3. Earthquae locations Data from permanent seismic networs located the 2001 Enola mainshoc within the crustal volume of the 1982-84 swarm. This initiated interest and concern that another 1982-lie sequence might develop in the Enola swarm area. Shortly after the mainshoc too place we deployed a portable seismic networ in the swarm area. The data collected by the portable seismic networ are used in this chapter to locate the earthquaes of 2001. Accurate earthquae locations and relative relocation procedures usually reveal smallscale features of the earthquae distribution. I propose that the swarm did not tae place on a fault/s but rather formed a dispersed cloud of seismicity. Out of ~2,500 earthquaes recorded on the continuously operating broadband seismograph (GUR) I selected the 100 largest earthquaes (based on the S phase recorded on the broadband N-S component. Numerous earthquaes were too small to trigger the accelerographs. The accelerographs were often redeployed; there were periods with only three usable instruments recording an event with the rest being too far from the swarm to be triggered (see Appendix D for a time-line of the networ development). Some accelerographs also experienced low levels of signal/noise ratio. I believe that this data set of the 100 largest earthquaes (about half of which were recorded on at least 4 stations) is representative and reveals basic properties of the 2001 earthquae sequence. 19
In this chapter I present and discuss three earthquae location techniques. The first one focuses on locating earthquaes based on P and S phase arrivals. The second describes a double-difference method (HYPODD) using operator piced phase arrivals (later referred to as: catalog data). The third technique explores a cross-correlation technique with HYPODD to better constrain relative earthquae locations. 20
3.1 Locating Earthquaes with HYPOELLIPSE It is a common approach to preliminarily locate earthquaes using one out of many earthquae location software pacages that implement Geiger s method (Geiger, 1910). I use XPICK (Robinson, 1990), a computer program to pic P and S arrival times and HYPOELLIPSE (Lahr, 1999) to locate earthquaes. The velocity model is derived from an analysis of reflection lines across the Enola swarm region (Chiu et al. 1984). The Vp/Vs ratio is assumed to be 1.73. Table 1. The velocity model derived from reflection lines across the Enola swarm region (Chiu et al. 1984). P-wave velocity (m/s) Depth (m) 3.73 0 4.73 1.22 5.68 2.89 6.13 6.23 6.6 13 8.18 40 Figure 8 shows earthquae locations using HYPOELLIPSE and the given (Table 1) velocity model. RMS values range from as little as 0.01 s up to 0.11 s. Horizontal uncertainties are as large as half a ilometer. Three earthquaes, denoted as asteriss, (also shown in Figure 8) are located using regional permanent seismic stations. I tried to improve the location of these three earthquaes by calibrating the networs (two second largest aftershocs, not shown here, 21
were recorded by both regional and the portable networ) and HYPODD. Unfortunately, only three regional permanent stations recorded the emergent principle phases of two aftershocs to be used in calibration. Figure 8. 2001 Enola earthquaes (blac dots), located using HYPOELLIPSE (SW cluster mared with the red circle). The mainshoc, denoted as a red asteris and the other two asteriss are earthquaes located using local and regional permanent stations (ellipses depict formal uncertainties). Blue triangles are portable seismic stations (up to 6 operational at a time). Except for GUR (broadband) station all the other stations represent K2 accelerographs. Some of the K2 sites were co-located with analog MEQ-800 seismographs. Grid spacing is roughly 2 m. 22
Even though I re-piced the arrival times of the principle phases of seismic waves for the total of 5 earthquaes recorded by the regional networ I could not identify more than two P arrival times for the reference earthquaes clearly. Because of this I could not obtain significantly different solutions than those presented in Figure 8. The magnitude of the mainshoc (asteris far to the left) was estimated to be M ~ 4.4. Magnitudes for the other two earthquaes are estimated to be M ~ 2.7 (CERI catalog). All three events too place within 25 hours, starting with the mainshoc on May 4 th, 2001 right after midnight, local time. These earthquaes, recorded only by the regional networ, seem to be located away from the strict Enola zone depicted using the portable networ data. Just by looing at the epicenters all together, I thin this is an unusual situation. However, the inter-distances between these three largest earthquaes (choosing convenient points on the formal error ellipses) are on the scale of the inter-distances of the swarm earthquaes recorded by the portable networ. Even though I cannot prove it at this point I thin that the largest earthquaes too place within the strict swarm area defined by the portable networ data. The HYPOELLIPSE locations give a very general picture of the 2001 sequence but it is still possible to observe a few characteristics of the swarm: The earthquaes occupy a volume of roughly 8 m 3 cube. Based on Figure 8, there seem to be two clusters of seismicity, one more condensed to the SW (mared with the red circle, Figure 8) and one more diffuse to the NE. 23
Figure 9. Two cross sections showing the HYPOELLIPSE locations. The major number of earthquaes seem to be confined within a 2 m thic layer. The boundary between Precambrian basement and overlying Paleozoic sedimentary rocs is roughly at the 5 m depth. Two hand-drawn red lines show possible faulting based solely on the NS crosssection. Two cross sections through the swarm area (Figure 9) reveal the depth profile of the 2001 sequence. The seismicity appears confined within a 2 m thic layer, close to the depth of the Precambrian-Paleozoic boundary. The sections do not reveal a strong, recognizable pattern but one thing should be pointed out: the hypocentral separations are very small, sometimes on the order 10s of meters. The uncertainties of these individual locations are on the order of half a ilometer. It is a challenge to hypothesize about the swarm s exact spatial structure. On the other hand it justifies the attempt to better constrain the locations using waveform cross-correlation as well as the double-difference 24
technique. Two hand-drawn red lines, solely based on the NS cross-section, propose possible faulting (Figure 9). Figure 10. Two location cubes reveal an arch-looing formation of the more concentrated earthquae cluster (far west) The dashed red line indicates a possible aseismic gap separating the major clusters (see text). I also examined these earthquae locations using the 3D plotting capabilities of MatLab. I present two snap shots of such plots (Figure 10). The figure reveals that these two clusters are at slightly different depths. The SW one appears to be closer to the surface. Due to uncertainties in the single-event locations, it is not conclusive whether the earthquaes form a single fault plane. I cannot observe, any sort of a lineation that would support a single fault hypothesis based on various 3-D sections, similar to those in Figure 10. 25
The earthquaes seem to assume a purely vertical but diffused orientation, when viewed looing SW or NE (not shown). At greater depths, 5 to 5.5 m, the seismicity becomes even more diffuse. The clusters lose the aseismic gap that separates them closer to the surface. Two hand-drawn red lines, shown in Figure 9, are not liely to be mapped into planes in the 3-D sections (Figure 10). HYPOELLIPSE implements Geiger s method for earthquae location. This method is by definition very insensitive to lateral velocity changes. It merely minimizes the errors in origin time, epicenter and depth for a horizontally layered velocity model. Pujol (1989) exercised this problem by assuming cylindrical velocity anomalies (velocities reduced by 40 %) of 1 m radius at each hypocenter, restricted to the third velocity layer for the 1987 Enola earthquaes. He obtained relatively small and unchanged RMS residuals, concluding that local velocity changes may go undetected when earthquaes are located individually. The single-event earthquae locations establish a general picture of the 2001 sequence locations. The location of the 2001 and 1982 sequence (comparing my HYPOELLIPSE locations with locations reported by Chiu et al., (1984) for the 1982 earthquaes closely coincide. The 2001 sequence too place in a 8 m 3 crustal volume showing an approximate NE-SW trend. This volume lies within a slightly larger volume containing the 1982 hypocenters Uncertainties in single-event locations, on the other hand, do not allow unequivocal interpretation of the swarm s spatial structure at hypocentral depths. To investigate this problem further I will explore relative location techniques in the following chapters. 26
3.2. The Catalog Data, Earthquae Locations and HYPODD 3.2.1. Method I use the HYPODD pacage that implements the double-difference (DD) algorithm of Waldhauser and Ellsworth (2000) with the goal of better constraining the spatial distribution of the 2001 Enola sequence. The DD algorithm requires a small hypocentral separation between two earthquaes so that the ray paths between the sources and a particular station are similar (Figure 11). Therefore, the difference in travel times for two events at a station is directly proportional to the hypocentral offset. If T i is the arrival time for an earthquae i at a station, t i is the origin time, p is the ray parameter and ds is a path element, we can write: i i T = t + p ds (4) i First we linearize (4) using a Taylor series expansion. Now we have to solve (5) where travel time residuals depend linearly on perturbations m i = ( x i, y i, z i, t i ), i t m m i = r i (5) 27
where i obs cal i r ( t t ) = and t obs, t cal are the observed and theoretical travel times, respectively. We can now define the double-difference equation between two events i and j and a station. dr ) ij i j obs i j cal = ( t t ) ( t t (6) When the hypocentral separation between a pair of earthquaes is small enough, the ray parameters at a station are assumed identical (Figure 11). Combining (5) and (6) we get: i j t i t j m m m m = r i r j = dr ij (7) m are adjustments in the hypocentral parameters to improve the model fit to the data. It is further possible to combine (7) for all the earthquae pairs for a station, and even further, for all the stations, in a system of linear equations (Waldhauser and Ellsworth, 2000). 28
Figure 11. Illustration of the double-difference method. Two earthquae are denoted as i and j, dt ij x is the travel time difference between the earthquaes i and j as observed at the station x. The ray parameter p is with respect to the stations. The arrows, x, are the relocation vectors for (7). (Modified from Waldhauser and Ellsworth, 2000) 29
3.2.2. Results The earthquae location procedure using HYPODD with operator-piced phase arrival times were subdivided into the following tass: Location procedure using catalog data using only P arrivals. Location procedure using catalog data using only S arrivals. Location procedure using catalog data using both P and S arrivals. Locations using HYPODD were initially performed using a priori assigned weights. The weights are assigned during the phase picing procedure and based on the waveform complexity of the phase arrival. The weights range from 0 for bad (hard to estimate the precise phase arrival) to 1 for good pics. In order to be able to compare locations based on different phase data sets I devised a set of default parameters among the many adjustable ones in HYPODD. These defaults include: setting the minimum number of observations (MINOBS) to 1, maximum number of observations (MAXOBS) to 11 (total number of stations) and event clustering to OFF. Again, using this raw mode I can easily compare earthquae location procedure results for the different data sets, see steps a, b, c and corresponding Figures 12, 13, 14. Figure 12 shows the result using only P arrivals. Formal location errors are on the order of meters (based on the HYPODD summary file). It should be noted that there are 90 HYPOELLIPSE individually located earthquaes (blac dots) but not all of them are relocated by the DD method. 30
Figure 12. 2001 Enola seismicity cloud. The blue triangle is station ENO. Blac dots are earthquaes located using HYPOELLIPSE with both P and S arrivals. Red dots are earthquaes located using only piced P arrivals with HYPODD. It appears now that the earthquaes are confined to a smaller area. Error bars of HYPODD locations are on the order of meters. This is not a surprise because the unfavorable station distribution with the respect to some earthquae locations may generate relocation uncertainties which will further lead to rejecting these earthquaes from the relocating procedure (remember that the hypocentral separation between an earthquae pair must be small compared to their distance to a station). Another fact is that a significant number of earthquaes were recorded on only three stations due to frequent station site changes. 31
Figure 13. The locations using S arrivals. The color scheme is the same as in Figure 12. The HYPODD location errors for S arrivals only are observable this time (on the order of 10 s of meters). Comparing Figure 12 and Figure 13 (DD algorithm with S arrivals) it seems as if P and S pics belong to different quality categories. Using only S arrivals it appears as if the locations migrate southeast for about a half a ilometer. Location uncertainties for S waves are greater by an order of magnitude than uncertainties of P locations. S phase picing procedures always have a larger window of error. Rotating (however, this cannot be done with the current version of XPICK v4.3) the horizontal components would help identifying at least the first arrival of the SV wave. This is not so 32
straightforward when a preliminary location and/or phase picing procedure is formed since the bac-azimuth and azimuth cannot be calculated without the earthquae location. Figure 14 reveals the results of the DD algorithm for both P and S arrivals. Location uncertainties for joint (P & S) DD relocations are several times the cumulative Figure 14. Earthquae locations using both P and S arrivals. Red crosses are earthquae locations due to the HYPODD procedure for both P and S arrivals where errors reach ~ 400 meters (~ grid size. This is also a scale for the HYPOELLIPSE location uncertainties). Even though both P and S arrivals have user assigned a priori weights during phase picing a refinement of the double-difference method is possible. This is conducted through HYPODD location parameters (re-weighting functions) adjustment (see text for discussion). 33
uncertainties from either P or S relocations alone (Figure 12 and Figure 13). This discrepancy might be reconciled by introducing data re-weighting functions in the HYPODD iteration steps. This either rejects bad S arrival data or diminishes their contribution to some arbitrary level, based on the user specified parameters. This would basically leave the P arrival data in control of the earthquae locations. The migration of locations using S waves (Figure 13) with the respect to the locations using P waves seems to be greater than the location uncertainties using S waves. S waves, because of their lower seismic velocity (as opposed to P waves) permit higher location resolution. Unfortunately this property is diminished by errors in S phase picing. One way of solving this problem is to pic S arrivals more precisely using rotated seismograms based on preliminary earthquae locations. Apparent waveform similarity of many events of the 2001 sequence maes a waveform cross-correlation technique an unavoidable procedure to obtain highly precise relative P and S arrival times. This exercise is the subject of the next section. 34