Oil and natural gas production from shale formations

Transcription

1 SPECIAL Passive SECTION: seismic Pand a s s ive microseismic Part and microseismic Part 2 2 Source characteristics of seismicity associated with underground wastewater disposal: A case study from the 2008 Dallas-Fort Worth earthquake sequence DELAINE REITER, MARK LEIDIG, SEUNG-HOON YOO, and KEVIN MAYEDA, Weston Geophysical Oil and natural gas production from shale formations using hydraulic fracturing or hydrofracking techniques has grown rapidly since 2008 and represents a vital and growing domestic energy resource. The waste fluids from increased production (called brine ) are typically injected into deep underground disposal wells to avoid discharge into streams and other sensitive drinking water supplies. The Environmental Protection Agency estimates that approximately 400 million gallons of brine are currently disposed of on a daily basis in more than 28,000 wells in the United States, and the volume of injection continues to grow dramatically. This process has recently been implicated as the source of some potentially induced seismicity in gas-producing states such as Oklahoma, Texas, Arkansas, and Ohio. The potential seismic hazard and risk associated with the injection of waste fluids into disposal wells has become an issue of concern to the petroleum industry, the general public, and regulatory agencies. Seismologists investigating these phenomena must obtain high-quality estimates of hypocenters and source characteristics for shallow events near disposal wells so that proximate causes of induced seismicity can be assessed. In this study we show results from reanalyzing a small swarm of earthquakes (all less than M3.3) recorded on a local network near the Dallas-Fort Worth (DFW) airport in late fall 2008, using best-in-class detection, location, and source characterization techniques adapted from the nuclear test ban monitoring community for use in induced seismicity applications. At the end of October 2008, Southern Methodist University (SMU) deployed a small local seismic network, following the onset of unexpected seismicity near a saltwater disposal well (SWD) situated on DFW airport property. The network consisted of six PASSCAL Rapid Array Mobilization Program (RAMP) broadband stations, which were installed at sites surrounding the primary area of seismicity (Figure 1). Even though the length of this particular deployment was short (~53 days), hundreds of local earthquakes were recorded. Frohlich et al. (2011) provide an excellent summary analysis of the higher-magnitude events that were recorded during the deployment. To supplement their efforts, we focused on applying advanced detection and location analysis to the waveform archive using software we have developed in recent years for seismicity observed at reservoir and local spatial scales. In addition, we studied the source characteristics of the largest events using seismic body-wave coda methods. Our purpose in following the footsteps of Frohlich et al. was to determine if some of the most successful explosion monitoring techniques developed to date could be used to better characterize the nature of induced seismicity that is suspected to be associated with oil and gas production activities. Detection and location analysis As a first step, we performed detection on the continuous archive DFW waveform data using a technique that examines a recorded signal s energy content over a broad frequency range. We have found that this technique allows us to consistently detect smaller-magnitude events compared to traditional methods (e.g., short-term average to long-term average or STA/LTA). The new detector also has the advantage of not requiring a prior example event waveform or events with specific source mechanisms, as is the case for other detection methodologies such as crosscorrelation. Our application of this technique yielded 203 event detections between 11 November and 14 December 2008, all of which were visually confirmed by an analyst. The sequence of detections is shown in Figure 2a, which shows that a significant cluster of seismic activity occurred on 20 November, followed by a smaller one on November 28. Most detected events were observed on only two stations and thus could not be located accurately. For a location to be Figure 1. Google Earth map showing the six-station SMU network around the Dallas-Fort Worth airport. The saltwater disposal well (SWD) is shown as a green triangle with a black dashed circle at a 10-km radius centered on the well. Sonic logs from the Trigg well (blue star) were used to derive a velocity model The Leading Edge December 2012

2 Figure 2. (a) Detections as a function of date after the SMU local network began operation. Two significant clusters of seismicity occurred on 20 and 28 November. (b) Velocity model used to predict P and S traveltimes for hypocenter determination. The shaded depth intervals indicate the stratigraphic boundaries reported for Trigg Well No. 1 (the Barnett Formation bottoms at approximately 2.7 km depth). performed we required that an event have picks on at least three stations, and that at least two P picks were available to supplement the more abundant S picks. Filtering of the detection bulletin produced 34 locatable events, which is significantly more than the 11 events located by Frohlich et al., hereafter referred to as FROH2011. To compute the hypocenters we used the grid-search multiple-event location algorithm (GMEL; Rodi, 2006). GMEL is a global optimization location technique that computes event locations and traveltime corrections from a set of arrival time data obtained from multiple events, stations, and seismic phases. For the DFW earthquake sequence, we ran GMEL in multiple-event mode, which is a type of relative event location that solves jointly for the location parameters (hypocenters and origin times) of seismic events in a cluster and traveltime corrections at the stations recording the events. Other examples of related location techniques include master event, joint hypocenter determination (JHD), hypocentral decomposition (HDC), and double-difference methods, which have all been shown to improve both absolute locations and relative locations between events in clusters. However, in our opinion, GMEL has some advantages compared to the other methods in its approach to reweighting data, resolving parameter trade-offs, and computation of location uncertainty. We tested the effects of different velocity models on the locations using a similar set of models to those in FROH2011, before selecting the smoothed velocity model shown in Figure 2b, which was derived from the sonic logs of a nearby well (shown as Trigg in Figure 1). Our phase pick errors ranged from 10 ms to 0.64 s for the P arrivals and from 20 ms to more than 1 s for the S arrivals. We allowed for station/ phase corrections in the solutions of up to ±75 ms. Our bestfitting hypocenters are shown in map and projected depth views in Figure 3. The overall root-mean-square (rms) fit to the traveltime data for the 34 events ranged from 5 to 94 ms, and the station/phase corrections ranged from 75 to +30 ms for S arrivals and +3 to +75 ms for P arrivals at five stations. When we compare our locations to the 11 events reported in FROH2011, we find that our hypocenters are on average 500 m shallower than theirs and within a kilometer laterally. In addition, our rms data fits for these same events are equal to or slightly better than those in FROH2011. The additional 23 events that we located cluster within the same local area, and none of them locate deeper than approximately 4.4 km. It is important to note that a seismic event location is only as good as the specific inputs and solution constraints used to produce it. Furthermore, event locations must always be accompanied by detailed information on their uncertainties otherwise the results are incomplete and can be misleading. For the locations presented here, we calculated 90% confidence ellipses and depth intervals using well-known, linear analytical formulas. To preserve clarity in Figure 3, we did not plot the confidence parameters for every event. Instead we plotted only the best and worst confidence results, which were chosen based on the area of the epicenter ellipse multiplied by the depth confidence interval. Epicenter confidence ellipses ranged in area from 240 m 2 to 5.1 km 2, (mean of 1.5 km 2 ) and the depth confidence intervals ranged from 600 m to as high as 1.9 km (mean of 1.2 km). Our confidence estimates clearly indicate that the depths of most events are quite poorly constrained. This is because of a number of factors, such as the sparseness of the local network, the lack of multiple stations within one focal depth s length of the mean epicenter, and the inherent trade-off between earthquake depth and origin time. Further compounding the depth uncertainties is the lack of clear P picks on many stations, and December 2012 The Leading Edge 1455

3 Figure 4. Comparison between direct-wave (blue) and coda-wave (red) amplitude-ratio measurements for the SMU11 versus SMU09 event pair. Error bars on the measurements indicate the significantly reduced uncertainty when coda measurements are used. focused in the interval of wastewater injection, the events can also be acceptably located shallower or deeper by up to ~1.0 km, depending on the velocity model used to compute the solution and the depth of the starting location. Figure 3. Maps showing locations of the detected seismicity, with colors denoting events on 20 November (blue stars), 28 November (red squares) and other dates (yellow circles). (top) A map view centered on the SWD well with epicenters plotted with respect to the well. (bottom) The depths of the events projected along the east axis. In addition, we plotted the 90% confidence ellipses (top) and depth intervals (bottom) for the best- and worst-constrained events (star and circle outlined in black). large pick uncertainties on both P and S picks for the smallest events. Unfortunately, there is little that can be done to improve absolute depth information once a local network has been deployed. Later in the article, we suggest some basic network design guidelines that can help improve the location capabilities of local networks. At the least, the quality of the final hypocenter estimates should be assessed by analyzing the sensitivity of the location solutions to velocity model and starting locations. For the DFW case, extensive sensitivity tests on the biggest events reveal that the epicenters are generally well constrained and in good agreement with FROH2011. However, while our best-fitting solutions indicate event depths that are Estimated source parameters using seismic coda mea sure ments In addition to applying detailed location analysis to the SMU recordings of the late 2008 earthquake sequence, we also analyzed the source characteristics of a set of events using the scattered seismic wavefield, or so-called coda. Seismic body-wave coda, which is composed of waves scattered by small-scale heterogeneities in the subsurface, has some unique properties that make it ideal for sparse network monitoring at local distances (Mayeda et al., 2007). Scattered waves average over local 3D crustal heterogeneity, source radiation pattern and directivity, and they behave predictably within a particular region. As a consequence the envelopes of coda waves steadily decay as a function of time in a manner that is independent of source size, and amplitude measurements made from coda envelopes have been shown to be significantly more stable than those made with direct waves such as P and S. Seismologists use these amplitude measurements to determine important source characteristics, such as seismic moment (moment magnitude), corner frequency, and stress drop, all of which are important measures of the size and the potential for damaging energy in an earthquake. To demonstrate the coda-wave methodology with the DFW data, we analyzed S-coda amplitude measurements for the 11 largest events, which correspond to the SMU events presented in FROH2011. Coda amplitudes were measured in 42 narrow, overlapping frequency bands ranging from 1 to 32 Hz, and the results were averaged over the three sensor components. The amplitudes are contaminated by site and path effects, and we remove these by computing amplitude ratios at a station for pairs of events that are nearly colocated The Leading Edge December 2012

4 Pa s s i v e s e i s m i c a n d m i c r o s e i s m i c Pa rt 2 Figure 5. The coda-derived absolute source spectra for the 11 events in our source characteristics study of the DFW sequence. We used the moment tensors of the two largest events, SMU10 and SMU11, to determine the spectral shape in the lower-frequency bands (< 2 Hz). Classical source models were used to determine the shape of the higherfrequency bands. Multiple studies performed over the last several years have shown this method to be significantly more stable and precise than direct-wave spectral ratios (Mayeda et al., 2007; Mayeda and Malagnini, 2010). Figure 4 demonstrates this by showing the log-amplitude ratios as a function of frequency band for one event pair (SMU11 and SMU09), measured using direct (blue) and coda (red) waves. The standard deviation on the ratios is shown with error bars, which clearly indicate that the scatter on the coda measurements is about 3 5 times smaller compared to the direct measurements. Further comparisons of the spectral ratios for all pairs of events indicate that SMU04, SMU10, and SMU11 were the largest events in the data set, which is in agreement with FROH2011. Following the coda spectral-ratio analysis, we calculated the complete source mechanisms of the two largest events (SMU10 and SMU11), using a time-domain moment-tensor inversion method (Ford et al., 2009). This analysis provides calibration information that allows us to tie our coda-derived relative spectra to absolute source spectra parameters. To invert for the moment tensors, we computed synthetic waveforms at stations AFDAD, CPSTX, and LKGPV, using a 1D layered velocity model in a frequency band of Hz. Centroid depths were allowed to range from km depth in the inversion. The results indicate that the two events Figure 6. Epicenters of the largest events in the study, found using multiple-event location analysis, color-coded by stress drop (pink star = high stress drop; red star = lower stress drop). The SWD well is shown as a green triangle; the yellow line depicts the strike of the best-fitting focal mechanism from the two highest-magnitude events (SMU10 and SMU11) The Leading Edge December 2012

5 have almost identical moment magnitudes of roughly M w 2.5 (±0.1 unit uncertainty), with a preferred solution depth of 5.0 km, although shallower solutions were equally acceptable. All preferred solutions showed a strike-slip mechanism with a ~25 strike angle, which is consistent with the alignment of the epicentral locations. Finally, to obtain estimates of the seismic moment magnitudes (M w ), stress drop, and corner frequencies for the events, we converted our source spectra from relative to absolute measurements. To accomplish this we performed a source model-fitting process, in which we used the moment tensor results from SMU10 and SMU11 to determine the spectral shapes in the lower frequency bands (< 2 Hz). For the spectral shapes at higher frequencies, we fit the spectra using classical earthquake source models such as Brune (1970) and Boatwright (1980). In Figure 5 we display the absolute source spectra in a plot of seismic moment versus frequency. The results show that all events exhibit similar corner frequencies near 8 10 Hz, with the exception of SMU04 and SMU10, which have slightly lower corner frequencies. In Table 1 we list the moment magnitudes (M w ) for the 11 events, along with the best-fitting hypocentral parameters from our location analysis. We estimate our coda-based moment magnitudes to be accurate to 0.01 magnitude unit, based on the low variance of the coda amplitude measurements. Our magnitudes are on average 0.2 units higher than the local magnitudes (ML) computed by FROH2011. This systematic bias is expected and due to well-known differences in magnitude estimation (Nuttli and Herrmann, 1982). In Figure 6 we plotted the epicenters derived in our location analysis, color-coded according to higher stress-drop (pink) and lower stress-drop (red) events. We note that SMU11 exhibits higher stress drop relative to the other events, radiating a larger amount of energy at a higher frequency band (10 20 Hz) compared to lower bands. This seems to further confirm FROH2011 s identification of SMU11 as exhibiting behavior unlike the other events. The yellow line plotted along the best-fitting strike of ~25 from the moment tensor analysis is consistent with the mapped epicenters of the events. There is significant spatial and temporal variability of the stress drops, but this may be because of the fact that our study events occurred during a brief snapshot of a longer sequence of events. Finally, in Figure 7, we show seismic moment versus corner frequency for the DFW earthquakes (red stars near the top left corner) compared to results from other recent earthquake sequences. The results indicate that the DFW events exhibited similar stress drop (~1.0 MPa) to many other recent earthquake sequences in the continental United States. In other words, their source characteristics are similar to those Figure 7. Comparison of the 11 largest DFW earthquakes (red stars) to other recent earthquake sequences. Results show that the DFW events had similar or slightly lower stress drop (~1.0MPa) compared to many other recent earthquake sequences. Frohlich et al. (2011) Event ID Origin Date (mm/dd/yyyy) Origin Time Latitude ( N) Longitude ( E) Depth (km) SMU01 11/20/2008 9:58: SMU02 11/20/ :00: SMU03 11/20/ :00: SMU04 11/20/ :12: SMU05 11/20/ :14: SMU06 11/20/ :20: SMU07 11/20/ :32: SMU08 11/20/ :23: SMU09 11/20/ :26: SMU10 11/28/ :49: SMU11 12/01/ :26: Coda-Derived M w Table 1. Hypocentral and magnitude parameters (computed using coda-amplitude methods) for the 11 largest events in the DFW earthquake sequence. The events in the table correspond to those published in Frohlich et al. (2011); the coda magnitudes average ~0.2 unit higher than their local magnitudes. December 2012 The Leading Edge 1459

6 of other naturally occurring tectonic events. We emphasize that this should not be considered conclusive evidence of the tectonic nature of the DFW sequence, because the number of events analyzed and the time period covered by the local network were small. Conclusions In this article, we demonstrated the application of advanced seismic location and source characterization techniques to a subset of small earthquakes that have been observed near the DFW airport in the last several years. Our study supports the findings of FROH2011, which indicated that the events were located at shallow depths near the SWD well on the DFW property, with temporal behavior that correlated with injection activities at the site. Our detection algorithm found a significant number of additional small events that could be located near the DFW focus, using an energy criterion tuned to the data characteristics. Grid-search locations of a subset of the detections, using a realistic velocity model for the area, placed the events on a general north-by-northeast trend, which corresponds closely to a mapped fault in the general area. Although the depth uncertainties are high because of the seismic network configuration, we found that the earthquakes were predominantly within the reported injection zone. Our final event locations are slightly shallower overall than FROH2011, but they are in reasonable agreement given the overall lack of absolute depth information. Moment tensor inversion of two of the largest events also supported event depths of 5 km or less. We used coda-based spectral-ratio methods to determine the source parameters for the 11 largest DFW events recorded during the brief local network deployment. In general, we have found coda methods to be a useful tool for extracting source information from small, locally-recorded earthquakes. The amplitudes from coda are more stable compared to direct wave measurements, and the coda spectral-ratio methodology can be applied in situations where few stations record events with a reasonable signal-to-noise ratio. In this study, we found that high-frequency S-coda measurements produced stable spectral amplitude ratios that did not vary with respect to station, in sharp contrast to measurements made using direct S-waves. Our final results, which included estimates of moment magnitudes, stress drops and corner frequencies, indicated that the source characteristics of the DFW earthquakes were similar to those of other recent naturally occurring earthquakes in the continental United States. Further analysis of potentially induced earthquake swarms is needed to determine whether stress drop or other seismically derived source parameters can serve as reliable indicators of triggering mechanism. To ensure the success of these studies, local networks must be specifically designed to capture the temporal and spatial character of small, shallow seismic events. There are some simple guidelines to follow in designing these networks, including deployment prior to the onset of injection activity, placement of shallow buried or borehole three-component short-period sensors, and high sample-rate digitization of the signals. In addition, high-quality hypocenters can be achieved if several sensors are placed within a focal-depth length from possible epicenters, with additional concentric rings of sensors surrounding the expected surface focus of seismicity to a radius to several focal-depth lengths. References Boatwright, J., 1980, A spectral theory for circular seismic sources; simple estimates of source dimension, dynamic stress drop, and radiated seismic energy: Bulletin of the Seismological Society of America, 70, Brune, J., 1970, Tectonic stress and the spectra of seismic shear waves from earthquakes: Journal of Geophysical Research, 75, no. 26, , Ford, S. R., D. S. Dreger, and W. R. Walter, 2009, Identifying isotropic events using a regional moment tensor inversion: Journal of Geophysical Research, 114, B1, B01306, org/ /2008jb Frohlich, C., C. Hayward, B. Stump, and E. Potter, 2011, The Dallas- Fort Worth earthquake sequence: October 2008 through May 2009: Bulletin of the Seismological Society of America, 101, no. 1, , Mayeda, K., L. Malagnini, and W. R. Walter, 2007, A new spectral ratio method using narrow band coda envelopes: Evidence for nonself-similarity in the Hector Mine sequence: Geophysical Research Letters, 34, no. 11, L11303, org/ /2007gl Mayeda, K. and L. Malagnini, 2010, Source radiation invariant property of local and near-regional shear-wave coda: Application to source scaling for the M w 5.9 Wells, Nevada sequence: Geophysical Research Letters, 37, no. 7, L07306, org/ /2009gl Nuttli, O. W. and R. B. Herrrmann, 1982, Earthquake magnitude scales: ASCE, Journal of the Geotechnical Engineering Division, 108, no. 5, Rodi, W., 2006, Grid-search event location with non-gaussian errors: Physics of the Earth and Planetary Interiors, 158, no. 1, 55 66, Acknowledgments: The authors thank Brian Stump for access to the DFW local network data and Cliff Frohlich for helpful discussions regarding the previous data analysis of the DFW events. We also thank William Rodi for use of his GMEL algorithm. Corresponding author: delaine@westongeo.com 1460 The Leading Edge December 2012