Seismic Monitoring of Egs Tests at the Coso Geothermal Area, California, Using Accurate Meq Locations and Full Moment Tensors

We studied high-resolution relative locations and full moment tensors of microearthquakes (MEQs) occurring before, during and following Enhanced Geothermal Systems (EGS) experiments in two wells at the Coso geothermal area, California. The objective was to map new fractures, determine the mode and sense of failure, and characterize the stress cycle associated with injection. New software developed for this work combines waveform crosscorrelation measurement of arrival times with relative relocation methods, and assesses confidence regions for moment tensors derived using linearprogramming methods. For moment tensor determination we also developed a convenient Graphical User Interface (GUI), to streamline the work. We used data from the U.S. Navy’s permanent network of three-component digital borehole seismometers and from 14 portable three-component digital instruments. The latter supplemented the permanent network during injection experiments in well 34A-9 in 2004 and well 34-9RD2 in 2005. In the experiment in well 34A-9, the co-injection earthquakes were more numerous, smaller, more explosive and had more horizontal motion, compared with the pre-injection earthquakes. In the experiment in well 34-9RD2 the relocated hypocenters reveal a well-defined planar structure, 700 m long and 600 m high in the depth range 0.8 to 1.4 km below sea level, striking N 20 ̊ E and dipping at 75 ̊ to the WNW. The moment tensors show that it corresponds to a mode I (opening) crack. For both wells, the perturbed stress state near the bottom of the well persisted for at least two months following the injection. HIGH-RESOLUTION HYPOCENTERS MEQs are traditionally located independently and individually by inverting measured seismic-wave arrival times picked either automatically or by an operator. This method, even with the benefit of an excellent seismic wave-speed model (Foulger and Julian, 2004) and highly skilled arrival-time picking, yields hypocentral locations with errors of many tens of meters relative to one another. Relative location accuracy can be improved with the use of modern programs that locate large clusters of MEQs simultaneously (Waldhauser and Ellsworth, 2000). The absolute location of the cluster is little improved by this technique, but the error in the locations of the individual MEQs relative to others in the cluster is much reduced. We applied this method, via the computer program hypocc (Julian, in preparation-b), to MEQs induced in EGS injection experiments at the Coso geothermal area. We improved the locations still further by picking seismic-wave arrival times automatically using waveform cross-correlation. For this we developed the program toonpics (Julian, in preparation-c). We repicked the MEQs using this method and relatively relocated them again. The resulting locations thus benefited from both refinements and provided the ultimate relative location accuracy currently available. SEISMIC MOMENT TENSORS Many earthquakes from geothermal areas have nondouble-couple mechanisms (Julian et al., 1998; Miller et al., 1998). These are thought to result from the involvement of fluids in the source process. Simply put, earthquake failure in geothermal areas involves not only shear movement on faults, but also opening and closure of cavities, probably as a result of the flow of abundant, hot, high-pressure fluids (Julian and Foulger, 2004). Earthquake source mechanisms are traditionally obtained by plotting P-phase polarity data, measured from seismograms, on a map of the focal sphere and analyzing the distribution of compressions and dilations. A “double-couple” (DC) interpretation, appropriate for pure shear faulting, implies that the compressions and dilatations can be separated by orthogonal planes (great circles on the map). If the source is not assumed to be solely shear, then the lines separating the compressional and dilatational fields are not generally great circles, but ellipses. Under some circumstances they may be small circles but this need not be the case. Non-DC source mechanisms are specified by moment tensors, which involve two more free parameters than DCs, and require more advanced data processing for their determination. We use compressional and shear seismic-wave amplitudes in addition to polarity data, to obtain general descriptions of the motions at the source. Amplitudes alone are subject to severe bias by wavepropagation effects such as focusing and attenuation. To counteract this problem we invert seismic-wave amplitude ratios, along with polarities, using a linearprogramming algorithm (Julian, 1986; Julian and Foulger, 1996). This combination of data and inversion approach can determine moment tensors for MEQs at least three magnitudes smaller than other methods can. A useful way to display moment tensors is on the orientation-independent “source-type plot” (Hudson et al., 1989). All DCs lie at the center of this plot. Mechanisms that plot above the center line involve explosive (volume-increase) components and those that plot below have implosive components. The horizontal position on the plot depends on the detailed type of shear involved. We recently developed a technique to assess uncertainties in derived moment tensors (Julian, 1986; Julian and Foulger, 1996). The linearprogramming method finds the moment tensor that best fits a set of observed seismic-wave polarities and amplitude ratios, in the sense of minimizing the L1 norm (the sum of absolute values) of the misfits to the observations (“data residuals”). We extended the method to determine what changes to this best-fit solution can be made while keeping the goodness of fit within a specified range. We formulate this task itself as a linear-programming problem, and solve it efficiently by standard methods. To use the new method, the user specifies a number of “objective functions”, linear combinations of the moment-tensor components that are to be maximized or minimized subject to keeping the L1 norm of the residuals within certain bounds that the user also specifies. Examples of such objective functions include the volume change, the amount of extension or compression in specified directions, and the similarity to particular chosen mechanisms. Because computing moment tensors is a complex and time-consuming processes, we also developed a convenient Graphics User Interface (GUI) to speed the work (Julian, in preparation-a). Earlier software tools amounted to a bundle of command-line-driven developer scripts, and determining each moment tensor might take several hours. Using the new GUI, determining a moment tensor typically takes ~ 20 minutes. The GUI cannot speed manual picking of the polarities and amplitudes, which must still be done by hand, but elimination of outliers and inversion of the data is greatly streamlined.