Receiver functions calculated from data recorded by the 1995 PASSCAL Toba Seismic Experiment in northern Sumatra, Indonesia and the permanent station PSI reveal the presence of a low seismic velocity zone at between 8 and 14 km depth, which is interpreted to be the top of a magma reservoir. The Moho under the Toba caldera lies near 30 km with a localized thickened region that reaches 39 km beneath the eastern edge of the caldera with an abrupt shallowing to the northeast. This thickened crust may be where hot gabbroic material underplates the base of the crust to serve as the heat source for shallower volcanism. Alternatively, the thickened crust and Moho offset may be due to localized crustal thickening when this region was situated in a sharp restraining bend of a now extinct segment of the Sumatran fault.
 While it is widely accepted that magmatism at convergent plate boundaries is principally from melting due to lowering of the solidus by the introduction of water into the mantle by the subducting plate [Perfit and Davidson, 2000], questions remain concerning the migration and levels of ponding of the melt, especially in the case of a continental upper plate. The Toba Caldera in Sumatra and the Taupo Volcanic Zone of New Zealand are two regions associated with ocean-continent convergent margins with some of the youngest and most voluminous caldera forming magmatism. Recent seismic studies from both volcanic regions identified possible upper crustal magma reservoirs [Masturyono et al., 2001; Bannister et al., 2004]. Further seismological studies have begun to decipher the larger scale crustal and upper mantle structures in the Taupo Volcanic Zone [Harrison and White, 2004; Stratford and Stern, 2006] but the relationship of the upper crustal magma bodies to Moho structures and the importance of crustal underplating remain unresolved in the Toba region.
 The presence of a magma reservoir beneath Toba caldera has been revealed by previous geophysical analysis that relied on an inversion of P-wave arrival times and gravity anomalies [Masturyono et al., 2001]. They identified two separate magma reservoirs in the northern and southern parts of the caldera in the 10–20 km depth range. The goal of this study is to refine the locations and depths of the magma bodies under the Toba caldera and examine their relationship with features in the lower crust and upper mantle using the receiver function method applied to archived broadband data from the 1995 PASSCAL experiment [Masturyono, 2000; Masturyono et al., 2001].
 Toba caldera in northern Sumatra is the largest known caldera of Quaternary age and is a part of the Sunda arc where the Indian-Australian plate subducts beneath the Eurasian plate along the Java trench. This region is seismically active and two recent great earthquakes (Mw 9.3 and Mw 8.6) have ruptured the plate boundary in December 2004 and March 2005 [e.g., Lay et al., 2005; Subarya et al., 2006] (Figure 1a).
 Toba caldera is a complex of calderas formed during several separate eruptions over the last 1.2 million years and the youngest and largest caldera was created by an eruption that occurred 74,000 years ago with an estimated pyroclastic volume of 2800 km3 [Knight et al., 1986; Chesner and Rose, 1991]. Samosir Island and the Uluan Block are resurgent domes that were uplifted following the last eruption and a NW-SE trending graben separates them (Figure 1b). The ignimbrites from the last three eruptions, the so called Oldest Toba Tuff (OTT ∼ 0.84 Ma), Middle Toba Tuff (MTT ∼ 0.5 Ma), and Youngest Toba Tuff (YTT ∼ 74 Ka), are all rhyolitic, high-K calk-alkaline rocks while the lavas from active Pusubikit and Tandukbenua volcanoes have dacite to basaltic andesite compositions [Chesner and Rose, 1991; Wark et al., 2000].
 The Sumatran Fault, a 1900 km-long right-lateral strike-slip fault system traverses the length of Sumatra between the Andaman and the Java Seas [Sieh and Natawidjaja, 2000]. This fault system accommodates a trench-parallel component of oblique slip along the plate boundary [Fitch,1972]. The volcanoes along the Sunda arc are located close to this fault system except near the Toba caldera where volcanism is offset to the east [Page et al., 1979]. The relationship between the Sumatran Fault, the volcanic chain, and the subducting plate has been analyzed [e.g., Bellier and Sébrier, 1994; Sieh and Natawidjaja, 2000; Wark and McCaffrey, 2001] but it is not yet fully understood. Page et al.  inferred that the change in the trend of the volcanic chain is due to the shallower dip angle of the slab segment east of the Investigator Fracture Zone (IFZ), a vertical fracture within the Indian-Australian plate that projects beneath the Toba caldera (Figure 1a).
2. Data and Methods
 We obtained 3-component broadband seismograms recorded by the Toba Seismic Experiment, a PASSCAL experiment deployed from February 11 to May 27 in 1995. Rensselaer Polytechnic Institute, Indonesian Meteorological and Geophysical Agency, and the Volcanological Survey Indonesia operated this seismic array [Masturyono, 2000; Masturyono et al., 2001]. Additional data were also obtained from the permanent station PSI in Sumatra for the time period from 1996 to 2002. The stations of the Toba experiment were arranged in two crossing lines along the long and short axes of the NW-SE elongated Toba caldera with the permanent station PSI located at the center of this array (Figure 1b).
 We used teleseismic events with epicentral distances between 25° to 100°, magnitudes (mb) greater than 5.5, and an adequate signal to noise ratio. The data were high-pass filtered for frequencies greater than 0.02 Hz and receiver functions were calculated from them using an iterative deconvolution approach [Ligorría and Ammon, 1999]. Low-pass Gaussian filter was applied with several widths (“a” = 1.0, 2.5, 5.0), and those filtered with “a” = 2.5 (corresponding to the frequency of 1.2 Hz) that showed variance reductions greater than 80% are presented in this study. The resultant dataset consisted of 386 traces from PSI and 334 traces from the stations in the PASSCAL array.
 As a priori information, the crustal P-velocity model by Masturyono et al.  was used to improve the inherent nonuniqueness of receiver functions [Ammon et al., 1990]. Their model estimated an average Vp of 4.90 km/s for the top 10 km layer, 5.76 km/s for 10–24 km, 5.73 km/s for 24–40 km, and 7.36 km/s for 40–65 km depth. Ps converted phase and two reverberation phases, PpPs and PpSs + PsPs phases were studied to estimate crustal Vp/Vs and the depth of discontinuities. The PpPs (upper case denotes downward waves and lower case denotes upward waves) reflects from the free surface and again at the velocity boundary to convert from P to S waves. The PpSs + PsPs are also surface-reflected phases, reaching the surface with opposite polarity of Ps and PpPs. While it is known that these phases can be greatly affected by lateral heterogeneities, anisotropy, near-surface complexities and non-planer interfaces [e.g., Levander and Hill, 1985; Clouser and Langston, 1995], numerous studies in the past two decades have successfully utilized receiver functions to investigate lithospheric structure. In particular, modern methods of deconvolution [Ligorría and Ammon, 1999], analyzing phase moveout for multiples identification [Gurrola et al., 1994], and migration and stacking methods [Zhu and Kanamori, 2000; Gilbert et al., 2003] have greatly improved the utility of the receiver function technique.
 First we analyzed receiver functions for each individual station. We used the stacking procedure from Zhu and Kanamori  and the results were used to construct moveout model curves, which were visually compared to arrivals on receiver functions stacked in ray-parameter bins [Gurrola et al., 1994] of 0.016 s/km width. We also stacked radial and tangential components, respectively, as a function of backazimuth (BAZ) with 10° bins to see the extent of lateral heterogeneity. Stacking receiver functions of different ray parameters and BAZ, with carefully set bin sizes, helps to constrain laterally heterogeneous structure such as dipping and anisotropic layers [Owens et al., 1984, 1988; Ammon, 1991; Cassidy, 1992; Levin and Park, 1998]. However, the estimation of Vp/Vs and discontinuity depth with these methods still suffer from non-uniqueness, partly due to the skewed BAZ distribution of the data.
 In order to reduce the noise that is incoherent from station to station, receiver functions from all the stations were stacked into Common Conversion Point (CCP) bins [Dueker and Sheehan, 1997, 1998; Gilbert et al., 2003]. The piercing points of receiver functions for each 1 km depth interval down to 100 km depth were binned into a grid of stacking points separated by 10 km. Bins were set as circles with a radius of 15 km centered at each stacking point. In finding the depth of the discontinuities, the mean depth and its error was found using bootstrap resampling with 200 estimations in the depth range of 3–20 km for shallow magma systems and 20–45 km for the Moho. Stacked receiver functions from multiple stations are smoother, but it should be noted that CCP stacking also averages out some useful information about lateral heterogeneities, provided by different ray parameters and BAZs [Ammon, 1991; Cassidy, 1992]. The depths given in this paper are referenced from the station elevation that in this region is approximately 1 km above sea level.
3. Results and Discussion
 The estimation of an average Vp/Vs and thickness of the crust under each station is presented in Table S1 (in the auxiliary material). Stations other than PSI, TS004, TS005, TT07A, and TT008 do not provide clear estimates of Vp/Vs. Stations other than TS005 show high Vp/Vs values ranging from 1.85 to 1.91 and the average Vp/Vs value is 1.87. This average Vp/Vs was used to compute the CCP stacking, except for the station TS005 where we used a Vp/Vs value of 1.73 because this station was outside the caldera and its estimated Vp/Vs was much lower than other stations (Table S1). A Vp/Vs value of 1.87 is considerably higher than the average continental value of 1.78 [Zandt and Ammon, 1995] but it seems reasonable for a region that has active volcanism where partial melting and the presence of fluids generally increase Vp/Vs. If we used the lower Vp/Vs value of 1.78 from Zandt and Ammon,  this would increase our largest crustal thickness (below station TT02A in Figure 2) of 39 km by 5 km.
Figure 2 displays the two cross-sections along each profile of stations. Along AA′, receiver functions detect a strong negative phase near 10 km depth except under the station TT008. The same negative phase is noticeable in BB′ where the largest amplitudes are restricted to the region below the caldera. The amplitude of this negative phase (proportional to the magnitude of the impedance drop) and its depth distribution is presented in Figure 3. The largest amplitude negative phase of up to −30% relative to the initial P-wave amplitude is found under the southern Samosir Island, corresponding with the area of the slowest P-velocity in the 10–24 km depth range from P-wave tomography [Masturyono et al., 2001]. Somewhat higher amplitudes are also observed in the northwestern and southeastern ends of the caldera. At the northwestern end, the high amplitude negative phase bisects into two separate depths (between 0 and 20 km offset along AA′ in Figure 2) and may continue further outside of the area of our data coverage. We picked the deeper arrival to be the primary negative phase for the purpose of producing our contour plot (Figure 3) because it possesses a slightly stronger amplitude compared to the shallow one. Also noticeable is a positive phase beneath the negative phase (Figure 2, cross sections AA′ and BB′) at 15 ∼ 20 km depth. The juxtaposition of these two phases suggests that the shallower negative phase marks the top of a thin low velocity zone while the deeper positive phase could correspond the sharp bottom of the low velocity layer. The positive phase is not as obvious to the southeast, suggesting the bottom of the low velocity zone becomes more gradational to the southeast. Partial melting, compositional layering, and presence of a liquid phase can form such a low velocity layer, and it probably represents a magma reservoir that provided rhyolitic magmas to earlier caldera forming eruptions and the present active volcanoes [e.g., Masturyono et al., 2001; Chimielowski et al., 1999; Bannister et al., 2004]. High Vp/Vs, and low Vp, under this area also agrees with this interpretation. The magma reservoir does not appear to be separated by a high-velocity boundary as suggested by Masturyono et al. , instead at the level we are able to resolve, it seems to be a continuous reservoir with a few zones of focused lower-velocities. However, this apparent continuity may be due to the smoothing effect of the stacking procedure. In general, the location of the largest negative arrival correlates well with the location of the magma chamber observed by Masturyono et al.  from the strong low P-wave anomaly in a traveltime tomography study and a negative gravity anomaly centered over Samosir Island. The largest negative amplitude areas also coincide with the locations of major past caldera eruptions [Knight et al., 1986; Chesner and Rose, 1991].
 A strong positive P-s conversion is detected around 30 km that varies in depth across the caldera. We interpret this phase as the Moho based on its depth and amplitude. From the northwest to southeast (A-A′ in Figure 2), the thickness of the crust increases abruptly from 33 km to ∼37 km at horizontal offset of 20 km and then abruptly thins to ∼28 km depth at the horizontal offset of 60 km. Variations in crustal thickness are also observed along the perpendicular BB′ cross-section (Figure 2), where the crust appears to thin from 39 km to 25 km thick at a horizontal offset of 50 km. Such a change in crustal thickness sometimes is an artifact resulting from large variations in Vp/Vs when using constant Vp/Vs for CCP stacking. However, we believe this thickness change is not entirely due to a Vp/Vs change because we used a lower Vp/Vs of 1.73 for the rays sampled by TS005, and other stations have comparable Vp/Vs as previously discussed.
 It is also important to note that in receiver function analysis, the depths of features can be migrated to incorrect depths due to the presence of unaccounted velocity heterogeneities at shallower depth [Wilson et al., 2003]. In our study, however, we consider this local depression to be a robust observation because: 1) a low velocity zone is observed not only above the depressed region of Moho, but throughout the study area (Figure 2, cross section AA′); 2) the locations of the largest negative phase, indicative of largest negative impedance contrast is offset 15 km from the region of deepest Moho, and 3) other features at depths greater than the Moho do not appear to be shifted in depth a similar amount as the Moho.
 The map of crustal thickness (Figure 1b) clearly illustrates that the thickest crust is located beneath the central portion of the eastern wall of the Toba caldera with Samosir Island (resurgent dome) and Pusubukit volcano within the region of the Moho depression at >32 km depth. The graben between Samosir Island and the Uluan Block appears to bisect the small crustal welt. There are several possible explanations for mechanisms that thicken the crust locally. One plausible mechanism is that this highly localized depression of the Moho is caused by underplated mafic material from the mantle. If the underplated material is gabbroic, and still contains partial melt, we would expect a prominent P-s conversion to mark the boundary between the underplated gabbros and the underlying mantle beneath the original Moho. Thus increasing the apparent thickness of the crust.
 Here, we use the term “Moho” from the geophysical view that defines it as a level of sharp increase in seismic velocities from typical crustal values to mantle values, rather than the petrological and compositional crust/mantle boundary [Mengel and Kent, 1992; Giese et al., 1999]. It is also interesting and supportive to note that the locations of thickest crust correlates well with the swath of increased seismic activity associated with the Investigator Fracture Zone of the subducting Indian-Australian Plate [Fauzi et al., 1996] (Figure 1a). It is possible that the presence of a large fracture zone within the slab can be the source of increased fluid release that induces partial melting of the overlying mantle wedge [Fauzi et al., 1996]. The underplating material could become a potential heat source for crustal melting that produces the silica rich magmas. Masturyono et al.  has suggested that a WSW-ENE elongated low P-velocity column below Pusubukit volcano and Samosir Island, associated with low-frequency earthquakes, is a possible modern magma conduit from Moho depths to the upper crustal magma chamber. Petrologic and geochemical investigations of the material erupted from the Toba caldera by Wark et al.  also suggests a process of magma-recharge from the upper mantle. However, this interpretation is not favored by seismic studies of the crust and upper mantle structure of the Taupo Volcanic Zone in central North Island, New Zealand, that reveals a mafic underplate where the large impedance contrast is at the top of the underplate, not the bottom [Stratford and Stern, 2006].
 An alternate interpretation of the observed crustal welt and Moho offset under the northeast caldera wall (Figure 2, cross section BB′) is that it reflects tectonic crustal thickening. Wark and McCaffrey  suggested that the unique characteristics of the Toba magmatic system, in comparison to the rest of the arc, may be related to its unique large offset from the Sumatran fault. They noticed that Toba Lake is located near a bend in the strike of the Sumatran fault and that small circles that trace the path of the fault to the northwest and southeast intersect along the eastern margin of the Toba caldera. They suggested that the fault may have been located along this position in the past, producing a large restraining bend. This localized compressional setting could have stalled rising mafic magmas leading to increased crustal melting and assimilation. Then, a westward jump to its present location ∼2 Ma could have partially released the compression and initiated caldera-forming eruptions. The Moho offset we observe beneath the eastern edge of the caldera could be interpreted as this relict Sumatran fault segment with localized crustal thickening on the southwest side (Figure 2, cross section BB′). This structure could have been a pathway for mantle melts to intrude the crust and feed the upper crustal magma reservoirs. In this scenario the mantle melts migrate directly to the upper crustal reservoir without substantial residence time near the base of the crust. If there is no underplate beneath Toba, it would be in contrast to the Taupo Volcanic Zone where the presence of a crustal underplate is better resolved [Stratford and Stern, 2006]. Although somewhat counterintuitive, the difference between Toba and Taupo could be related to the extensional back arc regime for Taupo and the neutral to slightly compressional backarc regime of Toba.
 Receiver function analysis confirms the presence of a magma reservoir under the Toba caldera with its top at depths between 8 and ∼14 km. The largest negative receiver function arrival, which may mark a zone of low shear-wave velocities due to the presence of magma, is detected under southern Samosir Island and two other relatively strong negatives are found under the northwest and southeast ends of the caldera. The average Vp/Vs of the crust under the Toba caldera is estimated to be 1.87, higher than that found outside the caldera (1.73), also supporting the presence of magma. The crust is found to be between 29 and 40 km thick with localized thickening under the eastern wall of the Toba caldera. The center of the region, where the crust is the thickest, is found above the subducted Investigator Fracture Zone. One possible interpretation for the localized area of thick crust is that it is underplated hot gabbroic material from partially melted mantle. An alternate interpretation is that the crust was locally thickened when it was located inside a tight restraining bend of a now abandoned segment of the Sumatran fault.
 The facilities of the IRIS Data Management System, and specifically the IRIS Data Management Center, were used for access to waveform and metadata required in this study. The IRIS DMS is funded through the National Science Foundation and specifically the GEO Directorate through the Instrumentation and Facilities Program of the National Science Foundation under cooperative agreement EAR-0004370. We thank two anonymous reviewers for their comments and suggestions that improved the final manuscript.