Corresponding author: M. S. Karplus, School of Ocean and Earth Science, University of Southampton, Waterfront Campus, European Way, Southampton, UK SO14 3ZH. (M.S.Karplus@soton.ac.uk)
 We use ambient-noise tomography to map regional differences in crustal Rayleigh-wave group velocities with periods of 8–40 s across north Tibet using the International Deep Profiling of Tibet and the Himalaya phase IV arrays (132 stations, deployed for 10–24 months). For periods of 8–24 s (sensitive to midcrustal depths of ~5–30 km), we observe striking velocity changes across the Bangong-Nujiang and Jinsha suture zones as well as the Kunlun-Qaidam boundary. From south to north, we see higher velocities beneath the Lhasa terrane, lower velocities beneath the Qiangtang, higher velocities in the Songpan-Ganzi and Kunlun terranes, and the lowest velocities beneath the Qaidam Basin. Maps at periods of 34 and 40 s (sensitive to the middle and lower crust at depths of ~30–60 km) do not show evidence of changes across those boundaries. Any differences between the Tibetan terrane lower crusts that were present at accretion have been erased or displaced by Cenozoic processes and replaced almost ubiquitously by uniformly low velocities.
 Since the pioneering deployment of long-period seismographs within Tibet in 1982 [Jobert et al., 1985], numerous studies of surface-wave dispersion have been performed within Tibet using regional and teleseismic earthquakes [e.g., Huang et al., 2003; Levshin et al., 1994; Rapine et al., 2003; Shapiro et al., 2004] and ambient noise [e.g., Guo et al., 2009; Li et al., 2009; Yang et al., 2010; Yao et al., 2008]. However, the majority of regional studies reporting high-resolution crustal models are in southeast [e.g., Yao et al., 2008] and southern Tibet [e.g., Guo et al., 2009]. In this study, we use new data collected from stations in north Tibet deployed by project INDEPTH (International Deep Profiling of Tibet and the Himalaya), phase IV, to create a high-resolution crustal Rayleigh-wave model.
 In southern Tibet, researchers observe a low-velocity zone from surface waves in the middle crust [e.g., Caldwell et al., 2009; Rapine et al., 2003] and suggest that it supports the hypothesis of fluids in the middle crust first imaged using INDEPTH II active-source seismic data [Nelson et al., 1996] and perhaps also supports the crustal-flow models proposed by Royden et al.  and refined by Beaumont et al. . These crustal-flow models propose that the middle and/or lower crust is decoupled from the upper crust and upper mantle, allowing the middle-lower crust to flow outward from central Tibet, releasing gravitational potential energy of the high plateau and continued compression from the India-Asia collision [Klemperer, 2006]. However, recent studies suggest that low-velocity zones as well as electrically conductive zones in the middle and lower crust beneath south and southeastern Tibet may be discontinuous, with significant lateral variations and complex structure [Hetényi et al., 2011; Yang et al., 2012; Yao et al., 2010].
 In north Tibet, recent wide-angle reflection and refraction results from INDEPTH IV [Karplus et al., 2011] show that middle- and lower-crustal P wave velocities beneath the Songpan-Ganzi terrane and Kunlun Mountains are as low as those in southern Tibet, indicating the possibility of partial melt or other fluids within the crust and thus the possibility for flow. Reflector geometries were interpreted to indicate injection of Kunlun Mountains lower crust beneath Qaidam crust at or near the Moho [Karplus et al., 2011]. Magnetotelluric data have been interpreted as showing north-directed midcrustal melt penetration across the north Kunlun fault [Le Pape et al., 2012].
 We use ambient noise on continuous seismic records from north Tibet to construct empirical Green's functions (EGFs) and constrain Rayleigh-wave group velocities for periods of 8–40 s. Our model is well-resolved and sensitive to structure from ~5–60 km in the Lhasa, Qiangtang, Songpan-Ganzi (SG), Kunlun Mountains, and Qaidam terranes from approximately 91°E to 99°E and 31°N to 37°N.
2 Data and Methods
 Our data sources are vertical-component continuous recordings from the collaborative Chinese, U.S., German, and U.K. INDEPTH IV linear profiles and areal array, grouped according to the times they were deployed. Station group A (78 stations) was deployed nominally from June 2007 to September 2008. Station group B (24 stations) was deployed from August 2008 to June 2009. Group C (30 stations) was deployed for the entire time period from June 2007 to June 2009 (Figure 1). Stations with < 10 days of recorded data are not used. Previous studies have used portions of these data for ambient noise studies [Yang et al., 2010, 2012], but we add stations from dense, 1-D profiles crossing the Jinsha and Kunlun sutures thereby adding greater redundancy in north Tibet. Station pairs with interstation distances ⪅ 30 km do not produce reliable ambient noise results, even for our shortest periods of 8 s, due to scattering. However, the stations of the 1-D profiles paired with other stations of the regional array ⪆ 30 km away give many two-station paths crossing the area around the 1-D profiles. We only use the vertical component noise, so that the cross-correlations contain only Rayleigh-wave signals [Bensen et al., 2007].
 In the time domain, the cross-correlation of whitened ambient seismic noise at two stations yields a waveform resembling the EGF between them [Sanchez-Sesma and Campillo, 2006]. In this study, we obtain Rayleigh-wave dispersion measurements from interstation EGFs generated by the cross-correlation of seismic ambient noise using well-established methods [Bensen et al., 2007; Prieto et al., 2009] (Supporting Information).
 Briefly, we obtain a group-velocity dispersion curve corresponding to each station pair using the multiple-filter technique [Dziewonski et al., 1969] (Figure S1b–S1c). At periods shorter than 6 s the ambient field amplitude is sufficiently small that coherency does not persist for ≥ 50 km. Robust dispersion measurements can typically only be made up to a period equal to the interstation spacing (in km) divided by 12 [Bensen et al., 2007]. We did not have a sufficient number of station pairs with interstation spacing >500 km to measure dispersion at periods > 40 s.
 We then use an iterative conjugate-gradient least-squares inversion method [Paige and Saunders, 1982] to derive velocities for a block dispersion model with 0.1° × 0.1° lateral block size, and 2 s increments of group velocity from 6 to 40 s. We perform checkerboard resolution tests to examine velocity recovery for different period slices (Figure 2) and report the velocity maps for those slices (Figure 3). We display this model as velocity vs. period curves for individual surface locations within the model (Figure 4).
3 Group-Velocity Results
3.1 Group-Velocity Maps
 Rayleigh-wave group-velocity maps at periods of 8–40 s are plotted in Figure 3. The approximate depths sampled by waves of various periods are shown in Figure S5 and noted in the text below. Decreased velocity with increased period need not correspond to an actual decrease of velocity with increasing depth [e.g., Shapiro et al., 2004]. However, group-velocity variations with period approximate average shear velocities for different depth intervals of the crust but also strongly depend on velocity gradients within the depth range of sensitivity [Villaseñor et al., 2007].
 We observe striking velocity contrasts from 8 to 24 s (down to ~30 km depth) across many of the major regional terrane boundaries and faults. At periods less than 24 s, the Bangong-Nujiang Suture appears to separate higher Lhasa terrane velocities from lower Qiangtang velocities, but the velocities become similar for longer periods. This suggests a faster Lhasa upper crust compared to the Qiangtang, confirming and extending into new geographic areas similar observations from INDEPTH III P wave modeling at 89°E [Haines et al., 2003]. Rapine et al.  observed the opposite trend in surface-wave group velocities (slower Lhasa upper crust compared to the Qiangtang) using earthquakes, albeit averaging along E-W paths over a larger area from ~80°E–90°E. They also observed lower group velocities at almost all periods, perhaps suggesting lateral heterogeneity or anisotropy.
 The Qiangtang terrane has several localized, shallow low-velocity anomalies (~0.1–0.15 km/s below average for the period slice). From 8–24 s, the Jinsha suture separates lower-velocity Qiangtang terrane from higher-velocity Songpan-Ganzi, but velocities at longer periods, 34–40 s (depths ~30–60 km), are similar across the Jinsha suture.
 A strong high-velocity anomaly of > 0.15 km/s above the period average is observed beneath the SG terrane between ~93°E–96°E, at periods <24 s. There appears to be a narrower N-S–oriented high-velocity anomaly beneath the SG at 34–40 s. This anomaly may represent a subterrane within the SG, although lack of basement outcrop in this region of Tibet limits our geologic interpretation of this anomaly.
 High-velocity anomalies of 0.05–0.10 km/s are seen in the Kunlun Mountains between the north Kunlun thrust (NKT) and the north Kunlun fault for periods < 24 s. These high velocities are likely due to intrusives of the Kunlun batholith.
 The slow velocity anomaly of up to ~0.15 km/s associated with the Qaidam Basin is clearly observed for periods < 18 s and less strongly, to 24 s, consistent with known thicknesses of Qaidam sedimentary rocks up to 15 km thick in the center of the basin [Yin et al., 2007], and relatively high shear-wave velocities beneath Qaidam compared to the Songpan-Ganzi between ~22–30 km depth [e.g., Ceylan et al., 2012; Mechie et al., 2012].
 Our group-velocity maps are broadly consistent with the phase-velocity maps of Yang et al.  at 10 s. At 40 s, the maps are also similar, except that we record a strong fast anomaly beneath the Songpan-Ganzi at 93°E–95°E and 34.5°N–35.5°N, near our densest station coverage.
 Our velocity anomalies are also broadly consistent with the phase-velocity maps produced by Ceylan et al.  using two-plane-wave tomography for waves with periods from 24 to 40 s. The one large difference between the results of our paper and Yang et al.  compared to Ceylan et al.  is that Ceylan et al.  showed significantly higher velocities (up to ~0.2 km/s higher) beneath the Qaidam Basin from 24 to 40 s compared to the Kunlun Mountains and Songpan-Ganzi to the south. This result is puzzling given that all three papers used the same INDEPTH IV regional stations across the Qaidam Basin, but may be an artifact of the different methods combined with generally poor resolution in the Qaidam Basin. Ceylan et al.  also did not image the strong fast anomaly that we find, seemingly well-resolved within our densest station coverage, beneath the Songpan-Ganzi at 93°E–95°E and 34.5°N–35.5°N.
3.2 Group-Velocity Dispersion Curves
 We extract Rayleigh-wave group-velocity dispersion curves from our velocity model every 0.5° latitude along a N-S profile along 95°E (Figure 4), along which we have good model recovery at all points from 32°N to 36°N (Figure 2). Our longest period Rayleigh waves, 40 s, are most sensitive to depths of 30–40 km, insufficient to sense the mantle in Tibet where the Moho is ~70 km deep south of the NKT and ~50 km deep beneath the Qaidam Basin [Karplus et al., 2011].
 The most striking feature of this dispersion curve profile is that almost all curves show group velocities decreasing with period between ~15–25 s (Figure 4). Group velocities decrease by as much as ~0.1 km/s at 34°N (just south of Jinsha suture) and 36°N (just north of NKT). In contrast, dispersion curves are flat in the same period range at 33°N (central Qiangtang terrane) and 37°N (Qaidam Basin).
 The decrease in group velocity at periods > 20 s is most evident where we find the highest group velocities at ~10 s period (upper-middle crust) of 3.11 km/s (~0.04 km/s higher than surrounding areas of the same period/depth) at 34.5°N (south of the south Kunlun fault in the Songpan-Ganzi terrane). Conversely, there is no midcrustal decrease in group velocity north of 36.5°N, within the Qaidam Basin, where we find the lowest, short-period (~10 s) velocity at ~2.8 km/s, presumably due to the sedimentary rocks in the basin known to reach depths > 15 km [e.g., Yin et al., 2007].
 Down to ~30 km (sampled by periods of ⪅ 24 s), changes in upper- to middle-crustal velocities occur across the suture zones. The velocity change from the Qiangtang to the SG terrane is 10–50 km south of the suture perhaps corresponding to the geologically inferred gentle southward dip [Kapp et al., 2003]; the other velocity changes are more closely aligned with surface faults implying more vertical structures. The sutures may have been erased in the lower crust, possibly by crustal flow, because the lateral velocity differences are much smaller at periods corresponding to the middle-lower crust than at periods representative of the upper crust.
4 Discussion and Implications for Geodynamics
 Our robust Rayleigh-wave group-velocity model for waves from 8 to 40 s for the north Tibetan Plateau was derived using ambient-noise tomography with stations of the INDEPTH IV linear and regional arrays. Our model adds new detail to our structural understanding of north Tibet while confirming known broad-scale features seen in previous studies of ambient noise and earthquake surface-wave tomography. The combination of the INDEPTH IV linear profiles and areal array gives us the densest ray-path coverage yet available across the north Qiangtang, Songpan-Ganzi, Kunlun, and Qaidam regions between 92°E–98°E.
 Above ~30 km depth, the Rayleigh-wave group-velocity contrasts across the Bangong-Nujiang suture (8–24 s), the Jinsha suture (8–24 s), the south Kunlun fault (18–24 s), the north Kunlun fault (18–24 s), and the north Kunlun thrust (8–24 s) likely mark terrane or block boundaries. If this is the case, the upper ~30 km of the crust may be deforming as coherent blocks, and any decoupling layer must be deeper than ~30 km. The major sutures are no longer evident in velocity maps at 34 and 40 s (below ~30 km), consistent with lower-crustal flow having erased or displaced velocity contrasts in the lower crust and suggestive that the middle-to-lower crust may be decoupled from the upper crust and deforming continuously [cf. Yang et al., 2012]. This is consistent with the observation that seismicity across north Tibet is restricted to the upper 20–30 km of the crust [Wei et al., 2010] and with the view from controlled-source seismic data that the north and south Kunlun faults do not extend below 20–30 km [Karplus et al., 2011], and the Bangong-Nujiang suture does not extend near-vertically below about 30 km [Haines et al., 2003].
Yang et al.  observed a low shear-velocity zone at ~30 km (sampled by waves with period 24 s) across most of the Tibetan Plateau marked by a velocity minimum in the 1-D vertical velocity-depth model. The low velocities are most marked near the boundaries of the Tibetan Plateau [Yang et al., 2012]. We also see a group-velocity minimum at ~24 s (sampling depths of ~30 km). In our study area, the most marked velocity minimum is seen near the Kunlun-Qaidam boundary.
 Previous studies have explored whether the presence of a low-velocity zone indicates midcrustal partial melting perhaps initiated by the presence of aqueous fluids [Caldwell et al., 2009; Yang et al., 2012]. Widespread fluids have been inferred from 25 to 50 km depth at 32.5°N–35.5°N, 92°E–94°E [Unsworth et al., 2004]. Hence, our velocity minimum, which may correspond to the low-velocity zones from previous studies, and that underlies a large area of north Tibet (~ 32°N–36°N), may indicate a region of partial melt or of aligned anisotropic minerals [Yang et al., 2012] and almost certainly indicates an area of significant crustal weakness [e.g., Klemperer, 2006] and substantial middle-to-lower crustal deformation that may continue today as a distributed shear zone [Yang et al., 2012]. The northern limit of the low group velocities is broadly consistent with the location where Karplus et al.  suggested northward injection of the Kunlun Mountains crust into the Qaidam Basin.
 INDEPTH activities are funded by NSF grants (EAR-CD-0409939, EAR-0634904, EAR-0409589), Deutsche Forschungsgemeinschaft, GFZ Potsdam, and CGS-1212010511809 in China. An NSF graduate fellowship supported M. Karplus. We thank the INDEPTH field teams. Instruments were provided by IRIS-PASSCAL, NERC-SEISUK, and GFZ-GIPP. The seismic data is archived in the IRIS Data Management Center (networks XO, X4, and Z1) and the GFZ GEOFON database (network XO). We thank N. Harmon for assistance with the group-velocity sensitivity calculation. We thank N. Rawlinson and an anonymous reviewer for suggestions to improve the manuscript.