Geophysical Research Letters

Characteristics of seismic and acoustic signals produced by calving, Bering Glacier, Alaska



[1] We recorded 126 calving and iceberg breakup events from the terminus of the Bering Glacier during five days in August 2008 using seismometers and three small-aperture arrays of infrasound sensors. The seismic signals were typically emergent, narrow-band, and lower-frequency, similar to records at other glaciers. The acoustic records were characterized by shorter-duration, higher-frequency signals with more impulsive onsets. We demonstrate that triangular infrasound arrays permit improved locations of calving events over seismic arrivals that rely on a relatively complicated, poorly known, velocity model. Twenty-six of 35 well-located events occurred on icebergs in Vitus Lake, rather than the glacier face. While our data do not permit a complete description of the source process, the distinctive frequency contents and durations in the seismic and infrasound data suggest that the two data types record different aspects of the same process.

1. Introduction

[2] The Bering Glacier, located in southeastern Alaska, is the largest glacier in North America by surface area at 3630 km2 [Beedle et al., 2008]. It terminates within Vitus Lake, and is subject to significant mass loss at its terminus due to calving. While the level of Vitus Lake is indirectly influenced by subdued tidal cycles through the Seal River outlet, the lake itself is protected from ocean waves. Calving has been studied elsewhere using passive seismology on a teleseismic scale [e.g., Ekström et al., 2003] and on a local scale [e.g., Qamar, 1988; O'Neel et al., 2007]. For example, at the Columbia Glacier in south-central Alaska, seismic records of calving have emergent onsets, a narrow frequency band (1–3 Hz), and long durations of 2–1000 s [e.g., O'Neel et al., 2007]. The emergent nature makes timing P-wave arrivals inaccurate and results in large uncertainties in event locations. The focus of the present study is to investigate the utility of infrasound sensors to record and locate calving events. The acoustic records of infrasound sensors deployed in small triangular arrays were used to estimate back-azimuths for the accurate location of calving events. We find that infrasound data are also useful for characterizing the calving source process, just as combined infrasound-seismic studies are important for modeling shallow volcanic processes.

2. Instrumentation

[3] In early June 2008, seven Mark Products L22 3-component geophones (2 Hz) equipped with RefTek 130 data acquisition systems and GPS time clocks, recording at a sample rate of 125 Hz, were deployed on small islands within Vitus Lake and on the adjacent mainland (Figure 1). No stations were deployed on the ice directly above the terminus due to the risk of unpredictable, large calving events displacing stations into the lake. In early August, three of the seismic stations were outfitted with infrasound arrays consisting of three Honeywell differential pressure sensor microphones (0.05 to 20 Hz), with identical digitizers, timing control, and sample rates as the seismic sensors. Each array was designed to be an azimuthally independent pointer, with one sensor coincident with the existing geophone and the other sensors forming an equilateral triangular array with sides of 30.5 m. The infrasound arrays and co-located seismic stations operated simultaneously for five days.

Figure 1.

Seismic stations (black dots), seismic and acoustic stations (yellow triangles), event locations (green, yellow, and orange dots and stars), and 99.7% confidence wedge intersections (red lines) for each hypocenter solution for thirty-five events (2007 Landsat image, MTRI). The event in Figure 2 is shown as the yellow star, and the event with spectrum indicated in Figure 4a is shown as the orange dot. The two events without confidence wedges (stars) did not have exact intersections, but the best solution was retained due to the high confidence in the approximate location. The yellow and orange dots and the yellow star had minimum seismic signal to noise ratios of three for all three acoustic co-located seismic stations. The 2008 ice edge is indicated with large blue circles where measured and approximated elsewhere with a dashed black line.

[4] The position of the glacier front during the recording period in 2008 was measured at discrete points using a GPS and rangefinder, with the continuous glacier edge interpolated as indicated in Figures 1 and 3. The overcast skies above the terminus during all Landsat passes in the summer of 2008 required the use of a 10 August 2007 Landsat image.

3. Seismic Record and Direct Visual Observations

[5] Typical seismic calving signals (Figures 2 and S1) had low-amplitude emergent onsets with several seconds of relatively high-frequency signal (1–30 Hz) followed by low-frequency (<5 Hz) narrow-band codas, with durations of 5–30 s. Typically coincident with part of this narrow-band signal, or shortly after its decay depending on the proximity of the event to the station, were several higher-frequency arrivals, apparently ground-coupled air-waves. Some seismic signals lacked the high-frequency arrivals and some signals lacked the long coda.

Figure 2.

Vertical seismic (S) and coincident acoustic (A) traces recorded at each site from a single calving event on 07 August 2008 at 08:06 UTC.

[6] Prior to the installation of the infrasound sensors, visual observations were made over a six-hour period in an effort to identify the sources of the different arrivals observed in the seismic traces from the island on which infrasound array 03 was deployed (Figure 1). The 70 observed events ranged from single blocks that broke free from the glacier face to one five-minute-long continuous calving sequence that produced many audible ruptures and large amplitude, long-lived seismic signals. Most calving events were too small and short-lived to be accurately located visually, although several long-duration event sequences resulted in large, freshly exposed blue ice faces hundreds of meters long. The direct observations made from the island confirmed that the higher-frequency seismic arrivals corresponded roughly to the sounds observed from calving.

4. Infrasound Sensor Record

[7] The acoustic signals recorded by the infrasound sensors clearly distinguished each of the higher frequency arrivals. The presence of a broad range of frequencies in the acoustic arrivals, with strong signals at 5 Hz and higher, suggests that these frequencies are present in the initial rupture, but are severely attenuated in the seismic record [O'Neel and Pfeffer, 2007]. While the seismic arrivals were always emergent, the infrasound sensors recorded impulsive arrivals coincident with the ground-coupled air-wave arrivals in the seismic records (Figure 2). Individual event sequences ranged from one discrete acoustic arrival lasting less than a second, to five discrete acoustic arrivals within one calving sequence, supporting the idea that one calving rupture may trigger subsequent ruptures [O'Neel and Pfeffer, 2007].

5. Data Processing

[8] We selected 43 events from the five-day recording period, based qualitatively on the amplitude of the air-wave arrivals and the presence of arrivals at all stations in all three arrays. For each array, we estimated signal coherence using the semblance method [Neidell and Taner, 1971] by performing a grid search over all azimuths (at half-degree increments) assuming an apparent velocity equal to the speed of sound in air. We varied the length of the time windows over which the analysis was performed from 0.5 s to 5.0 s, depending on the impulsiveness of the first acoustic arrivals. We then used a geometrical approach to identify the most likely source location based on the results from all three arrays at the intersection of the three back-azimuth solutions [e.g., Almendros et al., 2001].

[9] Formal uncertainties on event locations were computed using a bootstrap technique [e.g., Efron and Tibshirani, 1993]. We computed 1000 solutions for each event, by introducing quantities of variable noise to the original signal. We used a window of pre-event noise was that was randomly shifted and convolved with random Gaussian white noise [Sandvol and Hearn, 1991]. The maximum semblance value, corresponding to a best azimuth for each iteration, was computed and used to calculate the azimuthal mean and standard deviation at each array. By populating a spatial grid consisting of 10 m-square boxes, we were able to compute the region where the 99.7% confidence wedges (3 standard deviations) intersected for each event to determine the approximate event location and confidence region. Additionally, to increase the precision of the bootstrap statistics, 10 points were linearly interpolated between each time sample in the semblance calculations. Where significant noise at one or more arrays caused the confidence wedge to become too large to constrain a reasonably precise location, the poorly constrained solution was omitted. Two events, even though exhibiting very narrow 99.7% confidence wedges (ranging from 1.5° to 8°), had array setup errors that greatly exceeded those errors caused by noise in the data. In these two cases, the array confidence wedges did not intersect, so no confidence regions were assigned for those events. Errors of only ∼1.5 meters in the placement of one component of an array could have produced errors in the mean azimuth value of up to 3°.

[10] An example location is shown in Figure 3 for a calving event at 00:51 UTC 09 August 2008. We assume that the pressure wave at each infrasound array can be considered a plane wave, which is valid for nearly all events in this experiment due to the small apertures of the arrays with respect to the distances to the events. For the wavelengths (∼20 to 100 m) of interest, we assume a homogenous acoustic velocity and ignore wind, topographical effects, or velocity heterogeneity related to temperature/density differences. Also, although echoes of the initial rupture were audible as reflections off of the terminus wall, it was assumed that this energy was significantly reduced from the direct path energy and did not contribute to altering the best back-azimuth solution [Johnson et al., 2004]. We also assume that 1000 independent noise vectors sufficiently converge upon the true mean and standard deviation for each array [Sandvol and Hearn, 1991] (see Figure S2).

Figure 3.

An example of the resulting solution produced by the intersection of three 99.7% confidence wedges at each array.

[11] As a check on accuracy, we calculated the theoretical air-wave travel times to each array for each of the best-recorded events, assuming 333 m/s average speed of sound in air at 3.3°C. The arrival time differences between the arrays were then compared to the observed travel time differences, based on ground-coupled air-wave first arrival times, with RMS errors ranging .2 to 2.6 s. While a simpler method of determining event locations using seismic travel times alone may have worked on some of the larger-amplitude events, the variability of waveforms between arrays and low amplitude first arrivals prohibited precise and consistent picking for the majority of events from the seismic records alone. Furthermore, there is much greater uncertainty in the seismic velocity structure.

6. Results and Discussion

[12] Of the 43 hand-picked events, 35 were located using back-azimuths from all three arrays. The resulting 35 event locations reveal the presence of both terminus calving events and iceberg breakup events (Figure 1). The majority of the events, approximately 75%, are located in Vitus Lake away from the active calving front of the glacier, suggesting that failure of detached icebergs was the most probable source. Of these 35 events, 11 produced significant arrivals in the seismic records of stations 03, 06, and 07, other than the ground-coupled air-wave, with a minimum signal to noise ratio greater than three (Figure 1). Based on these results from the Bering Glacier, other studies which rely on seismic event counts alone to determine the frequency and number of glacier calving events might significantly overestimate these numbers by inadvertently including iceberg breakup events. Given the difficulty of measuring arrival times on the seismic records, using traditional location techniques result in large uncertainties.

[13] While this study was conducted primarily to test the utility of acoustic infrasound arrays for calving detection and characterization, we may draw some conclusions concerning source mechanisms. Many of the events produced both lower frequency seismic energy (1–5 Hz) and higher frequency infrasound energy (>5 Hz) on both the seismic and infrasound sensors, yet 24 of our 35 chosen events lacked significant seismic arrivals. We also observed seismic arrivals with no associated acoustic signals. Source mechanisms that have been suggested for the seismic-waveform details of calving and iceberg breakup events include cliff-face vibrations, single-force landslides, enlargement of a fluid-filled crack-tip, iceberg capsize, iceberg collision, rift opening, and iceberg grounding [e.g., Ekström et al., 2003; Müller et al., 2005; O'Neel and Pfeffer, 2007; MacAyeal et al., 2008, 2009].

[14] A number of studies have recognized similarity between volcano-related and glacier or ice-related earthquakes. Jónsdóttir et al. [2009] determined that 13,000 low-frequency earthquakes near Katla volcano since 2000, thought to be volcanic long-period earthquakes, were actually related to movement of the glacier that surrounds Katla. Métaxian et al. [2003] suggested low-frequency (<5 Hz) seismic events at Cotopaxi volcano were related to resonating water-filled cracks. O'Neel et al. [2007] conducted a seismic study of calving on nearby Columbia Glacier and found 1–3 Hz monochromatic seismic waveforms. They concluded that resonance and enlargement of a crack in the glacier could explain these data. These studies have ruled out site and path affects on the seismic signals so that the resonant character could be attributed to a source process.

[15] The events we observed at the Bering Glacier are consistent with the resonant crack model, where the air-wave suggests an initial broadband (impulsive) rupture of a small patch of ice, followed by attenuation of these frequencies into the resonant fluid, which becomes the dominant motion in the seismic record. This model is similar to a model of shallow crack resonance to explain seismic and infrasound records of long-period events that accompanied the recent eruption of Mount St. Helens [Waite et al., 2008; Matoza et al., 2009]. Given the similarity of the Bering Glacier data to the Columbia Glacier data described by O'Neel et al. [2007], we did not exhaustively investigate path and site effects on the data. However, the seismic records for any particular event have nearly coincident spectral peaks, which would be expected for a source characteristic (Figure 4a). Fluid-filled crack resonance is consistent with both the seismic and acoustic signals we observed, although our study lacks sufficient evidence to rule out all other potential mechanisms.

Figure 4.

(a) The individually normalized seismic spectra for a window around the signal prior to the arrival of the ground-coupled air-waves for an event occurring on 08 August at 02:44 UTC (orange dot in Figure 1), (b) the stacked acoustic spectra for all located events from every component of all three arrays, and (c) the stacked seismic spectra from the strongest eight events at each seismic station of all three arrays, with windows excluding any ground-coupled air-waves.

[16] Air-wave arrivals for the eight strongest acoustic and seismic events exhibit a peak at 11–14 Hz in the acoustic record and 2–4 Hz in the seismic record (Figures 4b and 4c, respectively). The short duration of acoustic arrivals suggests that only the broadband (impulsive) rupture couples well to the atmosphere, while the longer-lived seismic signal is dominated by resonance within a fluid-filled crack. The relatively high amplitude of the arrival of the ground-coupled air-wave in the seismic record compared to the time predicted for an arrival through the ground reveals that most of the energy partition of the initial rupture couples into the atmosphere, in turn suggesting that the failure point is shallow or at the ice surface, while the apparent resonance is deeper and most energy is partitioned into the ice (Figure 2).

[17] Calving events located along the western portion of the terminus generally provided coherent recorded seismic energy (Figure 1). This indicates that events along that face are either well-coupled to the ground, or perhaps that the total energy intensity is greatest at the western calving face of the glacier, allowing significant seismic energy in spite of poor coupling. Additionally, the isolated acoustic arrivals that were observed may be attributed to sub-aerial rupturing not leading to resonance. The presence or lack of seismic energy may also simply be due to a wide range of magnitudes, where only larger events overcome the seismic noise level.

7. Conclusions

[18] We accurately located 35 events related to glacier and iceberg calving or breakup over a period of five days in 2008. These events were located, using triangular infrasound sensor arrays, along the terminus of the Bering Glacier and within Vitus Lake. The acoustic records demonstrate numerous source locations that are located away from the terminus affiliated with iceberg breakup. While the rupture of isolated icebergs does not shed any additional light on the nature of calving as related to ice edge dynamics, we have shown that icebergs are a significant source of seismic and acoustic energy that may be falsely attributed to the calving front without accurate event locations. The acoustic arrivals are distinct from the low-frequency signals recorded seismically and provide improved locations over those obtained from seismic data alone, primarily due to the known and much slower acoustic velocity field and lack of picking bias assumed with the seismic first arrivals. We have demonstrated the utility of infrasound sensor arrays for locating calving events without the need for direct visual observations.


[19] The authors would like to thank the following people for assistance in making this project a success: Scott Guyer and Nathan Rathbun (Bureau of Land Management); Bob Shuchman, Christopher Roussi, and Liza Jenkins (Michigan Tech Research Institute); John Gierke and Kevin Endsley (Michigan Tech); and the anonymous reviewers. The seismic instruments were provided by the Incorporated Research Institutions for Seismology (IRIS) through the PASSCAL Instrument Center at New Mexico Tech. Data collected will be available through the IRIS Data Management Center. The facilities of the IRIS Consortium are supported by the National Science Foundation under Cooperative Agreement EAR-0552316, the NSF Office of Polar Programs, and the DOE National Nuclear Security Administration. We would also like to thank the Michigan Tech Remote Sensing Institute, the Michigan Tech Office of the Vice President for Research, and the Michigan Tech Fund for their financial support.