Geophys. J. Int. (2005) 163, 1180–1194 doi: 10.1111/j.1365-246X.2005.02716.x

Seismic tomography of the Tongariro Volcanic Centre, New Zealand

D. P. Rowlands, R. S. White and A. J. Haines Bullard Laboratories, Madingley Road, Department Earth Sciences, University of Cambridge, Cambridge CB3 0EZ, UK. E-mail: [email protected]

Accepted 2005 June 17. Received 2005 June 16; in original form 2004 November 19

S U M M A R Y Tomographic inversion of local earthquake P- and S-wave traveltime data is used to investigate 3-D P-wave velocity (Vp) and P-wave/S-wave velocity ratio (Vp/Vs) variations at the Tongariro Volcanic Centre (TgVC), New Zealand. P-wave model resolution is generally high throughout the TgVC at depths of around 3–9 km below sea level. Near-surface resolution is lower since rays at shallow depths are predominantly subparallel. Resolution decreases below 12 km depth due to clustering of seismicity at greater depths. The final 3-D Vp model shows low-velocity volumes from 3 to 7 and 8 km depth, respectively beneath the active volcanoes Ruapehu and Ngauruhoe. Smaller low-velocity volumes are observed beneath the active Tongariro volcano and the extinct Tama Lakes vents. Synthetic tests indicate that the resolution of the 3-D Vp/Vs model is insufficient to permit any reliable interpretation. Accurate earthquake depths are determined for the first time within the TgVC. Seismicity is cks largely restricted to two clusters west of Ruapehu, possibly with origins on eastward striking ro branches of the normal Raurimu Fault, two lineations near Waiouru in the south-east of the study area and a fault belt between Tongariro and Lake Taupo in the north of the study area. and Earthquake depths indicate possible shallowing of the brittle–ductile transition beneath Ruapehu and Ngauruhoe. fluids Key words: crustal structure, low-velocity zone, seismic velocities, tomography, Tongariro volcanic centre, volcanic structure.

the crust beneath the TgVC, together with accurate local earthquake

geothermics, 1 I N T RO D U C T I O N locations. , The Tongariro Volcanic Centre (TgVC) lies at the southern terminus of the Taupo Volcanic Zone (TVZ), a region of backarc rifting and 1.1 Previous geochemical/petrological studies of the TgVC intense volcanism formed in response to westward subduction of The Tongariro massif was generated during several periods of in- the Pacific plate beneath the North Island, New Zealand (Fig. 1). tense eruptive activity, beginning at least as far back as 275 ka with olcanology The TVZ exhibits predominantly andesitic volcanism in the north further bursts of growth at 210–200 ka, 130–70 ka and from 25 ka to V and south and extremely productive rhyolitic volcanism in the centre the present (Hobden et al. 1996). The most recent of these eruptive (e.g. Wilson et al. 1995). episodes has given rise to the relatively complex system of cones GJI The TgVC consists of several large, predominantly andesitic vol- seen today, with activity over the last 2500 yr confined principally canoes (Fig. 1): Ruapehu, Ngauruhoe and Tongariro are currently to Ngauruhoe. active, while Hauhungatahi, Pukeonake, Pihanga and Kakaramea Since modern records began (around 1830 AD) there has been are extinct. Other volcanic vents include the large flooded vents of an eruption of Ngauruhoe once every two or three years on average. Tama Lakes and several small craters near Ohakune. In this paper These eruptions have varied widely in style over short timescales, we will use the term TgVC to refer to the currently active portion as shown by the diverse chemical and isotopic compositions of of the area, namely the Ruapehu, Ngauruhoe and Tongariro massifs the erupted products (Hobden et al. 1999, 2002). No simple time- and surrounding ring plains including Hauhungatahi. composition relationship has been found for successive eruptions Much previous work on the TgVC has focussed on the geochem- and it is therefore unlikely that large, crustal magma reservoirs feed ical/petrological and seismological properties of the system. While the Ngauruhoe system. Small (<0.1 km3), short-lived batches of a great deal of knowledge has been gained concerning the eruptive magma are more likely to be responsible for the variability of the history of the TgVC volcanoes, seismological surveys have neces- observed eruptions. Such small volumes are extremely difficult to sarily been limited in scope due to a lack of instrumentation. As a resolve using seismic tomography, particularly if they contain low- consequence the crustal structure of the area is not known in any velocity material. detail. This paper aims to address this situation by presenting 3-D Vp Gamble et al. (1999) analysed the material erupted at Ruapehu and Vp/Vs seismic tomographic models of the uppermost 12 km of during the last 50 yr. Geochemical variations within this 50-yr time

C 1180 ! 2005 RAS Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1181

175.2˚E 175.4˚E 175.6˚E 175.8˚E 39.0˚S km

0 5 10

39.2˚S

35˚S

TVZ Distance along profile (km) Aus. 0 100 200 300 A 0 R A 100 40˚S 200

Pac. Depth (km) EQs 1994-2001 300 174˚E 178˚E

Figure 1. Main panel: Geographical map of the Tongariro Volcanic Centre. Tg—Tongariro; Ng—Ngauruhoe; Ru—Ruapehu; Hau—Hauhungatahi; Ka—Kakaramea; Pih—Pihanga; Puk—Pukeonake; KHS—Ketetahi Hot Springs; RC—Red Crater; TL—Tama Lakes; CL—Crater Lake. Lower left panel: tectonic setting of the TgVC. TVZ—Taupo Volcanic Zone; A—Andesitic sections of TVZ (black triangles denote large andesitic cones); R—Rhyolitic section of TVZ (light grey areas denote rhyolitic systems); Aus—Australian plate; Pac—Pacific plate. Small black rectangle shows area of main panel. Dotted line shows location of cross-section shown in lower right panel. Lower right panel: seismicity located within a 50-km-wide swathe by New Zealand National Seismograph Network, 1994–2001. White triangle marks location of TgVC. Note the large number of restricted depth locations indicating poor hypocentral constraint.

window were found to cover most of the range shown by lavas Further notable volcanic activity in the TgVC has occurred from erupted from the volcano over its entire history. This short timescale the Hauhungatahi and Tama Lakes vents. However, the most recent variability in magma chemistry was explained by influxes of fresh activity at Hauhungatahi dates from >340 ka (Hackett 1985), while magma into pre-existing pods of melt left over from previous erup- Tama Lakes are not thought to have erupted since the large 10 ka tions. The most recent eruptions of 1995–1996 were interpreted as Pahoka–Mangamate sequence of eruptions (Nairn et al. 1998). being due to one of these magma recharging events, an interpreta- tion shared by Nakagawa et al. (1999). Based on historical records, 1.2 Previous geophysical studies of the TgVC recent re-injections of magma are thought to occur every 20–30 yr. Gamble et al. (1999), therefore, concluded that Ruapehu is fed by Previous seismological studies of the TgVC have primarily focussed an open magma system which is, at least to first order, continually on Ruapehu and to a lesser extent Ngauruhoe. Following the up- active. grade of the permanent network in the TgVC in 1976 three vertical

C ! 2005 RAS, GJI, 163, 1180–1194 1182 D. P. Rowlands, R. S. White and A. J. Haines component stations operated at Ruapehu and Ngauruhoe. Latter model was then used to locate earthquakes recorded in the surround- (1981b) analysed seismicity from this early TgVC network and ing area over a 2 week period. Olson found several earthquakes at proposed a new system of earthquake classification for Ruapehu shallow depths beneath Ruapehu Crater Lake, preceding bursts of and Ngauruhoe. High-frequency events (>3 Hz) with sharp on- emergent, low-frequency volcanic events. Most of the remaining sets and clear phases were termed ‘tectonic’. Within 9.5 km of seismicity was located in the Waiouru area and around 35 km WSW Ruapehu and 8 km of Ngauruhoe such events were further de- of Ruapehu. Several events were also detected around National Park scribed as ‘volcano-tectonic’. Low-frequency events (<2 Hz) were village and around the southern flank of Ruapehu. All local seismic- termed ‘volcanic’ and were thought to be produced by a prolonged or ity, except that below Crater Lake, occurred between 9 and 31 km repetitive source mechanism within heat weakened material. These depth with deeper slab events located between 67 and 100 km depth. events were confined to smaller and shallower volumes beneath both However, the short recording period and the low level of seismic ac- volcanoes. tivity precluded the determination of a comprehensive picture of Latter (1981a) used earthquake data from the early TgVC network seismicity within the TgVC. and from two temporary seismometers to map highly attenuative A 2-month deployment of 14 broad-band seismometers in 1994 bodies within the upper crust near Ruapehu and Ngauruhoe. These in conjunction with the short-period permanent network stations bodies were interpreted as zones of partially molten rock. Five such permitted reasonable earthquake locations to be obtained through- zones were identified in the immediate vicinity of the two volcanoes. out the TgVC (Hurst & McGinty 1995). A southwards transition However, the uncertainties in the earthquake locations used (derived was found from the exclusively shallow (<10 km deep) seismicity from a maximum of only five stations) mean that the exact locations of the TVZ (e.g. Bryan et al. 1999), which persisted as far south and geometries of these attenuative bodies should be treated with as Lake Rotoaira, to deeper seismicity extending down to 20 km some caution. depth near Waiouru. This temporary deployment also enabled fur- An early investigation of the velocity structure of the southeast- ther investigation into the shallow events detected by previous re- ern ring plain of Ruapehu was made by Sissons & Dibble (1981). A search in the vicinity of Crater Lake. Volcano-tectonic events were 22-km NNE-striking profile was surveyed using several shots and located beneath Crater Lake and also in nearby areas to the south a combination of seismometers and geophone spreads. The loose and northwest, while low-frequency volcanic events were confined tephra of the near surface exhibited a very low Vp of less than to an area below the east side of Crater Lake (Hurst & McGinty 1 0.5 km s− . This was underlain by more consolidated tephra with 1995; Hurst 1998). Although the depth of the low-frequency vol- 1 Vp of 1.1–1.5 km s− and thicknesses of around 10–40 m. The next canic events could not be determined accurately, they were thought 1 layer down was around 600 m thick with Vp of 2 km s− , interpreted to occur within a 200 m thick zone somewhere between 100 and as lahar deposits emanating from Ruapehu Crater Lake. Beneath 1600 m below Crater Lake. the lahar deposits a layer of greywacke was detected with Vp of A series of earthquakes occurred some 15–20 km to the west of 1 4.9 km s− and a thickness of some 1.1 km. This layer was in turn un- Ruapehu a few months before the 1995 eruption (Hurst & McGinty 1 derlain by more consolidated greywacke with Vp around 5.4 km s− . 1999). The two strongest swarms took place shortly before rapid 1 In the southern section of the profile a 3 km s− refractor was also increases in the temperature of Crater Lake. Together with chem- found, and was thought to represent Tertiary marine sediments un- ical changes in the lake these rapid heating phases were taken as derlying the volcanic sediments. This feature was however, less well- evidence of interaction between fresh magma and lake water. How- constrained. ever, the precise relationship between the occurrence of the swarms More recently changes to the seismicity of Ruapehu following and the inferred upward movement of magma beneath Ruapehu re- the 1995–1996 eruptions were noted by Sherburn et al. (1999) and mained unclear. Relocation of the earthquakes using a 1-D velocity Bryan & Sherburn (2003). Prior to the eruptions volcanic tremor and model showed three clusters of seismicity with epicentres close to earthquakes were dominated by a 2 Hz spectral peak and some higher the normal Raurimu fault. However, composite focal mechanisms frequencies. After the eruptions, although these frequencies were for these clusters were not consistent with simple normal motion still present, the dominant frequencies were from 0.8 to 1.4 Hz. This on this fault. One of the clusters (cluster A) had an oblique-reverse change in frequency content was interpreted as being due to a change mechanism (predominantly reverse) with one nodal plane almost in the magmatic plumbing system of the volcano. Further evidence parallel to the Eastern Raurimu Fault. The mechanisms of the two for changes to the magmatic system beneath Ruapehu were provided remaining clusters were oblique-normal, again with strong normal by the results of Miller & Savage (2001) and Gerst & Savage (2002). components. They observed a change in the fast direction of polarized shear Sherburn (1993) provides a good summary of non-seismic geo- waves beneath Ruapehu before and after the eruptions of 1995– physical observations made in the area. Three main surveys have 1996. These observations were attributed to a change in the local been conducted. Firstly, sparse gravity measurements are available. stress regime caused by a pressurized tabular magma body being The most noticeable features are regional lows associated with the intruded beneath the volcano prior to the eruption, and a subsequent TVZ and the Wanganui basin. Sherburn calculated to first order that depressurization and re-adjustment of the stress field following the a negative anomaly associated with Ruapehu was due to low den- eruption. sity material within the cone itself, rather than beneath the base of Broader studies of the TgVC have also been conducted. Haines the volcano. Further interpretation, however, would require a higher (1979) analysed travel times of Pn and Sn phases recorded at a resolution data set. Secondly, data from two aeromagnetic surveys permanent station within the TgVC. He obtained extremely low found several positive anomalies over Tongariro, Ngauruhoe and 1 1 values of Vp 7.4 km s− and Vs 3.95 km s− in the uppermost Pihanga and a negative anomaly over Ruapehu. These were found mantle beneath= the TVZ, attributing=them to the presence of partial to be consistent with highly magnetized plugs beneath the three melt. northern cones and a reversely magnetized, or possibly a hot un- Olson (1985) used a traveltime graph obtained from recordings magnetized plug, beneath Ruapehu. Finally, a resistivity survey over of two shots fired in Ruapehu Crater Lake (Dibble et al. 1985) Tongariro delineated two conductive zones. The first zone was at- to construct a 1-D P-wave velocity model beneath Ruapehu. This tributed to a region of acid condensate, and was thought to feed both

C ! 2005 RAS, GJI, 163, 1180–1194 Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1183

Earthquake depth (km) Depth (km) -3 0 5 10 15 20 25 30 0 10 20 30

km R 0 10 20 39.0˚S

ERF

NPA Tg ? T Ng

39.2˚S NPB

RF

39.4˚S

TVZ W

39.6˚S 175.2˚E 175.4˚E 175.6˚E 175.8˚E

Ru 0

10 ? 20 Depth (km)

30

0 10 20 30 40 50 No. of earthquakes

Figure 2. Map and cross-sections of station and final earthquake locations. Stations are denoted as follows: Black triangles—Guralp¨ CMG-40T sensors; White triangles—CMG-3T sensors; White squares—vertical component NZSN stations (further stations located north and northeast of this map were also used); Inverted grey triangles—stations of the 1994 deployment; White diamonds—CNIPSE stations. Earthquake cluster labels are: R—Rotoaira cluster; T—Tongariro cluster; NPA—National Park Village cluster A; NPB—National Park Village cluster B; W—Waiouru cluster. Thick black lines denote faults (dashed lines are inferred faults) two of which are labelled: RF—Raurimu Fault; ERF—Eastern Raurimu Fault. Dotted lines mark location of topography profiles shown in cross-sections. 500-m topography contours are shown around the TgVC volcanoes from 1250–2750 m. Major roads are also shown. Dashed lines in the cross-sections mark the inferred brittle–ductile transition (see Section 4.5). Lower right panel: histogram of number of earthquakes versus depth.

Red Crater and Ketetahi Hot Springs. The second, deeper zone was Supplemental waveform and traveltime data were obtained from interpreted as a layer of hot brine. nearby permanent New Zealand National Seismograph Network (NZNSN) stations, from three stations of the adjacent CNIPSE ar- ray (Henrys et al. 2003) and from an array deployed in 1994. The 2 DATA A N D A N A LY S I S average station spacing over the target area (the central and northern TgVC) is around 5 km (Fig. 2). 2.1 Data The best 155 local earthquakes were selected after application of The data set used in this study is composed of local earthquake several criteria, namely distance to the closest observation (<1.5 arrival times recorded by an array of 20 Guralp¨ CMG-40TD and the hypocentral depth), number of high quality P observations (>8)× 3 CMG-3TD seismometers during the period 2001 January to June. and open azimuthal gap (<180◦). At the extreme north of the array

C ! 2005 RAS, GJI, 163, 1180–1194 1184 D. P. Rowlands, R. S. White and A. J. Haines several events with open azimuths of 180–200◦ were included in order to improve ray coverage. P-wave arrival times were picked on vertical components and 0 S-wave arrivals on horizontal components in an attempt to avoid mispicks. Consequently no S-wave picks are used from the verti- cal component NZNSN stations in the TgVC. Signal-to-noise ratios for the local earthquake data set were very high despite the preva- lent surface volcaniclastics and P-wave arrivals could routinely be picked to within one sample (0.01 s for temporary stations, typically 0.025 s for permanent stations). The presence of P-wave coda and 10 converted phases inevitably causes slightly greater uncertainties in picked S-wave arrival times but the mean uncertainty is nonetheless below 0.1 s. The final arrival time data set comprises approximately 3000 P and 1100 S picks.

2.2 Analysis Depth below sea level (km) The choice of initial reference model in local earthquake tomogra- 20 phy (LET) is a critical one (e.g. Kissling et al. 1994). In this study a minimum 1-D model is derived using the VELEST program, which simultaneously inverts arrival times for 1-D velocity structure, sta- 1-D initial Vp reference model tion corrections and hypocentres (Kissling et al. 1994). After per- forming several tests (convergence of different input 1-D velocity Minimum 1-D Vp model models, recovery of perturbed hypocentres and station corrections) the minimum 1-D model is considered to be well-resolved from 2 to 30 20 km depth (Fig. 3). Above and below this depth interval ray paths 2 3 4 5 6 7 8 are predominantly subparallel and hence provide little constraint on Velocity (km s−1) velocity. The minimum 1-D P-wave velocity model is broadly similar Figure 3. Minimum 1-D P-wave velocity model and 1-D initial reference to that obtained by Hurst & McGinty (1995) in the same region. P-wave velocity model used as starting model for the 3-D tomography. Velocity increases rapidly downwards to around 4 km depth and then more slowly down to the limit of resolution at about 20 km depth. Station corrections are almost all small (<0.3 s) with ex- Initially a coarse horizontal grid was used with nodes spaced every ceptions only at stations with few observations. All station cor- 15 km. Following resolution tests additional nodes at 10 km spacing rections are in good agreement with the surface geology. A lin- were added in the central 30 km of the model. The resolution ear interpolation of the minimum 1-D P-wave velocity model is of this medium parameterization± was then probed and extra nodes used as the 1-D initial reference model in the 3-D Vp tomography were placed every 4 km in the well-resolved central 20 km of the (Fig. 3). model. This fine parameterization represents the end± point of the Simul2000 (Thurber & Eberhart-Phillips 1999), a relatively recent graded inversion. Resolution tests show that no finer grid could be addition to the Simulps family of tomography programs (Thurber justified with the available data set. The final 3-D Vp model provided 1983, 1993), is used to perform the 3-D LET via an iterative, damped a 59 per cent reduction in data variance compared to that of the initial least-squares inversion. The inversion for Vp uses a graded inver- 1-D reference model. sion scheme (Eberhart-Phillips 1990, 1993). This approach com- An alternative strategy for the Vp inversion is to use a single- mences with a coarsely parameterized velocity model and pro- step inversion. In this approach the final parameterization from the ceeds to successively finer models, extra nodes being added where graded inversion is used, but with the 1-D initial reference model resolution tests indicate they are justified. The final output (ve- as the starting velocity model. Although a similar fit to the data locity model, relocated hypocentres, calculated traveltimes) from was obtained using this single-step method the final 3-D veloc- each such parameterization is used as input to the subsequent ity model was more complex than that obtained using the graded (finer) parameterization. For each parameterization it is necessary inversion approach. We therefore select the smoother final 3-D to choose a suitable damping value for the inversion. The approach model output by the graded inversion as our preferred final 3-D Vp of Eberhart-Phillips (1986) is followed, that is, a series of inver- model. sions using a wide range of damping values is performed and a Following the approach of Foulger et al. (1995), Husen et al. trade-off curve of data variance (data misfit) versus solution vari- (2000) and Haslinger (1998), we invert for Vp/Vs using the final ance (model complexity) is obtained. A suitable damping value 3-D Vp model output by the graded inversion and an initial Vp/Vs provides a good fit to the data without requiring too complex a ratio of 1.70 (as calculated from a Wadati diagram of S-P versus model. P traveltimes). This approach ensures that in well-resolved areas The 3-D grid was centred on 39.2◦S, 175.6◦E with x and y positive Vp/Vs is free to take the most appropriate value consistent with to the west and north, respectively. Depth nodes were set at 3, 0, 3, the data. Conversely, in areas of poor resolution Vp/Vs retains a 6, 9, 12, 16, 20, 25, 30 and 100 km below sea level. The 3,−30 and reasonable value, consistent with the average of the data set under − 100 km depth layers were held fixed in each inversion as they lay consideration. The final 3-D Vp/Vs model yielded a 48 per cent outside the resolvable depth range of this data set (these layers are reduction in data variance compared to the initial 1-D model with a nevertheless necessary to maintain the stability of the ray tracer). fixed Vp/Vs ratio of 1.70.

C ! 2005 RAS, GJI, 163, 1180–1194 Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1185

2.3 Resolution (a) P Several measures of resolution are commonly used in assessing the reliability of tomographic models. These fall into three main cate- gories: (1) ray-derived measures; (2) resolution matrix-derived mea- sures and (3) measures based upon the recovery of synthetic input 39.0˚S models. No single measure of resolution provides an unambiguous indication of whether a particular part of a model is well-resolved. Each type of measure has strengths and weaknesses and all must be examined to gain as full an appreciation as possible of the resolution of a tomographic model.

A good first impression of resolution can be obtained simply 39.2˚S from a plot of the ray paths used in the inversion (Fig. 4). It is clear from Fig. 4(a) that high numbers of crossing P rays are present in the central area of the TgVC. High resolution in the Vp model may therefore be expected here. Conversely, P rays around the edge of the TgVC are predominantly subparallel with little cross-firing. These areas will be subject to considerable smearing, and resolution near 39.4˚S the periphery of the Vp model is, therefore, likely to be poor. For S–P ray pairs the coverage is much poorer. This is largely because no S picks from vertical component stations were made. Reasonable numbers of crossing rays are present only in the northern TgVC and so high resolution in the Vp/Vs model may be expected 20 km to be restricted to this area. In order to examine the resolution more quantitatively ray-derived 39.6˚S measures such as hit count and derivative weighted sum (DWS, (b) S-P Toomey & Foulger 1989) are useful. The hit count simply gives the number of rays passing a given node of the model. DWS is more informative as it weights rays according to their distance from 39.0˚S a node. However, neither of these measures takes account of the directions in which the rays are travelling. The full resolution matrix describes the influence that each node of the velocity model has on every other node. In an ideal case this matrix would be an identity matrix, indicating that the velocity at every node is independently resolved (i.e. independent of the veloc-

ity values at all other nodes). However, the deficiencies of real data 39.2˚S sets, the introduction of damping to the inversion and the approxi- mation of linearity made in solving the inverse problem inevitably lead to departures from this ideal case. A further problem is that the resolution matrix is far too large to be easily visualized. The res- olution matrix diagonal elements (RDE) provide a simplified view of the full resolution matrix. However, they contain no informa- 39.4˚S tion on the amount of smearing prevalent in the model. Toomey & Foulger (1989) and Michelini & McEvilly (1991) introduced the spread function, a sum of the elements in each row of the resolution matrix weighted by distance from the diagonal element. The spread function for a given node gives a good indication of the extent to 20 km which that node is independently resolved. However, no informa-

tion is preserved about the directionality of a set of rays passing a 39.6˚S 175.2˚E 175.4˚E 175.6˚E 175.8˚E node, nor about the position of the centre of resolution which, in the presence of smearing, may not coincide with the position of the Figure 4. Ray path coverage for (a) Vp and (b) Vp/Vs inversions. White triangles denote stations, white circles denote epicentres (note that epicentres node. Spread values in the horizontal plane of our final V model p for S–P are shifted due to the inclusion of S information). Crosses denote are shown in Fig. 5 (those in the north–south and east–west planes the positions of nodes of the velocity model. may be viewed online). Resolution contours (Reyners et al. 1999), also termed smear- ing contours, address this issue by providing a visual display of the examination towards the surrounding nodes. The line connecting smearing around a node. A row of the resolution matrix describes these points then forms the resolution contour. Compact resolution the dependency of the diagonal element node on every other node contours which do not extend beyond adjacent nodes are desired. within the model. For, say, a 60 per cent resolution contour, the row However, it should be remembered that the choice of the resolution of the resolution matrix is examined and the points at which the contour percentage is somewhat arbitrary. Typical values used range resolution falls to 60 per cent of the RDE are noted. These points from 50–70 per cent (Reyners et al. 1999; Sherburn et al. 2003), are then translated into distances, measured from the node under with lower percentages indicating higher resolution. In this study

C ! 2005 RAS, GJI, 163, 1180–1194 1186 D. P. Rowlands, R. S. White and A. J. Haines

Figure 5. Spread values and resolution contours for the final 3-D Vp model. Compact resolution contours are marked in green, broader contours in red.

60 per cent resolution contours are used. Those calculated in the satisfactory for all LET data sets. The problem then is to decide horizontal plane of the final Vp model are shown in Fig. 5 (resolu- what values indicate high resolution for the data set under consider- tion contours in the north–south and east–west planes are viewable ation. One approach to solving this problem is to compare areas of online). It is critical to inspect resolution contours in three orthogo- good recovery from synthetic tests with the above resolution mea- nal planes in order to identify nodes with significant smearing. For sures. Commonly employed synthetic tests include chequerboard example, in the southeast of our model a fan of subparallel P rays tests, spike sensitivity tests (Spakman & Nolet 1988) and charac- emanates from a cluster of seismicity (Fig. 4a). In both the horizontal teristic model tests (Haslinger 1998). The different synthetic tests and east–west planes the resolution contours in this area indicate lit- employ similar methodologies. Firstly a synthetic anomaly model tle smearing. Only in the north–south plane is the large degree of is created. Next the forward problem is solved and a synthetic trav- smearing evident. Failure to examine the resolution contours in all eltime data set is calculated using the same source-receiver dis- three planes could have led to this area of the model being erro- tribution as in the real data set. Gaussian noise is usually added neously assessed as well-resolved. to these synthetic traveltimes at a level equivalent to the esti- A problem common to all of the above resolution measures is mated noise in the real data. Finally the synthetic traveltimes (plus that they provide only relative measures of resolution, i.e. no abso- noise) are inverted in an attempt to recover the synthetic anomaly lute threshold values exist above which resolution can be deemed model.

C ! 2005 RAS, GJI, 163, 1180–1194 Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1187

Figure 6. Chequerboard test of the tomographic models. Synthetic input anomalies are 5 per cent in magnitude and 2 2 2 nodes in spatial extent. ± × ×

In the Vp inversion of this study a close correspondence was found the recovery of a synthetic chequerboard input model. The pattern between (a) areas with DWS >50 (viewable online), spread function of anomalies is well recovered in the central and northern TgVC <2 and compact resolution contours, and (b) areas of good recovery down to the 9-km-depth layer. Recovery in the 12-km-depth layer is of synthetic models. Together these results provide a firm indication considered fair and deeper layers are not well recovered. Within the of which parts of the model volume are reliably resolved, and hence central 12 km of each layer the amplitude recovery is reasonably which features may be safely interpreted. For example, Fig. 6 shows good. Reco± vered anomalies of magnitude >4.5 per cent are common

C ! 2005 RAS, GJI, 163, 1180–1194 1188 D. P. Rowlands, R. S. White and A. J. Haines in the 3, 6 and 9 km depth layers. Larger amplitudes >5 per cent are recovered at a few nodes in the 0-km-depth layer, perhaps as a result of vertical smearing from lower layers. In the 12-km-depth layer recovered amplitudes are smaller, at best around 4.3 per cent, indicating a lack of resolution at this depth. Spike sensitivity and characteristic model tests show good recovery in similar areas and are viewable online. Consideration of all of the above measures of resolution suggests that we may regard the central 12 km of the Vp model (i.e. the central and northern TgVC) down± to the 9 km depth layer as well- resolved. Amplitudes in this part of the model are reasonably reliable except in one or two isolated parts of the near-surface where slight overestimation is possible. Resolution in the 12-km-depth layer is best described as fair with significant smearing and underestimated amplitudes likely. Assessment of the resolution of the Vp/Vs model was made in the same way as for Vp. Initially the spread function values, the DWS and a simple chequerboard test (Fig. 6) indicated reasonably good resolution. However, analysis of resolution contours and poor recovery of spike sensitivity and characteristic models revealed a high degree of smearing. The resolution of the Vp/Vs model is, therefore, considered to be fair at best. This is due to several factors including the lower number of S–P ray pairs, the higher uncertainty in S picks and the lower number of stations used. The Vp/Vs model is accordingly omitted from the interpretation.

3 R E S U LT S

3.1 3-D Vp structure

Two prominent low-velocity volumes are observed in the Vp model beneath Ruapehu and Ngauruhoe volcanoes (Figs 7 and 8). The Ngauruhoe low-velocity volume (Profile 4, Fig. 8) appears to be centred on the volcanic vent and diminishes in magnitude between the 6 km (!Vp 9 per cent) and 9-km-depth layers. In contrast the Ruapehu low-v= −elocity volume (Profile 2, Fig. 8) appears to be asymmetric, with a steep western edge beneath the presently active Crater Lake vent on the summit and a gently sloping eastern edge un- derlying the eastern flank of the volcano. This apparent asymmetry may however, be partly a consequence of lower resolution beneath the western flank of Ruapehu (note the lack of crossing rays in this area in Fig. 4). The low-velocity volume reaches a magnitude of 8 per cent in the 6-km-depth layer and decays to 3 per cent in −the 9-km-depth layer. − Smaller low-velocity volumes are observed beneath the active Tongariro volcano and the extinct vents at Tama Lakes (Figs 7 and 8). At Tongariro the low-velocity volume is strong ( 9 per cent) at 3 km depth but disappears by around 5 – 7 km depth− (Profile 5, Fig. 8). A less pronounced, 15 km broad low-velocity volume is observed beneath Tama Lakes, again starting at around 3 km depth and petering out at 5 km depth (Profile 3, Fig. 8). The 0-km-depth layer is characterized by areas of large amplitude positive or negative velocity anomalies. As discussed previously the amplitudes of these near-surface anomalies may be slightly exagger- ated. High velocities correlate with the volcanic cones, whilst low velocities are observed at shallow depth under the Rangipo desert in the east of the TgVC (Fig. 7).

3.2 Distribution of relocated earthquakes Figure 7. Percentage Vp anomalies of final 3-D model. The thick black/white line shows the area considered to be well-resolved (dashed line The general pattern of seismicity is apparent in Fig. 2 with shal- indicates fair resolution). Grey circles show nodes held fixed during the low seismicity (<10 km depth) in the north and deeper seismicity finely-parameterized (i.e. final) step of the graded inversion.

C ! 2005 RAS, GJI, 163, 1180–1194 Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1189

hours on 2001 February 10, at the northwest edge of Lake Rotoaira (R in Fig. 2).

4 D I S C U S S I O N Before interpreting tomographic models it is prudent to remind one- self of some of the limitations inherent in the tomographic ‘imaging’ process. First and foremost it is essential to remember that the com- mon practice of referring to the absolute velocities calculated in a tomographic model is somewhat misleading. The velocities ob- tained by LET are not absolute in a literal sense because of several factors, including: the linearization approximation made at the out- set of the tomography; the restriction that final velocities have to remain linearly close to the 1-D initial reference model; the use of a velocity model with only a finite number of parameters; the in- evitable averaging of velocities over considerable volumes; and the introduction of damping into the inversion. Clearly the final veloci- ties at any particular point in a tomographic model are not absolute in a strict sense. They should rather be thought of as being the absolute velocities viewed through the series of ‘filters’ listed above. Secondly, and following on from the first point, the seismic ve- locities from a tomographic model provide only a limited constraint on the nature of the crust. Independent constraints from other mea- surements are extremely important and a good interpretation should take account of all available data. Thirdly, the limits of resolution must be borne in mind at all times. In this study resolution varies both spatially and in depth beneath the TgVC (see Section 2.3). The following discussion focuses both on features of the LET models and on the distribution of seismicity in and around the TgVC.

4.1 Ruapehu The large low-velocity feature apparent in the Vp model beneath Ru- apehu may be explained in several ways. Firstly, the low velocities may indicate the presence of remnant partial melt from the 1995, 1996 and 1998 eruptions. However, this explanation has two weak- nesses. Recent eruptions at Ruapehu were relatively small in vol- ume (approximately 0.1 km3) and are thought to have been sup- plied through a narrow (tens of metres wide) conduit—any melt remaining beneath Ruapehu is unlikely to be distributed over a suf- ficiently large volume to account for the large lateral extent of the low-velocity volume observed. Furthermore, eruptions over the last 2 ka have all occurred through Crater Lake (Gamble et al. 2003), situated at the summit of the volcano, yet the low-velocity volume is centred beneath the eastern flank of Ruapehu. It is, therefore, considered implausible that a leftover body of hot and/or partially molten rock could fully account for the large low-velocity volume Figure 8. Profiles of absolute Vp of final 3-D model. Thick black/white lines observed. outline areas of good resolution (dashed line shows limit of fair resolution). A second possibility is that velocities have been lowered by hy- Abbreviations as in Fig. 1. drothermal alteration. From geological observations this process is known to have occurred at Ruapehu in the past (Gamble et al. in the west (around 15 km depth) and southeast (around 20 km 2003) and is known to be occurring at present beneath Crater Lake. depth). Hydrothermal circulation down to depths of 5–8 km is common The following prominent features are also observed in the final further north in the TVZ (Bibby et al. 1995), albeit within rhyolitic 3-D hypocentre data set (Fig. 2): volcanic systems. However, it is not clear how hydrothermal alter- (a) two clusters of particularly high seismicity located near Na- ation beneath Ruapehu could account for the large lateral extent of tional Park Village, northwest of Ruapehu (clusters NPA and NPB the low-velocity volume observed, especially in view of the recent in Fig. 2); focus of activity beneath the summit. (b) three lineations of seismicity, two at Waiouru and one at Ton- A further possible source of the low-velocity volume is down- gariro (W and T respectively, in Fig. 2) and warping of the crust due to the topographic load imposed by the (c) a swarm of earthquakes which occurred over a period of a few Ruapehu edifice. Circumstantial evidence for this downwarping is

C ! 2005 RAS, GJI, 163, 1180–1194 1190 D. P. Rowlands, R. S. White and A. J. Haines provided by the orientations of faults in the area (Fig. 2). Faults to the west, south and east of Ruapehu are oriented tangentially to the volcanic massif. A slight difficulty for this explanation is that the thickest part of the low-velocity volume does not quite coincide with the summit of Ruapehu (i.e. the greatest load). However, as previ- ously mentioned this apparent asymmetry may be a consequence of low resolution beneath the west flank of Ruapehu. A final explanation is that thick sequences of volcanic products deposited around Ruapehu cause the low-velocity volume observed. Donoghue et al. (1999) summarize the geology of the ring plain around Ruapehu as consisting of thick sequences of andesitic tephras overlying and interbedded with lahar deposits, lava flows, debris avalanche deposits, fluvial and aeolian sands, and exotic rhyolitic tephras erupted from the central TVZ. Successive episodes of de- position, burial and compaction of material from all these sources may have built up a thick deposit of low-velocity material around Ruapehu. However, it is not clear whether this mechanism could pro- duce the full 6 km thickness of the low-velocity volume observed. Clearly no single explanation is able to explain fully the low- velocity volume beneath Ruapehu, but it is possible that each has played a part in its formation. The low-velocities observed directly beneath the Crater Lake area may be indicative of hot or partially molten rock remaining from the last eruptive episode. This would certainly be consistent with the seismic ‘soft centre’ and petrological open system models for Ruapehu, which propose a hot conduit un- derlying Ruapehu Crater Lake, incapable of significantly impeding the ascent of fresh magma. A degree of hydrothermal circulation in and around the conduit would be expected in such a system, serving to broaden the low-velocity volume. The most reasonable explanation for the low velocities beneath the eastern flank of Ruapehu is considered to be successive, prefer- ential deposition of volcaniclastics to the east due to the prevail- ing westerly wind direction. This probably occurs in conjunction with continuing downwarping of the crust due to the increasing vol- canic load at the surface. Some evidence for this downwarping is 1 seen in the gentle westward dip of the 5.5 km s− contour east of Ruapehu (Profile 2, Fig. 8). A possible weakness of this explana- tion is the lack of evidence in the Vp model for downwarping to the west of Ruapehu, although this may simply be an artefact of the aforementioned low resolution in this area (Figs 7 and 8).

Profile 5 20 km 39.0˚S Profile 4 Profile 3

Profile 39.2˚S 2

39.4˚S

Profile 1

175.2˚E 175.4˚E 175.6˚E 175.8˚E 176.0˚E Figure 10. Final 3-D Vp/Vs model. Symbols as described in Fig. 7. Note that the resolution in each depth layer may only be considered fair at best.

Figure 9. Map showing the locations of the Vp profiles shown in Fig. 8. Any interpretation of Vp/Vs anomalies is therefore unwarranted in this study.

C ! 2005 RAS, GJI, 163, 1180–1194 Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1191

4.2 Ngauruhoe intrusions, and/or increased permeability due to fracturing’. These results are compatible with the low-velocity region observed in this Similar arguments may be applied to the low-velocity volume ob- study, although the low seismic velocities extend somewhat deeper served beneath Ngauruhoe. However, the comparatively young age than the dense/conductive anomaly. of Ngauruhoe (2500 yr old) and the slow rate at which the crust flexes (a few centimetres per year) imposes a limit on the amount of downwarping possible. 4.5 Earthquake locations The most likely explanation for the low-velocity volume beneath The earthquake locations in this study are the most accurate yet Ngauruhoe is a complex of hot or partially molten material left over obtained within the TgVC. Several features of the local seismicity from previous eruptions. Such an explanation would agree with warrant further discussion. the interpretation of Hobden et al. (1999) of the extreme chemical Firstly, the two clusters of earthquakes located near National Park variability of different lavas from Ngauruhoe indicating a network Village (NPA and NPB on Fig. 2) may not originate on the large, of interconnected conduits and small magma chambers. normal Raurimu fault. From a comparison with the composite fo- The possible role of hydrothermal circulation is not clear at Ngau- cal mechanism of cluster A of Hurst & McGinty (1999), it seems ruhoe. Although the seismically active geothermal system beneath likely that cluster NPA originates on the Eastern Raurimu Fault Tongariro is only some 2 to 3 km distant, Ngauruhoe exhibits very (ERF). little seismicity. This aseismicity may indicate that no active geother- Cluster NPB is probably equivalent to cluster B of Hurst & mal system is present beneath Ngauruhoe. However, some active McGinty (1999), although cluster NPB is located a few kilome- geothermal systems are known to be largely devoid of seismicity, tres northeast of cluster B and is more tightly focussed at 15 km e.g. Ohaaki (Sherburn et al. 1990, 1993; Bryan et al. 1999). We, depth. Hurst & McGinty present an oblique-normal composite fo- therefore, cannot eliminate the possiblity that an active geother- cal mechanism for their cluster B that may be interpreted in two mal system underlies Ngauruhoe and that our 6-month recording ways. Firstly, the fault plane may be the nodal plane oriented ap- window may simply have coincided with a low level of seismic proximately parallel to the Raurimu Fault. Alternatively, the fault activity. plane may be the nodal plane oriented approximately east–west. Regardless of the present situation hydrothermal circulation has On the basis of the distribution of seismicity neither possibility certainly occurred in the TgVC in the past (Nairn et al. 1998). It may be eliminated. Although the latter possibility does not corre- is possible that a volume of rock that was hydrothermally altered spond to any previously mapped fault, Hurst & McGinty did state during previous eruptive episodes may underlie Ngauruhoe and con- that of the two possibilities the east–west fault plane was consis- tribute to the low-velocity volume observed. tent with the better located events. In the absence of any other ev- idence for this fault, however, they concluded that the events were 4.3 Tongariro probably associated with movement on the Raurimu Fault. The pos- sibility nevertheless cannot be discounted that the east–west fault Several active geothermal vents are found on Tongariro. The low- exists but has not been mapped, perhaps because it lacks a surface velocity volume observed at shallow depths is, therefore, most easily expression. attributed to the presence of a hot source body for the geothermal Should earthquake clusters NPA and NPB occur on small east- field and to the effects of hydrothermal alteration. A broad area ward branches of the Raurimu Fault, rather than on the main fault of low Vp/Vs ratios is also observed at shallow depths beneath itself, the potential for large magnitude earthquakes here in the fu- Tongariro (Fig. 10). As discussed previously, this feature cannot ture may be lower than has hitherto been expected. However, further be regarded as being particularly well-resolved. Nevertheless, low seismic experiments, particularly with improved station coverage to Vp/Vs is at least consistent with the presence of water within a the west of the Raurimu Fault, will be needed in order to localize geothermal system. these clusters on particular faults. The broad lateral extent of the low-velocity volume at Tongariro Secondly, three lineations of seismicity are observed. The first two may be explained in a similar way to that at Ruapehu. Large-scale occur near Waiouru (marked W on Fig. 2) with hypocentres concen- collapses of the Tongariro complex, known to have occurred in the trated between approximately 12 and 24 km depth and strikes sub- past (e.g. Lecointre et al. 2002), fall-out from eruptions and a degree parallel to nearby mapped faults. The seismicity in this area has been of downwarping of the crust offer ready explanations for the low- closely examined by Hayes et al. (2004) using double-difference re- velocity volume observed. location, focal mechanisms and a technique to determine the stress tensor orientation. We note that the locations of Hayes et al. are in reasonably close agreement with those presented in this study. 4.4 Tama Lakes Hayes et al. concluded that the seismicity near Waiouru is of tec- Volcanic activity from the Tama Lakes is known to have occurred tonic origin and is due to fluid movement within a critically loaded as recently as the 10 ka Pahoka–Mangamate sequence (Nairn et al. fault zone. 1998; Nakagawa et al. 1998). Nairn et al. (1998) cite unpublished The third lineation extends downward and northward, from very results of gravity and audiomagnetotelluric profiles over the Tama shallow depth beneath Ngauruhoe to around 9 km depth beneath the Lakes as indicating that a ‘dense “basement” high underlies the northern flank of Tongariro (marked T on Fig. 2). Again this lin- volcanics beneath Tama Lakes area, associated with a conduc- eation has a subparallel trend to nearby mapped faults. It is possible tive anomaly at 1–1.5 km depths. It is underlain by material of that these earthquakes also occur on an unmapped fault, perhaps lower conductivity—interpr∼ eted as Mesozoic greywacke. Because again lacking a surface break. the Tama Lakes area forms part of a positive magnetic anomaly, Further analysis of these features of local seismicity is neces- the conductive material cannot be due to present day high temper- sary in order to understand more fully the relationship between atures or magma at depth. The conductive anomaly is interpreted earthquakes, faulting, regional extension and volcanic activity in as formed by alteration/mineralization associated with magmatic and around the TgVC.

C ! 2005 RAS, GJI, 163, 1180–1194 1192 D. P. Rowlands, R. S. White and A. J. Haines

A further notable feature of the distribution of seismicity is the −0.6 km N lack of events beneath Ruapehu and Ngauruhoe. This aseismicity Prevailing wind was also observed prior to and during the 1995–1996 eruptions KHS Tg RC of Ruapehu, when only a few deep (>1 km beneath Crater Lake) volcano-tectonic earthquakes were recorded (Sherburn et al. 1999). This lack of precursory seismicity has been interpreted by several Ng Ru G authors (e.g. Hurst & McGinty 1999; Sherburn et al. 1999) as being indicative of a ‘soft centre’ beneath Ruapehu, that is, a volume 10 km of weak rock through which ascending magma can pass relatively easily and relatively aseismically. This model is in good agreement ? with ‘open system’ petrological models of Ruapehu and models derived from tephrostratigraphy of the TgVC as a whole, in which rapid ascent of magma from depth is a key feature (Nairn et al. 1998; Lahar deposits, Hot, avalanche debris, Nakagawa et al. 1998; Gamble et al. 2003). "soft", ? andesitic tephra, open lava flows, In contrast, during the volcanically active period of 1976–1983 conduit Complex of small, ignimbrite, etc. the New Zealand Volcanological Record and Latter (1981a) re- interconnected magma chambers ported several volcano-tectonic earthquakes at depths ranging from ? the summit of Ruapehu down to approximately 20 km depth. If these locations are genuine then the present day lack of deeper seis- Figure 11. Cartoon of our interpretation—see text for explanation. Abbre- micity may indicate a large-scale change in the thermo-mechanical viations are as follows: Ru—Ruapehu; Ng—Ngauruhoe; Tg—Tongariro; regime of the crust beneath the TgVC. However, earthquake loca- KHS—Ketetahi Hot Springs; RC—Red Crater; G—Geothermal source tions within the TgVC during this period were not well constrained body. (e.g. Latter 1978, p. 32) due to a lack of instrumentation and a con- sequent lack of information about the velocity structure of the area. (d) thick, low-velocity deposits from various volcanic sources Detailed comparisons between the seismicity presented in this paper form the ring plains around Ruapehu and Tongariro. and historical seismicity are, therefore, unwarranted and evidence Local seismicity is largely restricted to two clusters west of for any large-scale changes to the TgVC system must be treated Ruapehu, possibly originating on small eastward branches of the with caution. large Raurimu Fault, and lineations near Waiouru and under Ton- Seismicity recorded during this study reinforces the ‘soft centre’ gariro. The crust immediately beneath Ruapehu and Ngauruhoe is al- and ‘open system’ models. Earthquakes are located at shallower most aseismic, possibly indicating a shallowing of the brittle–ductile depths beneath both Ruapehu and Ngauruhoe (see cross-sections transition. of Fig. 2), perhaps indicating a shallowing of the brittle–ductile transition beneath these volcanoes. This is consistent with our in- A C K N O W L E D G M E N T S terpretation of low seismic velocities being at least partly due to the presence of hot or partially molten material beneath Ruapehu We thank Steve Sherburn and Regina Lippitsch for countless valu- and Ngauruhoe. It should be borne in mind however, that only a able discussions. DR was supported by the Natural Environment Re- 6-month snapshot of seismicity is available from the START array. search Council (NERC). Fieldwork was funded by the Royal Society Improvements to the permanent network in the area will allow more and NERC. Instruments were loaned by SEIS-UK. Supplementary accurate routine hypocentral depth calculation and will shed light data were generously made available from: (a) the New Zealand on whether this apparent shallowing of the brittle–ductile transition National Seismograph Network, operated by IGNS, New Zealand; is a long-term feature. (b) the CNIPSE array, by Martin Reyners (IGNS) and Graham Stu- art (Univ. Leeds, UK) and (c) the 1994 and 1995 seismometer de- ployments in the TgVC, by Tony Hurst (IGNS). Logistical support 4.6 Conclusions was provided by staff at the Wairakei offices of IGNS. We thank all those who provided assistance in the field: Alex Brisbourne We present 3-D tomographic models and accurately located local and Paul Denton (SEIS-UK); Tony Harrison and Jenny Maresh seismicity for the TgVC. The V structure of the upper-mid crust p (Univ. Cambridge); Steve Sherburn (IGNS); Michael Bourne (Univ. is well-resolved in the central and northern TgVC, revealing large Leeds) and Wanda Stratford (Victoria Univ., Wellington). We also low-velocity volumes beneath Ruapehu and Ngauruhoe and smaller thank all those who kindly gave permission for us to deploy ones beneath Tongariro and Tama Lakes. These features are shown seismometers on their land, particularly Harry Keys at the De- to be consistent with previous petrological and seismological models partment of Conservation in Turangi, local Maori iwi, Rangipo for the TgVC volcanoes. Prison and several farm owners. This paper benefitted from help- Our interpretation of the tomographic models is summarized in ful reviews by Tony Hurst (IGNS) and an anonymous reviewer. cartoon form in Fig. 11. The main features of our interpretation are: Department of Earth Sciences, Cambridge, contribution number (a) the conduit beneath Ruapehu remains hot from previous erup- 7946. tions, providing little resistance to ascending magma, and is largely aseismic; (b) a complex of small, interconnected magma chambers lies R E F E R E N C E S beneath Ngauruhoe; Bibby, H.M., Caldwell, T.G. & Davey, F.J., 1995. Geophysical evidence on (c) a shallow, hot body underlies Tongariro feeding the active the structure of the Taupo Volcanic Zone and its hydrothermal circulation, geothermal system there and J. Volc. Geotherm. Res., 68, 29–58.

C ! 2005 RAS, GJI, 163, 1180–1194 Seismic tomography of the Tongariro Volcanic Centre, New Zealand 1193

Bryan, C.J., Sherburn, S., Bibby, H.M., Bannister, S.C. & Hurst, A.W., 1999. of shallow subduction in north Chile: a combined onshore and offshore Shallow seismicity of the central Taupo Volcanic Zone, New Zealand: study, J. geophys. Res., 105, 28 183–28 198. its distribution and nature, New Zealand, J. Geol. Geophys., 42, 533– Kissling, E., Ellsworth, W.L., Eberhart-Phillips, D. & Kradolfer, U., 1994. 542. Initial reference models in local earthquake tomography, J. geophys. Res., Bryan, C.J. & Sherburn, S., 2003. Eruption-induced modifications to vol- 99, 19 635–19 646. canic seismicity at Ruapehu, New Zealand, and its implications for erup- Latter, J.H., 1978. Volcano-seismic activity at Ruapehu volcano during 1977, tion forecasting, Bull. Volcanol., 65, 30–42. New Zealand Volcanological Record, No. 7, 26–32. Dibble, R.R., Olson, J.A. & Sissons, B.A., 1985. A seismic travel-time graph Latter, J., 1981a. Volcanic earthquakes and their relationship to eruptions to 25 km from explosions in Ruapehu Crater Lake, New Zealand Vol- at Ruapehu and Ngauruhoe volcanoes, New Zealand, J. Volc. Geotherm. canological Record: Volcano and Geothermal Observations 1983, 13, Res., 9, 293–309. 62–65. Latter, J., 1981b. Location of zones of anomalously high S-wave attenuation Donoghue, S.L., Palmer, A.S., McClelland, E., Hobson, K., Stewart, R.B., in the upper crust near Ruapehu and Ngauruhoe volcanoes, New Zealand, Neall, V.E., Lecointre, J. & Price, R., 1999. The Taurewa Eruptive Episode: J. Volc. Geotherm. Res., 10, 125–156. evidence for climactic eruptions at Ruapehu volcano, New Zealand, Bull. Lecointre, J.A., Neall, V.E., Wallace, R.C. & Prebble, W.M., 2002. The 55- to Volcanol., 61, 223–240. 60-ka Te Whaiau Formation: a catastrophic, avalanche-induced, cohesive Eberhart-Phillips, D., 1986. 3-D velocity structure in northern California debris-flow deposit from proto-Tongariro volcano, New Zealand, Bull. Coast Range from inversion of local earthquake arrival times, Bull. seism. Volcanol., 63, 509–525. Soc. Am., 76, 1025–1052. Michelini, A. & McEvilly, T.V., 1991. Seismological studies at Parkfield. Eberhart-Phillips, D., 1990. Three-dimensional P- and S-velocity structure I. Simultaneous inversion for velocity structure and hypocenters us- in the Coalinga region, California, J. geophys. Res., 95, 15 343–15 363. ing cubic B-splines parameterization, Bull. seism. Soc. Am., 81, 524– Eberhart-Phillips, D., 1993. Local earthquake tomography: earthquake 552. source regions, in Seismic Tomography: Theory and practice, pp. 613– Miller, V. & Savage, M., 2001. Changes in seismic anisotropy after vol- 643, eds. Iyer, H.M. & Hirahara, K., Chapman and Hall, New York. canic eruptions: evidence from Mount Ruapehu, Science, 293, 2231– Foulger, G.R., Miller, A.D., Julian, B.R. & Evans, J.R., 1995. 3-D Vp and 2233. Vp/Vs structure of the Hengill Triple Junction and geothermal area, Ice- Nairn, I.A., Kobayashi, T. & Nakagawa, M., 1998. The 10 ka multiple ∼ land, and the repeatability of tomographic inversion, Geophys. Res. Lett., vent pyroclastic eruption sequence at Tongariro Volcanic Centre, Taupo 22, 1309–1312. Volcanic Zone, New Zealand: part 1. Eruptive processes during regional Gamble, J.A., Wood, C.P., Price, R.C., Smith, I.E. M., Stewart, R.B. & extension, J. Volc. Geotherm. Res., 86, 19–44. Waight, T., 1999. A fifty year perspective of magmatic evolution on Ru- Nakagawa, M., Nairn, I.A. & Kobayashi, T., 1998. The 10 ka multiple ∼ apehu volcano, New Zealand: verification of open system behaviour in an vent pyroclastic eruption sequence at Tongariro Volcanic Centre, Taupo arc volcano, Earth planet. Sci. Lett., 170, 301–314. Volcanic Zone, New Zealand: part 2. Petrological insights into magma Gamble, J.A., Price, R.C. & Smith, I.E. M., 2003. 40Ar/39Ar geochronology storage and transport during regional extension, J. Volc. Geotherm. Res., of magmatic activity, magma flux and hazards at Ruapehu volcano, Taupo 86, 45–65. Volcanic Zone, New Zealand, J. Volc. Geotherm. Res., 120, 271–287. Nakagawa, M., Wada, K., Thordarson, T., Wood, C.P. & Gamble, J.A., 1999. Gerst, A. & Savage, M., 2002. Testing Proposed Changes in Seismic Petrologic investigations of the 1995 and 1996 eruptions of Ruapehu vol- Anisotropy at Mt. Ruapehu Volcano in New Zealand, AGU Western cano, New Zealand: formation of discrete and small magma pockets and Pacific Geophysics Meeting, Wellington, New Zealand, 9–12 July, 2002, their intermittent discharge, Bull. Volcanol., 61, 15–31. SE31C-03. Olson, J.A., 1985. Earthquakes in the Tongariro Volcanic Centre area, 1983 Hackett, W.R., 1985. Geology and petrology of Ruapehu volcano and related February 19–March 6, New Zealand Volcanological Record: Volcano and vents, PhD thesis, Victoria University, Wellington, New Zealand. Geothermal Observations 1983, 13, 73–74. Haines, A.J., 1979. Seismic wave velocities in the uppermost mantle beneath Reyners, M., Eberhart-Phillips, D. & Stuart, G., 1999. A three-dimensional New Zealand, New Zealand J. Geol. Geophys., 22, 245–257. image of shallow subduction: crustal structure of the Raukumara Penin- Haslinger, F., 1998. Velocity structure and seismotectonics of northwestern sula, New Zealand, Geophys. J. Int., 137, 873–890. Greece between the Gulf of Arta and Zakynthos, PhD thesis, ETH, Zurich,¨ Sherburn, S., Allis, R. & Clotworthy, A., 1990. Microseismic activity at Switzerland. Wairakei and Ohaaki geothermal fields, Proc. 12th New Zealand Geother- Hayes, G., Reyners, M. & Stuart, G., 2004. The Waiouru, New Zealand, mal Workshop, 1990, 51–55. earthquake swarm: persistent mid crustal activity near an active volcano, Sherburn, S., Clotworthy, A. & Hunt, T., 1993. No induced seismic activity at Geophys. Res. Lett., 31, L19613, doi:10.1029/2004GL020709. Ohaaki geothermal field, Proc. 15th New Zealand Geothermal Workshop, Henrys, S., Reyners, M. & Bibby, H., 2003. Exploring the plate boundary 1993, 201–205. structure of the North Island, New Zealand, EOS, Trans. Am. geophys. Sherburn, S., 1993. A review of geophysical data in the Tongariro Volcanic Un., 84(31), 289, 294–295. Centre, Institute of Geological &Nuclear Sciences Science Report, 93/04, Hobden, B.J., Houghton, B.F., Lanphere, M.A. & Nairn, I.A., 1996. Growth p. 31. of the Tongariro volcanic complex: new evidence from K-Ar age deter- Sherburn, S., Bryan, C.J., Hurst, A.W., Latter, J.H. & Scott, B.J., 1999. minations, New Zealand J. Geol. Geophys., 39, 151–154. Seismicity of Ruapehu volcano, New Zealand, 1971-1996: a review, J. Hobden, B.J., Houghton, B.F., Davidson, J.P. & Weaver, S.D., 1999. Small Volc. Geotherm. Res., 88, 255–278. and short-lived magma batches at composite volcanoes: time windows at Sherburn, S., Bannister, S. & Bibby, H., 2003. Seismic velocity structure Tongariro volcano, New Zealand, J. geol. Soc. Lon., 156, 865–868. of the central Taupo Volcanic Zone, New Zealand, from local earthquake Hobden, B.J., Houghton, B.F. & Nairn, I.A., 2002. Growth of a young, fre- tomography, J. Volc. Geotherm. Res., 122, 69–88. quently active composite cone: Ngauruhoe volcano, New Zealand, Bull. Sissons, B.A. & Dibble, R.R., 1981. A seismic refraction experiment south- Volcanol., 64, 392–409. east of Ruapehu volcano, New Zealand J. Geol. Geophys., 24, 31– Hurst, A.W. & McGinty, P.J., 1995. Seismicity of Ruapehu volcano recorded 38. during the 1994 seismometer deployment, Institute of Geological & Nu- Spakman, W. & Nolet, G., 1988. Imaging algorithms, accuracy and resolution clear Sciences Science Report, 95/45, p. 13. in delay time tomography, in Mathematical Geophysics, pp. 155–187, eds Hurst, A.W., 1998. Shallow seismicity beneath Ruapehu Crater Lake: results Vlaar, N.J., Nolet, G., Wortel, M.J. R. & Cloetingh, S.A. P.L., D. Reidel of a 1994 seismometer deployment, Bull. Volcanol., 60, 1–9. Publishing Company, Dordrecht, The Netherlands. Hurst, A.W. & McGinty, P.J., 1999. Earthquake swarms to the west of Mt. Thurber, C.H., 1983. Earthquake locations and three-dimensional crustal Ruapehu preceding its 1995 eruption, J. Volc. Geotherm. Res., 90, 19–28. structure in the Coyote Lake area, central California, J. geophys. Res., 88, Husen, S., Kissling, E. & Flueh, E.R., 2000. Local earthquake tomography 8226–8236.

C ! 2005 RAS, GJI, 163, 1180–1194 1194 D. P. Rowlands, R. S. White and A. J. Haines

Thurber, C.H., 1993. Local earthquake tomography: velocities and marked in red. Topography is shown along the top of each cross- Vp/Vs—theory, in Seismic Tomography: Theory and practice, pp. 563– section. 583, eds Iyer, H.M. & Hirahara, K., Chapman and Hall, New York. Figure S3. Spread values and east–west resolution contours for Thurber, C.H. & Eberhart-Phillips, D., 1999. Local earthquake tomography the fine Vp inversion. Compact contours are marked in green, con- with flexible gridding, Comp. Geosci., 25, 809–818. tours which spread beyond adjacent nodes are marked in red. To- Toomey, D.R. & Foulger, G.R., 1989. Tomographic Inversion of Local pography is shown along the top of each cross-section. Earthquake Data from the Hengill-Grensdalur Central Volcano Complex, Iceland, J. geophys. Res., 94, 17 497–17 510. Figure S4. Input spike model for the fine Vp inversion. Anoma- lies are 5 per cent in magnitude relative to the 1-D initial refer- Wilson, C.J.N., Houghton, B.F., McWilliams, M.O., Lanphere, M.A., ± Weaver, S.D. & Briggs, R.M., 1995. Volcanic and structural evolution ence model and extend over two nodes in each direction. Adjacent of Taupo Volcanic Zone, New Zealand: a review, J. Volc. Geotherm. Res., anomalies are separated in each direction by 2 unperturbed nodes. 68, 1–28. Figure S5. Recovered spike model for the fine Vp inversion. The input anomalies are recovered well in the 3 and 6 km depth lay- ers. Vertical and horizontal smearing is generally of low amplitude compared to the input and recovered anomalies. S U P P L E M E N TA RY M AT E R I A L Figure S6. Input characteristic model for the fine Vp inversion. This model is constructed by changing the strikes and polarities of The following supplementary material is available for this article the anomalies observed in the fine V model. online: p Figure S7. Recovery of the characteristic model for the V inver- Figure S1. DWS for the fine V inversion. The DWS 50 contour p p sion. Within the central 12 km of the model recovery is reasonably is marked with a white/orange contour. = good in the 3-, 6- and 9-km-depth± layers. Figure S2. Spread values and north–south resolution con- tours for the fine Vp inversion. Compact contours are marked This material is available as part of the online article from in green, contours which spread beyond adjacent nodes are http://www.blackwell-synergy.com

C ! 2005 RAS, GJI, 163, 1180–1194