ANALYZING TAMSEIS DATA FOR SEISMIC EVENTS OF HIGH TEMPORAL REGULARITY AND LARGE MAGNITUDE (MW = 1.8) BENEATH DAVID GLACIER, AND FOR LONG-PERIOD PLATE-MARGIN EVENTS
Highly regular seismicity associated with the flow of David Glacier in the Transantarctic Mountains of Antarctica has been detected and analyzed. We used data from the TAMSEIS network which consisted of 42 broadband seismometers deployed from November 2000 through December 2003 ([13, 14, 26]). The seismic events recur at a regular time interval of approximately 20 min. Travel times suggest that the events originated from the base of the David Glacier (approximately 1.8 km deep in this area). are consistent with low-angle reverse faulting P-wave first motions. A fault strike of 185 degrees was calculated, which is normal to the flow of David Glacier. The events are likely caused by an asperity beneath David Glacier that regularly released stress, that accumulated as David Glacier flowed over the asperity. The regularity of the events is due to the constant and homogenous driving stress of the overlying ice as well as the weakness of the bed. Models of earthquake source regions that include a few asperities within a weak active fault are though to display similar behavior. The TAMSEIS network has also been used to analyze long period seismic events occurring around the Antarctic plate. Twenty one events have been located that previously had not been catalogued in the National
Earthquake Information Centers catalog. Mostly originating from the plate boundaries.
Table of Contents
The ice sheets and glaciers of Antarctica are of interest due to contributions they could make to global sea level changes (). Ice flows from the thick inland ice sheets to the coast in ice streams and outlet glaciers, resisted by drag along the beds and sides (e.g., Raymond et al., 2001)(). Basal drag may be localized at “sticky spots” at the base of the ice (), which have been located using passive seismic techniques (). Understanding the contribution of sticky spots to basal drag is key in characterizing overall flow of the ice sheets ().
Sticky spots may originate in various ways, related to basal topography, lack of lubricating water or till, or local strengthening of otherwise lubricating till (). The resistance to flow that sticky spots produce can often dominate the entire basal shear stress for a region (e.g., [11, 27]). Passive seismic evidence was presented by Danesi et al.,
(2007) indicating the occurrence of a sticky spot beneath David Glacier, Transantarctic Mountains (TAM), which potentially has notable effects on the glacier flow.
Our study presents additional data on sticky-spot behavior beneath David Glacier, collected by the TAMSEIS project (see below) during field seasons from 2000-2003, before the 2003-2004 deployment of Danesi et al. (2007). We find larger sticky-spot influence during that earlier time interval, and thus evidence for time-evolution of the sticky-spot behavior.
David Glacier flows through the Transantarctic Mountain Range (TAM) into the Ross Sea (going afloat as the Drygalski Ice Tongue). About one-quarter of East Antarcticas grounded ice is drained through the nine primary outlet glaciers flowing through the TAM. Of these, David Glacier drains approximately 4% of the total ice of East Antarctica() from an East Antarctic catchment area of approximately 212,000 km2
(). Velocities generally increase downflow, reaching an average speeds of 529 − 714 m a−1 near the grounding line ([8, 9]). Measured ice thickness varies greatly with position, from 250-1800 m () from the terminus to the head of the glacier.
The primary objective of TAMSEIS was to image crustal structure beneath the TAM, and therefor the seismometers were deployed so as to maximize detection of teleseismic events (). A secondary goal of TAMSEIS was to study the local seismic activity in Antarctica from tectonic and glacial sources.
The TAMSEIS array consisted of 42 seismometers, which were strategically placed throughout the TAM, the West Antarctica Rift System, and East Antarctica (Figure 1). The data were recorded from November 2000 through December 2003, and collected from three sub-arrays of broadband seismometers ([26, 13, 14]). The first sub-array was a 16-station linear deployment oriented East-West across the TAM, with approximately 20 km spacing between stations. The second sub-array was oriented NNE-SSW and consisted of 17 stations near Terra Nova Bay with approximately 80 km spacing. The third sub-array contained 9 stations which were oriented North-South and ran along the Ross Island coast, and there was no typical spacing with these stations (Figure 1). This sub-array also contained one permanent Global Seismice Network (GSN) station (VNDA) that was previously deployed. Eight stations were deployed in December 2000, and the other 33 were deployed in late 2001. All stations used Reftek 72A-08 Digitizers. Most used Guralp sensors (either CMG-3T, or CMG-40T), but Strekeisen STS-2 sensors were also used. The data acquisition system was set to record at a rate of 40 samples per second (.025 sec interval), corresponding to a Nyquist frequency of 20 Hz.
Not all of the solar-powered seismic stations operated simultaneously during the summer months, and during the winter months the stations operated on a triggered scheme as opposed to the continuous scheme used during the summer. Local events were not large enough to trigger recording during the winter months, but during the summer months local seismicity was recorded in addition to the teleseismic events. Continuously recorded data allowed for observation of the local background seismicity originating from regions of Antarctica not locatable by the GSN. The maximum number of stations were deployed and actively recording from Nov 2002 to Jan 2003, so we focus on this two and a half month interval.
We conducted an automated scan of the continuos data. Events of different types were observed, including teleseisms and local events. Interestingly, the local events repeated (Figure 2) with an inter-event time of 19.9 minutes (1010 seconds) as shown in Figure 3, with an event having similar or identical waveform. As discussed below, these events originate in one region beneath David Glacier. We have analyzed some events of other types (see Ch. 2 of Zoet, 2009), but this contribution focuses on the repeating events. Most events that originated beneath David Glacier were detected on 12 different stations, but depending on local noise levels that number increased to as many as 18 stations for some events.
The data processing and analysis of each event consisted of the following steps: (1) band-pass filtering to emphasize the local events; (2) identification of local events through application of a Short Term Average / Long Term Average (STA / LTA) auto picker, (3) estimation of source location and source time for each event, (4) estimation of focal mechanism and refined location from a 30-event stack, (4) estimation of seismic moment and stress drop from the 30-event stack. Finally, we place these results in a glaciological context by relating the event characteristics to the flow characteristics of David Glacier.
We analyzed 70 days of data (from 11-01-2002 to 01-15-2003), which had the greatest station coverage. Limited inspection of data from other times reveals that the events were not limited to these 70 days, but full analysis of the full data would be difficult because of varying coverage and occasionally poor signal-to-noise ratio (SNR).
Initial inspection revealed the existence of the repeated events from David Glacier, and showed that they had strong frequency content in the 5-6 Hz range. We thus bandpass filtered the data to emphasize this feature. We started with the station with best SNR, and applied an STA/LTA automated picker, which identified 5077 of the repeated local events. Events were detected throughout the entire record except for brief intervals during major storms when the SNR was too low to allow detection. A portion of the results were manually reviewed for quality control purposes.
A total of 5077 events were detected at station E018, and 4538 of the total events were detected on more than 3 stations allowing an epicenter location to be calculated, using a standard grid-search location scheme with a velocity model consisting of ice over rock. The source was assumed to be at the base of the ice, using an ice thickness determined by radio-echo sounding of the area (), following the method of Lee et. al (1976). The p-wave arrival times at the various stations were measured in one of two ways. For events with sufficiently high SNR, arrival times were picked from the seismograms (Figure 4). These events proved to be in the same location within the considerable uncertainties, and to exhibit essentially identical waveforms and almost metronomic timing, indicating that these are repeating earthquakes. We thus increased the size of the database to include events with lower SNR by cross-correlating the record of each low-SNR event with a chosen high-SNR event and picking the correlation peak, as shown in Figure 6. Locations were then estimated for these low-SNR events as well as for the high-SNR events.
All of the locations determined were near David glacier (within 50 km, at a distance of 200-400 km from the stations in the array), and the differences in locations of the events were not significant. To resolve the location further, 30 events were chosen that exhibited large SNR on all stations and uniformly spanned the 70-day observational period. Experiments with stacking showed that SNR increased up to ≈ 30-fold stacking with little change thereafter, so 30-fold provides a little redundancy without greatly increasing the effort. All arrivals for these 30 events were picked manually rather than through the autopicker. The greater precision in arrival timing led to a tighter cluster of positions, with the epicenters of all of these events falling within a 10-km-radius circle centered 75.4S,160.6E (Figure 7). This location is in the upglacier section of David Glacier. The closest measurement of ice flow speed is ≈ 510 ma−1 (), the ice thickness is 1.8 km based on radio-echo sounding (), and the surface slope is 2 degrees. This location is also within 10 km of the site where Danesi et al. (2007) located most of their
To test our assumption that the events originate at the bed, we first stacked the 30 chosen events. We then fixed the epicenter at the center of the circle enclosing the events, and varied depth through a five-layer model (4 layers in the ice resting on a constant-velocity half-space representing the upper crust). A family of theoretical traveltime curves was generated, for sources at depth Z varying from the surface to 10 km below the surface in 0.1 km depth steps. Minimizing the square of the residuals between the measured and modeled arrival times allowed us to estimate the source depth. We used a crustal-half space velocity of 6.5 km s−1 found in Stein 2003 for upper crustal rocks. We found that a depth of z = 2km results in the smallest error. This is consistent with a source at the base of the ice (zice = 1.8 km).
To determine the source focal mechanism, first motions from our 30-fold stack were plotted on a lower-hemisphere projection around the source at the appropriate azimuths and take-off angles. These data indicate that the events were produced by a low-angle thrust fault (dip, δ = 3◦±3◦, strike φ = 2◦±3◦, and slip λ = 90◦±3◦) (Figure
8). The slip direction is perpendicular with the mean flow direction of David Glacier. We thus suggest that the most plausible source mechanism is rupture at the base of the glacier due to the gravitational driving stress of the ice.
We estimate the source seismic moment from the P-wave spectrum using the method of Brune (). The spectrum is fitted by a function that includes a constant low-frequency (Ω0) level and an ω−2 decay at frequencies higher than a corner frequency fc. The seismic moment M0, and hence the moment-magnitude, are given by:
where Up is the angular-frequency. We determine Up(x,ω → 0) by calculating Ω0 through fitting the Fourier transform. FP is the radiation pattern correction, ρs =
1087kg m−3 is the density at the source, ρd = 917kg m−3 is the density at the detector (ice), Vps = 3870 m s−1 is the P-wave velocity at the source, Vpd = 3800 m s−1 is the P-wave velocity at the detector, and ` is the distance from source to the receiver, which varied between 100-300 km. The density and velocity of the source were calculated assuming that the event ruptured between the ice and the bed, but that the ice contained 10% rock (we discuss this assumption below). The density and the velocity at the source were inferred from bed morphology observations in ([9, 6]). We obtain a moment of M0 = 6.5 × 1011 N m, equivalent to a moment magnitude of 1.8. Figure 9 shows the weak dependence of these results on rock percentages between 0 − 20%. The corner frequency is fc = 0.7Hz, and Ω0 = 5.0×10−9 msec from Figure 10. Another technique used to determine magnitude is to compare the duration of events(). Using the empirical formula derived for this technique a Mw = 1.8 event should have a duration on the order of 15 seconds which is within a factor of 2 of the duration of the David
Assuming a simple rupture the source rupture area is related to the corner frequency fc through the rupture speed (effectively, interference between energy radiated from the center and from the margins of the rupture will reduce the total energy at high
where β = 660 m s−1 is the shear wave speed at the fault. The rupture velocity is found to be 500 m s−1 by Weins et al. and β is calculated to be 4/3 of the rupture velocity(). fc = 0.7Hz is the corner frequency. We use a rupture velocity appropriate for glaciers, vr = 500ms−1 for a close distance from the nucleation point (¡ 20 km) () . The fault radius for a vr = 500 m s−1 then is 330 ± 15m.
We next compare the force supported on the fault plane to the local gravitational driving stress, and find that the fault plane represents an asperity that is retarding the motion of a much larger region of David Glacier. Our model for the base of the ice is an isolated asperity at which the earthquakes are occurring, surrounded by a large area A of relatively low friction whence the glacial driving stress is concentrated on the asperity. We also assumed that during a slip, between 10%-1% of the total imposed stress was released and can be measured as stress drop ∆σ (). Thus τxz ×A×psd = ∆σ ×πR2. Where psd is a value representing the partial stress drop on the entire fault, which can vary between 0.1 (10%) and 0.01 (1%) of the force being released. The stress drop (∆σ) was calculated following (). The driving stress for ice flow, hence the basal shear stress τxz if the driving stress is supported locally on the bed, is given by
τxz = Fρghsin(α) = 2.5 × 105Pa
where F = 0.45 is a shape correction factor for drag from the glacier sides (), g ≈ 9.8m2s−1 is gravity, ρ ≈ 910kgm−3 is the density of ice, h ≈ 1800m is the ice thickness, and α ≈ 2◦ is the surface slope for the region.
To find the actual shear stress on the fault plane, we note that the stress drop in an earthquake can be estimated from the seismic moment M0 = 6.5×1011N m following
where A = pi ∗ (R)2 where R is the radius of the fault for, and a is a geometric factor often taken to be 0.1 (which we follow here), although it could be as large as 0.3.
A faulting event is not expected to relieve all or even most of the stress on a fault plane. Following Brune (1970), the fraction relieved is estimated as 10%, which would yield a stress on the fault plane
σ = 3.3 × 105 to 3.3 × 106 Pa for vr = 500 m s−1, or ≈ 1 to 10 times the local
driving stress. Thus, the driving stress from area that is the order of magnitude or, an order of magnitude larger than the 0.34 km2 fault plane is being supported on the fault
Using the equation for seismic moment M0 = µ × D × A. Where A is the area of the fault, µ = 3.46 × 109 N m−2 is the rigidity of ice and µ = 1.0 × 109 N m−2 for the bed. The amount of displacement over the span of a year on the fault if the rupture occurs in the ice is ≈ 20 m, and if occurs in the bed it is ≈ 65m.
The timing and size of the events occurring at David Glacier give us insight to the mechanics of slip beneath glaciers and ice streams. These events are large for glacial events, and unique in their repeatability. The fact that the events are large leaves one to conclude that there was a degree of rock on rock breakage occurring between the bed and the rocks within the ice. It would be difficult to generate events of this magnitude by simply breaking ice. This rock on rock interaction could result in increased rates of erosion beneath glaciers of this type. The seismic energy could go into breaking large sections of rocks there by allowing mechanisms for increased rates of erosion.
The focal mechanism representing the events leads us to infer that the events are the results of the flow of the glacier and not a pre-existing fault structure in the area. Glacially generated events often have a strike which is normal or nearly normal to the flow of the glacier(). The events originating at David Glacier have do have a focal mechanism strike which is normal to the flow of David Glacier, which flow primarily from West to East. In addition to the azimuth of the strike, the dip of glacially generated events is generally a shallow dipping thrust fault which has been resolved for this focal mechanism. This provides confidence that these events are originating at the base of the ice and are a result of the flow of David Glacier.
These events appear to rupture from the same location at each occurrence. The waveforms had a high correlation to one another, indicating a similar mechanism produced all events. This would suggest that the events are likely the result of a single asperity that ruptures repeatedly, as theorized by Danesi et al. (2007), and not a series of sticky spots that ruptured independently. The location and depth of the events recorded by TAMSEIS match the findings of Danesi et. al.(2007) within the limits of our error windows.
The mean magnitude of the events was estimated at Mw = 1.8 which is large for glacially generated events which are not long period events of Ekstrom et. al (2006). The duration of these events was less than 1 minute, and to produce Mw ≈ 2 events, a relatively short duration and small fault radius, it is theorized that the seismicity was the result of the bedrock faulting with debris laden ice. We hypothesize that these events are not simply the result of clean ice slipping over the bed, but that some concentration (≈ 10%) of rock was entrained in the overriding ice thereby increasing the magnitude upon slip. Mean magnitude calculated for the events was within the same order of magnitude that Danesi et. al. (2007 reported (Ml = 1.3).
The corner frequency (≈ 1 Hz) for the events was much smaller than would be expected for a Mw ≈ 2 event (≈ 10 Hz). This was the result of the rupture velocity in the ice quakes (500ms−1) differing by an order of magnitude of rupture velocities(Vs = 2.7 km s−1) in crustal rocks (). A larger rupture velocity in equation 5 would yield an unreasonable fault radius (larger than the width of the glacier). Instead, rupture velocities of the order proposed by Weins et. al (2008). provides a fault radius more in concert with those to be expected from a Mw ≈ 2 ice-based event. Providing further validation that when ruptures occur involving ice, a modified rupture velocity should be used in to attain results that more accurately represent the real values.
David Glacier is one of the largest outlet glaciers in the TAM and unlike most of the ice streams of Antarctica has a high surface slope (2◦). This in combination with the thick ice at this location (1500 m) results in the increased basal shear stresses in comparison with many other locations in Antarctica. The increased basal shear stress in combination with the rock on rock breaking at the base of the glacier contribute to the larger events occurring.
Partial stress drop theories propose that when a fault ruptures, only a fraction
(between 1 and 10%) of the stress is released. Assuming for this fault that between 1-10% of the stress was released upon rupture, the amount of area needed to load the fault was approximately A = 1 km2 to A = 10 km2. The calculated area upstream of the rupture location was approximately 40 km2. This exceeds the area needed to load the fault meaning this size rupture would be possible with just loading from the ice above the asperity. An additional amount of the loading would be included by the pull from downstream section as well as the push from up stream. This would imply that a considerable portion of the area upstream of the rupture location was being held back by the side wall drag, and the force at the asperity.
David Glacier provides an interesting look at glacial bed mechanics through repeating events. It incorporates stick slip faulting which has been seen at other locations in Antarctica(), but these events are orders of magnitude larger than those events normally observed. It also incorporates aspects of the much longer period, larger events that exhibit stick slip () on a shorter time scale.
The high regularity of the events at David Glacier provide an excellent real world representation of stick slip mechanics at laboratory time scales. The processes that are believed to be acting at the base of David Glacier may provide some insight into the cause of high erosion rates beneath the glacier. The seismic energy released beneath this glacier could be the result of rocks cracking producing higher rates of erosion.
The total displacement resulting from these events is on the order of ≈ 100myr−1. This is the same order of magnitude as the total amount of slip of the glacier for the entire year. It is not inconceivable that almost all of the basal motion during this time period was taken up in this stick slip motion, and that there was little basal motion from basal sliding. In the study by Danesi et al. (2007) that takes place on David Glacier during the 2003-2004 field season (1 year later) repeat events of this nature (20 minute spacing) are not observed. This indicates that sometime between the TAMSEIS study and the Danesi et al. (2007) study there was a shift in the subglacial mechanics from a stick slip regime to a mostly sliding regime. They estimate that the amount of slip from the events they detected would be 16m, an order of magnitude lower.
On possible explanation for the shift in regimes is that during the time TAMSEIS was being conducted a relatively debris rich section of ice was passing over the asperity increasing the friction to the point of driving the system into a stick slip dominated regime. Later when Danesi et al. (2007) conducted their experiment that debris rich section of ice had passed by the asperity lowering the friction level enough that the glacier returned to a basal sliding dominated regime.
The glacier in total exerts a constant force, which goes into overcoming the frictional drag of the sticky spot, the viscous drag of the freely sliding parts, and any accelerations if the velocity is changing during a slip event. The force to effect the sliding on the sticky spot is kinetic friction, which is always less than static friction. During stick-slip behavior under the TAMSEIS experiment, the force on, the sticky spot varies from low-kinetic to high-static to low-kinetic, whereas under Danesi et al. (2007) the force on the sticky spot is always low-kinetic, so Danesi et al. (2007) should have seen more force going into viscous, which means higher velocity. An interesting implication for this regime change would be that under the Danesi et. al. (2007) time period there would be an increase overall glacial velocity from the velocity during the stick slip dominated regime. This would change the flux leaving David Glacier, there by altering David Glacier’s effect on the global sea level.
Further work for this study will be conducted on the time evolution of these events. A detailed look at the waveform evolution though time will be conducted to see if in fact there are small changes in the waveforms resulting from small changes at the source. The inter-evernt spacing will be analyzed and compared to the magnitudes of each event to see if there is a correlation between the two. The GSN station VNDA data will be analyzed in an attempt to locate other periods of regular stick slip events from David Glacier, as well as studying the onset and end of the regular 20 min time frame.
Seismic activity in Antarctica has not previously been found in large amounts with large magnitudes ([21, 20]). In a 70 day period of observations using the TAMSEIS array ≈ 5000 events were located with an average size of Mw = 1.8. The events originated from the base of David Glacier, likely from an asperity that allows stress to build and then release, as the stress reaches a threshold point. The fault radius was found to be 200m and the loading zone, assuming a 1% drop in stress during the events, there is ample volume of ice to produce this size event. There was a change in subglacial mechanism from stick slip to sliding which would theoretically result in a change in velocity. The magnitude and repeatability of these events gives a mechanism for a possible erosional increase beneath glaciers.
Table 1.1. Layers of thickness and velocities for ice model
|Layer||Depth Range (m)||Thickness (m)||Velocity (|
Fig. 1.1. TAMSEIS array
Figure 1 The TAMSEIS array consists of two main segments, 1; 16-station linear array oriented East-West across the TAM, with approximately 20 km spacing between stations 2; oriented NNE-SSW and consisted of 17 stations near Terra Nova Bay with approximately 80 km spacing. Each station is represented by a red triangle in the figure and a red squared represents the location of the seismic events at David Glacier. The location of the MODIS image relative to the rest of Antarctica is outlined in the box in the upper right corner representation of Antarctica.
Fig. 1.2. 150 Minute sample period
Figure 2 Representative sample of 150-minute period from the TAMSEIS array. This seismogram was recorded at station E018 and shows the arrival of 8 events. There is similar spacing between them and the amplitudes are also relatively uniform.
Fig. 1.3. Spacing Histogram
Figure 3 The mean time between events is 20 min and this plot is the distribution of the spacing for all the events located in the 69 days of the TAMSEIS dataset.
Fig. 1.4. Wavefront
Figure 4 Propagation of a wavefront generated at David Glacier as it crosses a section of the TAMSEIS array. The seismometers in this sample section are those that cross the TAM from east to west. The P-wave is recorded at the beginning of the time period and marked by the flags, but is just above the background noise level, followed by the S-waves, which are well above the background noise level. The duration of the wavefront is approximately 30 seconds, which is typical from the events generated from the David Glacier region.
Fig. 1.5. Power Spectrum
Figure 5 The power spectrum plot representing the spectra of the emissions from David Glacier. The majority of the power is in the 3-5 Hz. range but there is also a high amount of power in the 5-6Hz range that allows for distinction between events originating from David Glacier and other regions.
Fig. 1.6. Cross-Correlation
Figure 6 Cross-correlation of a sample wavelet with a sample section of seismogram indicating high degree of correlation between events which are detected originating from David Glacier.
Fig. 1.7. David Glacier Map
Figure 7 This is a map of the David Glacier region. The green dot is the central location of the events that were detected by this study, and the large circle is region in which the precisely located event all fell within. The orange dot is the location of the Danesi et al. repeat event.
Fig. 1.8. Focal Mechanism
Figure 8 The focal mechanism that was resolved for the events originating from David Glacier was a shallow dipping reverse fault. The dip is 3◦± 3◦ degrees with a strike of 2◦ and a rake of 90◦. The strike is approximately normal to flow of David Glacier, which is to be expected if the events are to originate from the glacier.
Fig. 1.9. Source Augmentation
Figure 9 The source is being varied between 0 and 20% rock and displaying the effects on the source magnitude moment and seismic moment variation as the source changes.
Fig. 1.10. Source Spectrum with Brune fit
Figure 10 The thick line is a (Fast Fourier Transform) fft of a random section of noise that existed at this station. The thin line is a stack of ffts for the P-wave arrival at this station. The blue section where the P-wave fft has higher values than the fft of the noise is the range in which there was a good S/N ratio for the data. The red line is best fit approximation using the Brune model which falls off at a w−2 after a corner frequency (.7 Hz)for the range where there was a high S/N ratio.
ANALYZING TAMSEIS DATA FOR SEISMIC EVENTS OF HIGH TEMPORAL REGULARITY AND LARGE MAGNITUDE (MW = 1.8) BENEATH DAVID GLACIER, AND FOR LONG-PERIOD PLATE-MARGIN EVENTS