
[Figure 4 here] 176 177 In OK, on the other hand, the number of mainshocks increased significantly after 2009. This 178 increase was concentrated in central and northern OK (Figure5). Moreover, the newly 179 emerging mainshock activity exceeds the 95% confidence limit of Poissonian distributed 180 data in a large region (red contours in Figure 5b), which is in agreement with Llenos et al., 181 [2014]. The significant increase in λ0 occurred in the neighborhood of some large-‐volume 182 injectors with maximum monthly rates of more than 1 Mbbl/mo (i.e. million barrel per 183 month). The median distance between individual earthquakes and the closest high-‐volume 184 injection wells decreased significantly from 39 to 20 kilometers after 2009 (Figure 5b). 185 This decrease in distance is observed for wells with both intermediate and high peak 186 injection rates and primarily a result of a northward seismicity migration. The extensive, 187 newly appearing mainshock activity in OK after 2009, represents a strong contrast to the 188 lack of observable changes in seismic activity in CA. If much of this activity is induced it 189 may also indicate a substantial difference in injection parameters between OK and CA. In 190 the following, I perform a detailed comparison of injection operations between CA and OK 191 to identify possible systematic differences. 192 193
[Figure 5 here] 194
5. Comparison between injection operations and focal depths in CA and OK 195 196 Injection rates and pressures may significantly influence the seismogenic potential of fluid 197 injection. I determine the maximum wellhead pressures and injection rates from the 198 DOGGR and OCC databases for individual, active wells to determine whether statistically 199 significant differences in injection parameters can be identified. In CA, I examine both WD 200 and WF wells, the latter as a proxy for typical EOR wells. In OK, both well types are treated 201 jointly. As expected, much of the injection activity is conducted at comparably low 202 pressures and rates. The two study regions show strong similarities in median peak-‐203 pressures, i.e. 4.3±0.2 MPa in OK and 5.5±0.2 MPa in CA, and peak injection rates, i.e., 204 2900±200 m3/mo in OK and 4700±200 m3/mo in CA (Figure 6). Both peak rates and 205 pressures are higher for WD compared to WF wells in CA, but no systematic differences 206 between CA and OK could be identified that could be responsible for the observed 207 difference in seismic activity. 208 209
[Figure 6 here] 210 211
7
Besides rates and pressures, injection depth may control induced seismicity potentials. In 212 OK, the packer depth of injection wells, which is a minimum value for injection depth, is 213 readily available from the OCC website. In CA, on the other hand, detailed depth 214 information is only available in the form of non-‐searchable pdf files. I used these files to 215 extract the effective depth, which provides a maximum value for injection depth, using 216 open source optical character recognition software. The quality of the pdf files requires 217 visible inspection of each value, so that I limited the analysis to 300 randomly sampled WD 218 and 300 randomly sampled WF wells. In agreement with Jordan & Gillespie, [2014], I find 219 that WD injection occurs on average at shallower depth than WF injection in CA (Figure 220 7a). The median depth of the former is ~0.9 km which is comparable with the median 221 depth of 1 km of injection activity in OK (Figure 7b). The median depth of joint WD and WF 222 depths in CA increases to ~1.5 km, in agreement with values reported by the DOGGR. This 223 depth is substantially deeper than the value reported for OK. In summary, I find no 224 evidence that injection activity is deeper in OK, on the contrary fluid injection in CA occurs 225 on average ~0.5 km deeper than in OK. 226 227
[Figure 7 here] 228 229 While injection depths in CA and OK were generally limited to the upper 4 km, the focal 230 depths of seismic events in both regions are significantly deeper. To compare focal depths 231 in CA and OK, I use a waveform-‐relocated catalog [Hauksson, et al. 2012] and the single 232 event depth within the OGS catalog. In CA, much of the seismic activity occurs distributed 233 within the upper 15 km with a median of 10 km (Figure 8). Events between M4 and M5 and 234 above M5 show a wide depth-‐range, from ~1-‐27 km between 1980-‐2001. Much of the 235 shallow earthquakes are aftershocks of the 1994 Northridge earthquake, whereas deep 236 events occur close to Ventura basin. The focal depths in OK did not change significantly 237 between 1980-‐2001 and 2002-‐2014, exhibiting the same mean depth of 5 km. The more 238 abundant earthquakes above M4 in OK between 2002 and 2014 (Figure 8), occurred within 239 a relatively localized depth layer extending from ~2 to 7 km. While substantial 240 uncertainties in focal depth are expected especially is regions with limited station coverage 241 and poorly constrained velocity models such as OK, the observed systematic differences 242 between CA and OK seem to be a robust feature within the data. Focal depths are 243 significantly deeper in CA than in OK. 244 245
[Figure 8 here] 246 247
6. Discussion 248 249
In this study, I compared fluid injection operations and examined temporal and 250 spatial seismicity variations in CA and OK. CA showed no large-‐scale correlation between a 251 recent increase in injection volumes and seismic activity. Seismicity is generally deeper in 252 CA and surficial injection operations likely have only limited influence on earthquake 253 activity. This is expected considering the low upper crustal stresses and frictional 254 properties of shallow basin faults, which inhibit seismic activity and large earthquake 255
8
ruptures [e.g. Goebel et al., 2015]. OK, on the other hand, showed a significant increase in 256 background seismicity rates starting in 2009. This newly appearing seismic activity 257 encompasses a region that also hosts large-‐volume injection wells in central and northern 258 OK. The spatial-‐temporal correlation between injection and seismicity indicate that fluid-‐259 injection may contribute to seismic activity at a large-‐scale, as pointed out by several 260 previous studies [Keranen, et al. 2013, Ellsworth, 2013, Llenos & Michael, 2013, Keranen, et 261 al. 2014]. 262
The injection operations in OK and CA show many remarkable similarities. These 263 similarities include the overall well count, injection depths, well head pressures and peak 264 injection rates. Some of the differences in injection operations include: 1) The overall 265 density of injection wells; and 2) the generally larger annual injection volumes in CA; 266 however, none of these differences is likely to cause the significantly stronger seismogenic 267 response to fluid injection in OK. 268
Several mechanisms may explain the strong seismogenic response in OK. First, the 269 observed seismicity rate changes in OK may be part of natural rate variations that occur 270 over long time-‐scales. The large recurrence times of intra-‐plate earthquakes and the 271 limited temporal extent of the corresponding seismicity record support this possibility. 272 However, the strong correlation between injection and seismic activity in several areas in 273 OK make this scenario less likely. Second, large-‐scale, average injection parameters may 274 not be representative of the seismogenic potential of fluid-‐injection in an entire region. The 275 seismogenic potential may be controlled by individual wells that strongly deviate from the 276 average operational parameters. Nonetheless, the reported cases of likely induced 277 seismicity in OK and CA were associated with wells of average injection activity and do not 278 seem to support this hypothesis [Keranen, et al. 2013, Goebel, et al. 2015]. Third, perhaps 279 the considered set of operational parameters is not complete and additional factors such as 280 net production rate, reservoir pressure, or the rate of change of injection rate may affect 281 poro-‐elastic stresses resulting in seismic activity within a region. 282
Finally, the difference in seismogenic response to fluid injection in CA and OK may 283 be driven by an overall difference in geologic setting and crustal stresses between plate 284 boundary and intraplate regions. The existence of higher crustal stresses at shallower 285 depth in OK is supported by shallower focal depths, which are consistent over time. 286 Moreover, several studies suggest that the upper crust in OK is close to failure. For 287 instance, the passage of seismic waves from large teleseismic earthquakes and connected 288 small dynamic strains are observed to result in an uncharacteristically strong seismogenic 289 response in the form of triggered earthquakes [VanDerElst, et al. 2013]. Furthermore, some 290 areas in Oklahoma may produce induced earthquake sequences as a result of pressure 291 perturbations as small as 0.07 MPa [Keranen, et al. 2014]. 292
Besides the difference in crustal stresses, the larger scale geologic homogeneity in 293 OK likely results in extensive hydraulic connectivity of the upper crust. Keranen et al., 294 [2014] suggested that this hydraulic connectivity may result in lateral diffusion of pressure 295 perturbations up to 20-‐35 km. In CA, the upper crust is subject to constant tectonic 296 deformation, so that crustal heterogeneity and especially mature fault zones may limit 297 lateral migration of pressure perturbations. 298
In conclusion, my results suggest that operational parameters of surficial fluid 299 injection are likely of secondary importance for the resulting seismogenic response. The 300 primary controls on injection-‐induced seismicity are the specific geologic setting, e.g. 301
9
hydraulic connectivity, and the stress state on near-‐by faults. The view that injection 302 induced earthquakes have successfully been avoided in CA in the past because of less 303 invasive injection operations is likely erroneous. The scarcity of induced seismicity in CA 304 may simply be an expression of lower stresses at injection depth and lack of large-‐scale 305 hydraulic connectivity within hydrocarbon basins. Although less probable, earthquakes 306 may be induced in CA through injection in areas of active faulting as shown by a recent 307 study [Goebel et al, 2015]. The largely similar injection operations in CA and OK and 308 absence of noticeable seismogenic response in CA indicate a fundamental difference in the 309 state of stress between the two study areas. The specific geologic conditions responsible 310 for individual, induced earthquake sequences remain to be understood. 311
10
7. Acknowledgements 312 313 I would like to thank Andrea Llenos, Sebastian Hainzl and the editors, Robert Habiger and 314 Gregory Beroza for their detailed reviews of the initial manuscript. The present work 315 benefitted from discussions with Preston Jordan and Kyle Murray. I thank the Oklahoma 316 Corporation Commission and the California Division of Oil, Gas and Geothermal Resources 317 for making the injection databases publically available. The earthquake data sets were 318 provided through the USGS Advanced National Seismic System and by the Oklahoma 319 Geological Survey. I am grateful to Whenzeng Yang, Egill Hauksson and Peter Shearer for 320 creating high-‐quality earthquake catalogs, and for Jeremy Zechar who made his 321 declustering algorithm available online. Finally, I would like to thank the Statistical 322 Seismology Community for the many useful online resources on corssa.org. 323
324
11
8. Figures 325 326
327
12
328 Figure 1: Seismicity (red dots, squares, star; see legend for magnitudes) and injection well 329 locations (blue triangles) in CA (a) and OK (d). Gray dots are earthquakes that were not 330 used within the following analysis. b) Mc variations for moving windows of 200 events (thin 331 lines) and after applying a 25-‐point median filter to remove high-‐frequency variations 332 (thick lines). Major active faults in CA are highlighted by black lines in a). 333
13
334 Figure 2: Annual fluid injection (vertical bars) and number of earthquakes (thick curves) in 335 CA (a) and OK (b) between 1980 and 2014. Earthquakes above M4 are shown by red 336 squares. Actual and inflation-‐adjusted oil prices are highlighted by dashed and solid green 337 lines in a). The light blue bars in a) are annual injection rates for the largest oil fields in 338 Kern County, CA. The dark and light blue bars in b) are injection volumes from Murray, 339 [2014] and the OCC database. Mbbl/yr refers to million barrels per year. 340
14
341 Figure 3: Background seismicity rates (large dots) and fraction of mainshocks (squares) for 342 CA (a) and OK (b). Median values and 95% confidence interval assuming a stationary 343 Poisson process are shown by solid and dashed lines. I explored the influence of time bins 344 by varying the binsize between 4 to 15 years (small dots and gray error) and averaged over 345 these different intervals using a 2-‐year time step (large dots). Vertical error bars show the 346 corresponding standard deviations from bootstrap resampling. 347
15
348 Figure 4: Smoothed, spatial variation in background seismicity rates in CA basins between 349 1980-‐2001 (a) and 2002-‐2014 (b). Earthquake locations are shown by black dots, location 350 of high-‐volume injection wells (>40,000 m3/mo) by blue triangles and areas of significant 351 rate increase by red contours. The insets show the distance between earthquakes and 352 closest high-‐volume injection well. 353
16
354 Figure 5: Smoothed, spatial variation in background seismicity rates in OK between 1980-‐355 2008 (a) and 2009-‐2014 (b). Earthquake locations are shown by black dots, location of 356 high-‐volume (>50,000 m3/mo) injection wells by blue triangles and areas of significant rate 357 increase by red contours. Histograms show changes in distance between earthquakes and 358 closest high-‐volume injection well between the two time periods. 359 360 361
17
362 Figure 6: Maximum injection pressure (a) and injection rate (b) for all wells in OK (green) 363 and for WD (orange) and WD+WF wells (red) in CA. Median values and 95th percentiles are 364 shown by solid and dashed lines. 365 366
18
367 368 Figure 7: Injection depth in OK (green) and CA (orange for WD, red for WD+WF wells). 369 Median depths and 95% confidence intervals are shown by solid and dashed lines. Note 370 that water flooding occurs on average deeper than water disposal in CA. Depth in OK is 371 reported as packer depth whereeas depth in CA is reported as effective surface depth. 372 373
19
374 Figure 8: Comparison of focal depth between 1980-‐2001 (a) and 2002-‐2014 (b) in CA (red) 375 and OK (green). Mean values and 95% confidence intervals are highlighted by solid and 376 dashed lines. The depth of M4-‐5 events is shown by colored squares and for M>5 events by 377 colored stars. Seismicity rates in OK did not increase significantly until ~2009, however, 378 the focal depth in b) is still dominated by the more abundant events after 2009. 379 380 381
20
9. References 382 383 Clauset, A., Shalizi, C.R. & Newmann, M.E.J., (2009). Power-‐law distributions in empirical 384 data, SIAM Rev., 51(4), 661–703. 385
Ellsworth, W. L. (2013). Injection-‐induced earthquakes. Science, 341(6142). 386 387 Gardner, J. K., & Knopoff, L. (1974). Is the sequence of earthquakes in southern California, 388 with aftershocks removed, Poissonian? Bull. Seismol. Soc. Am., 64(5), 1363-‐1367. 389 390 Goebel, T.H.W., E. Hauksson, F. Aminzadeh, & J.-‐P. Ampuero (2015). An objective method 391 for the assessment of possibly fluid-‐injection induced seismicity in tectonically active 392 region in central California, J. Geophys. Res., doi:10.1029/2015, (in revision). 393 394 Hainzl, S., Scherbaum, F., & Beauval, C. (2006). Estimating background activity based on 395 interevent-‐time distribution. Bull. Seismol. Soc. Am., 96(1), 313-‐320. 396 397 Hauksson, E., Yang, W., & Shearer, P. M. (2012). Waveform relocated earthquake catalog for 398 southern California (1981 to June 2011). Bulletin of the Seismological Society of America, 399 102(5), 2239-‐2244. 400 401 Horton, S. (2012). Disposal of hydrofracking waste fluid by injection into subsurface 402 aquifers triggers earthquake swarm in central Arkansas with potential for damaging 403 earthquake. Seismol. Res. Lett., 83(2), 250-‐260. 404 405 Keranen, K. M., Savage, H. M., Abers, G. A., & Cochran, E. S. (2013). Potentially induced 406 earthquakes in Oklahoma, USA: Links between wastewater injection and the 2011 Mw 5.7 407 earthquake sequence. Geology, 41(6), 699-‐702. 408 409 Keranen, K. M., Weingarten, M., Abers, G. A., Bekins, B. A., & Ge, S. (2014). Sharp increase in 410 central Oklahoma seismicity since 2008 induced by massive wastewater injection. Science, 411 345(6195), 448-‐451. 412 413 Kim, W. Y. (2013). Induced seismicity associated with fluid injection into a deep well in 414 Youngstown, Ohio. J. Geophys. Res., 118(7), 3506-‐3518. 415 416 Jordan, P., & Gillespie, J. (2013), Potential Impacts of Future Geological Storage of CO2 on 417 the Groundwater Resources in California’s Central Valley: Southern San Joaquin Basin Oil 418 and Gas Production Analog for Geologic Carbon Storage. California Energy Commission. 419 Publication number: CEC-‐5002014-‐029 420 421 Llenos, A. L., & Michael, A. J. (2013). Modeling earthquake rate changes in Oklahoma and 422 Arkansas: possible signatures of induced seismicity. Bull. Seismol. Soc. Am., 103(5), 2850-‐423 2861. 424 425 426
21
Llenos, A. L., J. L. Rubinstein, W. L. Ellsworth, C. S. Mueller, A. J. Michael, A. McGarr, M. D. 427 Petersen, M. Weingarten, and A. A. Holland, (2014), Increased earthquake rates in the 428 central and eastern US portend higher earthquake hazards, AGU Fall Meeting, Abstract 429 U34A-‐02. 430 431 Murray K.E., (2014), Class II Underground Injection Control Well Data for 2010–2013 by 432 Geologic Zones of Completion, Oklahoma. Oklahoma Geological Survey Open-‐File Report 433 (OF1-‐2014). Norman, OK. pp. 32. 434 435 Van der Elst, N. J., Savage, H. M., Keranen, K. M., & Abers, G. A. (2013). Enhanced remote 436 earthquake triggering at fluid-‐injection sites in the Midwestern United States. Science, 437 341(6142), 164-‐167. 438 439 440
441
442