Erebus volcano, Antarctica has hosted a persistent convecting phonolite lava lake for over 40 years. The lake produces small (VEI 0–1) Strombolian eruptions resulting from gas slugs rising through the upper conduit system. High-resolution (to scale lengths of several hundreds of meters) three-dimensional P-wave tomographic velocity images were obtained to a depth of approximately 600 m below the volcano surface. Data were collected using 91 seismographs deployed over an approximately 4 by 4 km area of the summit region. Seismic illumination was provided by 12 chemical shots emplaced in shallow snow and ice boreholes. P-wave direct arrival travel-time measurements were used to invert for strong velocity anomalies (with spatial variations in Vp exceeding ±1 km/s) associated with the uppermost few km. Shallow anomalies correlate with fumarolic ice caves, a prominent radial chilled dike, and ring structures associated with the caldera rim. Conduit structures feeding the lava lake and other vents within the Inner Crater are evidently too small (e.g., less than many 10 s of meters) to be imaged under the resolution limits of this experiment. However, combined velocity and coda interferometry scattering intensity images identify near-summit regions with both low velocity and high scattering that are candidates for magma accommodation. Results indicate a nonaxisymmetric near-summit magmatic system that is likely constrained by heterogeneous structures in the uppermost volcano. The most extensive volume of near-summit magma likely resides approximately 500 m NW of the active Inner Crater vents at depths of 500 m and more below the surface.
 Erebus volcano (77.32oS, 167.10oE) located near the center of Ross Island, sits on thin (17–25 km) continental crust at the southern end of the Terror Rift within the West Antarctic Rift System (WARS) [Cooper et al., 1987; Behrendt, 1999; Finotello et al., 2011; Figure 1a]. The volcano has been attributed to a mantle plume [Kyle et al., 1992; Sieminski et al., 2003; Gupta et al., 2009] and/or to decompression melting of the upper mantle within a localized extensional or trans-tensional tectonic regime within the current active WARS margin [Ritzwoller et al., 2001; Rocchi et al., 2002]. A region of low-mantle P-wave velocities beneath Ross Island has been interpreted as an upper-mantle hotspot associated with WARS rifting [Gupta et al., 2009; Zhao2007], with an elevated upper-mantle temperature anomaly of 200–300 °K [Watson et al., 2006].
 Erebus is a polygenetic stratovolcano assembled from tephriphonolite and phonolite lavas overlying older basanite to phonotephrite lavas [Kyle et al., 1992]. The summit region of the volcano has been built atop two calderas formed over the last 95 ka [Esser et al., 2004; Harpel et al., 2004]. An older (758 ± 20 ka) prominent caldera/crater rim is exposed at Fang Ridge [Esser et al., 2004]. The summit caldera [“Old” (80 – 24 ka) and “Young” rim (25 – 11 ka)] [Harpel et al., 2004; Kelly et al., 2008] are shown in Figures 1b and 1c. The calderas are filled with a heterogeneous mixture of pyroclastic deposits and lava flows, which are seen in the crater walls to form the present summit cone. The youngest and exposed lava flows filled the caldera and formed the summit crater range between 17 ± 8 ka and 1 ± 5 ka [Kelly et al., 2008] (Figure 1c). Interbedded ice and pyroclastic fall deposits are observed in the walls of some ice caves. The summit region has numerous geothermal fumaroles that generate extensive ice cave systems and are exposed at the surface as fumarolic ice towers [e.g., Curtis and Kyle, 2011; Figure 1]. Small areas of ice-free warm (up to 65°C) ground are observed in the Side Crater and on the flanks of the summit crater (Figure 1b).
 The persistent convecting phonolite lava lake at Erebus is the exposed tip of a magma-filled conduit system. The multibranched conduit system also daylights within 10s of meters of the lava lake at several other long-lived vents in the Inner Crater region, particularly the Active vent, which produces ash-rich Strombolian eruptions [e.g., Aster et al., 2003; Csatho et al., 2008) and “Werner's fumarole” (e.g., Oppenheimer and Kyle, 2008b]. Over the last 20 years, a number of publications have examined the geophysics, petrology, tephra, and gas geochemistry of the volcano [e.g., Kyle, 1994; Oppenheimer and Kyle, 2008a] and raised interest in a better understanding of the subsurface magmatic plumbing and other internal structures. Molina et al.  used fluid dynamical models to examine convection in 4 to 10 m diameter conduit and tested the effects of crystals on convection at Erebus.
 Infrasound and radar observations of Strombolian eruptions from the lava lake show they result from rising slugs of gas with overpressures of a few atmospheres [Johnson et al., 2004; Gerst et al., 2008, Gerst et al., The first second of a Strombolian volcanic eruption: Energies, pressures, mechanisms, in revision in J. Geophys. Res., 2013]. Seismic source and other studies have revealed a dynamic lava lake/conduit system that self-reconstructs following eruptions, leading to quasi-repeating eruptive signals [Aster et al., 2003; Knox, 2012] across multiple years. Very long period (VLP) eruptive signals associated with the Strombolian eruptions [Rowe et al., 2000; Aster et al., 2003; 2008] indicate a high-connectivity conduit system with an associated VLP source centroid approximately 400 m to the WNW, and approximately 400 m below the lava lake (bll) that is excited during posteruptive refill. The extended, multiminute component of the VLP signal has been interpreted [Aster et al., 2008] in moment tensor studies to indicate a geometric feature in the near-summit magmatic system near the above mentioned location that generates strong forces during posteruptive recharge and re-establishment of lava lake equilibrium.
 To investigate the detailed internal near-summit structure of the uppermost volcano, a dense seismographic deployment and a three-dimensional (3-D) P-wave tomography survey of the summit region was conducted during the 2008–2009 Austral field season. Here, we describe the seismic experiment, the data analysis, and the tomographic inversion procedure, and present tomographic images of the seismic structure of the uppermost volcano to depths of approximately 1 km below the topographic surface. We interpret the slow seismic anomalies in part as partially molten bodies associated with the magmatic feeding system and near-surficial deposits, and high temperature intrusive bodies. High velocities and the presence of dense rocks may represent old buried crater and caldera structures. Our interpretations are complemented and corroborated by the results of a coda-wave interferometry [e.g., Lobkis and Weaver, 2001] study Chaput et al.  that utilized the same dense seismographic network, but which exploited repeating lava lake eruption signals for the requisite seismic illumination. Fundamentally, the interferometric methodology images strong seismic impedance contrasts, such as will exist at contacts between host rock and magma. Quasi-specular reflectors are detected using auto-station correlation-derived Green's function estimates derived from coda correlations. High-amplitude features within these Green's functions are subsequently localized using back-projection and stacking. Results of the scattering study included strong scattering structure, interpreted to be associated with a complex near-surface conduit system. A notable feature of this structure was the presence of substantial off-axis features, particularly several hundred m to the NNW of the lava lake at depths of several hundred meters, as well as more centralized features at depths of up to 3 km. Details on the method and results are found in Chaput et al. .
2 Seismic Experiment
 The Tomo Erebus seismic experiment [Zandomeneghi et al., 2010] was designed to obtain a high-resolution (to 100 s of m), 3-D seismic P-wave image of the shallow structure of the volcano. Eighty (not all ultimately usable) three-component short-period seismographs (referred to as the ETS network), spaced 300 to 500 m apart, were deployed over a roughly 4 by 4 km area surrounding the summit crater and lava lake to provide ray path coverage through the underlying magmatic system and encompassing structures. The distribution of stations was constrained by cliffs, snow and ice cover, and by rock outcrops. The snow coverage allowed good snowmobile access over the summit plateau region, which assisted in rapid deployment and retrieval of a dense network of seismometers (Figure 1b). The ETS stations were equipped with three-component short-period Sercel L-28 sensors (4.5 Hz) and GPS timing. Continuous recording was achieved using REF TEK 130 data loggers at a sampling rate of 200 Hz. The imaging experiment also incorporated data from the flanks and around the summit from a 23-station array (referred to as the ETB network) that consisted of three-component, Guralp CMG-40 T intermediate-period seismometers (natural frequency 30 s) and REF TEK data loggers continuously recording at a sampling rate of 100 Hz. The ETB stations were deployed the year prior to the tomography experiment, during the 2007–2008 field season, allowing for enhanced observations of the volcanic seismicity over two field seasons. Data from the permanent Mount Erebus Volcano Observatory network of near-summit intermediate-period stations [Aster et al., 2004] were also incorporated in this study. These three networks together ultimately produced a constellation of 91 usable stations for tomographic imaging (Figure 1b).
 Tomographic seismic sources consisted of 11 chemical blasts detonated on the summit plateau within the dense seismic network. A more distant shot located at the head of Fang Glacier about 4 km NNE of the lava lake generated deeper-diving rays (Figure 1b). Charges consisted of 75–300 kg of Ammonium Nitrate and Fuel Oil packed in 10 cm diameter holes drilled 5.5–20 m into the snow/ice cover [Zandomeneghi et al., 2010] (see Table S-1 in the Supplement for shot details). These shots produced network-spanning wavefields with a typical frequency content of 1 – 10 Hz (Figure 2a) from which first-arrival time estimates were made for P-wave tomography.
3 Methodology and Data
 We applied the tomographic method of Toomey et al. , originally developed to image the strongly seismically heterogeneous mid-ocean ridge environment, to invert P-wave travel times for 3-D velocity structure [Barclay et al., 1998]. The method is computationally efficient and is well suited for the rough topography and strong velocity contrasts encountered in terrestrial volcanic systems. Forward modelling utilizes least time ray tracing [Moser, 1991] through a velocity model defined on a 3-D nodal grid. The grid is conformably mapped to the topographic surface by shearing each column of nodes vertically. Velocity perturbations are parameterized on a similarly registered 3-D grid. The LSQR algorithm is used to least squares minimize the residual vector misfit between observed and calculated travel times. Forward and inverse calculations are successively performed for up to five iterations. In this study, we used ray tracing and perturbation grid intervals of 50 m and 100 m, respectively. The topography [Schenk et al., 2004] was parameterized on a 0.0005° grid, corresponding to 56 m in latitude and 12 m in longitude.
 The inverse problem consists of solving linearized sets of equations d = Gm, where d is a vector of travel-time residuals with respect to the current model, m is a vector of model slowness perturbations from that model, and G is a matrix of partial derivatives of travel times with respect to slownesses. The system of equations is rank deficient, and so it is augmented with regularizing constraint equations that specify prior assumptions about initial model and spatial roughness. Our approach follows that of Toomey et al. , where we set up the regularized problem as a minimization of
where Cd is the data covariance matrix, Cp is the prior model covariance matrix, and Ch and Cv are, regularizing covariance matrices for horizontal and vertical roughness, respectively. The Lagrangian multipliers λp.h,v control the trade-off between data misfit and regularizing constraints and were varied to select the smoothest and most damped velocity perturbation model that minimized the data misfit while yielding absolute P-wave velocities that were not appreciably less than the minimum P-wave velocities of near-surface Erebus phonolitic magma as estimated by Dibble et al. . In addition to exploring the regularization parameters, the stability of the inversion was examined by varying parameterization details, including the ray tracing parameters and inversion model grid spacing, and in tests using only subsets of the travel-time data.
 The data set comprised P-wave travel-time measurements from 12 sources, recorded at 100 stations selected for good data quality. Following manual picking and assignment of a priori picking uncertainties (weights), and to optimize the dimensions of the inverted volume considering the spatial distribution of ray paths, travel times recorded at 91 stations were ultimately used in the tomographic inversion. These sources and receivers encompassed an area of 4.6 by 5.4 km in NS and EW directions, respectively, and included the Fang Ridge explosion together with the 11 summit area sources. Travel times had an average estimated P-wave arrival uncertainty of 25 ms.
 We assessed the importance of the initial model on the final results by testing a range of initial 1-D models that were consistent with a priori knowledge of the geology, previous geophysical investigations [Dibble et al., 1988; Dibble et al., 1994], and the results of a 2-D large-scale model for Ross Island obtained during the Tomo Erebus [Zandomeneghi et al., 2010; Maraj, 2011].
 Regardless of the choice of starting model, the overall pattern of strong lateral velocity perturbations was stable, except for minor changes in the amplitudes and positions of lesser velocity anomalies. For all considered velocity models, the deepest ray turning depth was less than 2 km below the topographic surface (bs). For the final 3-D inversions presented here, we used a starting model with a P-wave velocity of 2.2 km s−1 at the surface and a constant gradient of 1.6 s−1 to the base of the model, at 2 km bs (Figure 2b). Although the starting model was somewhat slow relative to the inversion mean for depths below approximately 0.5 km, most ray paths are above this depth, and most of the variance reduction in the inversion thus arises from the strong lateral heterogeneity of the final model (Figure 3).
4 Resolution and Results
 The uncertainty and resolution of a tomographic image are affected by uneven ray distribution, uncertainties in travel-time picks, regularization choices, and ray tracing limitations. Recovery tests of synthetic features were used to assess the reliability of the velocity anomalies obtained in the final model. We calculated travel times for the same experimental source/receiver distribution through a variety of synthetic velocity models, including 3-D checkerboard patterns. In these tests, we added Gaussian white noise to the synthetic travel times with the same time standard deviation (25 ms) as the estimated arrival time pick uncertainties. These data were then inverted using the same regularization parameters and model parameterization as used for the true inversion.
 In the checkerboard tests (Figure S1), we used a synthetic model with alternating, spatially smoothed prisms of 0.2 km/s positive and negative velocity perturbation, with lateral and vertical lengths of 0.5 km, superimposed upon the initial 1-D model. The test shows good recovery of the anomalies in the area surrounding the crater, especially in the NW sector. The resolution degrades with depth, and there was no usable model recovery by approximately 0.75 km bs (Figure S1). The overall amplitude of recovered anomalies in this test ranged between 25% (of the original minima) and 180% (of the original maxima). This asymmetric recovery is typical of ray-based tomographic methods due to the refraction of rays away from the lowest velocity regions, which are thus less well resolved [e.g., Aster et al., 2012]. The area south of the crater is only partially recovered due to the limited ray coverage arising from difficult station/shot deployment conditions. The recovered 0.2 km bs checkerboard shows no significant artefacts except for an unsurprising noticeable streak of high velocities extending SW of the Fang Ridge shot due to the lack of crossing rays in that region. The resolution tests show a generally good recovery of synthetic anomalies above the network and allow us to express confidence on the reliability of primary features within the central area of the image down to at least 0.4 km bs.
 Final tomographic images are displayed in Figures 3, 4, and 5. These models include a number of distinct shallow higher and lower velocity anomalies with amplitudes of up to 1 km/s. Although this level of velocity heterogeneity is large, it is within the range reported for shallow tomographic imaging of active volcanoes such as Etna [e.g., Patané et al., 2003; 2006]. To facilitate interpretation and more easily associate results with surface features, we plot the results in two ways. Figure 3 shows velocity structure at “draped” depths at 100 m depth interval from 0.1 to 0.6 km bs. Figure 4 shows slices at depths bll and provides comparison with the scattering imaging obtained using the same network but which employed Strombolian eruption coda interferometry and signal stacking [Chaput et al. 2012]. Representative cross sections of the velocity structure are shown in Figure 5.
 The root-mean-squared data misfit for the final model was 32 ms, corresponding to a variance reduction of 90% with respect to the 1-D (depth-dependent) starting model (Figure 2b). We note highly heterogeneous lateral velocity structure, with maximum peak-to-peak amplitude variations of ~1.6 km s−1 over distances of ~500 m throughout the imaged volume. Areas with low or no ray coverage, where the derivative weight sum [a distance-weighted measure of the ray coverage near each perturbation node; Toomey and Folger, 1989] was less than 0.05, are masked out. Ray coverage is densest in a ~4 by 3 km wide region, surrounding, and extending to up to 0.6 km bs within a region to the NNW of the crater.
5 Interpretation and Discussion
 Crustal P-wave velocity variations in volcanically and tectonically active environments can be caused by the juxtaposition of different rock types, variations in density, porosity and microstructures, fracturing and geochemical alteration, temperature contrasts, and the presence of magma, fluids, and gases. At Erebus volcano, the lava lake indicates the presence of a persistent magma supply and suggests that structural, thermal, and geochemical changes arising from the long-term presence of shallow magma should strongly influence near-summit velocity heterogeneity. These factors can combine with other causes related to magmatic activity, such as a possible deeper geothermal system. Although the lava lake system indicates the presence of well-connected and persistent macroscopic magmatic conduit system at Erebus, the upper volcano may also accommodate partial melt zones, within which the seismic wave speed is influenced by the cuspate and other small-scale geometric characteristics of the partial melt volume, in addition to partial melt proportion [e.g., Johnston et al., 1995; Hammond and Humphreys, 2000].
 Other anticipated causes for velocity anomalies include density and elastic differences between high porosity pyroclastic deposits (e.g., volcanic bomb and ash) and lava flows [Panter and Winter, 2008] and/or solidified intrusions. At shallow depths, there are local variations in ice cover thickness, occurrence of geothermal heated areas and permafrost. The existence of widespread fumarolic activity in the form of ice caves and ice towers over the summit plateau, and areas of ice-free warm ground, indicates magmatic degassing over a significant portion of the summit plateau [Curtis and Kyle, 2011]. The rich structural history of the volcano is undoubtedly an important contributor to the present seismic velocity structure in addition to the effects of the current near-summit magmatic, geothermal, and degassing systems. The interpretations and discussion below consider the limited depth resolution of the P-wave imaging, which suggests that strong anomalies in the summit region of good ray path coverage with lateral spatial wavelengths of several hundred m should generally be resolvable to 0.4 km below the topographic surface, and possibly as deep as 0.5 km, but with degraded amplitude recovery (Figure S1).
5.1 Low-velocity Features
 Five prominent shallow low P velocity anomalies (a, b, c, g, and h) appear in the 0.1 km bs image (Figure 3a) at various azimuths from the lava lake. There are some correlations between these anomalies and high-scattering features observed in scattering intensity images, particularly features b, c, and h (Figure 4). Strong seismic impedance contrasts are expected at the boundaries between magma and host rock, and regions of both highest scattering intensity and low velocity are suggestive of the presence of magma.
 Anomaly a is ~2 km to the east of the summit crater. It persists strongly to 0.4 km bs on the tomography (Figure 3d) but can be seen to a depth of 0.6 km. This anomaly is not apparent in the scattering images (Figure 4). Although there is no surface expression of the anomaly (it underlies a major snowfield), fumarolic ice caves and ice towers occur in the near vicinity. The persistence of the anomaly and the somewhat circular structure suggest that it could be a small, hot, and recent intrusion. There is no evidence for any recent lava flow vents in the vicinity of the anomaly.
 Anomaly b is ~1.25 km NW of the Main Crater. It is a small circular feature at 0.1 km bs (Figure 3a) but disappears by 0.4 km bs as resolution degrades (Figure 3d and Figure 5). The region of anomaly b is associated with a high degree of seismic scattering that connects to the deepest scattering anomalies from Chaput et al.  (Figure 4). The shallowest portions of this anomaly sit close to Tramway Ridge (Figures 1b and 1c), which includes a uniquely prominent ice-free area of gently sloping warm ground, located between 3350 and 3400 m elevation. The area has significant gas emission, and its well-developed soil has high surface temperatures reaching over 65°C at 15 cm depth [Area 1 of Lyon and Giggenbach, 1974]. The site is also of special interest for its diverse thermophilic biological communities [Soo et al., 2009] and lies near the toe of a young Tramways lava flow [Harpel et al., 2004; Figure 1c]. The surface morphology of the Tramway flow suggests that it is one of the youngest flows in the summit area of Erebus, and it is possibly the apparent lava flow that Ross  reported when the volcano was discovered as an N-side, near-summit incandescence observed from the southern Ross Sea in 1841. Although the area of warm ground may be related to the associated lava flow, the P-wave tomography indicates that this anomaly extends to at least several hundred meters, and the scattering tomography suggests that it extends even deeper. We therefore interpret this seismic anomaly and its associated surface heat as arising from a shallow intrusion.
 Anomaly c is ~1.5 km NE of the Main Crater. It is imaged to 0.4 km bs in the P-wave velocity image (Figure 3d, Figure 4, and Figure 5) and to elevations 0.5 km bll in the scattering image (Figure 4). A strong velocity anomaly sign change near 0.4 km bs suggests that it may be shallow. In the 50 m bll scattering image (Figure 4a), anomaly c extends as a linear feature to the NNW. Numerous ice towers and ice caves occurring at the N end of this linear feature (Figure 1b) are amongst the most CO2 rich in the summit area [e.g., Harry's Dream; Wardell et al., 2003; Figure 1c]. Since the presence of high CO2 implies the presence of near-surface degassing magma, anomaly c is interpreted as a small shallow magmatic feature.
 Anomaly g is ~0.75 km SE of the crater (Figure 3a) and lies near the boundary of the model. There is a weak indication that the anomaly persists to 0.4 km bs (Figure 3d). Although the anomaly is not apparent in the scattering intensities, there are several major ice caves systems (Mammoth/Cathedral ice caves) in the area. This poorly resolved feature may thus be an additional magma intrusion that provides the heat and gases responsible for forming these ice caves.
 Anomaly h is ~1.5 km SW of the lava lake and, like g, lies near the edge of the velocity model. Resolution issues with the seismic tomography do not allow adequate coverage below approximately 0.3 km bs (Figure 3c), but the feature is colocated with a high-scattering feature to a depth approximately 0.45 km m below the lava lake in scattering imaging (Figure 4). The anomaly is situated below two of the largest fumarolic ice towers in the summit area. Sauna ice cave, which underlies the largest ice tower, is also the hottest (over 30°C) observed and is associated with an old lava tube. Given that this is a site of significant heat and gas transfer, we conclude that anomaly h may thus indicate yet another small magma body or hot intrusion.
5.2 High-velocity Features
 High-velocity structures may represent buried lava flows, cold intrusive bodies, or compact relic caldera structures. Three prominent shallow high-velocity seismic anomalies (d, e, and f) are seen on the 0.1 km bs image (Figure 3a). At ~0.4 km bs (Figure 3d), these remain present, but three additional anomalies labelled i, j, and k are seen. These features extend to the deepest extent of the model, but their deeper geometries are strongly affected by reduced resolution at ~0.6 km bs (Figure S1). None of these high seismic velocity anomalies show up as clearly colocated scattering intensity regions, with the modest exception of f at the shallowest depths (Figure 4). Anomalies i and j (Figure 3c) lie along an elliptical trend that extends toward the E side of the older rim and could represent the buried extent of the younger (25–11 ka) of the two partially exposed near-summit caldera rims [Figure 1; Harpel et al., 2004; Csatho et al., 2008]. This trend also tracks through the prominent ice tower trend on the N side of the Main Crater (Figure 1b), showing that these buried features exhibit structural control on the locations of present-day degassing features (another prominent association is obvious in the large cluster of fumaroles at the eastern side of the older caldera rim, which may be nearly colocated with the eastern portion of the newer caldera rim as suggested by this study, thus further enhancing outgassing paths from the magmatic system. Anomalies e and k lie near the N edge of the older of the two rims (Figure 1).
 The intermediate to high-velocity anomaly d follows a line of fumaroles that extend along Ice Tower Ridge and through the Side Crater [Curtis and Kyle, 2011; Figure 1b], which follows the trend of the Side, Western, and Main Craters [Figure 1; Panter and Winter, 2008]. The associated topographic ridge extends radially SW from the lava lake out to the exposed edge of the younger caldera. This linear feature and alignment is consistent with an underlying near-vertical dike. A chilled dike with this orientation is exposed in the SW wall of the Side Crater and is otherwise apparent in the topography and LIDAR imaging [Csatho et al., 2008]. We hypothesize that this dike corresponds to anomaly d and that the Ice Tower Ridge fumaroles are linked to magmatic degassing along the trend of this feature.
 The prominent shallow high-velocity anomaly f, also noted in Figure 4 as a strong near-crater scatterer, may be a solidified intrusive body associated with the emplacement of the young (6 ± 2 ka) collocated South lava flow or with the southern part of the younger caldera rim [Harpel et al. 2004; Figure 1c].
5.3 Conduit System Implications
 Directly imaging and/or otherwise constraining the geometry of the magmatic conduit feeding the lava lake was one of the primary goals of this study. However, we observe no clear conduit structures that can be linked directly to the vents. It is likely that most or all through-going conduit features at Erebus have characteristic widths of at most 10s of meters. This implies that shallow magma filled conduit system elements feeding the lava lake system are narrow enough to elude the limited resolution of this experiment (e.g., less than several hundred m in radius). This is not especially surprising; Calkins et al.  based on heat flux estimates and lava lake temperatures suggested the conduit diameter was about 4 m whereas others estimates based on visual and video observations have suggested the conduit was 5–10 m in diameter [Aster et al., 2003; Dibble et al. 2008; Oppenheimer et al. 2009; Molina et al., 2012].
 The imaging of narrow low-velocity zones is a generally difficult proposition in seismic tomography at all scales. This is due to the potentially small sizes of such structures relative to the wavelength of seismic illumination, which leads to wave front healing that can result in much earlier (faster) arrivals than predicted by the infinite frequency approximation [e.g., Lees, 2007; Aster et al., 2012], resulting in lessened or totally absent low-velocity features in seismic tomography. The effects of healing are convolved with the smoothing effects of regularization, which reduce high wavenumber content in the images and commonly produce under-recovery of anomaly amplitudes, as demonstrated by synthetic examples. However, the seismic structure of Erebus, co-interpreted in association with the scattering intensities, allows for more robust constraints to be placed on the location and geometry of low-velocity anomalies. We contend that some of these anomalies are associated with the upper conduit system feeding the lava lake and other eruptive vents in the Inner Crater.
 First, we note that prominent high velocities (anomalies f, i, and j) occur in close proximity to the lava lake and other vents of the Inner Crater. These structures, which we suspect are solidified intrusions and/or compact high-velocity buried rim structures, could inhibit and/or redirect the formation of significant conduit features in these areas. A number of earlier Strombolian seismic source studies of forces generated during eruptions and during posteruptive refilling also provide some insight into the location of conduit components. Centroid moment tensor inversions of posteruption VLP signals associated with recharge of the lava lake system [Aster et al., 2008] suggest an inclined sub-lava-lake conduit dipping to the N and W of the lake, towards the topographic center of the upper volcano and along the azimuth of anomaly b. The single force functions in this inversion study show corresponding WNW dipping components consistent with posteruption reaction forces arising from magma accelerating through an inclined near-summit conduit. Also, the single force and force couple centroid location itself is located several hundred m from the lava lake along this azimuth. We conclude that the most likely locations for significant near-summit magma-filled conduit components are away from the Main Crater towards the N and NW. In this case, the tomography is likely providing an underestimate of the true negative velocity anomaly due to regularization, and the velocity values themselves may represent a bulk average between host rock and conduit components.
 In the above interpretation, the most definitive imaged elements of the shallow magmatic system are anomalies b and c. To further localize probable regions hosting subvolumes of magma in the upper volcano, we produced isosurfaces for low velocity (∆Vp < −1 km/s) and high-scattering coefficient [S > 46, Chaput et al., 2012, Figure 3]. The common volume satisfying these constraints (Figure 6) indeed resides NW of the Main Crater and beneath the summit plateau, where it persists to depths of ~0.6 km bs (to elevations near 2500 m). The system is not generally resolved below this in the travel-time tomography because of the paucity of deeper ray paths (Figure S1). However, scattering images suggest that anomaly b continues below this depth, where it centralizes into a large high-scattering anomaly (I in Figure 4) that has been interpreted by Chaput et al.  as a possible larger dimension centralized conduit by approximately 650 m bll.
 It is interesting that the plan view of some of the most prominent imaged low-velocity anomalies (e.g., b and c) trends towards circularity, as opposed to linearity, which would be the case for steeply dipping dike emplacement. For this active lava-lake sustaining magmatic system, this suggests countervailing trends towards circular plan view structures in low seismic velocity and potential active subvertical transport zones. Another possibility is sill-like structures emplaced in a vertical least compressive stress regime rather than vertical cylinders. These structures would be difficult to discriminate here because vertical resolution is poorer than horizontal resolution. A further possibility is that at least some of these features are not magmatic, but are instead geothermally altered low-velocity zones of laterally limited high permeability. Such zones could be perched above deeper components of the magmatic system.
 The above results, combined with prior studies, support the contention that the shallow conduit system exhibits a tortuous geometry, potentially with a number of side lobes. This presents implications for both the convective system that prevents the lava lake from establishing a solid surface and for the parcelling of eruptive gas slugs that drive Strombolian explosions from the lava lake and other vents. Combining results from this study with previous studies, we suggest a unified interpretation that: (1) The combined results of travel-time and scattering imaging methods provide strong support for the contention that the largest near-summit volume of magma resides in a volume that is offset to the NW from the active vents by approximately 500 m and that presents a top surface near 500 m below the topographic surface. (2) Corroborative with earlier studies (Aster et al. [2003; 2008; Chaput et al. 2012], a complex shallow conduit system is responsible for modulating a high degree of variability in the frequency of Strombolian eruptions at the volcano over the last 40 years [e.g., Dibble et al., 2008; Knox, 2012]. Small variations in geometrical conduit system geometry/tortuosity can thus provide control on the formation of large pre-eruptive gas slugs and thus on eruptive frequency. An underlying conduit complex with multiple surface threads is additionally consistent with the presence of sporadically active additional magma-filled vents in the active Inner Crater such as Werner's fumarole, located approximately 50 m SW of the lake, that display independent intermittent eruptive behaviors [Aster et al., 2003; Calkins et al., 2008] and differ in their gas geochemistry [Oppenheimer and Kyle, 2008b; Oppenheimer et al., 2011]. Finally, the lava lake eruptions and their associated VLP signals [Rowe et al., 2000; Aster et al., 2003; Aster et al., 2008] are inferred to require persistent geometrical conduit system complexities to consistently generate repeatable seismic moment rate tensor components in response to gas or magma flow [e.g., Chouet et al., 2003].
 A P-wave seismic tomographic image of the uppermost approximately 600 m of Erebus volcano has been acquired with resolution as fine as ~300 m. The strong velocity anomalies represent both low-velocity thermal and/or magmatic features as well as high-velocity features that may be cooled intrusions of crystalline phonolite and/or buried remnants of postcaldera formation structures. Corroboration between seismic tomography and scattering intensity imaging supports the contention that heterogeneity and complexity are substantially driven by the presence and thermal influences of the shallow magmatic system. However, the uppermost lava lake conduit system components are too small to be resolved by travel-time tomography through to the surface vents, and low-velocity regions in the image could be hydrothermal zones or bulk velocity measurements across for regions with magmatic structures that are substantially below the resolution limit. Scattering tomography, which reveals strong seismic impedance contrasts, thus offers a useful complement to the travel-time imaging. Joint imaging of a distinct low velocity and high-scattering coefficient volume indicates a probable region of substantial near-summit magma less than 1 km NW of the lava lake to depths of at least 0.5 km below the volcano's surface. This suggests that the feature, strongly associated with velocity anomaly b, is a principal element of the summit region magmatic system. This region of inferred significant near-summit magma is also consistent with moment tensor estimates of the posteruptive VLP source centroid location, and with the inferred inclination azimuth of the near-summit conduit system in that region [Aster et al., 2008]. Because the posteruptive VLP source is excited during advective recharge of the lava lake, this suggests magmatic linkage between the lava lake and the NW low-velocity/high-scattering anomaly.
 Our overall conceptualization of the summit region of Erebus, incorporating results from this and prior studies, is of a multithreaded shallow volcanic conduit system that features constrictions, highly inclined elements, fracture interconnections, and multiple volumes that link to the several observed surface vents through narrow structures. The mapped extent of fumaroles in the summit region shows some association with near-summit velocity features that represent regions of enhanced hydrothermal or degassing connectivity. The prominent topographic and fumarole trend of Ice Tower ridge is associated with a moderate to high-velocity SE trending feature that is subparallel to an exposed dike and to a dike-like LIDAR feature. Several higher velocity anomalies are consistent with geomorphic, LIDAR, and geologically constrained partially exposed caldera rims contrasting with lower velocity fill, and to the aforementioned dike.
 Seismic images of magmatic systems on active volcanoes at resolution scales as fine as a few hundred meters, especially as validated by both tomography and scattering imaging, are presently few [e.g., Brenguier et al., 2006]. Using dense short-period seismic deployments of this type, and given sufficient seismic illumination, we show these two techniques can be jointly applied to provide complementary constraints that improve confidence in image interpretation. This study also corroborates the contention that the shallow geometry of the feeding system likely plays a major role in modulating eruptive activity in Strombolian systems (and that changes in deeper degassing are correspondingly not required). We suggest that small changes in a complex geometry dynamic conduit zone in the upper few hundred meters could explain large changes in eruptive frequency and variable time delays between surface eruptions and magmatic recharge of the lava lake [Knox, 2012]. These results also provide insights into the nature of lava lakes, suggesting that an extensive conduit system can be associated with the generation and stability of persistent lava lakes, and in maintaining magmatic rheology and necessary heat flux [Harris, 2008; Molina et al., 2012].
 This study demonstrates the feasibility of further advancements in imaging and understanding the structural complexity of active volcanic structures. Even under harsh environmental conditions, such as at Erebus, moderately large numbers (approaching 100 or more) of portable seismographs can now be efficiently deployed with current and upcoming technology. With these denser deployments, complementary imaging techniques utilizing both natural (e.g., eruptive or internal seismicity) and artificial (e.g., chemical explosion) sources can be integrated to reveal increasingly detailed internal volcanic structures. Longer-term deployments of dense networks, or the monitoring of a smaller number of stations within the context of a structural model constrained by a temporary dense deployment, pose the further prospect of time-dependent monitoring of active volcanic systems that approach the resolution limitation of seismic methods.
 We thank the Ice Core Drilling Office at the University of Wisconsin and drillers Terry Gacke, Jay Kyne, and Matt Zimmerer (NMT) for providing the shot holes. Galen Kaip (University of Texas El Paso) and blasters from Raytheon Polar Services Company executed the shots. Further assistance in the field during the 2008–2009 short-period seismometer deployment was provided by Tim Burton, Omar Marcillo, and John Wood. Catherine Snelson and Sandra Saldana provided early assistance to the shot program. Support from the Fleet Operations office of Raytheon Polar Services Company ensured a plentiful supply of explosive. Fieldwork on Erebus was critically supported by Petroleum Helicopters International (PHI) and Helicopters New Zealand. The IRIS PASSCAL Instrument Center at New Mexico Tech provided essential facility support, and Instrument Center Staff Pnina Miller and Bob Gretschke assisted greatly in the field. All data and associated metadata are archived at the IRIS Data Management Center as an assembled data set (09–015). IRIS facilities are supported by the National Science Foundation (NSF) under Cooperative Agreement EAR-0552316, the NSF Office of Polar Programs and the Department of Energy National Nuclear Security Administration. This research was supported by NSF awards ANT-0538414 and ANT-0838817. This project would not have occurred without the encouragement, suggestions, and enthusiasm of Tom Wagner, former program manager for Polar Earth Science at the NSF Office of Polar Programs. We thank the associate editor, as well as Michael West, and an anonymous reviewer for thoughtful and valuable comments during the review process.