Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
 We present a velocity model of the onshore and offshore regions around the southern part of the island of Hawaii, including southern Mauna Kea, southeastern Hualalai, and the active volcanoes of Mauna Loa, and Kilauea, and Loihi seamount. The velocity model was inverted from about 200,000 first-arrival traveltime picks of earthquakes and air gun shots recorded at the Hawaiian Volcano Observatory (HVO). Reconstructed volcanic structures of the island provide us with an improved understanding of the volcano-tectonic evolution of Hawaiian volcanoes and their interactions. The summits and upper rift zones of the active volcanoes are characterized by high-velocity materials, correlated with intrusive magma cumulates. These high-velocity materials often do not extend the full lengths of the rift zones, suggesting that rift zone intrusions may be spatially limited. Seismicity tends to be localized seaward of the most active intrusive bodies. Low-velocity materials beneath parts of the active rift zones of Kilauea and Mauna Loa suggest discontinuous rift zone intrusives, possibly due to the presence of a preexisting volcanic edifice, e.g., along Mauna Loa beneath Kilauea's southwest rift zone, or alternatively, removal of high-velocity materials by large-scale landsliding, e.g., along Mauna Loa's western flank. Both locations also show increased seismicity that may result from edifice interactions or reactivation of buried faults. New high-velocity regions are recognized and suggest the presence of buried, and in some cases, previously unknown rift zones, within the northwest flank of Mauna Loa, and the south flanks of Mauna Loa, Hualalai, and Mauna Kea.
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.
 Geophysical measurements, such as gravity and seismic velocities, have been used to probe the internal structures of the Hawaiian volcanoes [Fiske and Jackson, 1972; Thurber, 1984; Okubo et al., 1997; Kauahikaua et al., 2000]. In particular, seismic tomographic studies have yielded improved constraints on the 3-D velocity structure of the island. Passive source seismic tomography based on local earthquakes [Thurber, 1984; Okubo et al., 1997; Haslinger et al., 2001], have resolved velocities mainly around the seismically active volcanoes, Mauna Loa and Kilauea. A recent active source seismic tomography study inverted first-arrival traveltimes from offshore air gun data recorded at on-land seismic stations, to obtain a velocity model around the offshore and coastal regions of Kilauea, and parts of Mauna Loa and Loihi [Park et al., 2007]. The seismic sources for each of these tomography studies are restricted spatially, i.e., onshore for the earthquakes and offshore for the air gun shots. Thus, the resulting velocity models are also spatially limited, making it difficult to fully resolve the regional structure of the island and the relationships among the volcanoes that compose it.
 To extend the spatial coverage and improve model resolution, we adopted a method to jointly invert active and passive source seismic data [Ramachandran et al., 2005], incorporating local earthquakes into the offshore air gun data set to carry out a 3-D simultaneous tomographic inversion. The addition of earthquakes within and beneath the island yields a new velocity model with wider coverage of the onshore region than the previous model [Okubo et al., 1997; Park et al., 2007], and also extends the depth of the model (up to 35 km below sea level). The new model encompasses most of Kilauea and Mauna Loa volcanoes, as well as Loihi seamount, and also reaches the southern and southeastern edges of Mauna Kea and Hualalai volcanoes, respectively. Our new velocity model reveals several previously unknown or uncertain volcanic features within the edifices of Hawaii, which, along with the seismicity patterns and recent geodetic observations, invite a reconsideration of how the island volcanoes grew and interacted in the past, and how their volcano-tectonic evolution has influenced their current configurations.
2.1. Growth of Hawaiian Volcanoes and Rift Zones
 Hawaiian volcanoes are constructed by magmas derived from the mantle, rising through and erupting onto the slowing moving Pacific plate [Stearns, 1946; Clague and Dalrymple, 1987]. Each volcano exhibits several stages of growth, characterized by changes in lava chemistry. The early submarine stage is noted for modest eruptions of alkalic lavas, followed by submarine to subaerial tholeiitic lavas during the shield-building stage, and then a transition back to alkalic lavas near the end of their growth [Macdonald and Katsura, 1964; Macdonald, 1968; Clague and Dalrymple, 1987].
 As a volcano grows, rising magma perches in a magma chamber from which eruptions originate [Tilling and Dvorak, 1993]. Olivine cumulates precipitate within these magma chambers, and thus accumulate over time beneath the active volcanic centers, e.g., the summits and rift zones [Hill and Zucca, 1987; Clague and Denlinger, 1994]. Continuous magma intrusions and eruptions build up the volcanic edifice during the voluminous shield-building stage. After the magma supply peaks, the growth rate of the volcano wanes, and eventually, volcanic activity ceases. As the overriding Pacific plate moves northwestward, a new volcano usually forms to the southeast of the older one [Stearns, 1946; Clague and Dalrymple, 1987]. This growth sequence results in the progressive overlap of younger volcanoes upon older ones [e.g., Moore and Clague, 1992], with the growth of the later ones influenced by topographic stresses and buttressing of adjacent edifices [Fiske and Jackson, 1972]. Concurrent with growth, large-scale volcanic landsliding modifies the slopes of the volcanoes [e.g., Moore et al., 1989]. The final island configuration is the product of all of these effects.
 On the Island of Hawaii, the volcanoes of Mahukona, Kohala, Hualalai and Mauna Kea have completed their shield-building stage and are now extinct or dormant, and Mauna Loa, Kilauea and Loihi are still within their active shield-building phases (Figure 1) [Moore and Clague, 1992]. Interpretations of each volcano's evolution are based largely on lava geochemistry and surface geology [McDougall, 1964; Macdonald, 1968; Wolfe and Morris, 1996]. However, because the volcanoes are gradually buried by younger lava flows, their past histories are often difficult to discern.
 One way to understand how Hawaiian volcanoes developed and interacted with each other is to define their past configurations, in particular, of volcanic rift zones. Each volcano is characterized by topographic ridges representing rift zones along which frequent dike intrusions and eruptions occur. For an ideal isolated volcanic edifice, three rift zones radiate from the summit at ∼120° apart [Fiske and Jackson, 1972]. However, when volcanoes develop close to each other, the growths and orientations of their rift zones can change from this typical pattern, influenced by the gravitational stress field caused by neighboring volcanoes [Fiske and Jackson, 1972]. In particular, as a propagating rift zone from one volcano approaches an adjacent volcano, it might be blocked, and prevented from extending further due to the resulting compressive stress field [Ryan, 1988; Dieterich, 1988]. Rift zones can also undergo much more dramatic changes due to flank failures [Lipman et al., 1988; Morgan et al., 2003], which may modify the orientation of the least compressive stress and thus, the preferred orientation of dike intrusions [Ryan, 1988]. Also, the rift zone could be abandoned and buried, thus lack the distinctive ridge-like features produced by active rifts zones in the present morphology of the volcano.
2.2. Seismicity and Velocity Structure of Hawaiian Volcanoes
 Earthquakes are relatively common within and beneath the Island of Hawaii, in particular around the active subaerial volcanoes, Mauna Loa and Kilauea (Figure 1). The dense seismographic network around these active volcanoes has been established to monitor ongoing seismicity, providing us with an enormous data set [e.g., Klein et al., 1987; Wright and Klein, 2006; Okubo et al., 1997]. Shallow seismic events around Kilauea are usually located near the summit and along its east rift zone (ERZ) and southwest rift zone (SWRZ) from 1 to 4 km depth below sea level (Figure 1) [Klein et al., 1987]. Deep events occur beneath Kilauea's south flank due to decollement slip along the Hilina seismic zone, from 6 to 13 km depth [Klein et al., 1987]. Around Mauna Loa volcano's southeast flank, relatively shallow to intermediate depth (4–11 km) earthquakes, related to near-vertical strike-slip faults, are observed along the Kao'iki seismic zone, and deep flank earthquakes associated with slip along a basal decollement occur around both Kao'iki and Hilea seismic zones from 6 to 12 km depth [Endo, 1985].
 Early 2-D onshore-offshore seismic refraction surveys yielded low-resolution velocity structure profiles for the volcanoes [Hill and Zucca, 1987]. These studies resolved high-velocity cores beneath the summits and rift zones of Mauna Loa and Kilauea, and low-velocity features along the offshore flanks. Subsequently, the abundance of earthquakes has allowed the inner structure of these volcanoes to be probed using passive source seismic tomography [Thurber, 1984; Okubo et al., 1997]; the 3-D seismic tomography improved the geometry and resolution of the high-velocity features beneath the summits and rift zones, interpreted to consist of intrusive rocks, such as magma cumulates and dike swarms [Thurber, 1984; Okubo et al., 1997; Haslinger et al., 2001; Park et al., 2007]. The low velocities associated with the Hilina and Kao'iki fault zones were interpreted to indicate brecciated fault rocks [Okubo et al., 1997] or hyaloclastites and volcaniclastic sediments deposited on the offshore flank [Haslinger et al., 2001; Park et al., 2007]. However, most of the previous velocity models were limited to certain regions dependent on their source and receiver locations, i.e., either onshore, or 2-D low-resolution onshore-offshore profiles.
3.1. Air Gun Data
 In 1998, ∼2000 km of multichannel seismic reflection data were collected around the south flank of the Island of Hawaii using the R/V Maurice Ewing [Morgan et al., 2003]. Most of the survey lines cover the offshore region of the south flank of Kilauea, and some lines extend to South Point, Puna Ridge, and offshore of Hilo (Figure 1). We were able to extract about 60% of the seismic data recorded by the on-land USGS-HVO Seismic Network, and picked ∼300,000 first arrival times manually. Pick uncertainties of 50 to 200 ms were assigned based on an empirical relationship with signal-to-noise ratio [Zelt and Forsyth, 1994]. Picks were then averaged into 0.5 × 0.5 km bins, yielding ∼41,000 traveltimes for tomographic inversion. Further details of the air gun data set are given by Park et al. .
3.2. Earthquake Data
 We used a relocated earthquake data set recorded by the USGS-HVO Seismic Network between 1970 and 2004. Initial earthquake locations were derived from a 3-D seismic tomographic inversion [e.g., Okubo et al., 1997]. The hypocentral locations were adjusted to the same local Cartesian coordinate system as the one used for the air gun data. Most earthquakes are located between 0 and 15 km depth below sea level, with others extending down to 35 km depth below sea level; the shallow ones with depths between 0 and 5 km mostly occur along Kilauea's summit and upper rift zones; those between 5 and 15 km depth are observed around Kilauea's south flank and Mauna Loa's east and southeast flank (Figure 1). In order to use the highest quality earthquake data, we required them to have magnitudes 2.5 and larger, an azimuthal station gap of less than 90 degrees, and to be recorded by more than 20 seismic stations. These criteria reduced the data set to 6628 events (Figure 1). The selected earthquakes were recorded by 42 seismic stations, resulting in ∼159,000 first arrival times. The first arrival times from earthquake data were assigned an uncertainty of 150 ms, and then combined into the air gun data set for the simultaneous inversion.
 The data were inverted for the 3-D P wave velocity structure using the regularized first-arrival time seismic tomography method of Zelt and Barton , with a modification for simultaneously relocating hypocentral parameters (locations and origin time) [Ramachandran et al., 2005]. The data misfit between the calculated and observed traveltime Δtij, for the jth earthquake recorded by the ith seismic station, can be written as
where ∂Tij/∂ml represent the partial derivatives of the slowness model parameters (l = 1, L), and ∂Tij/∂xjk are the partial derivatives of the hypocenter location parameters for the earthquakes (j = 1, J). The slowness and hypocenter parameter updates are represented by Δml and Δxjk, respectively, and the computed earthquake origin time updates are given by Δτj.
 In addition to the data misfit, prior constraints are applied to the model parameters as regularization factors [Jackson, 1972; Scales et al., 1990]. In this study, the prior constraints are chosen to yield a “minimum structure” solution, which can be measured in terms of model perturbation and roughness [Zelt and Barton, 1998], but no regularization is applied to the hypocentral parameters. Finally, the problem can be formulated by the following objective function, Φ(m), which includes data misfit and model roughness constraint.
where m is the current model vector; Δm = m − m0 is the model perturbation, m0 is the starting model, Δt is the data residual vector; Cd is the data covariance matrix; Ch−1 and Cv−1 are equal to DhTDh and DvTDv, in which Dh and Dv are the horizontal and vertical roughening matrices using the second-order spatial derivative, respectively; λ is the trade-off parameter which determines the relative importance between data misfit and model roughness; sz is the weighting of vertical roughness relative to horizontal roughness; and α is the relative weighting of perturbation versus roughness constraint. By minimizing the objective function in a least squares sense, the updated model, m is calculated by m = m0 + Δm.
 A simple 3-D starting model was developed using a 1-D velocity model for the subaerial edifice [Klein, 1981], combined with a velocity model for the offshore region developed from 2-D refraction data [Hill and Zucca, 1987]. These two velocity models were smoothly interpolated and extrapolated throughout the whole modeled region in such a way that the velocity smoothly increases with depth, without a sharp velocity jump across the Moho. Equation (2) is solved using the LSQR variant of the conjugate gradient algorithm [Paige and Saunders, 1982; Nolet, 1987]. The velocity of water was fixed at 1.5 km/s above the known bathymetry. The model was updated at each iteration, in which new raypaths were calculated by solving the Eikonal equation based on the finite difference algorithm of Vidale , with a modification for sharp gradients by Hole and Zelt . The final velocity model was obtained when an appropriate data misfit to the observed data was achieved according to the assigned picking errors [Zelt and Barton, 1998]. We used a uniform horizontal and vertical node spacing of 1 km for the forward calculation and a square cell size of 1 km for the inverse step.
5. Resolution Estimates
Figure 2 shows raypaths from air guns and relocated earthquakes, calculated using the final velocity model. Rays from air gun shots start at sea level along the offshore survey lines, and arrive at the onshore seismic stations. They provide dense ray coverage around Kilauea and Mauna Loa volcanoes, and also the offshore and coastal regions, up to 18 km depth below sea level (Figure 2). In contrast, most earthquakes occurred within the onshore region, e.g., near Kilauea's summit and rift zones, and Mauna Loa's southeast flank, and were recorded by the onshore stations (Figure 2a). The abundance of raypaths from earthquakes, which have depths ranging from 20 km to near sea level, compensate for the sparse on-land ray coverage from the air gun data. As shown in Figure 2b, the two data sets complement each other, thus improving the model resolution both nearshore and in the deeper onshore region.
 In order to estimate the lateral resolution of the velocity model, we applied checkerboard tests using the method of Zelt . Synthetic data were created using the real source-receiver geometry and checkerboard models, in which 32 different velocity squares of laterally alternating 10% positive and negative velocity anomalies with respect to the initial model are superimposed on the initial model. Data were inverted for the 3-D velocity structure using the regularized seismic tomography method of Zelt and Barton . The true earthquake locations were assumed to be known and were fixed. This means the resolution estimates may be somewhat optimistic. Comparison of the recovered model with the known checkerboard pattern was made using a semblance measure, and subsequently converted into lateral velocity resolution [Zelt, 1998]. Thus, for example, 10 km lateral resolution means that >10% velocity anomalies of more than 10 km size can be resolved by the current geometry of sources and receivers given the assigned picking errors.
Figure 3 shows horizontal slices of the estimated lateral velocity resolution from 2 to 16 km depth below sea level every 2 km. On the basis of the resolution test, most of the major geologic features in our study area, such as the summits and rift zones of the island's volcanoes, as well as fault zones and offshore structures to the southeast of the island are expected to be resolved with at least 20 km lateral resolution up to 16 km depth (Figure 3). In particular, better than 10 km lateral resolution is expected around Kilauea and Mauna Loa volcanoes, the Kao'iki, Koae, and Hilina fault zones, and the offshore region encompassing Kilauea's upper flank and outer bench, and Loihi seamounts up to 10 km depth (Figure 3).
 In order to examine the vertical resolution of the model, we also carried out checkerboard tests in which horizontally and vertically alternating 10% positive and negative velocity anomalies with respect to the initial model are superimposed on the initial model. As in the method of Zelt , the alternating anomalies are recovered with the real source and receiver geometry. Figure 4 shows the horizontal and vertical slices of the recovered anomalies using an anomaly size of 5 km in the vertical direction, and 20 km in the horizontal direction. The slices of the resolved anomalies show that features of this size can be recovered well around Mauna Loa's south and east flank, Kilauea's summit, upper rift zones and southeast flank, and offshore flanks at depths above about 20 km. The amplitude and shape of the anomalies are more smeared and distorted below 15 km depth and outside the central region (Figure 4).
6. Tomographic Results and Interpretation
 The final velocity model was obtained after six iterations, providing a 128 ms RMS traveltime residual, a misfit reduction of 60% from the starting model (Figure 5). Average absolute perturbations of earthquake traveltimes and hypocentral locations are 20 ms, and 15 m, 8 m, 14 m in the x, y, and z directions, respectively. These values are relatively small because we used an already relocated earthquake data set provided by the USGS-HVO Network.
 Compared with the lateral resolution of seismic tomography using the air gun data set alone [Park et al., 2007], the additional earthquake data expands the onshore region with improved lateral resolution to include Mauna Loa's west flank, the north flank of Kilauea's ERZ, and the southern and eastern parts of Mauna Kea and Hualalai volcanoes (Figure 3). Also, lateral resolution is improved beneath Kilauea and Mauna Loa volcanoes (Figure 3), due to more raypaths from deep seismic sources in the earthquake data set.
 The velocity structure resolved by our new model (Figure 6) shows many similarities with the previous active source only velocity model [Park et al., 2007]. In particular, the velocity structures around the offshore and coastal regions do not change significantly. The previous results from the tomographic inversion of the air gun data set showed high-velocity anomalies underlying the summits and rift zones of Kilauea, Mauna Loa, Loihi seamount and also beneath Kilauea's offshore bench [Park et al., 2007]. Low-velocity structures were imaged beneath Kao'iki and Hilina fault zones that bracket Kilauea's summit and upper rift zones, and along the upper offshore volcano flanks [Park et al., 2007]. However, the joint inversion resolves more inland features to better establish the connections between offshore, coastal and inland structures (Figure 6).
 In the following, we emphasize velocity features that are newly resolved and/or show improved resolution after the simultaneous tomographic inversion, and interpret the distribution of the relocated earthquakes within the velocity structures. Figure 6 shows horizontal slices of the velocity model at every 2 km from 2 to 16 km depth. Figures 7 and 8 show the vertical profiles of the velocity model and lateral resolution along Kilauea's rift zones (profile AA′) and across the rift zones (profiles BB′ and CC′). Figures 9 and 10 depict the velocity and lateral resolution profiles around Mauna Loa and surrounding volcanoes. In Figures 7–10, the interpreted oceanic crust is indicated by dashed black lines; the position of the Moho is based on the smoothed 7.5 km/s contour and a thickness of ∼5 km is assumed for the oceanic crust [Park et al., 2007]. On the basis of the vertical resolution test (Figure 4), we are confident that the oceanic crust and Moho are well resolved around Mauna Loa, Kilauea, and the offshore flank, up to 15 km depth. Compared with the previous studies about the depth to the Moho [Hill and Zucca, 1987; Li et al., 1992], the inferred depths of the Moho beneath Kilauea and Mauna Loa's summits match quite well with less than 2 km difference, i.e., Moho depths beneath Kilauea and Mauna Loa are ∼14 and ∼17 km from Hill and Zucca  and Li et al.  and ∼15 and ∼16 km in our model. All depths are referenced to kilometers below sea level unless otherwise stated.
6.1. Kilauea and Adjacent Volcanoes
 Kilauea's summit and upper rift zones are underlain by high-velocity materials (6.0–7.0 km/s) that extend through the oceanic crust to ∼12 km depth (Figures 6 and 7b). A shallow high-velocity (4.5–5.5 km/s) anomaly is visible near sea level beneath the summit and upper ERZ (Figure 7b, 35–55 km), and may record a shallow magma chamber recognized by others [Dawson et al., 1999; Pietruszka and Garcia, 1999]. The highest velocity materials within the edifice (6.5–7.0 km/s) occur beneath the summit below 4 km depth (Figure 7b, 35–50 km along profile), and are attributed to a deep magma reservoir dominated by olivine cumulates [Clague and Denlinger, 1994; Brandsdóttir et al., 1997; Okubo et al., 1997; Park et al., 2007]. This deep high-velocity region is also largely aseismic, and is bounded by two clusters of earthquakes, one shallow (2–4 km below sea level) and the other deep (15–18 km), near the base of the oceanic crust (Figure 7b). These two clusters are associated with magma migration from the mantle and into the summit magma chamber, respectively [Klein et al., 1987].
 Moderately high velocities (6.0–6.7 km/s) are distributed beneath Kilauea's upper ERZ (Figure 7b, 50–70 km) and along the upper SWRZ (Figure 7b, 20–35 km) below 3 km depth, both reflecting intrusive complexes [Okubo et al., 1997; Park et al., 2007]. A cluster of earthquakes occurs near the southwestern edge of high-velocity region along the SWRZ (Figure 7b, 20–25 km). Beyond these regions, velocities decrease to 5.5–6.0 km/s and 6.0–6.5 km/s within the SWRZ and ERZ, respectively (Figure 7b), consistent with the background velocity structure of extrusive lavas [Hill and Zucca, 1987; Klein, 1981]. A secondary velocity high is observed along the lower subaerial ERZ (Figure 7b, 87–100 km), suggesting a distinct secondary magma reservoir [Tilling and Dvorak, 1993] or cumulate body [Park et al., 2007]. The base of the interpreted oceanic crust is relatively smooth, and dips toward Kilauea's summit from both directions (Figure 7b).
 Along the vertical profile BB′ across Kilauea's ERZ (Figure 7c), several high-velocity anomalies stand out. The most seaward one occurs within the oceanic crust beneath Kilauea's outer bench, and is underlain by a broad low-velocity anomaly within the upper mantle (Figure 7c, 80–95 km). This feature resembles a similar anomaly pair found beneath Loihi (Figure 6), attributed to an intrusive core and mantle melt beneath the young volcano [Park et al., 2007]. A zone of high velocities (6.5–7.0 km/s) lies beneath the axis of the Kilauea ERZ below 5 km. This feature protrudes seaward above the ocean crust (7–9 km depth), which may be indicative of the southward migration of the ERZ over time [Swanson et al., 1976].
 A steep boundary separates the high-velocity core of the rift zone of Kilauea from a broad region of low velocities (4.0–6.0 km/s) that defines the upper submarine flank (Figure 7c, 60–80 km). These low velocities are attributed to accumulations of volcaniclastic sediments [Haslinger et al., 2001; Park et al., 2007], also well imaged in seismic reflection profiles across the submarine flank [Morgan et al., 2000, 2003]. A frontal bench constructed by overthrusting at the toe of submarine flank ponds these sediments within a midslope basin [e.g., Morgan et al., 2003]. Earthquakes are densely clustered near the top of the ocean crust (7–10 km depth) below the nearshore velocity boundary (Figure 7c).
 This vertical section also shows two other high-velocity features that rise from the oceanic crust landward of Kilauea; A broad southeast dipping zone with velocities of 6.5–7.0 km/s correlates with Mauna Loa's northeast rift zone (NERZ) (Figure 7c, 20–40 km), and the edge of a second high occurs within Mauna Kea's eastern flank (Figure 7c, 0–20 km). The intervening regions of reduced velocities are observed within the volcanic edifice (Figure 7c, 10–20 km).
 A third vertical profile, CC′, crosses Mauna Loa's SWRZ, the distal end of Kilauea's SWRZ, and the offshore flank of Kilauea, including the submarine high referred to as Papau seamount (Figure 7d). The highest velocities in the edifice (6.5–7.3 km/s) occur within the east flank of Mauna Loa (Figure 7d), where an old south rift zone of Mauna Loa was recognized from active source tomography [Park et al., 2007]. Profile CC′ cuts the old rift zone obliquely, but shows its general eastward dip (Figure 7d, 25–50 km). Lower velocities occur beneath and to the west of this feature (Figure 7d, 10–30 km). Earthquakes are clustered near the top of the oceanic crust along the eastern boundary of Mauna Loa's high-velocity region (Figure 7d, 40 km), defining a pattern very similar to that of Kilauea's south flank (Figure 7c).
 Compared with velocity structures beneath Kilauea's ERZ, the SWRZ of Kilauea is underlain by a relatively modest velocity anomaly (5.5–6.5 km/s) at 3–8 km depth (Figure 7d), which coincidentally occurs immediately above a sharp downward step in velocities in the seaward direction that was previously correlated with the faulted scarps of the Ninole Hills [Park et al., 2007]. The submarine flank again shows low velocities (Figure 7d, 55–95 km). This includes Papau seamount, which is not underlain by any significant velocity anomaly (Figure 7d, 70–80 km), consistent with a thick pile of volcaniclastic strata near the convergence of the submarine flanks of Kilauea and Mauna Loa [Morgan et al., 2003].
6.2. Mauna Loa and Adjacent Volcanoes
 Profile DD′ extends across the southeast flank of Mauna Kea, Mauna Loa's NERZ, and the southeast flank, passing through a large cluster of seismicity (Figure 9b). The summit and south flank of Mauna Kea are underlain by high velocities (6.5–7.0 km/s) that peak ∼5 km below sea level (Figure 9b), although the resolution is relatively poor along the edge of the model (Figure 10b). Another set of high-velocity (6.5–7.0 km/s) features occurs beneath and to the south of Mauna Loa's NERZ (Figure 9b, 35–55 km). A broad scatter of earthquakes spans the southern edge of the rift zone velocity anomaly, and deepens to the south (Figure 9b), similar in character to that recognized beneath Kilauea (Figure 7c). The interpreted top of oceanic crust beneath Mauna Loa and Mauna Kea volcanoes dips toward the island, reaching 12 km depth beneath Mauna Kea's summit (Figure 9b).
 Profile E-E′ (Figure 9c) stretches across Mauna Loa's summit, transverse to the SWRZ. The high-velocity materials (6.0–6.8 km/s) to the southeast (Figure 9c, 40–50 km) are imaged with better than 10 km resolution (Figure 10c), and deepen with distance from the SWRZ, reaching the top of the oceanic crust and below. This feature corresponds to the eastern edge of the buried rift zone shown in Figure 7d [Park et al., 2007]. High velocities (6.6–7.0 km/s), peaking between 2 and 7 km depth, also extend up to 20 km to the northwest of Mauna Loa's summit (Figure 9c, 20–40 km), although the resolution decreases to ∼15 km lateral resolution in this region (Figure 10c). This is the first seismic velocity evidence for an intrusive complex aligned with what could be Mauna Loa's northwest rift zone (NWRZ), as previous proposed by Baher et al. . This high-velocity material within Mauna Loa's northwest flank is overlying a relatively low-velocity zone within the southeast flank of Hualalai (Figure 9c, 0–20 km). Notably, the subaerial southeast rift zone of Hualalai, which parallels this transect, is not characterized by increased velocities at shallow depths (Figures 6 and 9c).
 Profile FF′ (Figure 9d) traverses the west flanks of Mauna Kea and Mauna Loa, parallel to the SWRZ of Mauna Loa. Lateral resolution in this area is relatively poor, but still resolves features less than 20–40 km in dimension, and locally less than 10 km (Figure 10d). High velocities (6.5–7.0 km/s) occur beneath Mauna Kea's west flank, and extend to Mauna Loa as well (Figure 9d, 0–25 km). The shallow high-velocity body interpreted as Mauna Loa's NWRZ also appears at 4–8 km depth (Figure 9d, 25–35 km), with slightly reduced velocities at 8–12 km depth. A seismic cluster is distributed vertically between 5 and 9 km depth along the southwestern boundary of the NWRZ (Figure 9d, 38 km).
 A third region of slightly increased velocities (6.5–7.0 km/s) lies near the southern edge of this profile, just above the top of ocean crust (Figure 9d, 55–70 km), and is also separated from the northern low-velocity zone (6.0–6.5 km/s) (Figure 9d, 40–55 km). The origin of this southern high-velocity structure is not clear. It could reflect an inactive, buried secondary intrusive zone, similar to that noted around the lower ERZ of Kilauea (Figure 7b), or perhaps a buried landslide block detached from the upper flank [Morgan et al., 2007]. A second cluster occurs near the top of the ocean crust at ∼30 km along the profile, and a third lies near the top of ocean crust at ∼60 km (Figure 9d).
 The resulting tomographic model shows distinct patterns of high- and low-velocity anomalies that appear to designate magma cumulates or dike complexes beneath the summits and rift zones, and brecciated rocks or volcaniclastic sediments within the offshore flank [Okubo et al., 1997; Haslinger et al., 2001; Park et al., 2007]. However, some of the prominent anomalies are not consistent with surficial geologic features, indicating more complicated volcanic structure and evolution than previously assumed. These are the primary focus of further discussion, and lead to new insights into how Hawaiian volcanoes grew and interacted, and explanations for present-day volcano-tectonic behavior.
7.1. Rift Zone Growth and Evolution
 Our velocity model shows high-velocity structures along rift zones, associated with underlying magma cumulates and dike swarms [Hill and Zucca, 1987; Okubo et al., 1997; Haslinger et al., 2001; Park et al., 2007], but some structures lie far from the axes of the current rift zones (Figures 7 and 9). These high-velocity structures, in association with other geophysical data (e.g., gravity and magnetic surveys and GPS measurements), demonstrate that the internal structure of volcanic rifts is much more complex than we would infer from the surface geology alone, raising the possibility of large-scale changes in volcano geometry and processes over time.
7.1.1. Kilauea Volcano
Swanson et al.  suggested that the ERZ of Kilauea was initially inboard of its current position, and has migrated southward over time, producing a conspicuous bend in the rift's axis near the summit. Rift zone migration is explained by younger dikes intruding to the south of older dikes, accounting for an offset in the peak gravity anomaly to the north [Swanson et al., 1976; Hill and Zucca, 1987]. Tests of this model would lie in the presence of deep high-velocity rocks to the north of the present track of the ERZ, and also in the asymmetry of the associated high-velocity body.
 Our velocity model, however, does not show a high-velocity anomaly to the north of Kilauea's ERZ (Figures 6 and 7c), although we do see a steeper velocity boundary to the south than to the north, and locally, a seaward shift of the shallower intrusions (Figure 7c). These observations appear to be consistent with more recent gravity models that suggest a persistent southward bend in the position of high-density materials along Kilauea's upper ERZ [Kauahikaua et al., 2000]. We suggest that Kilauea's ERZ might have initiated with curvature similar to that observed today, although subsequent dike intrusions south of the rift zone may have enhanced the southward bend [Swanson et al., 1976]. As Kilauea was built upon the preexisting southeast flank of Mauna Loa [Swanson et al., 1976; Lipman et al., 2000], it is possible that rift intrusions near Kilauea's summit followed a weakness parallel to the sloping flank.
 The shallow portion (3–8 km depth) of Kilauea's middle SWRZ exhibits a zone of relatively low-velocity materials (6.0–6.5 km/s), implying that this region has a lower concentration of intrusive rocks than beneath Kilauea's summit and upper rift zones (Figure 7b). The lower portion of Kilauea's SWRZ shows low-velocity materials (5.0–6.0 km/s) down to the top of the ocean crust at 7 km (Figure 7b), also suggesting a lack of intrusions. This latter observation is consistent with the previous interpretation that Kilauea's SWRZ is a surficial feature, built upon Mauna Loa's volcanic edifice [Swanson et al., 1976; Lipman et al., 2006]. The low-velocity region might reflect volcaniclastic sediments or porous surface flows on Mauna Loa's southeast flank, with Kilauea's SWRZ flows defining a thin veneer.
7.1.2. Mauna Loa Volcano
 Mauna Loa volcano shows a much more complicated internal structure than Kilauea, consistent with its greater age and potential interactions with adjacent volcanoes. As noted previously [Park et al., 2007], Mauna Loa's south flank contains a large abandoned south trending rift zone. This feature stands out as a high-velocity lineament prominently in our velocity model (Figures 6 and 7d), and must define an earlier configuration of this large volcano. The presence of this great mass may influence Kilauea's SWRZ propagation, as well as the mobility of Mauna Loa's southeastern flank [Park et al., 2007].
 The deeper portions of Mauna Loa's middle SWRZ is characterized by low velocities (6.0–6.5 km/s) between 4 and 10 km depths (Figure 6). Thus, this region appears to lack the dense intrusive core observed beneath other rift zones. The most likely explanation is that Mauna Loa's SWRZ is a relatively young feature, which became active after the older south rift zone was abandoned. Thus, SWRZ vents were built upon surficial lavas erupted from the older rift zone, accounting for their low velocities. Alternatively, intrusive rocks once present in this region were removed by catastrophic landsliding [Lipman et al., 1988, 1990]. The large body of increased velocity along the southern edge of profile FF′ (Figure 9d, 55–70 km) may represent a buried landslide block derived from this rift zone, similar to the giant landslide blocks found offshore of Mauna Loa's western flank [Lipman et al., 1988; Moore et al., 1995].
 The shallow high-velocity feature noted within Mauna Loa's northwest flank appears as one of three high-velocity prongs that radiate from the summit of Mauna Loa (Figures 6, 9c, and 9d). We interpret it to be a third rift zone, referred to as Mauna Loa's NWRZ, not clearly expressed at the surface. This interpretation is consistent with other observations of radial vents in this region [Lipman, 1980] and unusual seismicity [Baher et al., 2003]. Significantly, the high velocities are limited to the shallow region, and are underlain by relatively low-velocity materials (Figures 9c and 9d), suggesting that the rift zone grew outward on top of the underlying flanks of Hualalai and Mauna Kea. The length and activity of Mauna Loa's NWRZ was probably limited by buttressing effects of the adjacent volcanoes [Baher et al., 2003].
 Mauna Loa's NERZ stands out as a prominent high-velocity region, suggesting a long-lived rift zone. However, deeper, high-velocity anomalies are offset to the south of the current NERZ axis (Figures 9a and 9b). This could be explained by the progressive northward migration of eruptive centers, as most historic eruptions along Mauna Loa's NERZ have occurred on its north flank [Lockwood, 1990; Kauahikaua et al., 2000]. Such northward migration may have been induced by the growing edifice of Kilauea. Alternatively, the high-velocity bodies, presumed to consist of cumulate-rich magma, may have spread southward over time, similar to the process inferred within Kilauea's south flank today [Clague and Denlinger, 1994].
7.1.3. Mauna Kea Volcano
 The surface distribution of volcanic vents on Mauna Kea suggests one distinct rift zone, which extends northeastward, previously thought to continue offshore as the Hilo Ridge [Fiske and Jackson, 1972; Wolfe and Morris, 1996]. Few data exist to map out the subsurface signature of these features, although gravity data show southeastward trending highs suggestive of an old rift zone [Kinoshita et al., 1963; Fiske and Jackson, 1972]. Our velocity model shows the east flank of Mauna Kea to be a comparatively low-velocity region for depths greater than 4 km (Figures 6 and 7c). These data confirm the absence of deep-seated northeast rift intrusives, allowing for a Kohala origin for the Hilo ridge [Holcomb et al., 2000; Kauahikaua et al., 2000]. High velocities do extend to the south-southeast beneath Mauna Kea (Figure 6), with the top of this body deepening from 3 km near the summit to 8 km within Mauna Loa's flank (Figure 9b). The southward extent of this rift zone might have been limited by the growing volcanic edifice of Mauna Loa [Fiske and Jackson, 1972], and subsequently buried by later lava flows from both volcanoes.
Fiske and Jackson  argued persuasively that the geometry and growth of rift zones in Hawaii are affected by concurrent gravitational stress fields induced by nearby volcanic edifices as they developed. Their analysis depended on the surficial expressions of the rift zones, however, so it did not take into account the past configurations of the volcanoes. Our new velocity model provides compelling evidence for prior rift zone configurations that force us to rethink volcano growth and evolution in Hawaii. For example, the seismic velocity and gravity data suggest that several of the older volcanoes possessed south to southeast trending rift zones that are now buried, i.e., Mauna Kea, Hualalai, and Mauna Loa (Figure 11). All of these were most likely active during tholeiitic shield building to account for their large dimensions. These rift zones were subsequently abandoned, perhaps in the case of Hualalai and Mauna Kea because their growth was restricted by adjacent edifices, i.e., Mauna Loa. Later eruptions (postshield for Hualalai and Mauna Kea) followed new vent trends favored by the associated stress state [Wanless et al., 2006].
 The abandonment of Mauna Loa's massive south trending rift zone may have been influenced by other processes as well, for example, large-scale landsliding along the volcano's western flank [Lipman et al., 1988, 1990]. Preferential dike intrusion along the western edge of Mauna Loa's south rift zone, possibly in response to landsliding and away from the growing volcanic edifice of Kilauea, may have caused westward bending of the rift zone, and eventual abandonment in favor of the current SWRZ trend. The old south rift zone of Mauna Loa may now restrict the westward growth of Kilauea's SWRZ, deflecting its trend to the south [Park et al., 2007], more parallel to the topographic slope of the underlying volcanic edifice [Fiske and Jackson, 1972].
7.2. Seismicity and Velocity Structure
 Our simultaneous inversion provides a volcanic velocity model with improved resolution that is consistent with the earthquake data recorded from 1970 to 2000. Although the average perturbations of the hypocentral relocation are relatively small, we found that some earthquake clusters, previously addressed by various studies [Klein et al., 1987; Denlinger and Okubo, 1995; Gillard et al., 1996; Endo, 1985; Walter and Amelung, 2006] have spatial correlations with the resolved velocity structures and connections with geodetic flank motions (Figure 11). We infer that the relationships between seismicity and volcanic structure might provide us with a better understanding of volcano-tectonic processes.
7.2.1. Seismicity at Kilauea Volcano
 Shallow seismicity at Kilauea occurs primarily above the high-velocity regions beneath the summit and along the two rift zones (Figure 7b). This activity can be related to brittle deformation in response to shallow magma intrusions [Klein et al., 1987] or high stress concentrations [Gillard et al., 1996]. The underlying high-velocity regions are largely aseismic (Figure 7b), consistent with previous interpretations of ductile cumulate rich magmas, perhaps due to the presence of interstitial melt [Denlinger and Okubo, 1995; Park et al., 2007]. The interpreted ductile state of this material probably allows this cumulate mass to flow outward [Clague and Denlinger, 1994], pushing the volcano's south flank seaward (Figures 7c and 11) [Denlinger and Okubo, 1995]. Earthquakes located near the base of the edifice (Figure 7c) are recognized to result from frictional sliding of Kilauea's edifice along a decollement zone, which accommodates low-angle reverse faulting [Denlinger and Okubo, 1995].
 The dense cluster of earthquakes near the high-velocity region beneath Kilauea's middle SWRZ is more enigmatic (Figure 7b). Klein et al.  suggested that these earthquakes reflect restricted magma intrusion due to a structural barrier. Failure mechanisms responsible for seismicity may be analogous to shallow (3–5 km depth) dike emplacement, interpreted beneath Kilauea's ERZ [Klein et al., 1987; Gillard et al., 1996]. However, the earthquakes along the middle SWRZ extend to greater depths (5–10 km depth) than the ones along the ERZ and summit (Figure 7b), and are associated with comparatively low-velocity materials (6.0–6.5 km/s). This seems to confirm the absence of ductile cumulates in this area, allowing for brittle behavior of the volcanic edifice. Thus a different origin than magmatic intrusion may account for the intermediate depth seismic events beneath Kilauea's middle SWRZ. We propose that these earthquakes are derived from deeper volcano-tectonic processes, for example, (1) interactions between Kilauea's SWRZ and the buried extension of the Kao'iki fault zone [Swanson et al., 1976], (2) edifice cracking due to lithospheric flexure [Thurber and Gripp, 1988] or rupture [Got et al., 2008] in response to volcanic loading of Mauna Loa, and/or (3) interactions between Kilauea and Mauna Loa volcanoes at depth, triggering seismicity at their boundaries.
7.2.2. Seismicity at Mauna Loa Volcano
 Compared with Kilauea, Mauna Loa exhibits fewer shallow earthquakes (Figure 11), probably related to the lower rate of recent volcanic activity [Lipman, 1995]. However, seismic activity is substantial within and beneath the southeast flank of Mauna Loa (Figure 11), particularly in the low-velocity regions referred to as the Kao'iki and Hilea seismic zones [Endo, 1985]. The deeper earthquakes in these seismic zones (9–13 km) form a background of seismicity, with larger earthquake magnitude, in contrast to the shallow seismic events [Thurber et al., 1989; Walter and Amelung, 2006]. This observation suggests that the deep earthquakes have a volcano-tectonic rather than magmatic origin [Koyanagi et al., 1988, 1989; Jackson et al., 1992], i.e., due to seaward sliding of Mauna Loa's volcanic edifice over a basal decollement zone [Endo, 1985].
Park et al.  noted that these decollement-type earthquakes are highly clustered around the eastern and southern boundary of the high-velocity region (6.8–7.3 km/s) beneath Mauna Loa's southeast flank, a correlation that also persists for the relocated earthquakes (Figures 7d and 11). At first glance, this correlation is consistent with the concentration of decollement-type earthquakes at the seaward edge of Kilauea's high-velocity region (Figure 7c) but, as noted by Park et al. , probably has a different explanation. The high seismic velocities in Mauna Loa's southeast flank are attributed to the ancient south trending rift zone of Mauna Loa, which presently shows little seaward displacement, in particular on its west side (Figure 11). This buried rift zone corresponds to a relatively stable portion of Mauna Loa's flank, in contrast to the more mobile region above the Kao'iki fault zone. Seismicity is localized at the boundary between the two regions (Figure 11).
 Mauna Loa's northwest flank shows two distinct clusters of seismicity; one lies at intermediate depth (5–8 km depth) near the southwestern boundary of the high-velocity region beneath Mauna Loa's northwest flank at 27 km along the profile EE′ of Figure 9c, interpreted as the failed NWRZ [Baher et al., 2003]. The other lies near the top of the oceanic crust at ∼10 km depth between 60 and 65 km along profile FF′ of Figure 9d, just west of Mauna Loa's middle SWRZ. Similar spatial relationships of intermediate-depth seismicity concentrated at the boundaries of high-velocity anomalies are observed near the Hilina fault zone (Figure 7c) and Kao'iki fault zone (Figure 7d). Both of these regions are interpreted to have experienced flank failures across seaward facing fault zones [Swanson et al., 1976; Lipman et al., 1985, 1990; Lipman, 1995], which may account for the intermediate seismicity. As Mauna Loa's west flank is thought to have experienced giant flank failures in the past [Lipman et al., 1988], these intermediate-depth earthquakes also might indicate a hidden landslide scar or detachment fault [Lipman, 1980].
 The deeper (∼10 km) earthquakes around the western flank of Mauna Loa (Figures 9d and 11) lie in an area referred to as the Kona seismic zone [Walter and Amelung, 2006; Wolfe et al., 2004], characterized by low-angle, west directed thrust earthquakes. This focal mechanism is consistent with seaward slip of the western flank away from Mauna Loa's SWRZ, probably along a basal decollement, similar to that modeled for the southeastern flank [Walter and Amelung, 2006].
7.3. Volcano-Tectonic Implications
 Seismicity in Hawaii can be classified according to depth [Klein et al., 1987] but now also according to distribution relative to the volcanoes' internal structure. Many of the shallowest events (2–5 km depth) occur in relatively low-velocity materials near volcano summits and rift zones, indicative of brittle deformation during magma migration [Wright and Klein, 2006]. Deeper earthquakes (8–13 km) lie near the volcano–oceanic crust interface, and accommodate seaward spreading of the volcanoes along a frictional decollement [Koyanagi et al., 1989; Tilling and Dvorak, 1993; Denlinger and Okubo, 1995]. The absence of earthquakes within the highest velocity materials in young volcanoes, i.e., Kilauea, is predicted by the ductile state of the deep cumulate-rich magmas [Clague and Denlinger, 1994; Denlinger and Okubo, 1995]. However, there are also clear deviations from this relationship, for example, beneath Mauna Loa's southeast flank (Figure 7d), which may be explained by the more solidified and frictional state of the flank cored by this ancient rift zone [Park et al., 2007].
 We also see distinct trends of intermediate depth (5–8 km) earthquakes, often located near the boundaries of high-velocity regions (Figures 7b and 9d). These earthquakes may reflect interactions between edifices subject to different volcano-tectonic stress fields, for example, Kilauea and Mauna Loa (Figure 7b), or alternatively, they define reactivated detachment or transfer fault zones. In either case, they represent zones of potential earthquake risk. These less well understood earthquakes are ideal targets for future relocation efforts, similar to recent work that has been carried out along the south flank of Kilauea volcano [Wolfe et al., 2007].
 One of the most enigmatic zones of seismicity is the Kao'iki seismic zone, which exhibits a deepening trend of seismicity to the south. This region also defines the most rapidly displacing portion of Mauna Loa's edifice (Figure 11), which could account for the abundance of earthquakes. The high-velocity materials at depth allow for a ductile cumulate-rich mass that can flow outward to accommodate such deformation. However, the cloud of earthquakes seems to penetrate into the high-velocity features (Figure 9b), suggesting the correlation observed at Kilauea does not hold. Perhaps the earthquakes in this region also follow buried fault trends, e.g., deeper versions of the Kao'iki fault zone [Lipman et al., 2006].
 We have derived a 3-D P wave velocity model of the combined subaerial and submarine portions of the Island of Hawaii using a simultaneous inversion of first-arrival times from air gun shots and earthquakes, recorded by the USGS-HVO seismic network. This model has provided a unique opportunity to clarify the past configurations of the island volcanoes, in particular, their rift zones, and furthermore, to correlate current volcano-tectonic processes with internal volcanic structures within the active Hawaiian volcanoes.
 The velocity structure beneath Kilauea's ERZ suggests that the early rift intrusions might have propagated southeastward near the summit over the seaward sloping flank of Mauna Loa. The middle SWRZ of Kilauea, resolved by relatively low-velocity features (6.0–6.5 km/s), has intermediate-depth (3–8 km) seismic events, indicating that this region might be brittle and less dominated by magma cumulates than the region beneath Kilauea's summit and upper rift zones. The lower SWRZ of Kilauea is underlain by low-velocity materials (5.0–6.0 km/s) at 3–7 km depth, attributed to volcaniclastic sediments deposited on the southeast flank of Mauna Loa.
 The high-velocity region (6.8–7.3 km/s) within the south flank of Mauna Loa, where little seismic activity occurs, probably represents cooled magma cumulates of Mauna Loa's old south rift zone. The middle part of Mauna Loa's SWRZ, underlain by low velocities (6.0–6.5 km/s), has experienced westward flank failures, and subsequent dike intrusions into the lower SWRZ may have developed along hidden scars of landslides or slumps. Mauna Loa's NERZ is interpreted to have migrated northward away from the growing edifice of Kilauea. A shallow high-velocity anomaly (6.5–6.8 km/s), related to the buried NWRZ of Mauna Loa overlies the low-velocity materials (6.0–6.5 km/s), that define the buried flanks of Hualalai and Mauna Kea. Hualalai and Mauna Kea have high-velocity anomalies (6.5–7.0 km/s) within their south flanks indicative of buried rift zones, whereas there is no indication of a high-velocity core beneath Hualalai's southeast rift zone and Mauna Kea's east rift zone despite observed surface vents.
 The observed correlations of recent seismicity with buried, and in some cases abandoned, rift zone structures indicates that the latter can play a defining role in controlling present-day volcano-tectonic motions in Hawaii. Thus, predictions of future volcanic behavior depend on an understanding of the deeper and older configurations of Hawaiian volcanoes.
 We thank L. Peters and N. Benesh for helping with extraction and initial processing of the wide-angle seismic refraction data. The reviews by C. Thurber and D. Hill improved the manuscript. Plots were generated using Generic Mapping Toos (GMT) [Wessel and Smith, 1998]. This research was supported by NSF grants OCE 0221951 and 0551750.