Volume 123, Issue 4 p. 3178-3196
Research Article
Free Access

Multiyear Shallow Conduit Changes Observed With Lava Lake Eruption Seismograms at Erebus Volcano, Antarctica

H. A. Knox

H. A. Knox

Geophysics and Atmospheric Sciences, Sandia National Laboratories, Albuquerque, NM, USA

Search for more papers by this author
J. A. Chaput

J. A. Chaput

Department of Mathematics, Colorado State University, Fort Collins, CO, USA

Search for more papers by this author
R. C. Aster

Corresponding Author

R. C. Aster

Department of Geosciences and Warner College of Natural Resources, Colorado State University, Fort Collins, CO, USA

Correspondence to: R. C. Aster,

[email protected]

Search for more papers by this author
P. R. Kyle

P. R. Kyle

Department of Earth and Environmental Science, New Mexico Institute of Mining and Technology, Socorro, NM, USA

Search for more papers by this author
First published: 14 February 2018
Citations: 11

Abstract

We analyze near-repeating broadband seismograms from nearly 3,000 eruptions (2003–2011) from the Erebus volcano lava lake to investigate temporal changes in the shallow eruptive dynamics of an open conduit volcano. Cross-correlation analysis reveals progressive variable time lags between correlation-aligned short period (SP) and very long period (VLP) seismogram components ranging over approximately ±1 s and evolving over weeks to months. Lava lake eruptions both excite an SP explosion seismic signal and gravitationally unload the conduit. After a delay of several seconds, this unloading excites a posteruptive, minute-long VLP seismic signal that persist for several minutes until the lava lake is refilled. VLP-SP lag variations are consistent across multiple seismic stations and are independent of eruption size, spectral characteristics, eruption frequency, and lava lake morphology changes. Lag changes are interpreted in terms of variable communication time between eruptions and the subsequent elastic and gravitational response of a tomographically imaged near-summit magma storage and VLP source region several hundred meters below the lava lake. Tomographic and VLP moment tensor studies, combined with modeling, suggest that the elastodynamic communication time between the distinct SP and VLP source regions is mediated by conduit-guided Stoneley waves within the uppermost magma-filled conduit system that are sensitive to small shallow conduit geometry changes. Conduit geometry changes may be driven by internal or external processes, including conduit wall melting and refreezing, repeated eruptive slug erosion, and deformation from inner crater inflation or collapse.

Key Points

  • We note progressive changes in the timing of short period (SP) and very long period (VLP) signals from lava lake eruptions at Mount Erebus
  • SP and VLP have distinct source processes and locations, but are linked by elastodynamic communication through the shallow conduit
  • Our preferred interpretation is that the timing changes are indicative of changes in the uppermost conduit geometry

Plain Language Summary

We observed 8 years of eruptions from the lava lake of Mount Erebus, Antarctica, with near-summit seismographs. Analysis of these generally similar seismograms reveals progressive changes that reflect changes within the near-summit volcano magma system and conduit. We model these changes to show how subtle evolution in the conduit geometry can explain the observed observations. This methodology shows how near-repeating seismic signals can be used to monitor progressive structural evolution within the interior of an active volcano.

1 Introduction

1.1 Short Period and Very Long Period Seismic Signals from Active Volcanoes

The panoply of transient physical processes at active volcanoes radiate a remarkable variety of elastic wave signal types (e.g., Chouet & Matoza, 2013; Gasparini et al., 1992) that span the full frequency/period range of broadband seismology (from hundreds of hertz to thousands of seconds) (e.g., Diaz, 2016). These source processes encompass subareal or contained gas-driven explosions (e.g., Hidayat et al., 2003), stick-slip faulting (e.g., Kendrick et al., 2014), tractions arising from turbulent (including eruption column) fluid flow (e.g., Matoza et al., 2009), diverse types of elastic resonance (e.g., O'Brien & Bean, 2008), reaction forces arising from fluid (e.g., Aster et al., 2003; James et al., 2004; Nishimura et al., 2000) and/or gas (e.g., D'Auria & Martini, 2009; Prejean & Brodsky, 2011) transport, pressurization, and volumetric injection or withdrawal (e.g., Lough et al., 2013; Prejean et al., 2003).

Very long period (VLP) volcano seismograms (encompassing a general period range from several to hundreds of seconds) have been recognized for many decades in both eruptive and noneruptive systems (e.g., Sassa, 1935) but have become increasingly identified and studied, as broadband seismic network and array observations have become much more tractable (Aster et al., 2000). Studies of VLP source processes have been greatly aided by the moment tensor inversions for couple and single force time histories, which are facilitated by these data (Waite, 2015). The timing, duration, moment tensor characterization, and spectral characteristics of VLP signals provide information on magma and gas storage and transport, and have thus been studied at diverse volcanoes worldwide and in laboratory and computer modeling (e.g., Aster et al., 2000; Chouet et al., 2003; D'Auria & Martini, 2009; James et al., 2006; Karlstrom & Dunham, 2016; Lyons & Waite, 2011; Nishimura et al., 2000; Rowe et al., 1998; Waite, 2015). Because observations of elastic waves are ultimately dependent on the time history of causative forces applied to the source region (e.g., Waite, 2015), interpretations from seismic and/or infrasonic data alone frequently have nonunique interpretations, and robust interpretation typically requires additional constraints based on other observations.

VLP signals commonly display a high degree of repeatability. Because of the long elastic wavelengths involved, these signals are commonly observed in the near field, insensitive to small changes in the surrounding medium, and consequently dominated by source processes. Repeatability indicates approximately self-reconstructing source processes and location. However, VLP waveforms in some cases show progressive changes in their character and/or frequency of occurrence that may reflect progressive or sudden changes in conduit geometry, in the ability of a conduit transport gas and/or magma, in magmatic or volatile conditions, and/or in dynamical processes (e.g., Aster et al., 2003; Chouet et al., 2003; D'Auria & Martini, 2009; Karlstrom & Dunham, 2016; Rowe et al., 1998).

VLP source processes are thus diverse, and have been variously attributed to processes that include gas and magma transport (Chouet & Matoza, 2013; Nishimura et al., 2000; Waite et al., 2013), the excitation of slow elastic/viscous fracture-guided (Ferrazzini & Aki, 1987) or slow-trapped body elastic waves (Garces, 2000), and/or pressure and volumetric transients (Lyons & Waite, 2011). Particularly well-documented and studied examples, in addition to Erebus volcano, include Stromboli volcano, Italy (interpreted as gas slug transport forces interacting with a shallow dike and conduit system) (Chouet et al., 2008), Aso volcano, Japan (interpreted as a slow elastic wave resonance within a near-summit hydrothermal reservoir) (Kawakatsu et al., 2000), and Kilauea, Hawaii (interpreted as unsteady magmatic flow within an intricate shallow plumbing system below the summit caldera) (Dawson et al., 2010).

1.2 Erebus Volcano

Erebus volcano is a 3,794 m high stratovolcano on Ross Island, Antarctica (Figure 1) that exhibits persistent low-level eruptive activity. The summit region consists of an approximately 500 by 600 by 120 m deep elliptical Main Crater, an Inner Crater that contains a long-lived convecting phonolite lava lake, and other explosive and fumerolic vents. The lava lake and its underlying near-summit, magma-filled conduit system has been a persistent feature of the volcano for over 45 years (Giggenbach et al., 1974; Kyle et al., 1982; Peters et al., 2014). Lava lake systems commonly exhibit bubble-/slug-initiated surface disturbances (Bouche et al., 2010) and thermal-/magma-/gas-driven convection (Oppenheimer & Kyle, 2008; Oppenheimer et al., 2011; Peters et al., 2014) and can generate a wide variety of seismic signals arising from explosive, advective, elastic, gravity, and convective forces, for example (Aster et al., 2000; Chouet & Matoza, 2013; Dawson & Chouet, 2014; Gerst et al., 2013; Karlstrom & Dunham, 2016). Volcanoes hosting strombolian eruptions may allow for unusually detailed study of upper conduit and eruptive processes because of their frequent, and often quasi-repeating, eruptive activity. Observations of strombolian systems can also commonly be made under less hazardous environments and over what may be extended periods of activity. Here we conduct a comparative analysis of just two types of eruptive signals that are strongly manifested at Erebus volcano, Antarctica, in the frequent and well-observed eruptions from its long-lived lava lake. Short period (SP) eruptions from this volcanic system have been well observed using seismometers, infrasonic microphones, Doppler radar, and during clear weather, with video cameras, so that the source mechanism is unusually clear and has been modeled in detail as bubble growth and explosive bursting at the lava lake surface (e.g., Gerst et al., 2013) arising from large pressurized buoyantly transported gas slugs (e.g., Ilanko et al., 2015).

Details are in the caption following the image
Elevation map of Erebus volcano on Ross Island. Circles show summit plateau seismic and infrasound stations. Video data from the camera site (VID) was used to monitor eruptive behavior and the state of the lava lake and is the viewpoint from which the still photos in Figure 3 were obtained (only approximately 4 years of RAY seismic and infrasound data exist because the station was destroyed by an eruption in 2007).

The Erebus lava lake exhibits both continuous passive degassing and infrequent and irregular eruptive bursting of pressurized gas bubbles ranging from relatively small (meter diameter scale) to large (up to >15 m diameter) and highly explosive (Ilanko et al., 2015). During especially active periods, the volcano has commonly produced 50 or more large lava lake eruptions per week. These eruptions are characterized by the emergence of gas slugs that decompress explosively and collapse at the surface of the lava lake, commonly ejecting bombs to distances of up to many hundreds of meters. This eruptive process has been observed and analyzed using visible and thermal infrared imaging, Doppler radar, LIDAR, and monitored with infrasound and seismic instruments (Aster et al., 2003; Gerst et al., 2013; Johnson et al., 2004; Jones et al., 2015; Peters et al., 2014). Fourier transform infrared (FTIR) observations (Ilanko et al., 2015) show that eruption slugs exhibit enhanced CO2/CO (factor of 3–6) and CO2/H2O (factor of 2–10) relative to passive degassing, indicating that these large gas slugs grow and are delivered from a greater depth than the passive near-surface gas source region.

2 The Eruption Model and VLP Mechanism

Aster et al. (2003) proposed an eruption model for the Erebus strombolian system that consisted of five phases. The first stage is gas slug sequestration and growth (Figure 2: Stage 1). Knowledge of slug coalescence and sequestration is not seismically observable and has largely come from laboratory models (e.g., Jaupart & Vergniolle, 1988).

Details are in the caption following the image
Schematic depiction of the strombolian eruption mechanism at Erebus volcano, Antarctica with a representative vertical-component seismogram (E1S). The short period signal is shown in black and the very long period signal is shown in red. Linkage between stages and the seismogram are represented by either dashed or solid lines. The solid lines indicate seismically interpreted phases, and dashed lines indicate aseismic precursory phases.

The second stage begins when the gas slug rises through the upper conduit system and ascends into the lava lake (Figure 2: Stage 2). During this phase, the rising slug displaces magma, which induces flow and produces pressure and traction forces on the conduit walls. Also, the slug is growing during this phase due to decompression and possible scavenging of gas bubbles in the conduit (Chouet et al., 2003; Gerst et al., 2013). This phase is at best marginally observed in the seismic data, though given uncertainties in rising velocity, it is unclear where in this stage the slug begins to produce observable seismic signals. This is supported by the short duration (<5 s) of the preeruptive VLP signal at Erebus volcano and by results from laboratory experiments (e.g., James et al., 2006).

The third stage begins when the slug emerges from the conduit into the base of the lava lake, continues through the slug's rapid growth due to final decompression, and ends as the slug ruptures at the lava lake surface (Figure 2: Stage 3). This phase is seismically observed at the closest stations (<700 m from the lava lake) in the polarity and amplitude-variable preeruption segment of the VLP signal, and the final expansion of the lava lake system has been well observed with radar (Gerst et al., 2013). We believe that this signal variability evidences a complicated process and that numerical and laboratory models hold the key to understanding this complexity. James et al. (2004, 2006) conducted laboratory experiments that investigated slug passage through a flare (significant increase in diameter) and showed that large-amplitude pressure signals are generated. They also showed that during this passage, the individual slugs may fragment or bifurcate. This finding was further supported by D'Auria and Martini (2009), who attributed the breakup in their numerical models to turbulence. Both studies observed a positive pressure pulse as the slug passed through the flare. Keeping these findings in mind, we propose that the preeruptive signal variability at Erebus volcano may be attributed to flow complexities induced by the following processes: (1) rapid decompression of the slug and its interaction with the conduit and lava lake wall rock; (2) turbulence generated when the slug passes through a constriction at the bottom of the lava lake; (3) breakup of the slug due to turbulence; and (4) sinking of magma back into the conduit below the lava lake.

The fourth stage is the eruption, which we define as the rupturing of the magma shell above the slug and the venting of slug gases into the atmosphere (Figure 2: Stage 4). This stage is marked by the onset of the SP seismic signal, which is preceded by surface distention of the lava lake (∼3 s before SP; Figure 3). Gerst et al. (2013) estimated typical Erebus gas slug overpressures to be in the range of ∼100–800 kPa and noted that the slug's expansion just before rupture, typically affects the entire lava lake surface (Radius ≈20 m). After shell burst, shell fragment bombs are ballistically ejected from the lava lake to distances of up to many hundreds of meters (Caldwell & Kyle, 1994; Kelly et al., 2008), with initial speeds of ejecta as high as 150 m/s. Gerst et al. (2013) also estimated the erupted magma shell mass removed from the lava lake in larger recent explosions to range between 1,800 and 3,600 metric tons, with a corresponding volume of 900–1,800 m3, assuming a bulk erupted magma density of 2,000 kg/m3. Because the uppermost magma is highly vesicular (as observed in cooled bomb fragments), the actual volume removed from the lava lake is probably several times this value. Posteruption, the lava lake is observed in video footage to be emptied to a depth of tens of meters, revealing a funnel-shaped uppermost geometry that is consistent with late-stage slug stretching or even bifurcating (Figure 3).

Details are in the caption following the image
A characteristic Erebus lava lake eruption (25 January 2001) viewed in false infrared color from the VID video camera site (Figure 1), and closely corresponding to the viewpoint used in Figure 10b. Times are indicated relative to the short period origin time defined by the bubble wall burst (Gerst et al., 2013) (frame c). From Aster et al. (2003).

Eruption seismic monitoring prior to 2001 was largely accomplished with a robust telemetered short period (SP) (1 Hz seismometers) seismic and infrasonic network (Dibble, 1994) which remained operational until December 2016. Construction of a broader-band digital seismic and GPS network using Guralp 40-T (30 s) seismometers began in 2001. By the 2003–2004 field season the telemetered network of five near-summit and one-flank stations became active with unified sample rates, locations, state of health and environmental data streams, and real-time telemetry to McMurdo Station (Figure 1; Aster et al., 2004). We utilize data from the near-summit stations of the network in this study (one of these stations, RAY, was destroyed by bombs during the study period).

Here, we document and discuss seismological observations of temporal changes in eruption activity at Erebus volcano, and interpret these in terms of geometry changes within the terminal magma-filled conduit system. The key observation is systematic variability in the relative timing of SP and VLP seismic signals across a period of several years. Both the SP and VLP signals individually exhibit high degrees of event-to-event similarity and can thus be correlated to quantify their relative timing changes. These changes occur progressively with time scales of weeks to months. We analyze these changes using a near-summit five-station broadband seismic network and conclude that they reflect gradual alterations to the geometry of the terminal shallow magma-filled conduit over a length of several hundred meters and may represent changes in the degree of constriction, tortuosity, and/or freezing/melting modification of conduit walls.

3 Eruption Seismic Characteristics and Detection

Two distinct eruption-related seismic signals, occupying distinct frequency bands (referred to as SP and VLP) are associated with Erebus lava lake eruptions (Aster et al., 2003). Because these eruptions are small (VEI 0 to 1), and background noise levels on Ross Island due to wind and ocean waves can be high (Anthony et al., 2015), SP and VLP signals are best recorded at ranges of less than about 2 km from the lava lake (e.g., Rowe et al., 1998).

3.1 SP Seismic Signals

The SP eruption signal at Erebus consists of body waves and surface waves generated by highly impulsive lava lake eruptions generated by the bursting and collapse of large gas slugs (Chaput et al., 2012; Gerst et al., 2013; Rowe et al., 2000); Figure 3. These signals have dominant energy at frequencies between 1 and 8 Hz (Figure 6), a frequency band that is also strongly scattered by the heterogeneous upper volcano (Chaput et al., 2015). Lava lake eruptions additionally produce strong infrasonic signals (up to tens of pascals in peak pressure amplitude at ranges of 1 km) with similar frequency content and amplitude scaling relative to SP seismic signals (Gerst et al., 2013; Jones et al., 2008). When filtered above ∼1 Hz to maximize signal-to-noise levels, SP seismograms display emergent and variable-first motions that reflect fine-scale differences in the initial rupture characteristics of eruptive gas slugs (Gerst et al., 2013). Initial SP motions are followed by a robust and quasi-repeatable coda that persists above background noise for 30 s or longer (Aster et al., 2003). Event-to-event vertical-component SP waveform (60 s) correlation values are ≈0.3–0.7 with a median of 0.42 (Knox, 2012).

3.2 VLP Seismic Signals

VLP events at Erebus volcano are ubiquitously associated with lava lake eruptions.

VLP signals associated with strombolian eruptions at Erebus volcano show both preeruption and posteruption components. The preeruption, which can last up to several seconds, is interpreted as being generated by the ascending and decompressing gas slug interacting with conduit walls and displacing magma as it ascends to the lava lake surface. Polarity and amplitude characteristics of this signal are variable (Aster et al., 2003), indicating a varying balance between compressive and decompressive forces within the uppermost volcano. The later, postexplosion, VLP signal is much more repeatable and exhibits stable and distinct nonharmonically related modal spectral components with periods near 8, 11, and 21 s. Stacking large numbers of events reveals shorter period VLP spectral components to approximately 2.5 s (Aster et al., 2003; Rowe et al., 1998). These spectral modes have been hypothesized (Aster et al., 2003, 2008) to represent superimposed low-Q resonant periods of an oscillatory refill process that becomes excited when the uppermost conduit is gravitationally perturbed by eruptive mass removal from the conduit tip. This model is supported by the observation that stacked VLP signals persist for over 400 s and cease when the lava lake refills to its preeruptive level and returns to gravitational equilibrium (Aster et al., 2003).

3.3 VLP-SP Lag Measurements

A catalog of lava lake explosions was generated by scanning the continuous seismic and infrasound data using an SP band (1–8 Hz) correlation-based multichannel (six seismic stations and three infrasound stations) filter with a template generated from stacking 20 high signal-to-noise eruption seismograms (Knox, 2012). Event review was conducted using move-out criteria and by verifying the presence of a diagnostic eruption infrasound signal waveform. This produced a catalog of 2,974 irregularly temporally spaced lava lake eruptions occurring between 2003 and 2011 (Figure 4) for which the correlation of every SP signal with the stack of the entire event group (using 60 s of data beginning 5 s before and ending 55 s after the event onset) was 0.4 or greater. The number of usable events at each station was variable depending on individual station uptime and noise level.

Details are in the caption following the image
Weekly distribution of 2,974 eruptions detected during the study period. Gray-shaded time intervals indicate periods when three or more broadband stations were contemporaneously operational, with uptime percentage on a weekly basis noted on scale at left. Network gaps generally correspond with austral winter periods when power resources were insufficient to maintain operation.

The distinct SP and VLP signals were isolated on the vertical seismogram components by band-pass filtering between 1–8 Hz for the SP and 0.03–0.2 Hz for the VLP bands (Aster et al., 2003, 2008). VLP signal to noise was further improved by integrating the VLP seismograms from ground velocity to displacement with a high-pass filter above the seismometer corner period (30 s).

Following filtering and event selection, relative event-to-event timing of the SP and VLP signals was estimated via independent cross correlation of the SP and VLP signals at one sample (0.025 s) time resolution, with the VLP-SP lag being the difference between the two lags. The VLP-SP lag was also independently estimated from the SP-aligned seismogram matrix using a peak tracking algorithm (Knox, 2012), which was found to be consistent with cross correlation-based relative lag estimation. These relative correlation lags were then examined on a station-by-station basis across more than 8 years of seismograms to examine progressive VLP-SP lag variations. The principal result of this detection and lag estimation process is depicted in Figure 5.

Details are in the caption following the image
Overview of eruption waveforms and measurements. (a) Vertical-component E1S short-period (SP) signals (1–8 Hz dominant energy) aligned by (60 s) SP cross correlation. Velocity seismogram at right shows the corresponding SP seismogram stack. (b) Very long period (VLP) velocity seismograms (0.03–0.2 Hz) aligned using the SP signal correlation alignment shown in Figure 5a. The red line shows peak tracking of the VLP signal variation, with the VLP correlation-aligned signal stack shown at right. Progressive time shifting of the VLP signals indicate change in the relative timing of SP and VLP source origin times. (c) Broadband velocity seismogram stack for closest station E1S. (d) Broadband stack displacement power spectral density from Figure 5c, showing characteristic VLP and SP frequency bands. Note that low signal-to-noise for some events produces some outliers and null determinations in the peak-tracking algorithm, as reflected in Figure 6.

3.4 VLP-SP Lag Variations

Systematic shifts in VLP signal timing relative to the SP alignment are seen to occur over days to months between 2003 and 2011. Distinct time shifts encompass several episodes (Figure 6) and are necessarily best sampled during times of greater eruption activity. We observe a notable smooth multimonth VLP-SP delay decrease and subsequent increase with approximately ±1 s lag variations during late 2005 and early 2006, and another comparable variation between November 2006 and March 2007. Less well-resolved decreasing trends are evident between late 2007 and mid-2008, and between late 2008 and early 2009. Figure 7 displays the nature of the relative seismogram SP/VLP lag changes during late 2005 and early 2006 using five eruption seismograms recorded at E1S across a 7 month period.

Details are in the caption following the image
(a) VLP-SP delay (respective station median delays subtracted) observed at the five near-summit seismic stations, showing variation and station-to-station consistency for the most active eruptive time period (2004 to mid-2009). The period 2006–2007 was well sampled due to a high rate of lava lake eruptions (Figure 4). Three primary time periods of variation are indicated with black bars: (1) A decrease and increase between late 2005 and early 2006; (2) A decrease and increase in late 2006 and early 2007; (3) Two episodes of less resolved decrease between late 2007 and early 2009. Vertical lines (A–E) indicate times of representative seismograms shown in Figure 7. (b) VLP-SP delay detail (with respect to eruption event number to provide equal sampling) at high signal-to-noise station E1S, showing consistency between maximum tracking (black) and seismogram cross-correlation (red) estimates. Black bars indicate the same time periods as in Figure 6a. Sporadic outliers are due to poor signal-to-noise events.
Details are in the caption following the image
Representative vertical-component short period (SP) and very long period (VLP) velocity seismograms recorded at station E1S (Figure 1) between November 2005 and May 2006 corresponding (a–e) times in Figure 6). The selected VLP zero crossing, indicated by a vertical line in each case, illustrates VLP-SP delay changes relative to the whole-waveform (60 s) SP (1–8 Hz filtered) correlation alignment. The variation in seconds relative to the previous seismogram is noted at the top of each panel. Note that the early SP signal is emergent and variable due to small-scale eruptive slug variability from event to event.

Station-specific effects can easily be ruled out to explain the VLP-SP timing differences by the similarity across all five stations (Figure 6a). The noisiest trends are seen at LEH, which is the most distant station from the lava lake analyzed here (Figure 1) and has a lower VLP signal-to-noise ratio relative to the other summit plateau stations (Aster et al., 2008). For this reason, we disregard LEH from further analysis. VLP-SP trend comparisons between other station pairs show correlations varying from 0.69 to 0.91 (Tables 1 and 2).

Table 1. Matched Filter: Event Correlation Statistics
Station Mean correlation coefficient Variance
E1S 0.5398 0.0173
CON 0.4869 0.0229
NKB 0.6175 0.0288
RAY 0.5483 0.0319
LEH 0.4816 0.0282
Table 2. Station-to-Station Correlation Matrix of the Temporal Trends Shown in Figure 6
E1S NKB RAY LEH CON
E1S 1
NKB 0.703 (314) 1
RAY 0.851 (247) 0.687 (227) 1
LEH −0.213 (376) −0.204 (280) −0.496 (194) 1
CON 0.914 (406) 0.768 (369) 0.755 (217) −0.108 (367) 1
  • Note. Number of events used for each measurement is shown parenthetically.

3.5 VLP Waveform Stability

VLP-SP timing changes were quantified by tracking the maximum peak in the VLP signal relative to the SP-determined alignment, as in Figure (6b) and directly from VLP cross-correlation lags, with both methods of analysis yielding very similar results. Knox (2012) searched for, but did not find, VLP frequency-dependent time shift, that is, if one or more of the spectral components near 21, 11, and 8 s (see Figure 5d) were to shift in phase relative to the other(s). The observed shift is thus a broadband timing shift between SP and VLP source origin times, rather than a more complex frequency-dependent change in the VLP signal. We additionally examined VLP analytic signal envelopes and the relative amplitudes of VLP spectral components and found no systematic changes in the decay of the VLP signals or in their constituent spectral components. The modal excitation of the three principal (21, 11, and 8 s) VLP components can consistently be modeled as three highly damped oscillations with Q values of approximately 4, 18, and 11, respectively (Aster et al., 2003).

4 Discussion

VLP-SP relative timing changes occur across a gradual (days to weeks) time scale and are a seismic frequency-independent shift between distinct VLP and SP components of eruption seismograms. Interpreted in terms of the cyclically repeating eruption process (Aster et al., 2003, 2008; Gerst et al., 2013), we infer that this observation shows a variable delay between the SP-associated eruption and gravitational unloading due to the removal of mass at the crack tip and the subsequent refilling of the deeper conduit as reflected by the excitation of the VLP force centroid region.

Prior to focusing on conduit changes, we investigated, and ruled out, a range of factors that might drive the lag change observations. We investigated possible correlations with instrumentation changes, eruption sizes, and the slug exit location within the lava lake. We then examined changes in the surface morphology of the Inner Crater region. These are briefly summarized below and are addressed extensively in Knox (2012).

4.1 Instrumentation Temperature

Surface instrumentation on the Erebus summit plateau experiences large seasonal temperature variations that typically range between −20 and −55°C. Seismometer vault temperatures recorded on network tiltmeters (Aster et al., 2004) were examined and found to be approximately 10–15°C warmer than this. Lack of any annual trend in the lag observations enabled us to rule out a uniform systematic temperature effect on the Guralp 40T seismometer response as affecting our observations.

4.2 Eruption Size

Erebus eruption sizes, quantified via either maximum or root-mean-square SP seismic or infrasonic amplitudes, vary by approximately 2 orders of magnitude and show a strong linear correlation that breaks down only for the smallest events (Rowe et al., 2000). These measures of event size correlate with gas slug size and with the volume of magma ejected from the lava lake during an explosion (Aster et al., 2003; Gerst et al., 2013; Johnson & Aster, 2005; Rowe et al., 2000). Comparison between the VLP-SP delay times to the event size metrics by Knox (2012) found no significant trend or correlation. Event sizes increased modestly between 16 December 2005 and 23 February 2006, corresponding to the VLP-SP change described by phase 1 in Figure 6, but was generally stable throughout the catalog.

4.3 SP/Infrasound Locations

Variation in the location of the first burst for eruptive slugs is tightly constrained by the dimensions of the (maximum ∼50 m diameter) lava lake. A subset of eruptions between 2 January 2005 and 8 April 2009 were examined, which were well recorded by crater rim-sited infrasound sensors and thus well-located acoustic hypocenters determined using semblance-based methods (Johnson et al., 2011; Jones et al., 2008). No strong correlations were found between explosion origin and lag variations, although there exists a small (few meters; Figure 8) average jump in origin along the N-S direction in early 2006. However, any SP delays due to such small variations in explosion origin location would be much smaller than the observed range of lag variability (several seconds) that is observed to occur uniformly across the seismic network for a range of source station azimuths (Figures 1 and 6).

Details are in the caption following the image
(top) Very long period-short period (VLP-SP) delay as a function of time at representative station E1S (from Figure 6). (middle and bottom) Infrasonically determined locations of the SP source on the surface of the lava lake from Jones et al. (2008), relative to the center of the lava lake.

4.4 Lava Lake and Crater Morphology and Possible Effects on the Shallow Conduit

We evaluated available video and still observations (Aster et al., 2003; Dibble et al., 2008) (Figure 1) of the lava lake to seek evidence of VLP-SP delay-correlated lava lake elevation, location, and/or size changes. Only sporadic video data are available for periods of particular interest (e.g., late 2005 through early 2006) due to difficulties in making long-duration visual recordings from the crater rim in Antarctic conditions. Available observations can be further complicated by the common occurrence of a plume of volcanic gas emitted from the lava lake, which often fills the crater. A potentially illustrative period of available data spans the end of 2005 through the end of 2006. The largest observed VLP-SP change occurs in the time period between the two series of images (Figures 9, top and 9, bottom), across which the lava lake shows no large changes in its morphology but moderate (approximately 30%) changes in apparent surface area.

Details are in the caption following the image
Comparative lava lake images (false IR color) recorded at the VID site (Figure 1), showing general stability of the lava lake surface across two field seasons (also compare to preeruptive image of the lava lake from 2001 in Figure 3). Dates and times of each image are indicated in the upper right-hand corner (top row) and upper left-hand corner (bottom row).

Over longer periods of time, however, the topography of the Inner Crater undergoes vertical and other changes in geomorphology on the scale of up to tens of meters (Jones et al., 2015). Figure 10 shows this dynamic aspect to the crater surface from differencing repeated laser scans. These changes suggest that both the surface vents and the underlying shallow conduit system undergo geometric changes that may be relevant to the interpretation of VLP/SP timing changes. The circular geometry of the observed elevation changes is suggestive of an active subcircular ring dike/fault system that moves in response to gravitational forces and shallow magmatic system changes. The volcano has not been observed to host large-scale flank tilt, inflation, or deflation (Otway et al., 1994). These large dynamic effects are thus predominantly confined to the Inner Crater.

Details are in the caption following the image
(a) Differential elevation within the Erebus Inner Crater between December 2001 and December 2010 determined by comparative laser scanning, showing changes of up to tens of meters. (b) Photograph of the Inner crater from December 2010, identifying features corresponding to Figures 10c and 10d, from Jones et al. (2015). (c, d) Elevation change along the two transects indicated in Figure 10a. From Jones et al. (2015).

4.5 Variable VLP-SP Delay Mediated by Guided Conduit Elastic Waves

Incorporating the above analysis of possible other factors, we suggest that Erebus VLP-SP lag variations arise from changes in the elastically mediated communication time between the eruption disturbance at the conduit tip and the underlying VLP source region (Figure 11). In this model, eruptive depressurization is elastically communicated to the VLP region, and the deeper magmatic supply from the near-summit magma chamber then responds by refilling the uppermost conduit in an oscillatory fashion (Aster et al., 2003). This process produces the forces that drive the VLP signal and that persist until the system reestablishes its gravitational (magmatic head) equilibrium. Upon completion of this cycle, the terminal conduit system is restored to its preeruptive condition and is primed to generate another highly similar eruption and suite of seismic signals with the passage of the next eruptive gas slug.

Details are in the caption following the image
Conceptualization of the Erebus terminal conduit system in approximate geometric context (which could be crack-like or subcircular in cross section) constrained by crater imaging and tomography (Figure 12). Indicated events, shown referenced to a vertical-component seismogram filtered in the short period (SP) and very long period (VLP) bands at the bottom, are as follows: (1) preeruptive storage or transport of eruptive gas slugs through the near-summit magma chamber; (2) Buoyant ascent of the gas slug into and up the eruptive conduit; (3) emergence of gas slug into the near-surface lava lake and origin of the preeruptive VLP signal; (4) surface explosion, mass removal, and origin of the SP signal; (5) elastic communication of surface explosion to deeper conduit system and VLP source region; (6) posteruptive excitation of VLP signal (with shaded source region indicated) during conduit refill through reestablishment of system equilibrium. VLP-SP relative lags can be robustly estimated because both seismogenic processes produce similar and correlatable seismograms. However, the true onset of the posteruptive VLP signal is difficult to determine because of its long period spectrum and its overlap with the preeruptive slug emergence signal.

Multiple lines of evidence suggest that the uppermost conduit is narrow and inclined. Moment tensor inversion and polarization analysis indicate that the VLP force centroid lies approximately 200 m to the west-northwest and less than approximately 400 m below the lava lake (Aster et al., 2008). This centroid location maps closely to the seismically imaged summit magma chamber edge at its nearest point to the lava lake (Figure 12; Zandomeneghi et al., 2013). Neither scattering nor travel time inversion resolves a conduit structure directly below the lava lake, and a visually observed terminus diameter of just 5–10 m is visible below the lava lake after eruptive evisceration (Aster et al., 2003). The preferred moment tensor inversion solutions of Aster et al. (2008) also include a significant (necessary for ≈25% of total seismogram fit variance reduction, and with a typical amplitude of 1 × 107N) steeply dipping single-force term that lies predominantly in the vertical/west plane, and is particularly important for fitting the later extent of the VLP waveform. This solution component is consistent with oscillatory advective transport along a similarly inclined conduit during VLP excitation. The lava lake, VLP source, and magma chamber geometry imply an average terminal conduit dip of approximately 60°. Localization of the VLP source centroid at the edge of the magma chamber and near the presumed initiation of the terminal conduit suggests that the VLP source may occur in a region of constrictive transition from the magma chamber to the narrow terminal conduit.

Details are in the caption following the image
Shaded summit relief map showing inferred boundaries of the Erebus near-summit magma body (purple), as jointly constrained by P velocity anomaly (ΔVp) and high-frequency scattering (S), jointly constrained by scattering and travel time tomography (Zandomeneghi et al., 2013). Red star indicates the location of the lava lake and yellow circle indicates the very long period (VLP) force centroid constrained by moment tensor inversion (Aster et al., 2008) and VLP particle motion (Aster et al., 2003) studies. Lack of direct imaging of the uppermost conduit, moment tensor, and tomographic results consistently suggest that the terminal conduit is narrow and inclined with an average dip near 60°.

We next consider several models for posteruptive pressure communication between the surface and the VLP region and for generating temporal variability in this process: (a) Direct communication through a compressional conduit wave propagating at P wave velocity, (b) a conduit-guided Stoneley wave model of propagation along a fluid-filled conduit, and (c) a nonlinear downward propagating shock wave.

Nonexclusive temporally varying processes relevant to these models may include variations in magma gas content, in conduit width or more general geometry, in conduit wall elastic properties. Such changes could arise due to wall melting or freezing, to subsurface strain associated with Inner Crater collapse and inflation (possibly focused along ring faults or other areas of strain concentration; Figure 10), or to mechanical damage induced by the progressive passage of gas slugs and/or convecting magma and gas (Ilanko et al., 2015; Jones et al., 2015).

First-order calculations relevant to an elastically communicated variable VLP-SP lag mechanism can be performed using a simplified (straight and constant inclination) conduit consistent with the seismically/visually constrained SP and VLP source locations (Chaput et al., 2012; Jones et al., 2008; Zandomeneghi et al., 2013) (Figure 12). The corresponding surface-to-VLP source centroid conduit length is l≈450. This length is a minimum, and the effective l, leff, could be substantially longer than this, given a more complex conduit geometry. In this case, modeled VLP-SP delays may to a first approximation be scaled as leff/l.

4.5.1 Model A: Acoustic wave

A conceptually simple process for transmitting a gravitational and elastic disturbance from the conduit tip to the VLP source region is via a compressional acoustic (P) wave traveling through the magma-filled conduit at the P wave velocity. Appropriate compressional wave velocities can be estimated from the suite of upper conduit velocity profiles estimated by Dibble (1994) for an Erebus phonolitic magma column with vertically varying water content and vesicularity. One-way travel times for various P wave velocity models were evaluated for a straight conduit with l = 450 m to assess if the ≈±1 s lag change might be due to water content and/or vesicularity changes based on this modeling. The slowest Dibble (1994) model (1% water content by weight and 75% surface vesicularity) had a calculated travel time of 1.95 s for this length, while the fastest model (0.5% water content and 50% surface vesicularity) had a travel time of 0.84 s. The preferred model had 0.5% water content, and 75% surface vesicularity. The full span of these models produces an acoustic travel time difference that is still insufficient by a factor of ≈2 to account for the observed variability in VLP-SP delays. These calculations employ a minimum conduit length of l = 450 m, and a more complicated and lengthy conduit geometry could proportionately improve agreement with the observations.

4.5.2 Model B: Stoneley-Guided Conduit Wave

In regions hosting a high-density and elastic moduli (seismic impedance) contrast, such as at the boundary of a magma-filled conduit embedded within volcano host rock, elastic energy will propagate as trapped dispersive boundary waves, particularly for wavelengths that are comparable or longer than the conduit diameter. A more realistic candidate for elastically transmitting the pressure disturbance from the surface to the VLP source region than the simple P wave model above, is thus a Stoneley (boundary) seismoacoustic wave. These elastic wave types can also propagate at much slower seismic wave velocities than those of the bulk (magma and surrounding rock) elastic constituents, and be highly sensitive to conduit or conduit wall conditions (Ferrazzini & Aki, 1987).

Korneev (2008, 2010) derived analytical solutions for elastic wave propagation bound to a thin (relative to wavelength) fluid-filled fracture. We will utilize this model to estimate elastic propagation speeds and the degree of conduit change required to produce the observed lag variations for a conduit length in the range constrained by the Erebus geometry. The Korneev formulations can be shown to be asymptotic to those derived by Ferrazzini and Aki (1987).

The fundamental mode phase (dispersive) velocity as a function of frequency f given by (Korneev, 2008) is
urn:x-wiley:jgrb:media:jgrb52579:jgrb52579-math-0001
where h is fracture thickness, μ is the shear modulus of the conduit wall, ρf is the fluid density, and γ is the conduit wall Vs/Vp ratio, and the viscous adjustment factor, C is given by
urn:x-wiley:jgrb:media:jgrb52579:jgrb52579-math-0002
where
urn:x-wiley:jgrb:media:jgrb52579:jgrb52579-math-0003
and η is the kinematic fluid viscosity.

We investigated this model for fracture width ranging between 3 and 15 m using a magma density of 2,000 kg/m3, conduit wall vs/vp of 0.4, wall vp of 2,000 m/s (Zandomeneghi et al., 2013), wall density of 2,700 kg/m3, and magmatic kinematic viscosity of η = 104 Pa s (Dibble et al., 1984). The propagation of guided waves in this system is highly dispersive. Although the explosively generated SP seismic and acoustic signals from eruptions have most of their power in the 1–8 Hz band (Aster et al., 2003), decompression of the conduit system due to eruptive mass removal from lava lake eruptions will occur on a time scale that is commensurate with the eruptive removal of mass and relaxation of the overpressured eruptive gas slug, which occurs on a time scale of 1–2 s (Gerst et al., 2013). The relevant frequency for the unloading signal to be propagated down the conduit is thus approximately 0.5–1 Hz. We thus investigated a frequency range of 0.1 to 8 Hz.

Stoneley wave velocities for the above parameters range between 60 m/s (h = 4 m f = 0.1 Hz) and 778 m/s (f = 8 Hz and h = 15 m). Corresponding minimum conduit length (450 m, assuming a tortuosity of 1.5) travel times from surface to the VLP source region thus vary between approximately 7.5 and 0.6 s, respectively. The kinematic viscosity value has a large effect on the calculated wave speed at lower frequencies and for higher viscosities.

The modeled travel times displayed in Figure 13a show that, for this model to explain the observed lag variations of up to approximately 1 s, and under these values of material properties, an appreciable portion of the conduit between the surface and the VLP source region must have a characteristic dimension of 5 m or less. This contention is consistent with posteruptive observations of a narrow conduit below the surface of the lava lake, with a lack of terminal conduit resolution in tomographic studies, and with the variable proclivity of the system to sequester preeruptive gas slugs (assuming that sequestration occurs within the shallow conduit system). However, the travel time calculations displayed in Figure 13 are made under the simplifying assumptions of a minimum length and uniform crack-like conduit, both of which are likely to be poor. If recently estimated higher values of 105 Pa s or greater for the Erebus phonolitic magma experimentally estimated using recovered bomb fragment material from the lava lake (Le Losq et al., 2015) (Figure 13b) are adopted, the conduit elastic wave travel time estimates at lower frequencies are increased by several seconds. Decreasing the vs/vp, such as might be appropriate for the heated wall of a magmatic conduit also will decrease wave speed and thus increase travel time.

Details are in the caption following the image
(a) Representative Stoneley-guided wave travel time variations calculated using the crack wave formulations of Korneev (2008) as a function of frequency and (logarithmically spaced) crack width. Calculations used a 450 m long Erebus conduit with a tortuosity factor of 1.5, a magma density of 2,000 kg/m3 (Dibble et al., 1984), conduit wall vs/vp of 0.4, wall vp of 2000 m/s (Zandomeneghi et al., 2013), wall density of 2700 kg/m3, and magmatic kinematic viscosity of η = 104 Pa s (Dibble et al., 1984). Note the rapidly increasing sensitivity of the system to conduit thickness change as h and f decrease. A variation of ±1 s in the travel time in the lower-frequency range is achievable for variations in h of one to several meters. Travel time curves are normalized to the h = 7.7 m curve to show the representative variation in h required to span ±1 s of travel time variation relative to the central curve. (b) Same calculations as in Figure 13a, but using a higher kinematic viscosity of 105 Pa s (Le Losq et al., 2015).

A more realistic scenario, rather than varying the width of a uniform planar fracture, is that the geometry of the VLP source region to the surface conduit system is more complex and that smaller changes in the thickness between individual fractures and/or along narrow segments of the conduit give rise to the required variability. This view is further supported by tomographic imaging efforts of the upper edifice (Chaput et al., 2012; Zandomeneghi et al., 2013) that indicate complex larger-scale magmatic geometry.

4.5.3 Model C: Nonlinear Shock Wave

A third process for mediating conduit elastodynamic communication is suggested by the numerical experiments of Nishimura and Chouet (2003). This study used finite-difference methods to investigate elastodynamic effects of the abrupt depressurization of a generalized cylindrical magma column due to removal of a restrictive cap and its interactions with a deeper reservoir. This modeling effort found a downward propagating nonlinear pressure shock wave that initiates following the cap removal and propagates slightly faster than the associated P wave magma column velocity (i.e., 170–180 m/s). The velocity of this shock wave is invariably greater than the P wave velocity of the magma column, but is controlled above this threshold by the magnitude of the pressure and density steps at the head of the shock wave. In the limit where the shock wave pressure step falls to zero, the shock wave speed converges to the P wave velocity. This implies that the lower limit of shock wave velocity is controlled by the acoustic velocity of the magma column, which is in turn affected by the quantity of dissolved gas in the magma, and will not be strongly affected by conduit geometry. Thus, temporal changes in the VLP/SP delay time would be driven primarily by gas content and/or vesicularity changes within the conduit. Furthermore, Gerst et al. (2013) constrained the pressure in the gas slug at the onset of the explosion to be rather low (100–600 kPa). This means that a downgoing shock wave would need to be caused by the sudden unloading of magma mass at the surface, and without any preeruptive pressurization or other high-pressure process that would be conducive to the direct explosive generation of a nonlinear disturbance. Finally, propagation speeds exceeding the magma P wave velocity are too rapid to explain the observed range of lag variability given the SP and VLP source separation, even assuming a highly convoluted conduit.

4.6 Volatile Variations

The three models above depend on dynamic magma properties that are affected by the gas content and vesicularity of the conduit magma. Short-term (tens of minutes) cyclic changes in degassing characteristics and lava lake elevation (≈1.5 m) have been observed at Erebus (Ilanko et al., 2015; Molina et al., 2015), but progressive change in gas concentrations over the time scale of the VLP-SP lag variation is difficult to constrain, given that such measurements at Erebus have only been made over relatively short time intervals. Oppenheimer et al. (2011) performed FTIR spectroscopic measurements of lava lake CO2, H2O, and other gases and found that the gas budget at Erebus is consistent with two-component mixing of passive degassing (mass CO2/H2O≈1.8) and distinct eruptive slug degassing (mass CO2/H2O≈4.3). They further conclude that the more CO2-rich slugs partially retain deeper gas signatures under stable fluxing CO2-rich that originates with the degassing of parental basanite at pressures greater than ≈4 Kbar.

Although we cannot fully discount the possibility of weeks-to-months time scale variations in gas concentrations, we suggest that appreciable changes in the gas content of the shallow magma system would create changes in eruption frequency or size. A change in eruption energy could also cause a substantial increase in the velocity of the downgoing pressure shock wave, thus reducing the lag between VLP and SP signals. However, as already noted, analysis of eruption sizes (Figure 14) shows very weak to no correlation with VLP-SP lags, and the temporal arrangement of event sizes does not show the progressive trends evident in VLP-SP lag variations.

Details are in the caption following the image
Short period (SP) explosion root-mean-square amplitude with respect to very long period (VLP)-SP timing lags at E1S.

More recently Moussallam et al. (2013) experimentally reproduced the stable Erebus phonolitic mineral assemblage studies to conclude that terminal Erebus magma is very dry, with an H2O weight percentage of less than 0.5%, which substantially reduces associated theoretical compressional wave velocity variation to the very near surface (≈50 m in the Dibble, 1994, formulation), and to substantially less than that of the Dibble (1994) end-member models. Finally, strong changes in degassing rate or gas content (including water) are thought to be unlikely due to the observed long-term chemical stability and scale of the magmatic system and its emitted gases (Caldwell & Kyle, 1994; Kelly et al., 2008; Sweeney et al., 2008).

4.7 Preferred Model

We favor a Stoneley conduit wave model, given the ability to accommodate observed lag changes with a conduit model that is consistent with tomographic and source constraints. In this model, internal geometric readjustments of the shallow conduit system produce the observed delay variations between lava lake eruptions and the initiation of magmatic flow and associated reaction and pressurization forces in the VLP centroid region. Tomographic (Zandomeneghi et al., 2013) and scattering inversion results (Chaput et al., 2012, 2015) for the upper structure of Erebus volcano indicate that the uppermost conduit system and the general structure of the summit region is indeed geometrically complex. Differences in H2O/CO2 ratios and high-frequency content measured between the lava lake and a second, ephemeral, lava lake, Werner's Fumerole, further support a complicated multithreaded conduit system in the summit region (Oppenheimer & Kyle, 2008). Closely spaced vents, suggestive of bifurcations and other complex uppermost conduit features, are common features of summit crater strombolian systems (e.g., Ripepe et al., 2007). We thus propose that the observed long-term Inner Crater region deflation may be associated internal changes in conduit system geometry occurring on week-to-month time scales. However, we recognize that this speculation is circumstantial; changes that effect VLP-SP delay and eruptive frequency could also occur within a magma-filled conduit system due to assimilation, expansion, or partial, slow collapse of structural elements that would not necessarily couple to surface morphological changes.

Geometric adjustments in the uppermost conduit system could also be driving the observed dramatic changes in eruptive frequency (Figure 4) by affecting slug formation and sequestration in the upper system, though both observations are not necessarily codependent. The strong variability of eruptive frequency for the Erebus lava lake system is contrary to its otherwise extraordinarily steady state behavior, especially compositional studies that show that the magma has been uniform for thousands of years (Iverson et al., 2014). This suggests that eruptive frequency in this system is highly sensitive to changes in conduit geometry or other physical features of the conduit system (Jaupart & Vergniolle, 1988), such as those proposed here to explain the VLP-SP delay changes (as opposed to changes in the geochemical or vesicular properties of the magma). A secondary effect of near-lava lake conduit variations might be to slightly change the exit trajectory of eruptive gas slugs and the origin of the SP source within the lava lake (Figure 8).

5 Summary and Conclusions

We observe progressive temporal timing changes between lava lake eruption initiation (characterized by the SP origin time) and associated VLP seismic signals for eruption observations spanning approximately 8 years (January 2003 to May 2011). These timing changes are quantified using waveform correlation and peak tracking as an oscillatory, repeating, and minute-long VLP signal relative to the surface eruption-generated SP signal. We note approximately ±1 s of variation from the mean for progressive changes spanning up to several months. We examine these lag changes with a number of potentially contributing factors, including eruption location within the lava lake, event size, and eruptive frequency. Within the limits of our observations, no clear association with these phenomena, other than a tenuous correlation with eruption location within the lake, can be established.

We conclude that the variable delay reflects internal adjustments of the shallow conduit system that vary the elastodynamic communication time between the surface eruption and the magmatic system at the VLP centroid depth. We conclude that the most compelling mechanism is that these changes are mediated by trapped conduit boundary (Stoneley) waves propagating through the uppermost conduit system that communicate the unroofing and associated pressure reduction of the conduit system from the lava lake to the VLP source region. For this modeling to be relevant, parts of the uppermost conduit likely must have effective (crack-like) thicknesses as thin as several m. In contrast, the deeper magmatic system geometry responsible for the extended VLP signal remains stable during this observation period. LIDAR imaging and other summit observations indicate that the lava lake system has been generally sinking into the Inner Crater since 2001, and we hypothesize that deflation-associated internal readjustments to conduit geometry may be occurring within a shallow in a dynamic zone above the VLP centroid depth. Changes within this zone that affect the VLP-SP delay could include width, inclination, lengthening/shortening/tortuosity, melting or freezing at the conduit walls, and/or partial obstruction. Such upper conduit changes might also influence the highly variable eruptive frequency of the volcano and the change in eruption size distribution, by altering the ability of the uppermost conduit to grow and sequester the characteristic (up to lake spanning) gas slugs that drive Erebus lava lake eruptions.

Acknowledgments

We thank UNAVCO and the IRIS PASSCAL Instrument Center at New Mexico Tech for facility support and field assistance on Mount Erebus. Data are archived and available at the IRIS Consortium Data Management Center. The facilities of the IRIS Consortium are supported by the National Science Foundation under Cooperative Agreement EAR-1261681, the NSF Office of Polar Programs and the DOE National Nuclear Security Administration. We thank the many Raytheon Polar Services Company individuals and groups at McMurdo and within USAP who made this field effort possible. This research was supported by National Science Foundation awards OPP-0229305, ANT-0538414, ANT-0838817, and ANT-1142083, by the U.S. Antarctic Program and its support contractors, and by New Mexico Institute of Mining and Technology Research and Economic Development. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia LLC, a wholly owned subsidiary of Honeywell International Inc. for the U.S. Department of EnergyÕs National Nuclear Security Administration under contract DE-NA0003525.