ABSTRACT

The largest recorded earthquake in Kansas occurred northeast of Milan on 12 November 2014 (M w 4.9) in a region previously devoid of significant seismic activity. Applying multistation processing to data from local stations, we are able to detail the rupture process and rupture geometry of the mainshock, identify the causative fault plane, and delineate the expansion and extent of the subsequent seismic activity. The earthquake followed rapid increases of fluid injection by multiple wastewater injection wells in the vicinity of the fault. The source parameters and behavior of the Milan earthquake and foreshock–aftershock sequence are similar to characteristics of other earthquakes induced by wastewater injection into permeable formations overlying crystalline basement. This earthquake also provides an opportunity to test the empirical relation that uses felt area to estimate moment magnitude for historical earthquakes for Kansas.

INTRODUCTION

The largest recorded earthquake in Kansas occurred about 5 km northeast of Milan on 12 November 2014 with M w 4.9. It was felt in at least six states, with felt reports from cities as far as Memphis, Tennessee (720 km to the east), Omaha, Nebraska (470 km to the north), and Norman, Oklahoma (220 km to the south). Although the maximum intensity reported by the U.S. Geological Survey’s (USGS) “Did You Feel It?” (DYFI) program was VI at Argonia, Kansas (about 8 km away from the epicenter), most of the reported damage occurred in the city of Milan (about 5 km away from the epicenter). Reports included instances of cracked plaster, items thrown from shelves, and structural damage to some buildings constructed of unreinforced masonry.

From 1974 to 2012, Kansas had experienced only 15 events of M w >3.0, yielding a rate of seismicity of less than one earthquake every two years. In 2014 and 2015, the number of events climbed sharply to 108 earthquakes (M w ≥3.0). These events were located almost exclusively in Harper and Sumner counties, a region in southern Kansas where wastewater injection from oil and gas extraction operations had dramatically increased, starting about 2010, when 19 million barrels of wastewater were injected, to 2013, when 61 million barrels were injected (Kansas Corporation Commission Docket Number 15‐CONS‐770‐CMSC). Understanding the nature of this earthquake and its foreshock–aftershock sequence is important to characterizing seismic hazard in Kansas. In this study, we examine the source characteristics of the mainshock, analyze the expansion of the Milan sequence to delineate previously unmapped faults, and evaluate whether the sudden injection of wastewater by disposal wells into a deep geological formation contributed to the dramatic increase in seismic activity. The known M w and felt reports from this earthquake also provide an opportunity to test the efficacy of an empirical relation that has been applied to historical earthquakes to estimate moment magnitude from felt area.

TECTONIC BACKGROUND

The major structural features of Kansas include the Nemaha and central Kansas uplifts (Fig. 1; Baars and Watney, 1991; McBee, 2003). The Nemaha uplift is a 600‐km‐long feature that extends from eastern Nebraska to central Oklahoma. The Nemaha fault system is a right‐lateral, strike‐slip fault zone that throughout its length is transected by cross‐cutting northwest‐trending, left‐lateral shears (McBee, 2003). The Nemaha uplift and its attendant fault system are the dominant tectonic features near Harper and Sumner counties of south Kansas where the surge in seismic activity has been concentrated. In the study area of Harper and Sumner counties, the subsurface consists of thin horizontal sedimentary strata that are predominantly Paleozoic layers of limestone, shale, and sandstone from 50 to 200 m thick. The bottommost formation, the Cambro‐Ordovician Arbuckle, is a permeable layer overlying a Precambrian basement that contains numerous pre‐existing faults. The depth to the Precambrian basement for the study area is ∼1.7 km.

DATA

The Milan sequence consists of events within a radius of 10 km around the mainshock epicenter that occurred in the months preceding and in the year following the mainshock. Initial locations of earthquakes were obtained from the USGS/COMCAT system (see Data and Resources). In response to the increase of earthquakes in Harper and Sumner counties starting in 2013, the USGS deployed a local seismic network in southern Kansas now called the Induced Seismicity Project Menlo Park (ISMP)‐Kansas network. The network, which began with five stations in March 2014, has since evolved into its current 15‐station configuration that is composed of broadband seismometers and accelerometers (Fig. 2). Although the station distribution currently provides excellent near‐source and azimuthal coverage for the area of the Milan sequence, early station coverage was uneven. Coverage prior to the mainshock was limited to stations 10 km or more south or west of the earthquake sequence. Within one week of the mainshock, three additional stations were deployed. Events located by the ISMP network report M L magnitudes; unless otherwise stated, all magnitudes in this article are moment magnitudes using conversions described in Petersen et al. (2015).

Earthquakes are initially processed in Advanced National Seismic System (ANSS) Quake Monitoring System (AQMS) with magnitudes and locations computed in HYPOINVERSE (Klein, 2002). Solutions for earthquakes within a prescribed area, which meet quality criteria and are above M L 1.5, are manually processed and submitted to COMCAT. Using these solutions, we refine the earthquake locations using the Hypocentroidal Decomposition (HD) algorithm (Jordan and Sverdrup, 1981) as implemented by McNamara, Benz, et al. (2015). In contrast to the conventional single‐event methods used by most reporting services, the HD algorithm employs a multiple‐event approach similar to methods such as Joint Hypocentral Determination (Dewey, 1972) and Double Difference Relocation (Waldhauser and Ellsworth, 2000). The multiple‐event approach allows data from newly deployed stations to improve location estimates of earlier earthquakes. Using dense local networks, McNamara, Benz, et al. (2015) demonstrated that the scatter in locations can be reduced by a factor of two, allowing details of small fault lineations to be discerned in clusters of earthquakes in Oklahoma. The velocity model is from Herrmann et al. (2011) modified to fit surface‐wave dispersion observations and local earthquake arrival‐time observations. Depth to basement in the model is 2.0 km. Shifts in epicenters between initial COMCAT locations and the HD locations are shown in Figure 3.

SOURCE CHARACTERISTICS OF THE MILAN EARTHQUAKE SEQUENCE

The relocations of 193 events of the Milan sequence from October 2014 through December 2015 are shown in Figure 4. Analyzing the progress of the sequence in discrete time windows enables us to derive the geometry of the causative fault as well as recognize nearby faults.

Five earthquakes prior to the M w 4.9 mainshock were recorded by the ISMP network. Two events with M w 3.4 and 2.9 occurred on 8 September, two months before the mainshock, at virtually the same location as the eventual mainshock (Fig. 5a). Three other events, one in February and two in July, were also identified in the area, but the limited quantity of stations at the time of these events precluded their inclusion in the HD relocation. Their single‐station locations are shown with gray circles. As their locations are within the main cluster that later develops, they do not affect the choice of a fault plane. However, their timing indicates that the state of stress in the small region changed as early as nine months before November’s mainshock. A large increase in earthquakes preceding the mainshock has been observed in other suspected induced earthquake sequences in Prague, Fairview, and Cushing in Oklahoma (McNamara, Benz, et al., 2015; McNamara, Hayes, et al., 2015; W. L. Yeck et al., unpublished manuscript, 2016; see Data and Resources) as well as Guy–Greenbrier, Arkansas (Horton, 2012), Trinidad, Colorado (Rubinstein et al., 2014), and Azle, Texas (Hornbach et al., 2015).

The aftershocks that occurred in the 31 days following the mainshock (Fig. 5b) are confined to a well‐delineated area (orange‐shaded zone) with dimensions about length 4 km and width 3 km about the mainshock hypocenter. We interpret this zone to be the coherent fault that ruptured dynamically. The north–south nodal plane of the focal mechanism for the mainshock is the preferred fault plane as it is consistent with the narrow north–south trend of aftershocks. Its right‐lateral motion is also consistent with the Nemaha fault system. Like fault planes of earthquakes associated with the Nemaha fault system further to the south in Oklahoma, the P axis of this nodal plane was optimally oriented with the maximum horizontal compression direction for fault activation (Holland, 2013; Alt and Zoback, 2014). The identified fault plane appears to be an extension of a mapped Nemaha fault but its connection to the fault system is not known. Many of the rupture zones delineated by clusters of earthquakes in the Nemaha fault system further to south in Oklahoma have also been observed to occur in the proximity of, but not on, mapped faults (McNamara, Hayes, et al., 2015). For the foreshock, the rotation in the T axis relative to that of the mainshock makes the choice of rupture on the north–south nodal plane incompatible with right‐lateral slip of the mainshock. The choice of the faulting on the east–west nodal plane implies that conjugate faulting preceded the mainshock. Conjugate faulting has been observed in cratonic settings, for both natural events (e.g., Wetmiller et al., 1988; Choy and Bowman, 1990) and induced events in the Wilzetta‐Whitetail fault system (e.g., McNamara, Hayes, et al., 2015).

Aftershocks in the fifth and sixth weeks following the mainshock (Fig. 5c) subside substantially within the dynamic rupture zone, but activity develops outside the periphery of the dynamic rupture zone. At seven weeks through five months (Fig. 5d), we see a near absence of activity within the rupture zone. Finally, aftershocks from the seventh week through the following year (Fig. 5e) show two predominant types of aftershock activity. The first type is clustering at the northern and southern ends of the dynamic rupture zone. The second type of activity is diffusely spread about, but clearly not on, the inferred fault for the mainshock.

The pattern of aftershocks suggests that the main rupture and dominant release of energy occurred on a well‐delineated rupture zone (orange area in Fig. 5b). The absence of seismicity within that zone between seven weeks and five months indicates near‐complete release of strain (Fig. 5d). Gradual crack instability at the periphery of the dynamic rupture zone would explain time‐dependent continuation of the rupture at the northern and southern ends of the rupture zone (Das and Scholz, 1981). The concentrated aftershocks at the northern and southern ends of the rupture zone may be at high‐strength regions that were stressed where dynamic rupture stopped.

The extent and geometry of the dynamic rupture are also seen in cross section (Fig. 6). The one‐month aftershocks (yellow circles) delineate the dynamic rupture zone. Nucleation of the mainshock (red circle) at the bottom of dynamic rupture zone implies that rupture was predominantly up‐dip. The later aftershocks (blue circles) expand outward from the main rupture to delineate the quasi‐static growth over the next year. However, in conjunction with the complementary map views (Fig. 5d,e), a number of the later aftershocks are observed to cover a diffuse area far from the main fault. Although relatively sparse, off‐main fault aftershocks seem to define at least one en echelon fault that is almost 5 km away from the mainshock fault (dashed line parallel to the inferred mainshock fault in Fig. 5e).

Assuming that the distribution of early aftershocks represents the coseismic rupture area, the mainshock rupture surface has an area of 12 km2. This is significantly larger than the 5.3 km2 estimated from the empirical relationships between subsurface rupture area and M w for tectonic source regions (Wells and Coppersmith, 1994). Rupture zones of earthquakes in induced‐earthquake fault zones of Oklahoma have also been found to be larger than expected from empirical estimates for tectonic earthquakes. If we approximate the area as a circular asperity with radius ∼1.7 km, the relationship between corner frequency and asperity radius (Das and Kostrov, 1986) yields a corner frequency of ∼0.9 Hz. A Brune stress drop based on the same radius yields 1.7 MPa. This is similar to 0.95 Hz for the lower corner frequency found by O. S. Boyd et al. (unpublished manuscript, 2016; see Data and Resources) by applying the method of spectral ratios to the mainshock, three foreshocks, and an aftershock. The Brune stress drop derived by O. S. Boyd et al. (unpublished manuscript, 2016; see Data and Resources) ranges from 2 to 4 MPa. C. Cramer (personal comm., 2016) found a similar corner frequency at ∼1.2 Hz and a Brune stress drop of 5 MPa using velocity shear‐wave spectra of body waves. These stress‐drop estimates for the Milan earthquake are consistent with the range of stress drops observed for induced sequences in Oklahoma (Hough, 2014; Sun and Hartzell, 2014; Sumy et al., 2014). This stress‐drop range is much smaller than the median value of 14.0 MPa obtained by Atkinson and Boore (2006) for tectonic earthquakes of eastern North America. However, tectonic earthquakes usually occur at depths deeper than those of induced earthquakes. Yenier and Atkinson (2015) and O. S. Boyd et al. (unpublished manuscript, 2016; see Data and Resources) have suggested that dependence on depth as well as rupture model can account for the difference in stress drop between shallow and deeper earthquakes.

CHARACTERISTICS IN COMMON WITH INDUCED EARTHQUAKES

It is well documented that the introduction of fluids near a critically stressed fault can induce failure by increasing pore pressure and reducing the normal stress across a fault surface (Healy et al., 1968, Raleigh et al., 1976). The Milan 2014 earthquake sequence has a number of characteristics consistent with the hypothesis that it was induced by fluid injection. These characteristics broadly follow the criteria of Davis et al. (1995), and indicate that the earthquake sequence was induced. First, the surge of earthquakes with M w ≥2.5 is unprecedented in the Milan region of south Kansas. In the vicinity of Harper and Sumner counties, only two earthquakes, both with M w 2.7, were reported by the USGS/National Earthquake Information Center for the period of 1964–2010. With a historic rate of less than one earthquake of M w >2.5 per year, this region was effectively aseismic. The quiescence was shattered in 2014 and 2015 as 374 earthquakes with M w ≥2.5 occurred. Of these, 24 were in the Milan sequence. Second, the onset of seismicity started soon after the initiation of injection of large volumes of wastewater. The injection volumes at monthly intervals from 2013 to 2015 of wells in the region are compared with the occurrence of the mainshock and foreshocks in Figure 7. The total amount injected by close‐in wells within 10 km of the mainshock is shown by the solid black line, whereas the total amount injected by wells between 10 and 20 km of the mainshock is shown by the dashed black line. Both curves show injection activity was low‐level in 2013 (averaging ∼100 kilobarrels(kbl)/month) but accelerated dramatically during the next year. For the close‐in wells, activity suddenly shot up to more than 400 kbl/month in July 2014, which was sustained for several months. The mainshock occurred in November 2014. The depth of the mainshock, as well as all of the earthquakes in the Milan sequence, is in the Precambrian basement (Fig. 6). The depths of the wastewater injection wells in the Milan region typically bottom out at depths of 1.3–1.7 km, placing the injection within a few hundred meters of, if not on or into, the basement. The observation that the earthquake depths are deeper than the depths of wells is consistent with suspected induced earthquakes in Oklahoma where wells inject into the Arbuckle formation, and induced earthquakes are in the basement (Keranen et al., 2014; McNamara, Benz, et al., 2015). Earthquakes associated with fluid injection have also been observed to occur in basement beneath the Ellenburg formation, Texas (Hornbach et al., 2015) and the Ozark Aquifer, Arkansas (Horton et al., 2012). Finally, as delineated by the pattern of well‐located aftershocks, a fault exists whose activation could be attributed to fluid injection.

Rizzo and Dane are the wells within 10 km of the mainshock that injected the highest volumes in 2014 (solid red and solid blue lines, respectively, in Fig. 7). When considering injection volumes alone, Rizzo would be considered the dominant well that is influencing seismic activity. Each month the amount injected by Rizzo was at least twice as much as Dane. Moreover, Rizzo reported using injection pressures between 200 and 400 psi (about 1–3 MPa). In contrast, all other wells reported 0.0 psi injection pressure, presumably relying on gravity feed for wastewater disposal. However, the higher injection pressures used by Rizzo are not necessarily an overriding factor in inducing the mainshock. Gravity‐feed (zero pressure) injection wells have been known to induce earthquakes in underpressure structures such as the Raton basin between Colorado and New Mexico (Rubinstein et al., 2014). Although it injected lower volumes than Rizzo, Dane is located much closer to the susceptible fault (about 4 km vs. 8 km for Rizzo). Moreover, the drill depth of Dane is 1.7 km, whereas the drill depth for Rizzo is 1.4 km. The depth to the Precambrian basement can be as shallow at 1.7 km (see Data and Resources). Hydraulic connectivity between injection fluid and basement rocks has been suggested as a mechanism of inducing basement earthquakes (Ellsworth, 2013; Zhang et al., 2013; Walsh and Zoback, 2015; Chang and Segall, 2016). Thus, the pore‐pressure alteration in the basement due to wastewater injection from Dane might have been enhanced by the well’s closer proximity to the fault and by injection in or near the Precambrian basement. Determining the relative effect on pore pressure from injection at these two wells warrants further hydrogeological modeling.

Diffusion of pore pressure away from injection wells can result in induced seismicity far from the injection points. Keranen et al. (2014) demonstrated with hydraulic diffusivity modeling that small pore‐pressure perturbations (∼0.07 MPa) are sufficient to trigger earthquakes in the Jones, Oklahoma region at distances of 10–20 km from high‐volume injection wells. The hydrologic property controlling pore‐pressure diffusion is hydraulic diffusivity. Talwani et al. (2007) demonstrated that the hydraulic diffusivity c is c=r2/4Δt, in which Δt is the time lag between the start of fluid injection in a well and r is the distance between the well and the onset of seismicity. We first consider wells within 10 km of the mainshock. For Dane and Rizzo (4 and 8 km from the mainshock, respectively), we take Δt as the five months between the sharp increase in total injection volume in June 2014 to the time of the mainshock (black solid line, Fig. 7). For these wells, the hydrologic diffusivity c varies from 0.3 to 2.0 m/s2. This is within the 0.1–10 m/s2 range found by Talwani et al. (2007) for induced earthquakes and by McNamara, Hayes, et al. (2015) for Cushing, Oklahoma. We next consider the distant injection wells, Dowis and Perth, which are ∼16 km from the mainshock. Their combined injection volume shows a sharp jump about September 2014, a few weeks before the mainshock. If these wells were to be inducing the earthquakes, a hydraulic diffusivity c of about 13 m/s2 would be required. Although this value is outside the range associated so far with induced earthquakes, the wells might still have been a contributing factor.

Following the mainshock in November 2014, the total volume of injected fluid for both close‐in and distant wells rapidly decreased as most wells (including Dane) ceased or significantly reduced injection (Fig. 7); however, Rizzo continued to inject steadily between 200 and 300 kbl/month. Concurrent with the volume decrease in 2015, the number of earthquakes with M w ≥2.5 has for the time being decreased. Although this is likely due to the decreasing rate of aftershocks, it may indicate a link between injection operations and seismicity. We note, however, that aftershocks of other induced‐earthquake sequences have persisted for years after cessation of fluid injection (e.g., Healy et al., 1968). Moreover, as we noted in the description of the rupture process, the periphery of the rupture zone may have been stressed as strain was relieved within the rupture zone. If ambient tectonic stress keeps faults within intraplate Precambrian basements at near‐critical state of stress (Townsend and Zoback, 2000), it would be premature to conclude that the hazard has been completely eliminated by a reduction of injection volume.

FELT AREA AND M w

(1) 2. The recent Milan earthquake provides the first opportunity to evaluate the efficacy of the empirical formula for Kansas by comparing a known M w with felt area. In lieu of traditional intensity surveys, a felt area can be approximated from the data in the DYFI program run by the USGS, which collects information reported by people who felt an earthquake (2), M w is 4.5 from equation 2), M w is 4.7. For the measured M w 4.9, the expected empirical felt area is 510,000 km2. Depending on the interpretation of the felt area used in equation w could be different from the actual M w by a few tenths of a magnitude unit. In assessing seismic hazard, a recent empirical formula developed for the central and eastern United States from the Electric Power Research Institute et al. (2012) that uses felt area to estimate moment magnitude is often applied to historical earthquakes. For historical earthquakes, the moment magnitude can be estimated using in which FA is the felt area in km. The recent Milan earthquake provides the first opportunity to evaluate the efficacy of the empirical formula for Kansas by comparing a known Mwith felt area. In lieu of traditional intensity surveys, a felt area can be approximated from the data in the DYFI program run by the USGS, which collects information reported by people who felt an earthquake ( Dewey et al., 2002 ). Although the Milan earthquake generated intensities up to VI, we are interested in the extent of the felt area. This is generally measured at the periphery of intensity II and III reports (J. Dewey, personal comm., 2016). Figure 8 shows population centers that are more than 200 km from the epicenter and that are regarded as reliable as DYFI received a minimum of five reports from them (see Data and Resources ). For instance, Dodge City, Kansas (220 km to the west) had 39 reports; Omaha, Nebraska (470 km to the north) had 15 reports; Norman, Oklahoma (224 km to the south) had 138 reports; and Memphis, Tennessee (720 km to the east) had 9 reports (see Data and Resources ). The felt area can be approximated by a polygon enclosing the periphery of reporting cities. Moreover, as Memphis is isolated from the preponderance of reporting cities, we consider two alternative felt areas as approximately the upper and lower bounds. A smaller felt area excludes Memphis (Fig. 8 , dotted line), whereas a larger area includes Memphis (dashed line). For the smaller area (218,000 km), Mis 4.5 from equation (1) . For the larger area (388,000 km), Mis 4.7. For the measured M4.9, the expected empirical felt area is 510,000 km. Depending on the interpretation of the felt area used in equation (1) , the empirical Mcould be different from the actual Mby a few tenths of a magnitude unit.

SUMMARY

The Milan earthquake of 12 November 2014 (M w 4.9) was the largest recorded earthquake to occur in Kansas. Data from the local stations of the ISMP network in tandem with multistation processing provided excellent resolution for identifying the causative fault plane, for detailing its rupture geometry, and for delineating the expansion and extent of the subsequent seismic activity. The first‐month aftershocks define a well‐delineated zone that aligns with the north–south nodal plane of the USGS/COMCAT focal mechanism. The causative fault is inferred to be an unmapped basement fault possibly related to the Nemaha fault zone. This earthquake has the characteristics associated with other earthquakes induced by fluid injection: (1) the unprecedented occurrence of regional earthquakes with M w ≥3.0, (2) wastewater injection wells in the vicinity (about 10–20 km) of the mainshock epicenter, (3) seismicity following episodes of increased fluid injection, (4) a geological structure that could be activated by changes in pore pressure and hydraulic connectivity, and (5) a low stress drop. Although the spatial and temporal association of earthquake activity with fluid injection is clear, wells with different characteristics (i.e., different depths to basement, injection volume, injection pressure, proximity to activated fault) make it difficult to pinpoint a single overriding causative factor. As evidenced by pockets of late seismicity at the extremities of the rupture zone, stress may have been transferred to the ends of the fault, implying some seismic‐hazard remains. The constraints on rupture geometry, the transmission of stress to adjacent fault segments, and the quasi‐static expansion of the aftershock zone are factors that may be useful for estimating maximum earthquake size and seismic hazard. Application to the Milan earthquake of an empirical formula that has been used to derive moment magnitude from felt area for historical earthquakes shows that the empirical M w may differ from the instrumentally determined M w by a few tenths of a magnitude unit.

DATA AND RESOURCES

Earthquake epicenters, magnitudes, and focal mechanisms are available at the U.S. Geological Survey’s (USGS) Advanced National Seismic System (ANSS) comprehensive catalog (http://earthquake.usgs.gov/earthquakes/search/, last accessed April 2016). Records of the wastewater injection wells are publicly available on http://maps.kgs.ku.edu/oilgas/index.cfm (last accessed January 2016) and injection volumes from the Kansas Corporation Commission (KCC; http://kcc.ks.gov, last accessed April 2016). USGS “Did You Feel It?” data for the Milan 12 November 2014 earthquake are available at http://earthquake.usgs.gov/earthquakes/eventpage/usc000swru#dyfi (last accessed April 2016). All other data used in this article came from published sources listed in the references. Some plots were made using the Generic Mapping Tools v.3.4.3 (www.soest.hawaii.edu/gmt, last accessed May 2016; Wessel and Smith, 1998). The unpublished manuscript by O. S. Boyd, D. E. McNamara, S. Hartzell, and G. Choy (2016) is “Comparison of stress drops for tectonic and induced earthquakes in North America,” submitted to Bull. Seismol. Soc. Am. The unpublished manuscript by W. L. Yeck, M. Weingarten, H. M. Benz, D. E. McNamara, R. B. Herrmann, J. Rubinstein, and P. S. Earle is “Far‐field pressurization induces Oklahoma’s third largest recorded earthquake, reactivating a large pre‐existing basement fault structure,” submitted to Geophys. Res. Lett.

ACKNOWLEDGMENTS

This research was supported by the U.S. Geological Survey’s (USGS) National Earthquake Hazards Reduction Program. The authors are grateful to Morgan Moschetti, Jill McCarthy, and an anonymous reviewer for helpful and constructive comments that greatly improved the article; to Robert Herrmann for providing his velocity model for the region; to Robert Williams, James Dewey, Chris Cramer, and Elizabeth Hearn for insightful discussions; to Lynn Watney of the Kansas Geological Survey for details of southern Kansas geology; and to Harley Benz for foresight and management in deploying additional seismographs in southern Kansas in response to increasing seismicity. We thank Ryan Hoffman and Lane Palmateer of the Conservation Division of the Kansas Corporation Commission (KCC) for their prompt response to data requests. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

© 2016 by the Seismological Society of America