The Seismic Array Hikurangi Experiment (SAHKE) investigated the structure of the forearc and subduction plate boundary beneath the southern North Island along a 350 km transect. Tomographic inversion of first-arrival travel times was used to derive a well-resolved 15–20 km deep P wave image of the crust. The refracted phases and migrated reflection events image subducting slab geometry and crustal structure. In the west, Australian Plate Moho depth decreases westward across the Taranaki Fault system from 35 to ∼28–30 km. In the east, subducted Pacific Plate oceanic crust is recognized to have a positive velocity gradient, but becomes less distinct beneath the Tararua Ranges, where the interface increases in dip at about 15 km depth from <5° to >15°. This bend in the subducted plate is associated with vertical clusters in seismicity, splay fault branching, and low-velocity high-attenuation material that we interpret to be an underplated subduction sedimentary channel. We infer that a step down in the decollément transfers slip on the plate interface at the top of a subduction channel to the oceanic crust and drives local uplift of the Tararua Ranges. Reflections from the Wairarapa Fault show that it is listric and soles into the top of underplated sediments, which in turn abut the Moho of the overriding plate at ∼32 km depth, near the downdip end of the strongly locked zone. The change in dip of the Hikurangi subduction interface is spatially correlated with the transition from geodetically determined locked to unlocked areas of the plate interface.
 Subduction zones produce the largest earthquakes and tsunamis on Earth, as evidenced by the 2011 Tohoku Mw 9.0 earthquake [Simons et al., 2011] and 2004 Sumatra Mw 9.3 earthquake [Ammon et al., 2005; Lay et al., 2005]. The fault-slip behavior on subduction thrust faults varies greatly between subduction zones and along strike at the same subduction zone. Our study aimed to image and resolve the physical properties of a locked subduction thrust so that we may better understand the mechanics of slip that may fail in a future large subduction earthquake in the Wellington region of New Zealand [Wallace et al., 2009].
 New insights into the physical mechanisms of subduction process and megathrust earthquakes have been obtained from geological and geophysical studies of representative subduction systems such as Nankai and the Japan Trench [e.g., Bangs et al., 2009; Hayakawa et al., 2002; Kodaira, 2002], Cascadia [Audet et al., 2010], and Central and South America [Bilek, 2010; Ranero et al., 2003]. Seismic reflection imaging can directly determine physical properties at depth, reveal subduction interface geometry, and in conjunction with geodetic and seismology studies, is an important tool in mapping stick-slip versus stable sliding patches on the shallow (<40 km) part of the interface [Calvert, 2004; Kimura et al., 2010; Kodaira et al., 2004; Nedimovic et al., 2003; Sage et al., 2006; Sato et al., 2011]. For example, Nedimovic et al.  and Calvert  identified a thin band of high-amplitude reflections in what they propose to be the locked unstable zone at the Cascadia margin, and a broader band of high-amplitude reflections in the zone of stable sliding that experiences slow slip.
 To investigate the physical parameters that control locking on the Hikurangi plate interface and to determine geometrical relationships with secondary faults, we undertook a controlled-source and passive (earthquake) seismic imaging experiment in the Wellington region. The Seismic Array Hikurangi Experiment (SAHKE) imaged the plate interface beneath southern North Island with onshore-offshore marine active-source seismic data from three sides, onshore shots, and local and teleseismic earthquakes. We present a preliminary analysis of this large data set. Onshore-offshore wide-angle seismic-reflection and refraction data are analysed to constrain the first detailed two-dimensional (2-D) tomographic model of P wave structure across a transect of the southern Hikurangi margin orthogonal to the Australia/Pacific plate boundary. This model places constraints on key parameters such as Moho geometry, subducting slab geometry and forearc structure, which may modulate the location of plate coupling. The available ray paths provide strong constraints on the velocity of the shallow plate boundary to a depth of 20 km. The SAHKE transect data provide additional insight into plate interface coupling and upper crustal splay faults that are fundamental to our understanding of subduction zone architecture.
2. Tectonic Setting
 The Wellington region in southern North Island lies at the southern end of the Tonga-Kermadec-Hikurangi subduction zone (Figure 1a). Thick and bathymetrically elevated 120 Ma oceanic crust of the Hikurangi Plateau (Pacific Plate), a large igneous province, is being subducted westward beneath the Australian Plate [Davy and Wood, 1994; Wallace et al., 2009]. The plate motion of ∼42 mm/yr is partitioned in the near surface between (1) folding and thrust faulting offshore east of North Island in a submerged accretionary wedge, and onshore in emergent parts of the forearc [e.g., Lamb, 1988; Nicol et al., 2002, 2007]; (2) strike-slip faulting near to the axial ranges onshore; and (3) vertical-axis fault-block rotations [e.g., Beanland, 1995; Mouslopoulou et al., 2007; Van Dissen and Berryman, 1996; Wallace et al., 2004]. Near Wellington, the margin-parallel component of plate motion is accommodated by dextral slip on the Wellington and Wairarapa Faults (Figure 1) [e.g., Beanland, 1995; Van Dissen and Berryman, 1996]. A smaller amount of convergence (less than several mm/yr) is accommodated along reverse structures in the upper plate, such as the Kapiti-Manawatu Fault System [Lamarche et al., 2005; Nicol and Wallace, 2007].
 New Zealand's largest historic earthquake, the Mw 8.2 Wairarapa earthquake of 1855 [Grapes and Downes, 2010], resulted in about 150 km of surface rupture on the Wairarapa Fault, where horizontal displacements of up to 18 m were accompanied by regional uplift and tilting to the northwest [Rodgers and Little, 2006]. Elastic dislocation models suggest movement on a west-dipping listric fault [Darby and Beavan, 2001].
 The Hikurangi subduction system exhibits a range of slip conditions along strike, with strain release by earthquakes, slow slip events (SSEs), and repeating microseismic events [Wallace and Beavan, 2010]. Contemporary GPS data and seismicity indicate that the subduction interface beneath the southern North Island is currently strongly coupled or locked (Figure 1a) to depths of 25–30 km over a 90–110 km wide zone perpendicular to the strike of the margin [Darby and Beavan, 2001; Walcott, 1984; Wallace et al., 2009]. This is in contrast to the northern Hikurangi margin, where the maximum locking depth is much shallower at <15 km [Wallace et al., 2009]. Although there is no historic record of great megathrust earthquakes (Mw > 8.0) at the Hikurangi margin [Webb and Anderson, 1998], the crustal strain accumulation measured by cGPS techniques suggests that the southern Hikurangi plate interface fault will eventually undergo slip as Mw 8.0–8.5 or greater earthquakes [Wallace et al., 2009].
 Interplate slip is broadly reflected in long-term geological strain rates within the overriding plate. Crustal strain is concentrated above the downdip end of the locked plate interface - approximately 30 km deep beneath Wellington - and near to the surface expression of the subduction thrust in the Hikurangi Trough [Nicol and Beavan, 2003]. Long-term geological coupling on the subduction interface downdip of this locked patch has caused the Wanganui Basin to subside at an average rate of ∼1 km/Ma, and the basin depocenter has migrated southward and eastward with time [Stern et al., 1993] (Figure 1a). Seismic stratigraphy in the Wanganui Basin records ongoing subsidence for ∼4 Ma [Anderton, 1981; Lamarche et al., 2005] and suggests subduction mechanics have been consistent since the Pliocene.
 Leeds University deployed an array of seismometers in the early 1990s [Stuart et al., 1995] that were used to invert direct P, S, and converted SP and PS phases from local earthquakes for shallow P wave velocity structure and depths to the slab interface [Reading et al., 2001]. Topography on the converting interface shows that the plate boundary, under the Tararua Ranges, changes westward and downdip from shallow-dipping (<5°) to more steeply dipping (>15°). Receiver function analysis of structures beneath Wellington was also undertaken by Savage , Savage et al. , Boyd et al. , and Ewig . Earthquake tomography studies show spatial variations in material properties that may reasonably be expected to influence plate coupling and rupture behavior during large earthquakes. For example, part of the region of strong plate coupling beneath Wellington coincides with high Vp/Vs and low Qp of surrounding material [Eberhart-Phillips and Reyners, 2012; Reyners and Eberhart-Phillips, 2009]. However, elsewhere on the margin, the same properties (high Vp/Vs, low Q) are indicative of weak coupling, suggesting that either these physical properties do not correlate with changes in interface behavior, or that greater resolution is required near the interface [Eberhart-Phillips et al., 2005, 2008].
3. SAHKE Experiment
 The SAHKE experiment was carried out in two field seasons. The first phase (SAHKE-I) was a wide-angle onshore-offshore reflection-refraction survey and passive instrument deployment between November 2009 and April 2010, consisting of several subarrays deployed at various times. The second phase (SAHKE-II) was an onshore active-source experiment in May 2011 recorded on a single array. The experiment was designed intentionally to collect crossover recording of active sources and earthquakes. The total data volume provides regional three-dimensional (3-D) ray coverage beneath the southern North Island at high spatial resolution and a NW-SE multichannel and wide-angle crustal transect in the dip direction of the subducting Pacific Plate. A summary of experiment components and line-naming conventions is given in Table 1 and Figure 2.
Table 1. Experimental Nomenclature Used in the Text and Figure 2
Name for the whole 2 year and two-phased SAHKE-I and -II experiment of onshore-offshore recording to determine the structure of the plate boundary
Pegasus Basin MCS survey
Pegasus MCS line numbers
SAHKE 2-D array
Forty-eight short-period sensors deployed to record PEG09 airguns and also local and teleseismic earthquakes
Central profile of 37 short-period sensors redeployed from the SAHKE 2-D array and forming the onshore-offshore array recording airguns from SAHKE01 and SAHKE02
A line of 10 broadband seismometers operating during the period November 2009 through May 2011
Eastern MCS line and recorded also by 16 OBS and onshore stations of the SAHKE transect
Western MCS line and recorded by four OBS and onshore stations of the SAHKE transect
Northern MCS line and recorded by nine onshore stations north of Wanganui
Eastern MCS line and recorded by an array of eight shore stations
Eastern MCS line and recorded by an array of six land stations
Second phase of SAHKE, in which 871 land stations recorded 12 land explosions
 During SAHKE-I 57 short-period (L-22, 2 Hz) three-component (3C) land seismometers (RefTek RT130 from the IRIS-PASSCAL instrument pool) and 10 broadband instruments were deployed as a distributed array along the SAHKE transect in the Wellington-Wairarapa region between November 2009 and the beginning of March 2010 (Figure 1a) [Seward et al., 2010]. The broadband array remained in place until May 2011. The primary focus of this deployment was to record offshore airgun seismic sources during a multichannel seismic survey of the Pegasus Basin (PEG09), which lies at the eastern (frontal thrust) end of our transect. The distributed array had an instrument spacing of ∼7 km and was complemented by 25 permanent national network (GeoNet) stations. A total of 92 instruments were used to record >69,000 airgun sources from a grid of 17 offshore lines.
 This array was then redeployed as four lines of seismometers in March 2010, when the survey ship (M/V Reflect Resolution) returned to complete the PEG09 survey. An additional 400 km of marine multichannel seismic-reflection (MCS) data (10 km streamer length, 6000 in.3 source, and shot spacing of 100 m) were acquired along a central double-sided onshore-offshore transect across the margin (SAHKE01 and SAHKE02). These were also recorded with 38 recorders from the distributed array (redeployed along a line with L-28 4.5 Hz sensors) and co-linear with 20 ocean-bottom seismographs (OBS; 16 on the eastern side spaced 5 km apart and 4 on the west spaced 10 km apart). The transect also reoccupied stations of the Leeds array. An additional 340 km of MCS data along three secondary (one-sided) onshore-offshore transects (SAHKE03, PEG09–23, and PEG09–25) were also acquired. Details of the field deployment and processing steps are given by Seward et al. . MCS data were processed by Geotrace . Sections presented here have 12.5 m common depth point reflection (CDP) trace spacing and are prestack time migrated.
 The distributed and transect arrays recorded >100 local earthquakes and 72 teleseismic events (>6.0 Mw) over the November to April period, including the 27 February 2010 Mw 8.8 Chile earthquake and aftershocks.
 In SAHKE-II, 12 borehole explosive sources (350–500 kg) were distributed approximately 8 km apart along the central transect and detonated between 10 and 13 May 2011 (Figure 1b) [Seward et al., 2011]. The energy was recorded on 835 seismic stations (interleaved 277 three components and 558 vertical sensors) deployed at a nominal 100 m spacing, except for a denser section of 50 m spacing on the central part of the transect to image the shallow part of the Wairarapa Fault. The 3C stations were spaced 300 m apart and comprised 831 single component RefTek RT125 “Texan” instruments configured as three recorders connected to L-28 4.5 Hz sensors, which were oriented to magnetic N, E, and vertical. The remaining 1C stations consisted of Texan and Hakusan LS8200SD instruments connected to 4.5 Hz high-frequency vertical component sensors. Additional RT130 3C seismometers were deployed at each shot point and colocated with Texan and Hakusan stations. Data from collocated RT130, RT125, and LS8200SD instruments show similar instrument responses.
 We present transect results of the SAHKE active-source first-arrival data, using the results of both SAHKE-I and II onshore-offshore and onshore explosive components along the central transect.
 Offshore-onshore receiver gathers from many stations have clear arrivals at offsets >150 km. East and west coast receiver gathers for the central station TS030, coincident with shot point 9, are shown in Figure 3 as one “supergather” [Okaya et al., 2002]. Wide-angle phases are received from both airgun profiles SAHKE01 and SAHKE02 and can be viewed as a split-spread shot gather, that is, as if the source was at TS030 and receivers at each shot point. The inclusion of shot point 9 from the second phase of SAHKE (collocated close to station TS030) completes the supergather.
 Onshore and from the eastern (trench) side, prominent first arrivals at transect km 200 are interpreted as turning waves associated with sedimentary basins (Pg). In the west, first arrivals visible to offsets of 100 km (km 100) are considered to be diving waves in the Australian continental crust (Pg). A few seconds after Pg, the top of a bright and deep band of crustal reflectivity (Figure 3b) is considered to be caused by precritical wide-angle reflections (PintP). These are observed at 8 s vertical incidence under shot 9 (Figure 3b). On the east coast, this phase arrives before a deeper, weaker phase that does not appear on near offsets but is prominent on stations located east of the Wairarapa Fault and at offsets >50 km. We consider that PintP is a reflection from the plate interface, from the weaker phase is a reflection from the top of the oceanic Hikurangi Plateau (PtopP). Offshore MCS data (SAHKE01; Figure 3e) across the Hikurangi Trough record phases at 4–6 s two-way travel time (twt), which we consider to be a continuation of PintP and which image the active plate décollement (PintP) at ∼10 km depth. Beneath the décollement, reflectors suggest that a sequence up to 2 s twt thick being subducted along with underlying oceanic rocks of the Hikurangi Plateau. Elsewhere on the margin this distinctive seismic sequence is mapped by Davy et al.  as Mesozic sediments (MES).
 Phases arriving at a time consistent with reflections from the base of the downgoing Hikurangi oceanic crust (PbaseP) are identified on OBS data from the Hikurangi Trough and on land receivers from shots west of the Wairarapa Fault and across the Tararua Ranges but are less clear on onshore-offshore gathers across the west coast (Figure 3). Prominent phases arrive on west coast onshore-offshore gathers at offsets >100 km at times consistent with reflections (PmP) from Australian crust Moho. The crossover of PintP and PmP merge with other crustal phases at transect km 110–120, resulting in a region of complexity in the coda of the receiver gathers. Also, at km 60–70, arrivals with moveout consistent with diffractions (PdP) appear to originate from the Moho reflection. Phases at ∼10 s twt are visible on MCS data from the western offshore sector of the transect, and these are consistent with being Australian Plate Moho reflections (SAHKE02, Figure 3e). Onshore-offshore data from SAHKE02 also record arrivals at offsets >150 km (km 25–40 in Figure 3b) with an apparent velocity of 8.0 km/s, which we interpret as Pn traveling in the mantle of the Australian Plate. In the east, OBS stations (not shown) record large-offset (>200 km) Pn phases with very high (>8.5 km/s) apparent velocity interpreted as refracted phases through the mantle (and/or lower crust) beneath the Hikurangi Plateau. Especially prominent on onshore data are phases observed at 7–8 s twt beneath shot points 9–12 and between distances 140 and 240 km along the transect (Figures 3 and 7), which we consider to be precritical lower crustal (PcP) reflections.
4.1. First-Arrival Tomography
 We picked first arrivals from all onshore-offshore SAHKE01 and SAHKE02 profile shots into all OBS and transect surface seismometers. These arrival data were combined with first arrivals from 12 SAHKE-II land explosion shot gathers, totaling 17,773 travel times. We model this collection of first-arrival travel times from all available transect data by inverting for P velocity structure using existing tomographic methods [Eberhart-Phillips, 1990, 1993; Thurber and Eberhart-Phillips, 1999]. The tomographic method solves along ray paths in three dimensions, but we choose a large out-of-transect plane grid-node spacing to effectively create a 2-D parameterization of velocity structure. In the vertical (z) dimension, we use near-surface nodes at 1 km intervals from +1 km to −1 km throughout the inversion scheme. At greater depths, we progressively reduce the node spacing from 15 km in the x direction and 5 km in the z direction to 2 and 1.5 km, respectively.
 We start with a one-dimensional regional velocity model (Figure S1, supporting information).1 Offshore, this starting model is modified to account for bathymetry and sediment velocities. From the seabed to a depth of 12 km, we set our initial velocity model to a spatial average of velocities determined by optimal migration and stacking of MCS data.
 Between each of the six grid refinements, we perform three progressive inversions, allowing each node to vary by a maximum of 0.3 km/s. As the inversion steps progress, nodes that exceed velocities >8 km/s are forced to 8 km/s and held fixed for subsequent inversions. The root-mean-square (RMS) travel time residual was reduced from 0.45 s to 0.04 s after 18 iterations (Figure 4). A 2-D grid of derivative weight sum (DWS), which is a relative measure of the ray distribution [Toomey and Foulger, 1989], was constructed and used to discriminate between constrained and unconstrained regions of the first-arrival model (Figure S2b). The DWS weights each ray path length according to its spatial separation from the grid node. Larger DWS values indicate better data coverage. A DWS of <10 was used to mask the velocity model in Figure 3d.
4.1.1. First-Arrival Resolution Test
 A series of checkerboard tests was used to estimate the spatial resolution of the first-arrival tomography results. Using the source-receiver geometry of the data along the SAHKE transect, synthetic data were generated for a series of known models to test the recovery of velocity perturbations by the inversion at various resolutions. These models consist of alternating anomaly patterns of ±0.3 km/s perturbations superimposed on the starting model used for first-arrival tomography. The anomaly pattern function has the form sin(x)sin(z). We test both shallow and deep resolution by imposing fine and coarse checkerboard grid perturbations, respectively (Figure 5). The synthetic data were then inverted in the same manner as the real data. In order to quantify how much of the ±0.3 km/s velocity perturbations was recovered, the starting model was subtracted from the final models.
 The 15 km × 10 km (horizontal and depth) checkerboard analysis (Figure 5b) shows that the shallow structure is well resolved in the upper crust down to about 15–20 km depth beneath the SAHKE-II land explosions and profile SAHKE01 offshore shots. At greater depths, fewer rays (Figure S2a) result in poorer resolution, but a broad scale (35 km × 20 km) structure is recovered. However, while the overall perturbation pattern is preserved, the inversion has difficulty in recovering positive velocity anomalies at depth.
4.1.2 Previously Published Velocity Model Comparison
 Our first-arrival velocity model along the SAHKE transect is similar to but also differs in several respects from the 3-D earthquake tomography model previously published [Eberhart-Phillips and Reyners, 2012]. Figure 6 shows the two velocity structures. The active source first-arrival velocity structure has resolved more detail in the shallow crust and images higher velocities in the Australian Plate. In general, we do not observe the broad lateral velocity variations that mark terrane boundaries as proposed by Eberhart-Phillips and Reyners , but instead we attribute fine-scale lateral variations to faulted structures. The significant difference is the low velocities (<6.5–7.0 km/s) in the lower crust and in the crust of the subducting Pacific Plate observed in our model to depths of 40 km. The central region of the transect is not well resolved below 15 km, but low Vp in the subducting crust was not observed in earthquake tomography. This study and the Eberhart-Phillips and Reyners  study use the same tomography code but different input data. It is outside the scope of this paper to do a 3-D inversion of all active-source (reflected and refracted phases) and earthquake data combined from the region.
4.2. Seismic Reflections
 We converted the first-arrival velocity image to twt and displayed it with the offshore CDP section in Figure 3e. In addition, we constructed a CDP stack of the crust along the transect profile using only precritical reflections from SAHKE-II land explosions (Figures 7a and 7b). A detailed section of the stacked image beneath the Tararua Ranges is displayed in Figure 7c and individual shot stacks are shown in Figure S3. Details of the processing scheme are included in Table S1. Reflection phases identified in Figure 3b were positioned in depth by 2-D depth-migration [Van Avendonk, 2004] using the first-arrival velocity field. Those phases identified on TS030/SHOT 9 are shown in Figure 3d and additional PcP reflection arrivals picked on shot and receiver gathers are shown in Figure 8. All picked reflection arrivals are also indicated in Figure 9a. Tomography and prestack depth migration (PSDM) of the reflection coda are subjects of separate studies.
4.2.1. Plate Interface
 Our interpretation of the plate interface reflection (PintP), identified as Y (SAHKE01) in Figure 3e, is marked by a distinct change in character and dip beneath shot point 8, just east the Wairarapa Fault, at model distance km 205 (Figures 7 and 9a). To the east, PintP is a weak reflection at ∼6 s twt (Figure 7), but dips westward as a distinct band of reflectivity to 8–10 s twt beneath the Tararua Ranges. Using velocities obtained from first-arrival tomography and depth migration of picked PintP reflections from shot point 9 (Figures 3b and 3d), the plate interface is at a depth of ∼10 km beneath the east coast, deepening to >25 km beneath Kapiti Island and the west coast. Our PintP picks are close to or <1 km shallower than the surface identified by Reyners and Eberhart-Phillips  as the plate interface on the basis of an upper envelope of intraslab seismicity (Figure 10). Our crustal model has velocities that are about 10% slower than the 3-D model of Reyners and Eberhart-Phillips . This is sufficient to account for the difference and results in good agreement between their interpretation of earthquake hypocenters and our pick of plate interface reflections (Figure 9a).
4.2.2. Hikurangi Plateau Oceanic Crust
 Deep near-vertical incidence reflections coinciding with the top of the Hikurangi Plateau large igneous province (PtopP) can be traced offshore to the east on SAHKE01 but are weak beneath onshore shots 1–7. PtopP reflections appear to merge and are equivalent to PintP phases beneath shot point 10 (km 180), where there is also a change in the velocity structure (Figure 9). To the east, the top of the Hikurangi Plateau has a strong velocity gradient, but this gradient is lower under shots 8–12 (km 170–205). We interpret this step change in velocity gradient as a ramp thrust that dips >10° to the west (km 180). PtopP phases map ∼3 km deeper than the plate interface reflection (PintP) beneath the eastern onshore region (200 < km < 250), and the region is marked by a distinct boundary in seismicity (Figure 10). Here, earthquake hypocenters in the oceanic plateau are characterized by vertical clusters, suggesting near-vertical faults [Du et al., 2004], whereas earthquakes to the east are sparse. We interpret the base of the Hikurangi Plateau crust as a band of discontinuous reflectors (PbaseP) parallel to and up to 4 s twt later than PintP (Figures 7 and 9). PbaseP maps to the lower envelope of seismicity and implies that the Hikurangi Plateau is ∼13 km thick beneath the southern part of the North Island, which is slightly less than has been suggested offshore [Davy and Wood, 1994].
4.2.3. Australian Plate Moho
 The base of the crust at the western end of the line is identified from PmP arrivals identified on OBSs located west of Kapiti Island, on MCS line SAHKE02, and on onshore stations that recorded offshore shots. PmP reflections migrate to ∼30 km depth and define the Moho in the Australian Plate as an east-dipping (∼10°) interface. However, the region near to where the subducting Hikurangi Plateau intersects the Australian plate Moho is complex and additional reflected phases are identified in Figure 3. A coda of apparently diffracted arrivals (PdP) seems to originate from the Moho, since this event merges with PmP (km 80 in Figure 9). Migration of the picked PdP event focuses to a single steeply dipping (>60°) plane that corresponds to the deepest extent of the Taranaki Fault (km 90), which offsets the Moho. More detailed modeling of reflection data is required to constrain the change in crustal thickness across the Taranaki Fault, but our data indicate that the crustal offset is approximately 5 km in agreement with previous seismic reflection data in this region [Stern and Davey, 1990].
4.2.4. Crustal Reflections
 Crustal reflections (PcP) above the plate interface are observed in the CDP stack (Figure 7) as a zone up to 3 s twt (∼8 km) thick and at least 30 km in width (Figure 7c and Figure S3) beneath the Tararua Ranges (shots 8–12; 170 < km < 205). Velocities in the highly reflective zone are lower than surrounding regions and range from 4.5 to 5.0 km/s. The top of the prominent high-amplitude reflection zone at 12–15 km depth is cut by a prominent west-dipping splay-fault reflector (A in Figure 7) that is semicontinuous for 50 km eastward from where the reflection merges with the plate interface beneath Kapiti Island, to where its dip steepens from 15° to >40° beneath shot point 9, and projects to the surface where the Wairarapa Fault is mapped. Other, less-prominent splay faults (B and C in Figure 7) also develop from near the step in the plate boundary beneath shot 8 (km 205) and also appear to project to the surface near known faults that are mapped east of Martinborough [Lamb and Vella, 1987; Nicol et al., 2002].
4.3. Receiver Functions
 Receiver functions from Tararua array stations show prominent positive arrivals at 2, 4, and 5 s (TAR in Figures 9b and 9g), and have been, respectively, interpreted as conversions from a relatively flat boundary at 12 km, possibly caused by a change in metamorphic grade, and the top of subducted oceanic crust at 28 km [Savage et al., 2007]. Negative polarities and back-azimuthal variations of receiver functions before the 4 s arrival were interpreted as conversions from a 4 km thick 4% anisotropic low-velocity (Vs = 2.55 km/s, Vp = 5.1 km/s) layer of metamorphosed sediments at the top of the subducting plate (between 24 and 28 km depth). The oceanic Moho was modeled at a depth of 38 km, yielding a thickness of 10 km for the oceanic crust in this region. However, the deepest arrival was weak, and therefore the thickness is not well constrained. A further arrival at 7–9 s could possibly be caused by a deeper Moho. Absolute velocities and depths of interfaces are less well constrained by receiver function studies than relative velocities [e.g., Ammon et al., 1990].
 Figures 9f–9i compares the measured receiver functions at station W3 (see Figure 1 for location) from Savage et al.  at the most common back azimuths with two synthetic models (Table 2) calculated using a ray-theoretical method [Frederiksen and Bostock, 2000]. The anisotropic model TAR [Savage et al., 2007] was devised to match all the Tararua stations. An isotropic model, SAHKE-W3 (Figure 9h), was determined from the tomographic modeling to represent the material through which rays passed to reach W3, and using a constant Vp/Vs ratio of 1.78 and a model, SAHKE-RCVR (Figure 9i), which was forward modeled using the constraint that the boundaries and P wave velocities were kept as in model SAHKE-W3, the S wave velocities were revised to try to match better the amplitudes and arrival times of the data at W3. All the models match the general character of the data (Figure 9f), with the initial peak at 0 s, another early 2 s peak, and a trough followed by two more peaks and further troughs. TAR (Figure 8g) has the 2 s peak arrive late, at 2.5 s, and the trough is also late by ∼0.5 s. The following two peaks are separated, making the last peak arrive a full 2 s late. SAHKE-W3 has the 2 s peak and the trough arriving 0.5 s early, and the amplitude of the next peak is too high relative to the following peak, but their arrival times are about right. For SAHKE-RCVR, the amplitudes are smaller than the other two synthetics and closer to the data. Interestingly, in model SAHKE-RCVR, the low-velocity layer at the top of the slab has a relatively low Vp/Vs ratio of 1.70 rather than the expected high Vp/Vs ratio of a fluid-filled layer. To create a 2 s peak with the constraints on the Vp and depth models would require a very low Vs and an unrealistically high Vp/Vs ratio. Ewig  reinterpreted Tararua array data and added new receiver function calculations for KIW (Figure 9b), which is on Kapiti Island and is the westernmost land station on our transect. KIW did not yield any conversions at 2 s, which suggests that it might be a local feature beneath the Tararua Ranges. There was a strong arrival at 5.5 s preceded by a negative arrival, which he interpreted as the extension of the pair slab/above-slab low-velocity zone to the west [Ewig, 2008]. He interpreted the thickness of subducting oceanic crust beneath KIW to be also 10 km.
Table 2. Isotropic Velocity Models Discussed in Texta
ρ (kg m−3)
ρ is density. St is the strike of the top of the interface. Dip is the dip of the top of the interface, measured from horizontal according to the right-hand rule in the convention of Aki and Richards . Dep is the depth to the bottom of the interface.
5.1. Subducting Sediment Channel and Crustal Underplating
 Marine seismic line SAHKE01 images the trench slope and active plate décollement (PintP) at ∼10 km deep (Figure 3), coinciding with the upper part of a condensed sequence of strongly reflective Late Cretaceous-Early Oligocene (70–32 Ma) marine nanofossil chalks, mudstones, and subordinate chert—referred to in previous work as sequence Y (Seq Y in Figures 3e and 10a) [Davy et al., 2008]. Beneath the décollement, up to 3.5 km of Mesozic sequences are being subducted along the Hikurangi margin (Figures 3e and 10a, MES of Davy et al. ) together with underlying oceanic rocks of the Hikurangi Plateau. PSDM of PEG09–19, south of SAHKE01 (Figure 1), shows that the Cretaceous and Paleogene sedimentary rocks constitute a ∼3.5 km thick subduction channel beneath the active plate boundary thrust dipping ∼2° landward and roofed by the top of sequence Y [Plaza-Faverola et al., 2012].
 The lower crust beneath the Taraura Ranges is imaged from first-arrival tomography and receiver functions as a low-velocity triangular zone with prominent high-amplitude reflectors and is interpreted as an imbricated sequence of layers ∼8 km thick above the subduction interface at ∼25 km depth. The stacked reflection section (Figure 7c) suggests this zone contains multiple west-dipping layers up to ∼2 km thick with reflective boundaries. Such stacked slices of Vp = 5.2 km/s from 15–23 km depth are also consistent with forward modeling of receiver functions. A ramp structure in the deepest slice coincides with a 10° increase in dip of the subducted plate, which we interpret as a step down in the active subducting plate decollément at transect distance km 180, where west-dipping splay faults are inferred (B and C in Figure 7). Another pronounced splay fault (A) cuts the top of the imbricate sequence at a depth of 15–20 km and projects to the surface as the Wairarapa Fault.
 We propose a continuation beneath the North Island of the subduction channel imaged at the eastern end of the transect. The step down of the active plate interface at km 180, inferred from reflector geometry, may represent a transfer of slip to the top of the oceanic crust and would allow low-velocity and low-density Cretaceous and Paleogene strata to be underplated. We propose that continued plate convergence has stacked imbricated sheets and driven uplift of the Tararua Ranges.
 We expect subduction channel material, probably sediments, to have low velocities due to relatively high porosity and enhanced fluid pressure that could lead to high attenuation (i.e., low Qp). Low velocity is identified from our tomography, and low Qp (<400) is identified along the SAHKE transect beneath the Tararua Ranges and near the top of the subducting plate (yellow line in Figure 10) from earthquake tomography [Eberhart-Phillips et al., 2005; Reyners and Eberhart-Phillips, 2009]. The 7–8 km vertical and 11 km horizontal resolution used in the Q analysis is coarser than our tomography, but the location of the Qp < 400 contour (Figure 10) corresponds closely to the region we identify as a subduction channel and underplated duplexes in the lower crust. Low Qp values extend downdip into the inferred source region of slow-slip events west of Kapiti Island [Wallace and Beavan, 2010].
 A low-velocity layer above subducting Hikurangi Plateau crust was previously identified from a 300 km long SW-NE wide-angle reflection/refraction experiment between Palliser Bay and Hawke Bay [Chadwick, 1997]. At the southern end of this profile, where it crosses the SAHKE transect (Figure 8d), a ∼2.0 km thick low-velocity layer of 3.6 km/s is interpreted to lie immediately above 7.1 km/s oceanic crust at 17 km depth. Models of aftershocks from the 1990 Cape Palliser earthquake sequence similarly find a <2 km thick 3.6 km/s layer south of where the SAHKE transect intersects the east coast [Luo, 1992]. South of SAHKE01, PSDM of MCS line PEG09–19 indicates that the subduction channel is 3.5 km thick offshore Cape Palliser, and the velocity of Cretaceous and Paleogene rocks in the channel is between 4.4 and 4.7 km/s at the deformation front [Plaza-Faverola et al., 2012]. The impedance contrast across the active plate interface and sequence Y is small (∼0.05), but thin (<100 m) layers are not resolved by PSDM velocity analysis or our tomography.
 Our first-arrival tomography indicates that material within the 3 km thick channel below the plate interface has a consistently slow velocity (<5 km/s) east of transect distance km 240 (Figure 9). Farther west beneath shots 4–8 (200< km < 240), we were not able to distinguish a low-velocity zone or separate reflectors associated with the plate interface (PintP) from those associated with the top of oceanic crust (PtopP). We speculate that a 40 km wide seamount may be subducting beneath the margin at this location. Downdip of the putative seamount, we again find evidence for a low-velocity channel and high-amplitude crustal reflectivity (Figure 9). The base of this reflectivity zone appears to coincide with the top of the oceanic crust (PtopP).
Reading et al.  suggested that the relatively flat slab at a depth of 18 km underneath the central section of the Wairarapa (200 < km < 240) develops a pronounced increase in dip to 15° at ∼km 200, with the plate deepening to 30 km beneath Kapiti Island (km 150). Our study images the top of oceanic crust at 15 km beneath the east coast (km 250), but it dips ∼5° and reaches 22 km at km 200. Prominent reflections are interpreted as the top of a subducting sedimentary channel and faults in underplated duplexes, and at 18 km, they are about 3–4 km shallower than the top of the slab. One possibility is that the phase that was converted from S to P occurred not at the plate interface, but from this shallower underplated region, and hence the velocity used to determine 17–18 km may be too high. Farther north, Reading et al.  modeled a 2–4 km thick layer of underplated sediments, and this confirms that our suggestion of a seamount subducting beneath the Wairarapa may be correct, but the seamount may extend <30 km to the north.
 Underplating is observed elsewhere along the margin off Hawke Bay [Henrys et al., 2006] and Raukumara [Bassett et al., 2010]. The spatial correlation observed in the northern Hikurangi margin between the inboard edge of the low-velocity prism and the intersection between the subduction interface and forearc Moho is also observed on the SAHKE transect. This suggests that the location of lower crustal underplating along the whole margin is modulated by Moho depth. This is consistent with the previously proposed cyclical forearc crustal dynamic in which the density of subducted and tectonically eroded strata is sufficient for it to escape from the subduction channel near the base of the crust, driving underplating and surface uplift [Scherwath et al., 2010; Sutherland et al., 2009].
5.2. Crustal Faults
 Our stacked reflection image (Figures 3 and 7) provides a cross section through the Wellington region. Together with our Vp tomography and previous work, the relationships of crustal faults to the subduction interface beneath can be examined.
 At the western end of the transect, MCS line SAHKE02 images reveal sunken topography and faults beneath Wanganui Basin (Figure 3e). The largest two-fault systems correspond to the Kapiti-Manawatu fault system [Lamarche et al., 2005], and farther west, a probable continuation of the Taranaki Fault [Holt and Stern, 1994; King and Thrasher, 1996; Rattenbury et al., 1998; Stagpoole and Nicol, 2008]. Both faults dip moderately or steeply eastward with the Taranaki Fault intersecting the Moho at a depth of approximately 28 km and offsetting the crust.
 East of the Tararua Ranges, we recognize two fault strands (B and C in Figure 7) that splay off the subduction interface beneath the Wairarapa Fault. We infer these to be active thrust faults that bound the edge of late Neogene sedimentary basins. The western fault (B) has surface expressions as the Martinborough or Huangarua faults [Begg and Johnston, 2000; Lamb and Vella, 1987; Nicol et al., 2002]. The eastern fault crops out just offshore near the coast, and movements on this fault may be responsible for uplift of the coastal range and for coseismic uplifts of Holocene beach deposits [Berryman et al., 2011].
 Reflections from the Wairarapa Fault show it is a listric fault ramping up eastward from depths of 12–15 km beneath the Tararua Ranges, where it dips at ∼15°, to a dip of >40° beneath shot point 9 and projects even more steeply (>60°) to the surface (Figure. 9). PcP arrivals on receiver gathers (Figures 3 and 8) indicate the fault appears to bound the upper surface of reflective low-velocity material, possibly underplated sediment, and sole into the region of crust where the plate interface intersects Australian Plate Moho at about 32 km depth—near the downdip end of the strongly locked coupled zone (Figures 10a and 10b) at 29 ± 1 km depth [Darby and Beavan, 2001]. The CDP stacking method does not image steeply dipping structures, and PSDM is required to help better resolve dipping events. We see no clear signature of the Wellington Fault at depth. We hypothesize that the Wellington Fault either follows a similar listric geometry or soles steeply into the same splay-fault strand (A) that we identify as the Wairarapa Fault. The 10 km shot spacing between shots 8, 9, and 10 means that data from regions shallower than 12–10 km are missing between shots. Closer shot and receiver spacing is needed to image these faults in the near surface.
 Our observation of a listric geometry for the active strike-slip Wairarapa Fault at depths below 10 km indicates the fault is poorly orientated with respect to the direction of maximum horizontal stress [Townend et al., 2012] and implies the fault must be weak at depth.
5.3. Relationship to Locking and Slow Slip
 The relationship between our seismic observations and geodetic data are summarized in Figure 10.
 Relocated hypocenters, as part of 3-D tomographic inversion [Reyners and Eberhart-Phillips, 2009] along the SAHKE transect, reveal increased intraslab seismicity beneath the strongly coupled part of the interface, where we infer underplating. Slab seismicity also correlates with an increase in reflectivity of the subducting Hikurangi Plateau (Figure 7), where our velocity model observes Vp < 7.0 km/s (Figures 3 and 9) and Vp/Vs > 1.8 (red in Figure 10) [Eberhart-Phillips and Reyners, 2012]. The highest seismicity is observed within the subducting plate in a 10−12 km thick zone between km 170 and 220. Normal faulting involved in the 2004 Upper Hutt swarm beneath the Tararua Ranges (km 180), and the subsequent 2005 ML 5.5 main shock, represent unbending of the subducted plate [Reyners and Bannister, 2007]. Further updip, normal faulting in the top of the subducted plate represents bending of the subducted plate [Du et al., 2004; McGinty et al., 2000]. These faults, which are likely inherited from normal faults as observed on the Hikurangi Plateau [Plaza-Faverola et al., 2012], are now buried at a depth of >20 km beneath the forearc and are reactivated by bending stresses and increased pore pressures from dehydration of the subducted crust at temperatures of 100°C–350°C [McCaffrey et al., 2008]. Overpressured fluids might be trapped in the subducting channel, particularly if sequence Y acts as an impermeable seal. We suggest that the presence of underplated material may be inhibiting any slab fluid release, as an alternative to the concept of an impermeable terrane boundary [Eberhart-Phillips and Reyners, 2012]. In either case, the change in plate dip in the region of Vp/Vs > 1.8 reflects deformation of the slab, probably caused by dehydration [e.g., Yamasaki and Seno, 2003]. With dehydration, there is a phase change to much denser material (eclogite) and likely to expect an extra torque being imposed at that point.
 The relationship between this underplating and slow slip, however, remains unclear. For example, at Nankai, high Poisson's ratios > 0.30 (Vp/Vs > 1.9) in the subducted oceanic crust are interpreted as high pore fluid pressure that likely generates slow slip [Kodaira et al., 2004]. Also at Cascadia, slow slip and tremor may be facilitated by trapped fluids and high pore fluid pressures in underplated subducted sedimentary rocks [Calvert et al., 2011]. But at southern Hikurangi, Vp/Vs > 1.8 and low Qp (<400) [Eberhart-Phillips et al., 2005], the underplating observed on the SAHKE transect are found in the subducting crust below the locked part of the interface, east of the slow slip region. However, the listric geometry of the Wairarapa Fault implies the fault plane at depth is mechanically weak, and we cannot rule out the deeper surface as a source of slow slip observations beneath Kapiti Island. Furthermore, underplating is observed further north along the margin off Hawke's Bay [Henrys et al., 2006] and Raukumara [Bassett et al., 2010] in unlocked regions of the margin. This would suggest that underplating of sediments is not a strong discriminant of geodetic behavior on the plate interface and calls into question the relationship of high Vp/Vs to the occurrence of slow slip.
 Changes in subduction dip may provide a first-order structural change along the strike of the Hikurangi subduction zone that best explains the locking pattern where increase in dip to angles greater than 8° defines a transition from locked areas of the plate interface to unlocked and partitions stable and unstable slip regimes. Along the margin, we have found this change in dip also corresponds to a steepening of the topographic slope to greater than 3° seaward of the kink [Barker et al., 2010]. The dip change may be a locus of inherent weakness in the subducting slab that relates to a northward increase in subduction rate that controls initial slab dehydration and fluid release. If the locked zone slips, either as an earthquake or aseismic event, then it will cause compressive stresses in the upper plate updip of the slip zone. Indeed, we observe thrust faults ramping up from where the dip of the plate boundary interface increases and faults step down, allowing underplating of sedimentary channel material to stack up. Permanent inelastic strain is manifested in the Wellington region as faulting and folding in the overriding plate for the last 5 Ma with localization of relatively high strains within the Tararua Ranges [Lamb and Vella, 1987; Nicol and Beavan, 2003]. We suggest the high uplift rates within the ranges are the result of underplating further, implying that the position of plate bending and the locus of thrust faulting have been sustained for the last 5 Ma, as suggested by Nicol and Beavan .
 In this study, we have developed a 2-D P wave velocity model across a 350 km long transect of the Hikurangi subduction zone beneath the Wellington region with the aim of determining properties of the locked subduction thrust that may fail in a future large earthquake. Our analysis uses a well-constrained first-arrival ray tracing inversion technique to model SAHKE active-source onshore-offshore and onshore explosion data. In addition, we have used the tomography velocity model to migrate picked reflection events and stacked low-fold land explosion data to image subducting slab geometry and crustal structure.
 We propose that plate convergence in the last 5 Ma has underplated and stacked imbricated sheets of sedimentary material from the top of the Hikurangi Plateau, on the overriding Pacific Plate, into the footwall of the Wairarapa Fault, driving uplift of the Tararua Ranges, and may be acting as a seal-trapping fluid in the oceanic crust. Reflections from the Wairarapa Fault show it is a steeply dipping listric fault that appears to bound the upper surface of reflective low-velocity underplated sediment and soles into the intersection of plate interface and the Australian Plate Moho at a depth of about 32 km and near the downdip end of the strongly locked coupled zone.
 Previously we have documented that there is a long-lived along-strike change in the dip of the Hikurangi subduction interface at 10–20 km depth that is spatially correlated with changes in surface slope and the pattern of sediment accretion, deformation, and underplating within the outer forearc crustal wedge [Barker et al., 2010]. We can now extend that correlation to the southern Hikurangi margin and further suggest that increase in dip to angles greater than 8° defines the transition from locked areas of the plate interface to unlocked and partitions stable and unstable slip regimes. As a result, mechanical behavior and seismic hazard of the subduction interface may also be spatially correlated with this along-strike change of interface geometry.
 Finally, abrupt slab-dip changes are inferred to have implications for seismicity, where bending points may act as barriers to rupture propagation [Contreras-Reyes et al., 2012; Ito et al., 2005]. This appears to be the case for Mw ≤ 7.5 earthquakes offshore from Miyagi [Ito et al., 2005] but step changes in subduction dip in the same area proved no barrier to slip from the 11 March 2011 Tohuku-Oki Mw >9.0 earthquake [e.g., Ammon et al., 2011; Kiser and Ishii, 2012; Lay et al., 2011a, 2011b; Simons et al., 2011]. When slip deficit is relieved on the southern North Island segment of the plate boundary, it may rupture in a single Mw 8.5–8.7 event [Wallace et al., 2009] and, given this scenario, we cannot rule out slip propagating all the way to the trench.
 The SAHKE project was supported by public research funding from the Government of New Zealand, the Japanese Science and Technology Agency, and the National Science Foundation (NSF OCE-1061557). Additional analysis support was provided by the Earthquake Research Institute visiting Researchers program (SH), a New Zealand Earthquake Commission Postdoctoral Fellowship (AW), the New Zealand foundation for Research, Science, and Technology (VUW) and by a VUW research and study leave grant (MKS). Acquisition of the MCS data was supported by the New Zealand Ministry of Business, Innovation, and Employment. We are grateful to the officers and crew of the M/V Reflect Resolution for acquisition of the marine seismic data and the M/V Mariner and M/V Resolution for deployment and retrieval of OBS instruments. Mark Henderson was instrumental in executing shot firing and in supporting instrument deployment. Explosives and technical support was provided by Orica New Zealand Ltd. Individual land owners, Greater Wellington Regional Council, Transpower, and forestry companies kindly allowed us onto their land. The IRIS/Passcal instrument pool provided the land instruments for the active and passive seismic experiments, and ANSIR provided the broadband instruments. Donna Eberhart-Phillips provided simul2006 tomography codes. We acknowledge the use of GLOBE Claritas® seismic processing software and support from the Claritas team. We also acknowledge the two anonymous reviewers who helped to improve the manuscript.