Seismic interferometry is a recently developed theory that allows for the recovery of a medium's impulse response between two points should randomly distributed sources of white noise, or equivalently, a multiply scattered equipartioned wavefield, be present throughout the medium. We exploit the extremely scattering nature of volcanic media and seismic illumination from impulsive Strombolian eruptions to extract single-station body wave Green's tensors at an unusually dense array of stations on Erebus volcano, Antarctica. We optimally rotate these Green's tensors for each notable arrival and back project them to construct a 3-dimensional scattering map of the magma-filled volcano conduit system, also corroborated by an independent active source tomography experiment using the same station distribution. This approach not only favors highly scattering media, contrarily to most conventional methods, but its passive nature (i.e., non-anthropogenic sources) allows for studies of structural temporal variability, and possible extension into real-time monitoring of active volcanoes.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 Widespread application of seismic interferometry [Claerbout, 1968] as a tool to exploit multiply scattered or equipartioned wavefields [Wapenaar and Fokkema, 2006; Snieder, 2004] has produced a wide range of ambient noise related tomography efforts on various scales [Campillo and Paul, 2003; Shapiro et al., 2005; Brenguier et al., 2007]. Many applications to date have focused on surface wave methods, while body waves studies are relatively few [e.g., Tonegawa et al., 2009; Tonegawa and Kiwawu, 2010; Roux et al., 2005; Chaput and Bostock, 2007; Draganov et al., 2009]. Conceptually, if a medium is excited by a sufficiently scattered wavefield, the correlation of wavefield components at any two points in the medium provides an estimate of the corresponding seismic Green's tensor terms between those two points. In the case of near-surface seismic noise sources such as wind and ocean waves, the wavefield and corresponding retrieval of Green's function tensors are typically dominated by surface waves. Recognized situations with suitably equipartitioned body waves are fewer, but, if present, offer complementary and unique seismic structure information.
 Erebus volcano (77.32′S, 167.10′E), is a 3794 m, approximately 1.3 My old [Esser et al., 2004; Harpel et al., 2004] persistently active stratovolcano situated on Ross Island, Antarctica [Oppenheimer and Kyle, 2008b], and was observed to be in eruption at the time of its discovery in 1841 by the James Clark Ross expedition. The volcano features a summit crater hosting one of Earth's rare long-lived lava lakes, which lies at the terminus of a magma-filled conduit system and has produced small (VEI 0) highly repeatable Strombolian eruptions with variable frequency of occurrence and size for over four decades [Giggenbach et al., 1973; Rowe et al., 1998; Aster et al., 2003; Jones et al., 2008; Aster et al., 2008] from multiple vents within an 80 m-radius inner crater. The uppermost volcano (above 3000 m), approximately 70 ka in age, is composed of highly heterogeneous layered lava flows, bomb accumulations, and minor pyroclastic deposits [Panter and Winter, 2008], and some near-surface permafrost layers, firn, and small glaciers. Volcanic activity on Ross Island arises from rift-related (and possibly plume abetted) processes near the western edge of the West Antarctic Rift System [Winberry and Anandakrishnan, 2004]. The volcano has been relatively well monitored by a multidecadal network of short period and broadband seismic stations [Aster et al., 2003], abetted by denser temporary deployments [Rowe et al., 1998; Aster et al., 2003; Zandomeneghi et al., 2011] as part of the Mount Erebus Volcano Observatory (MEVO) project. Studies of very long period (VLP) seismic signals associated with eruptive gas slug ascent and lava lake re-equilibration [Rowe et al., 1998; Aster et al., 2003, 2008] identify a VLP source centroid location that is significantly offset by several hundred meters to the northwest and lies at a depth of ∼400 m below the lava lake. The VLP sources are excited by the passage of large gas slugs that exit at the lava lake, thus suggesting that the near-summit magmatic system has significant geometrical complexity near the VLP centroid region. Shallow complexity in the magmatic system is also supported by comparative spectroscopic plume gas studies of H2O, CO2, CO, SO2, HF, HCl and OCS [Oppenheimer and Kyle, 2008a] between the principal and an episodically active secondary lava lake that suggest sequestered magmatic system elements and partially decoupled vent systems separated by just a few hundred m. Such small-scale details could in theory be resolved at higher frequencies with scattering methods, given the frequency dependence of resolution in such scenarios. Furthermore, the use of diffuse coda to emulate reflection studies through seismic interferometry is one such method which shows promise in highly scattering media.
 The theory of seismic interferometry states that the auto- or cross-component Green's function between two seismic stations, or for a single station, can be recovered if the medium of interest is illuminated by dense and irregularly spaced broadband sources of either impulsive or uncorrelated continuous character. In such cases, the Green's function can be recovered by stacking the cross-correlated contributions from all discrete sources in the medium. Features that are coherently retained in this stacking operation will correspond to stationary regions of the averaging integrand [Snieder, 2004]. The necessary seismic illumination state may also be achieved if the wavefield is sufficiently equipartitioned via strong local scattering [Lobkis and Weaver, 2001], such that source information is effectively lost, thus rendering the Green's function estimate insensitive to any source type or location. Within highly heterogeneous volcanoes, where the scattering mean free path of seismic waves is particularly short, equipartition can occur by ∼5 s or less after excitation by an impulsive source [Yamamoto and Sato, 2010].
 Here, we utilize optimally rotated six (three auto-component and three cross-component) correlations of Strombolian eruption coda from individual seismic stations to image strong seismic impedance (density time velocity) contrasts, such as will exist between magma and host rock, or between buried lava flows or intrusions and lower seismic velocity bomb, ash, or clast deposits. Features presenting a quasi-specular face to a given station, thus generating a high energy bidirectional raypath set with stationary phase, will be especially well-imaged with this technique (see auxiliary material).
2. A Seismic Image of the Erebus Shallow Magmatic System
Figure 1 shows a digital elevation model of Ross Island and the locations of the 92 three-component seismic stations used in this study. The persistent lava lake lies approximately 250 m below the crater rim summit. Figure 2 shows a series of depth slices of scattering intensity within a roughly 5 by 5 km grid extending down to sea level with a map view origin at the position of the lava lake. Active-source travel times recorded by the dense near-summit seismographic deployment shown in Figure 1 have recently been used to tomographically image strong shallow P wave velocity anomalies [Zandomeneghi et al., 2011]. The amplitudes of the low velocities imaged (corresponding to absolute velocities below 2 km/s) in the regularized tomographic inversion are consistent with near-summit magma or partial melt, and are displayed in Figure 2 (left). Coda seismograms from 92 seismic stations and over 2400 eruptions (for the permanent MEVO stations) were filtered between 1 and 8 Hz and correlated to produce single-station Green's tensors, which were then event-stacked and rotated to produce maximal amplitude arrivals (see auxiliary material). These time series were then back-projected into the volcano using rotation angle particle motion solutions and Born scattering kernels (see auxiliary material). The scattering intensity value is dimensionless as we normalize the rotated Green's functions prior to back-projection in order to avoid particularly loud stations. Recovering single station Green's tensor estimates this way amounts to performing single station reflection experiments with co-located sources and stations. Most of the eruptive energy falls between 1–8 Hz, thus setting theoretical constraints on the spatial resolution of the method (see auxiliary material). The strongest scattering and low velocity anomalies occur along two prominent trends (Figure 2, anomalies A and B), that extend to the west-northwest and north-northeast of the lava lake, respectively, and that merge into a very strong (the strongest scattering feature overall), centrally located structure (Figure 2, anomaly C) by approximately 1 km below the summit, or 750 m below the surface of the lava lake. The conduit and other features necessarily become increasingly difficult to image with increasing depth because of geometrical spreading, seismic attenuation and increased likelihood of higher-order multipathing.
 The combined scattering and tomography images indicate a complex and narrow near-surface conduit system with significant off-axis and highly inclined elements that simplifies centrally at depth (Figure 3), obviating simple models of a central conduit extending straightforwardly to the vent at this volcano. A common conceptual model for the generation of large gas slugs in Strombolian systems requires a constricted and/or inclined conduit geometry that provides locales where exsolved gas can coalesce until a critical slug size is achieved that can ascend buoyantly to eruption. The complexities of the shallow structure recovered here are consistent with such a view, with the occurrence and gas state of multiple vents in the crater, and with dispersed recent lava flows on the Erebus summit plateau that may have arisen from off-axis vents [Harpel et al., 2004]. Additional corroborating evidence for a geometrically complicated conduit system [Aster et al., 2008] is presented by the generation of near summit VLP signals during final slug ascent and subsequent lava lake refill. The centroid location of these signals, which are believed to be excited by flow through a prominent upper magmatic system constriction is also off axis, west to north-west of the lava lake, and at a depth of approximately ∼400 m below the lava lake [Aster et al., 2008], or ∼3200 m elevation. This places the VLP centroid along the azimuth of the most prominent shallow high scattering and low velocity anomaly (Figure 2, feature A).
 Several practical improvements are suggested to this scattering imaging methodology that could improve feature sharpness and integration. Notably, back projecting Green's function arrivals through a tomographically derived velocity model instead of the simple velocity model used here would be more accurate, thus highlighting the usefulness of complementary scattering and direct-ray analysis methods. The deployment of small localized arrays within a larger array would also enhance the discrimination of scattered arrival wave types in the Green's function estimates [Yamamoto and Sato, 2010], thus reducing the ambiguity of scattering kernels and improving the accuracy of scatterer mapping.
 We present a novel body wave seismic interferometry seismic imaging method and apply it to the highly scattering medium of an active volcano that is illuminated by repeating impulsive Strombolian eruptions. Using 92 short-period and broadband stations deployed during the 2007–2008 field seasons, and seismograms from ∼50 near-repeating Strombolian eruptions, we calculated single-station cross- and auto-component Green's function estimates and rotated them with time-varying rotation parameters into absolute maximal vectors with associated angle solutions. We back-projected each optimally rotated Green's function determination into the volcano, using the time-varying angle solutions to define the raypath sensitivity kernels of each arrival, to reveal a map of scattering potential for the volcanic edifice. Results show a complex near-surface conduit system which bifurcates from the west to the northwest in the first 500 m below the lava lake, and strongly centralizes beneath the lava lake near 1 km below the summit. The low dip of the shallow conduit beneath the lava lake, as little as 15°, indicates a terminal structure that might control variable gas slug sequestration and thus Strombolian eruptive frequency, and link to the centroid VLP source excited by eruptions and subsequent magmatic flow during lava lake refill [Aster et al., 2003, 2008; H. A. Knox et al., Mulityear timing variations between associated short period and very long period Strombolian Eruption Seismograms: Evidence for upper conduit geometry changes at Erebus volcano, Antarctica, manuscript in preparation, 2012]. This view of a complex uppermost magmatic plumbing system is additionally consistent with modeling [Lahaie and Grasso, 1998; Shaw and Chouet, 1991] of volcanic systems that suggests that the substantial variation in scales of observable phenomena are best modeled by a system of connected micro-chambers exhibiting nonlinear interactive behavior.
 These results further suggest new opportunities for active volcano monitoring and for the general imaging of strong subsurface seismic impedance contrasts. Given a sufficiently dense network of seismic stations and a suitably broadband natural or artificial noise source or collection of sources, such as internal natural seismicity, scattering images of an active volcano would facilitate 4-d monitoring of internal structure and time-lapse changes, especially when complemented with ancillary geophysical and geochemical data, such as provided by seismic tomography, GPS geodesy, thermal and gas emissions, and internal seismicity. Realization of this vision would require development of low-cost, robust (and perhaps expendable), telemetered sensor networks that could be readily deployed at high densities on active volcanoes.
 Funding for this study was provided by the National Science Foundation Office of Polar Programs grant 0538414. The seismic instruments and field support were provided by the Incorporated Research Institutions for Seismology (IRIS) through the PASSCAL Instrument Center at New Mexico Tech and d Data collected are 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.
 The Editor thanks the two anonymous reviewers for their assistance in evaluating this paper.