Characteristics of seismic radiation during the 1994 Bolivian earthquake and implications for rupture mechanisms

1. Introduction

[2] The seismological and physical characteristics of a deep-focus earthquake have long been the topic of debate among many investigators. Several mechanisms have been proposed to explain the occurrence of deep earthquakes [Green and Burnley, 1989; Kirby et al., 1991; Silver et al., 1995]. To test the validity of different earthquake mechanisms at depths where brittle failure should not normally occur, observations from large, deep events are critical. The Bolivian earthquake of 9 June 1994 [location = 13.86° S, 67.54°W; depth = 637 km; M_w = 8.3 after NEIC] remains the largest deep-focus earthquake ever instrumentally recorded, and it, therefore, provides the best opportunity to date for the study of these mechanisms. This earthquake was well recorded by the global digital broad-band seismic network (GDSN), and those data enable the source parameters and rupture area to be constrained with a high degree of confidence even when different approaches are used [Kikuchi and Kanamori, 1994; Beck et al., 1995; Chen, 1995; Antolik et al., 1996]. Aside from this, some temporal broadband seismic stations at regional epicentral distances provided further information with regard to rupture characteristics [Silver et al., 1995]. Studies of this event have in fact led to the conclusion that the source mechanism included a horizontal nodal plane and that a slip occurred on it. The estimated static stress drop was quite large while rupture velocity was relatively slow [Kikuchi and Kanamori, 1994; Silver et al., 1995]. Furthermore, it has been estimated that there was a low ratio of radiated seismic energy to total strain energy. However, the seismic data from these observations do not provide sufficient spatial resolution with which to analyse complex short-period waveforms and discuss the detailed properties and implications of different rupture mechanisms.

[3] Many dense regional short-period seismic networks have recorded this event at teleseismic distances. Unfortunately, the records from those networks were mostly over-scale and presented an obstacle to any further analysis of the detailed rupture properties. In this study, the on-scale recorded PKP waves, which penetrated the Earth core and arrived at the other side of the earth surface, were recorded by a dense regional short-period network in Taiwan (Figure 1a). The aim of this study is to analyse the short-period array waveforms to clarify the rupture physics of large, deep earthquakes. The analysis of this study indicates that, compared with the others, the two major subevents suppressed high frequency seismic energy. This phenomenon can be explained by the elastohytrodynamic lubrication of fault [Brodsky and Kanamori, 2001], and implications here are support for the hypothesis of “frictional melting during deep-focus earthquakes” [Kanamori et al., 1998; Kanamori and Heaton, 2000].

2. Data

[4] Designed to monitor local earthquakes, the Central Weather Bureau Seismic Network (CWBSN) is a regional seismic network composed of 75 digitally telemetered seismic stations entirely covering Taiwan and its offshore islands (Figure 1b). Each station is equipped with a set of matching 3-component short-period velocity sensors (Teledyne Geotech S-13) with peak amplitude response near one Hz. The small spacing (about 25 km) between the CWBSN stations allows for array studies of the teleseismic short-period waveforms. The network data are simultaneously recorded in the Institute of Earth Sciences, Academia Sinica (IESAS) which is set in continuous mode to report major teleseismic events and to study the deep structures of the Earth [Huang et al., 1996]. As a rule, therefore, large, deep events at a distance range of greater than 140° from South America are well recorded in the IESAS. Illustrating this are the data of the 1994 Bolivian earthquake from an epicentral distance of near 167°. Figure 1a displays the direct seismic raypaths which penetrated the Earth. The array seismograms of the 1994 Bolivian earthquake show that the PKP waves from two different branches were well separated (Figure 2). It is found that the PKP waves radiated its seismic energy with take-off angles near the vertical and its first arrived waves penetrated the Earth's inner core and arrived at each station with an incident angle near the vertical (Figure 1a). The vertical component seismograms of the PKP(DF) wave can be considered as far-field source time functions of a deep-focus earthquake because their receiver side converted seismic waves are limited to the horizontal components, and the source side surface-reflected phases arrived long enough after the direct PKP(DF) wave and did not contaminate this phase. However, although the same case for the phase PKP(AB) in both source and receiver side, its waveforms are more complex. The AB branch has near-grazing incident to the Earth's outer core and its waveforms are sensitive to the lateral heterogeneity at the base of the mantle. Herein, only PKP(DF) waves are employed to analyze the source time functions.

[5] Besides the short-period seismic sensors, seven downhole wide-band seismic instruments (Teledyne-Geotech 541000) are included in the CWBSN. These instruments have the flat amplitude response from the long period limit of 1 to 10 seconds. The observed PKP waveforms are shown in Figure 3a. It is found that these records include lower frequency content than do those of the short-period records (Figure 2). Because the spacing (about 80 km) between the wide-band seismic stations is larger than that between the short-period stations, the records do not indicate high-frequency coherence. However, these wide-band observations may provide, at least in part, additional information with regard to the frequency gap between the well studied GDSN broadband data and the high spatial density CWBSN short-period array observations. This can be examined by comparing the stacked seismic beam of the wide-band records with the filtered broadband velocity seismogram from station TATO of the GDSN (Figure 1b). The stacking process is described in more detail in the next section. Filtered by the same frequency band, both traces show consistent waveforms (Figure 3b). The original broadband waveform of TATO and the short-period array stacked beam are also shown. Because the data of TATO and other GDSN stations have been previously analysed in an attempt to retrieve the spatial rupture characteristics of this event [Antolik et al., 1996], the consistency between the filtered TATO and CWBSN wide-band waveforms provide a valuable constraint. Actually, the CWBSN array data can be employed to examine the high frequency rupture characteristics of major asperities which have previously remained unresolved when only global seismic network data were used.

3. Analysis and Results

[6] In this study, to recover the high-frequency source information of the 1994 Bolivian earthquake, seismograms from the regional seismic array (CWBSN) were processed according to a filtering and stacking technique to enhance seismic signals. As shown in Figure 1, although the range of the CWBSN spans more than three hundred kilometers, the ray paths of the PKP(DF) waves from the source to the CWBSN stations are similar. This suggests that those seismograms should have similar arrivals. However, crustal structure, which is different at each station, produces a tail of scattered waves after the arrival of a direct PKP(DF) wave, which therefore distorts and obscures the later part of each arrival (Figure 2). To eliminate these coda generated near the receivers and the background noise, selected clean seismograms were aligned by the eye with an easily identified feature in the initial part of the direct arrival. Then, those seismograms were normalized to a peak amplitude of unity and then added together. As the scattered energy was inconsistent between stations, in the stacked trace it was diminished by a factor of the square root of N where N is the number of traces stacked [Vidale and Houston, 1993]. Thus, the stacked waveforms were enhanced by eliminating the scattered coda. An example of this is found in Figure 3a which shows the aligned wide-band velocity traces of the CWBSN and in Figure 3b which displays their stacked beam. It is evident that the scattered coda have been effectively suppressed.

[7] When the same procedure is applied to the short-period velocity traces of the CWBSN (Figure 2), the same degree of effectiveness is well demonstrated. The bottom trace of Figure 3b shows the sum of 56 selected clean short-period traces of the CWBSN for the 1994 Bolivian earthquake. From the stacked beam of the short-period traces, the initial and ending of the source rupture are identified and represented as IR and ER in this trace. The rupture duration was estimated as 50 seconds in this study which is consistent with measurements using global data [Kikuchi and Kanamori, 1994; Silver et al., 1995]. In this stacked seismogram, the low noise before the initial rupture and its abrupt termination on the seismic trace confirms the effectiveness of the procedure in eliminating the background noise and the scattered coda. The stacked waveform represents the component of the PKP arrival that is common to all stations and it can be considered as the high-frequency source time function of this earthquake.

[8] To determine the temporal pattern of seismic radiation, an envelope of stacked beam was constructed. The envelope of this time series was determined by the square of the stacked velocity beam. The envelope is always positive and essentially passes through the peak of the stack. It represents the temporal distribution of seismic energy radiation from the earthquake source [Kanamori et al., 1993]. One limitation to this short-period stacked beam is that the radiated energy is band-limited. No information concerning long-period seismic energy radiation of this event is included. In this study, the broadband seismogram of TATO, which is a GDSN station located inside the CWBSN array (Figure 1b), is employed to extend the observed frequency band of the source rupture. In order to compare the short-period seismic radiation with lower frequency traces, the envelope of the short-period beam was smoothed by a moving average with a one-second window as shown in Figure 4 (middle trace). Based on the determined rupture time history, four asperities are identified from these array observations. As in the case of the short-period seismic beam, the broadband seismograms of TATO were processed following the same enveloping and smoothing procedures and is also shown in Figure 4 (bottom trace). To enhance the difference of seismic energy radiation during the rupture processes from high and low frequency bands, the amplitude ratio of the short-period to broadband radiated seismic energy was estimated and shown in the top trace of Figure 4. Just as in Figure 4, the integral of the squared velocities may provide another point of reference from which to estimate radiated seismic energy [Kanamori et al., 1993]. This represents the accumulation of seismic radiated energy from an earthquake with respect to time (Figures 5a and 5b). Such time histories indicate radiated seismic energy with regard to its filtered signal bands. This provides invaluable background information for a discussion on the temporal behaviour of the source rupture.

4. Discussion and Conclusions

[9] From both global and regional observations, the fault slip distribution of the 1994 deep Bolivian earthquake has been resolved as several major asperities [Kikuchi and Kanamori, 1994; Silver et al., 1995; Beck et al., 1995; Chen, 1995; Antolik et al., 1996]. With this spatial information of source rupture considered, the newly determined temporal frequency variations in seismic radiated energy (Figures 4 and 5) may provide some clues as to its detailed rupture processes as well as some implications in terms of rupture mechanisms. In Figure 4, the time history of the amplitude ratio of the high to low frequency energy presented the suppression of high frequency seismic energy during the rupture quantitatively. Furthermore, because the amplitude ratio of the high to low frequency waves is unaffected by the focal mechanism variation, it indicate that the high frequency suppression reported in this study do not affected by the possible differences in mechanisms between the subevents. In Figures 4 and 5b, the time histories of the radiated seismic rupture energy are marked by the initial rupture times of the four major asperities of this event as determined by Kikuchi and Kanamori [1994] using global data. It is found that these estimated rupture times correlate well with the onset of major seismic energy radiation. Both asperities E1 and E2 radiated more low frequency seismic energy than did asperities E3 and E4. Here, according to the spatial distribution of the rupture as determined by Kikuchi and Kanamori [1994], both asperies E3 and E4 are located near the end boundaries of the rupture plane, and asperity E2 is located in the central part of the rupture area, while E2 is identified as the largest asperity with a seismic moment release which is about double that of the other three. It is noted that the initial rupture occurred about 10 seconds before the major rupture and that it released limited seismic moment compared with the other major asperities. However, the frequency content of the initial rupture was resolved as being rich in its short-period band.

[10] After examination of the spatial and temporal frequency variations of seismic waves during the 1994 deep Bolivian earthquake, its short-period seismic radiation features can be reconstructed. In the initial stage of rupture, this event radiated a small amount of seismic moment which was richer in high frequency than the succeeding major subevents. The rupture was continuous for about 10 seconds, after which the major asperities of this event were sequentially triggered. During the rupture, the radiated seismic energy was major over a long period. Finally, when the rupture was near its ending stage, the propagation extended to its boundaries and large amounts of high frequency seismic energy were radiated. These interesting observations may point to some important properties of the source rupture. Few, if any, observations resembling these have been reported elsewhere. However, using the short-period seismic array observations, stress release, rupture duration and the envelope of seismic radiation from deep earthquakes have previously been examined and discussed for the rupture mechanisms of deep earthquakes [Houston and Williams, 1991; Vidale and Houston, 1993; Houston and Vidale, 1994]. In this study, the observed rupture behaviour of this suppressed high frequency seismic radiation from large asperities can be explained by the lubrication of frictional slip during fault rupture [Brodsky and Kanamori, 2001]. They have suggested that the fault zone fluid pressure is significantly increased in a large slip region during a rupture. The increased pressure may widen the fault zone gap and reduce the number of asperity collisions. Following this mechanism, the decrease in high frequency radiation can be predicted as a significant feature of this fluid lubrication process [Brodsky and Kanamori, 2001].

[11] The mechanism of a deep earthquake which is a friction-induced melting was first proposed by Kanamori et al. [1998]. They proposed that the melting during the 1994 Bolivian earthquake, the same as most deep earthquakes, formed a thin fluid layer that reduced friction in much and promoted fault slip. On the basis of their hypothesis, high frequency seismic waves radiated from a melted fault plan should be reduced due to fluid lubrication. Results of this study can be considered as partial evidence in support of frictional melting during a deep-focus earthquake. However, other hypotheses cannot be fully ruled out because the observations in this study can also be explained in part by several other hypotheses for the occurrence of deep earthquakes. The first model has been transformational faulting, whereby a phase transition triggers the rupture [Green and Burnley, 1989]. In that model, the atoms of the mantle rock, known as olivine, are somehow rearranged forming a more condensed mineral called spinel which transformations may occur suddenly and can be accompanied by shear dislocations. It is possible that the spinel are formed in the shape of fine grids which can then serve as a solid lubricant inside the fault slip zone. The same lubrication concept can be applied to another model which considered a deep earthquake to be a reactivation of a preexisting weak zone [Sugimura and Uyeda, 1967]. In that model, the premise is that the fluid in the subducted slab is captured by a preexisting normal fault created at shallow depth. Accordingly, the fluid can be preserved within hydrous minerals during plate subduction. When the fluid-filled fault is reactive in deep, the pore pressure of the fluid may serve as a liquid lubricant and thereby reduce its slip friction. Some evidence has shown that this may be the most plausible explanation for an intermediate depth earthquake occurrence [Jiao et al., 2000].

[12] In the present study, observations from a regional short-period seismic array uncovered new characteristics of seismic energy radiation during the 1994 deep Bolivian earthquake and have brought new implications as to its rupture mechanisms. However, due to limited knowledge of earthquake physics in the deep earth, the full cause of this event must still remain under debate.