3.1. Bangong-Nujiang Suture (BNS)
 Two important but unresolved questions about the BNS are how was it formed, and how is it related to the Moho step and change in mantle properties with which it is spatially associated [Kosarev et al., 1999]. The BNS is defined by a 150-km-wide belt of scattered Jurassic ophiolite and related melange material that was emplaced on the northern Lhasa block during the latest Jurassic or earliest Cretaceous [Girardeau et al., 1986]. The northern half of the Lhasa block was shortened as much as 60% during and after the emplacement of the ophiolite material [Murphy et al., 1997]. The rock types on either side of the BNS (the Lhasa and Qiangtang blocks) are not strikingly different (mainly Mesozoic shallow marine sedimentary units) [Kidd et al., 1988].
 INDEPTH III provides some of the first geophysical data to characterize the BNS within the crust. We find a region of low seismic velocity extending to about 30 km depth, located near 32°N. A minimum-structure magnetotelluric model [Wei et al., 2001] shows little signature associated with the BNS, whereas more finely parameterized models [Solon et al., 2000; Solon, 2000] allow complex near-surface resistivity and a low-resistivity body located near the BNS in the middle and lower crust. A change in lithology or the presence of a fractured/sheared zone could explain the low velocities at the BNS. Fluid accumulations in a fractured zone would be consistent with low resistivity values.
 Aeromagnetic data across the BNS [Liu, 1989] show distinct anomalies corresponding with known outcrops of serpentinized peridotite but are otherwise generally smooth (Figure 6; see locations of ophiolite in Figure 1). This magnetic pattern suggests that either peridotite at depth is not serpentinized (unlikely, given that peridotite at the surface is highly serpentinized) or that no major ophiolite is present in the subsurface. This observation is in contrast to the Yarlung-Zangbo Suture where a pronounced aeromagnetic anomaly is linked to the large bodies of ultramafic ophiolite believed to exist at depth [Makovsky et al., 1999]and mapped at the surface [Liu et al., 1986]. Better known examples elsewhere in the world include the Great Valley of California where a 10-km-thick ophiolite in the midcrust gives rise to a clear magnetic signature associated with the presence of the large ophiolite at depth rather than individual anomalies associated with bodies of serpentinite at the surface [Cady, 1975; Godfrey and Klemperer, 1998].
Figure 6. Aeromagnetic data [Liu, 1989] for eastern Tibetan Plateau. Data are not available for western plateau. Note pronounced anomaly associated with YZS and lack of anomaly associated with BNS. Scale of map is same as Figure 1.
Download figure to PowerPoint
 On the basis of geologic evidence, Girardeau et al.  and Yin et al.  hypothesize that the oceanic crust corresponding with the ophiolite masses of the BNS was formed above a subduction zone during the Late Triassic to mid-Jurassic (Figure 2b). That the suture zone includes only a minimal volcanic arc (evidenced by scattered small granite bodies, and little or no arc volcanic rock) suggests that the closure of this basin may have included only minimal subduction. INDEPTH III seismic data show no evidence for high-velocity subducted oceanic crust beneath or within the Qiangtang block. In addition, evidence for anomalously warm mantle material beneath the Qiangtang block [McNamara et al., 1997; Wittlinger et al., 1996] is not consistent with the presence of cold slab material that would be associated with a long-lasting subduction environment. Though the signature of a subducted slab could have been obscured by magma injection, the lack of this signature is at least consistent with the other evidence mentioned. Thus we conclude that the ophiolite associated with the BNS was formed in a narrow basin (forearc or back arc) that existed between the Lhasa and Qiangtang blocks during the early Mesozoic (perhaps created by the subduction of Songpan-Ganzi to the north) and that little subduction was associated with the closure of this basin (Figure 2c).
 A tempting explanation for why the ophiolitic fragments are located as much as 150 km south of the suture zone is that they were emplaced by lateral motion along the several prominent strike-slip faults located near the main outcrops of ophiolite at Xainxa and Gyanco/Donqiao. However, the few tens of kilometers of displacement estimated along these faults by Armijo et al.  is insufficient to account for the long distance spanned by the ophiolitic fragments to the south of the BNS. Another possible explanation is that the southern ophiolite bodies are indicative of another suture within the Lhasa block [e.g., Matte et al., 1996], but little supporting evidence exists for such a feature.
 The shallow extent of the signature of the BNS suggests that the BNS is unrelated to the upward step in the Moho and the changes in upper mantle properties that occur beneath the suture. If the two features were related, we might expect to find that the low seismic velocities associated with the shallow suture would extend to the Moho, or that seismic reflections from such a structure would be present in our reflection data. Low seismic velocities in the crust of northern Tibet [Zhao et al., 2001] indicate low density material there. Low Pn velocities and high attenuation of Sn phases beneath the northern Tibetan Plateau suggest an anomalously warm upper mantle. The thinner crust in the north and the laterally continuous elevation of the plateau can be explained by lower density crust in the north or support there by warm and less-dense mantle.
 In order to explore the issue of plateau isostatic support, we used velocity-density regression parameters [Christensen and Mooney, 1995, Table 8] to calculate lithostatic density columns for the Lhasa and Qiangtang blocks (Figure 7) from our velocity model (Figure 4)[Zhao et al., 2001]. For the mantle lithosphere beneath the Qiangtang block we use a density of 3320 kg/m3 [Ringwood, 1975]. The density for the uppermost 5 km of crust is taken to be the average of reasonable end members as determined for each block. For the Lhasa block, the lower plausible bound is ρ = 2440 kg/m3 (1 km unconsolidated sediments over 4 km sedimentary rock using velocity-density relations after Gardner et al. ) and the upper bound is ρ = 2700 kg/m3 (granite extending to the surface). For the Qiangtang block, the lower bound is ρ = 2430 kg/m3 (0.5 km of unconsolidated sediments above 4.5 km of sedimentary rock) and the upper bound is ρ = 2550 kg/m3 (sedimentary units extending to the surface). Lateral density variations for the upper 5 km can change the total calculated pressure by up to 10 MPa. Densities corresponding to the velocity model of Zhao et al.  indicate that crustal density differences alone can maintain isostatic balance despite the flat plateau topography and northward crustal thinning of Tibet because the Qiangtang velocity function is so much slower than the Lhasa block velocity function, particularly in the upper crust. The total pressure beneath the Lhasa block at the Moho (65 km) is 1.829 GPa, whereas the pressure beneath the Qiangtang at 65 km is found to be 1.832 Gpa, a difference of only 0.003 GPa (0.03 kbar) (Figure 7). Thus the northward step-up of the Moho near the BNS, with no accompanying elevation change, does not imply a change in mantle density and should not be used as corroborating evidence for the variations in mantle density argued for from teleseismic results [e.g., Wittlinger et al., 1996; McNamara et al., 1997; Huang et al., 2000]. Previous gravity studies lacking our constraints on density (via velocity) and crustal thickness have assumed constant crustal density and either changes in mantle density [Jin et al., 1996], or crustal thickness variations not supported by our data [Braitenberg et al., 2000] to explain the Tibetan gravity field. Our result also provides a reasonable explanation for the change in Moho depth that occurs beneath the surface trace of the BNS without indicating that the trace of the BNS extends to the Moho. In fact, the change in depth to Moho is explained mainly as a result of upper crustal density variations, rather than as a product of any sort of variations in the lower crust beneath the BNS.
Figure 7. Velocity (Vp) and density (ρ) columns for the Lhasa and Qiangtang blocks. Total pressure beneath the Lhasa and Qiangtang blocks at 65 km depth is the same to within 3MPa (0.03 kbar), indicating that isostatic compensation is achieved within the crust. Density was calculated using the velocity-density regression parameters of Christensen and Mooney [1995, Table 8]. Crustal velocities are from Zhao et al. , averaged over the swaths shown in Figure 4. Total pressure is calculated by summing the product ρgh for each column, where g is gravitational acceleration (9.8 m/s2), considered to be constant within the columns, and h is the thickness of each layer within the column. Layers in velocity and density columns were designed to best represent the velocity gradients within each layer of the Vp model of Zhao et al. .
Download figure to PowerPoint
 The underthrusting of southern Tibet by Indian lithosphere north to about 32°N (coincidentally close to the surface location of the BNS, at the present time) remains the best explanation for the change of physical properties of Tibetan lithosphere at that latitude [Owens and Zandt, 1997]. The dearth of any sort of signature of the BNS vertically (or sub-vertically) in the lower crust beneath its surface location may indicate that this feature (assuming that it once existed) has been eradicated by lower crustal processes. Thus, we advocate models for plateau development that involve large-scale motion of middle and lower crustal material beneath the suture.
3.2. Qiangtang Block
 The deepest-seated rocks exposed in the Qiangtang block are early Mesozoic blueschist-bearing metamorphic units exhumed by Late Triassic-Early Jurassic low-angle normal faults. Exposures of these units form an E-W trending metamorphic belt interpreted to represent a large volume of melange emplaced during the southward subduction of Songpan-Ganzi oceanic lithosphere beneath the Qiangtang block in the early Mesozoic (Figure 2b) [Kapp et al., 2000]. Field measurements at a melange outcrop near Shuang Hu, ∼10 km east of the INDEPTH III seismic array, show the bounding Falong Detachment dipping steeply at the surface [Kapp et al., 2000] (Figure 8).
Figure 8. Detail of Shuang Hu area, after Kapp et al.  and Yin et al. . a) map showing melange core complex, Shuang Hu normal fault, location of shot 97 (data in Figure 5c), black triangles marking locations of INDEPTH seismometers, dashed line showing 2-D approximation of INDEPTH III seismic transect; b) possible E-W cross section with Falong Detachment as the ∼3-km-deep interface required by seismic data; c) possible E-W cross section with base of Triassic strata as the ∼3-km-deep interface. Tr: Triassic, P: Paleozoic.
Download figure to PowerPoint
 The arrival of a distinct seismic phase at receivers in the northernmost part of the INDEPTH III seismic array from shots north and south of Shuang Hu indicates a transition from a velocity of <5 km/s to a velocity of 5.8 km/s at a depth of about 3 km (see Figures 4b, 5b, and 5c). In the vicinity of Shuang Hu, the INDEPTH III seismometers are located within a half graben down-dropped ∼6–8 km along the 13.5 Ma Shuang Hu normal fault (Figure 8a) [Blisniuk et al., 2001; Yin et al., 1999]. However, because of the crooked-line geometry of the INDEPTH III seismic array and because of the low-velocity material within the graben, most of the material sampled by ray paths in the Shuang Hu area from shots north and south of the map area (Figure 8a) lies to the west of the Shuang Hu normal fault. Thus the depth to the Falong Detachment or other geologic features indicated by the results presented here is the depth to those features on the footwall side of the Shuang Hu normal fault and inferences regarding offset on that normal fault can not easily be drawn from our data.
 Our preferred interpretation (Figure 8b) is that the velocity contrast corresponds with the contact between Triassic strata and the metamorphic core complex. This interpretation therefore suggests that the Falong Detachment flattens at a shallow depth [cf. Kapp et al., 2000], so that the melange remains at shallow depths to the north of the core complex exposure. Alternatively (Figure 8c), the velocity increase could represent a contact between Triassic and Paleozoic units within the Qiangtang crust, above the Falong Detachment (at unknown depth), and the melange would represent a smaller component of the crust. Although high-seismic-velocity limestone exists within the Triassic strata [Yin et al., 1999], we doubt that these units cause the observed velocity contrast because low-velocity shales continue beneath the thin limestone units.
 In our preferred model, the melange must be of regional extent beneath the northern plateau (Figure 9a), consistent with the conclusions of Yin and Harrison  and consistent with our observations of the relatively low seismic velocity of the Qiangtang crust. Hence our model for plateau development must account for the underthrusting of a large volume of Songpan-Ganzi material beneath the Qiangtang block in order to emplace the melange. The original Qiangtang lower crustal material in a later section has been displaced, presumably by lower crustal flow (see below).
Figure 9. Generalized cross sections of the Tibetan Plateau along transects shown in inset and Figure 1: a) A-A′, SSE–NNW in central Tibet, along the INDEPTH III seismic transect, b) B-B′, SSW-NNE in eastern Tibet, along the Lhasa Golmud Highway, c) C-C′, W-E across the central Qiangtang block. Songpan-Ganzi Terrane (SG), South Tibetan Detachment System (STDS), Main Boundary Thrust (MBT), Main Himalayan Thrust (MHT) and other abbreviations as in Figures 1 and 2 [after Nelson et al., 1996; Yin and Harrison, 2000; Owens and Zandt, 1997].
Download figure to PowerPoint
3.3. Crustal Fluids in Northern Tibet
 The combined data sets of INDEPTH III active- and passive-source seismic and magnetotelluric surveys reveal less evidence for fluid concentrations along the main INDEPTH III seismic transect in central Tibet (Figure 9a) than do data from southern and eastern Tibet. Minimum regional electrical resistivity values for the main INDEPTH III transect (∼20 Ωm) presented by Wei et al.  are much higher than those found in southern Tibet (∼3 Ωm) [Chen et al., 1996]. The more highly parameterized inversion of Solon  shows low electrical resistivity values (∼4 Ωm) locally beneath the BNS, possibly indicating the presence of fluids within the suture zone. Surface-wave analysis along the same transect shows no crustal low-velocity zone [Rapine et al., 2002], a result supported by velocity modeling of active-source seismic data [Zhao et al., 2001] and by the lack of pronounced high-amplitude reflections (bright spots) in reflection profiles. In contrast, magnetotelluric studies in northeastern Tibet along the Lhasa-Golmud highway (Figure 9b), detect highly conductive (∼3 Ωm) crust consistent with fluid accumulations more extensive even than those in southern Tibet [Wei et al., 2001]. Unfortunately, active-source seismic data have not been collected along the Lhasa-Golmud highway for comparison with the magnetotelluric data.
 The lack of bright spots in INDEPTH III seismic reflection data (CMP profiles and wide-angle data) suggests, but does not require, the absence of fluid concentrations in the central Tibet survey area. For comparison, individual bright spots imaged in southern Tibet span an in-line distance of as much as 25 km, though much of this apparent length may be due to diffraction from multiple smaller concentrations (1 to 2 km) [Makovsky and Klemperer, 1999]. The resolution of these concentrations was made possible by the collection of common midpoint data along ∼200 km of the 300-km INDEPTH II transect. Along the 400-km INDEPTH III transect, true common midpoint profiling (with 16 m midpoint spacing—Geometrics data) was limited by cost to only about 17 km length, though augmented by about 60 km of profiling with ∼125-m midpoint spacing (RefTek data). Thus the INDEPTH III data indicate only that fluid concentrations are not present on the 20% of the INDEPTH III transect that is well sampled (including much of the northern 80 km of the Lhasa block).
 On the basis of consideration of all the available evidence, we conclude that fluid bodies are not widespread along the main INDEPTH III seismic transect. The available data are consistent with either diffuse crustal fluids or a dry but hot crust for the INDEPTH III seismic transect [cf. Hacker et al., 2000]. We conclude that melt north of 31°N is most prevalent in eastern Tibet (along, and east of, the Lhasa-Golmud highway; Figures 1 and 9b), and may be related to the proposed flow of lower crustal [Clark and Royden, 2000] or upper mantle [Huang et al., 2000] material to the east. These fluids may be a result of shear heating caused by ductile flow [e.g., Huang et al., 2000]; or they may be a result of heating from a mantle diapir [Wittlinger et al., 1996], and a facilitator of ductile flow. A west-east variation in physical properties (Figure 9c) of the lithosphere is suggested by several different data sets. Magnetotelluric results from northern Tibet (previously described) and new results from southern Tibet [Unsworth et al., 2001] show that higher conductivity values (and likely also correspondingly greater amounts of crustal fluids) exist in the crust of eastern Tibet than western Tibet, at least to the northern end of the INDEPTH III seismic transect (∼200 km north of the BNS). That the zone of high Sn attenuation under northern Tibet (shown on Figure 1) extends eastward beyond the plateau margin but extends westward only to a distance of several hundred kilometers from the western plateau margin [McNamara et al., 1995] is also consistent with a relatively greater percentage of crustal fluids under eastern regions of the Tibetan Plateau. Thus we must take into account this apparent west-east gradient as we consider models for plateau uplift and evolution.
3.4. Crustal Thickening Mechanisms
 A comparison between velocity models for the Tibetan crust and the global-average crustal velocity models [Christensen and Mooney, 1995] (Figures 3a and 10a) shows that the most fundamental difference between the global average and the velocity functions from Tibet is crustal thickness. We consider the Tibetan crust north of the Himalaya to be an average of 65 km thick [e.g., Zhao et al., 2001], approximately 1.5 times the thickness of average continental crust. We apply the six crustal-thickening mechanisms of Figure 3 to the average velocity-depth function of Christensen and Mooney  to increase the thickness to 65 km. We ignore the increase in velocity due to higher pressure, and the decrease in velocity due to higher temperature, because the effects tend to cancel such that the net changes are small (approximately +0.01 km/s for granite moved from 5 km to 30 km depth, and approximately −0.10 km/s for mafic granulite moved from 30 km to 50 km depth [Christensen and Mooney, 1995]).
Figure 10. P wave velocity versus depth plots, a) showing velocity functions averaged over three areas (Lhasa Block, BNS, Qiangtang Block) from shallow velocity model of this paper plotted with published velocity-depth functions from 1Kola-Ojo and Meissner , 2crustal-scale model of Zhao et al. , and 3global average of Christensen and Mooney ; b) global-average velocity function thickened to represent 65-km-thick crust through each of six different mechanisms, with velocity functions from Lhasa and Qiangtang blocks for comparison (Qiangtang block velocity-depth function has been smoothly stretched to 65 km for ease of comparison).
Download figure to PowerPoint
 The thickening mechanisms tested represent the range of models proposed for the plateau, and give an estimate of the velocity structure that would be expected for each. They are not intended to stand individually as realistic models of plateau development, but rather to illustrate the velocity trends that would result from each mechanism. Wholesale crustal underthrusting (Figures 3b and 10b) would result in high-velocity lower crustal material at midcrustal depths, a pattern not seen in the velocity functions from Tibet. Injection models [e.g., Zhao and Morgan, 1987] predict normal thickness upper crust above pure shear thickened middle and lower crust, again requiring upper crustal velocities considerably higher than those found in Tibet (Figures 3c and 10b). In addition, these models imply the loss of Indian upper crustal material, a mechanism unsupported by geologic evidence.
 The fit of the global average velocity function to our measured Tibetan velocities (arithmetically averaged over ∼50-km swaths of the Vp model of Zhao et al. , shown in Figure 4c) is best when lower crust is lost from the system, as when the average crust is thickened by a pure-shear mechanism accompanied by removal of lower crust either via ductile flow or eclogitization (Figures 3e, 3g, and 10b). Fit is also reasonable for pure-shear thickening with (Figure 3f) or without (Figure 3d) lateral ductile flow of only the midcrust. In general, Tibetan crust shows somewhat greater proportions of upper (sialic) crust relative to lower (mafic-granulitic) crust when compared to the global average. Because the Qiangtang block upper crustal velocities are low to a great depth (∼30 km), it is unlikely that this crust is a result of tectonic alteration of previously average crust. The present velocity structure requires a lithologic difference from average crust, perhaps related to the large volume of melange material believed to make up much of the crust of northern Tibet [Yin and Harrison, 2000].
 Additional constraint can be placed on uplift mechanisms by the amount of convergence that has occurred between India and Asia since the ∼50 Ma collision. Yin and Harrison  use geological observations to estimate convergence of >1400 km, though 360 km of this convergence was accommodated in the Nan Shan, north of the plateau itself. Seafloor spreading indicates convergence of ∼1800 km in western Tibet and ∼2750 km in eastern Tibet [Dewey et al., 1989], though a portion of this convergence was likely accommodated north of the plateau. On the basis of an average north-south 1000-km width of the Tibetan Plateau, convergence of 1 to 1.5 times the width of the plateau must be explained. 150% pure-shear thickening of average crust (corresponding with 50% convergence) can produce 65-km-thick Tibetan crust, but any convergence >50% of the original width of the crust (∼500 km of convergence) necessitates removal of material in order to produce crust that is only 65 km thick. Even if the convergence was accommodated by underthrusting/thickening of blocks with rifted margins and an average thickness of only 32.5 km, convergence in excess of 1000 km still necessitates removal of material in order to leave crust that is only 65 km thick. Pure-shear doubling of normal thickness crust accompanied by loss of ∼50% of the initial volume of the lower half of the crust (see Figures 3e and 3g) can also accommodate only 100% convergence (1000 km). In order to achieve 1500 km of convergence while fitting our velocity profiles for the plateau, we must underthrust/thicken previously thinned crust and remove ∼50% of the lower crust. The loss of lower crust may occur partially as conversion from seismic crust to seismic mantle by eclogitization (Figure 3g) but is probably mostly due to lateral escape by ductile flow (Figures 3e, 3f, and 9c). We favor ductile flow over eclogitization because the presence of large quantities of eclogite in the upper mantle (a thickness of 15 to 20 km corresponds with loss of ∼50% of the lower crust) would inhibit plateau uplift and would be evidenced by high seismic velocities not compatible with the slow velocities and poor signal propagation found beneath northern Tibet. In addition, thermal models [Henry et al., 1997] suggest that while eclogitization may be common beneath some mountain belts, it is unlikely beneath the Himalayas due to cold underthrust material and fast convergence rates. The loss of crustal material via large-scale lateral motion along strike-slip faults [e.g., Molnar and Tapponier, 1975] can accommodate greater convergence, or can permit the original crust to have been of average 42-km thickness or even to have been tectonically thickened [e.g., Murphy et al., 1997] to >42 km prior to the India-Asia collision.
 While erosion has locally removed a great thickness of material near the plateau margins (e.g., 25 km in the southern Himalaya and 4 km in the Gangdese arc as discussed by Fielding ), the interior of the plateau has lost little material through erosion. Thus, as a percentage of the total plateau volume, the material removed by erosion is small and does not enter our discussion in a quantitative sense. Inclusion of erosion would permit our models to accommodate greater convergence or initially thicker crust, or to include a smaller volume of material lost via ductile flow.
3.5. Crustal Evolution in Tibet
 We propose that the Qiangtang and Lhasa blocks were originally parts of a single block (Figure 2a) with lower-than-average-velocity upper crust and average-velocity lower crust. During the early Mesozoic, a large quantity of Songpan-Ganzi material was subducted beneath the northern margin of what became the Qiantang block, and the Lhasa block rifted from the Qiangtang block (Figure 2b). The resulting rift closed during the Late Jurassic to Early Cretaceous with the formation of the BNS (Figure 2c). The Lhasa block was thickened by pure shear beginning in the Cretaceous, and was intruded by large bodies of granite in the Cretaceous and Tertiary (e.g., the Gangdese batholith and Bangoin granite) that increased the average velocity of the upper crust (Figure 2d). Lhasa-block lower crustal material was thrust beneath the Qiangtang block [Yin and Harrison, 2000], as the Lhasa block was underthrust by Indian lower crustal material (see Figures 2f, 9a, and 9b). This chronology fits well with the BNS-reactivation scenario that Roger et al.  propose in their interpretation of the Eocene magmatic belt that they identify across central Tibet. In order to achieve its current velocity structure, the Qiangtang block must have lost a large amount of lower crust relative to upper crust (assuming that lower crustal material was originally present) in order to accommodate underthrusting by Lhasa-block lower crust from the south and underthrusting by Songpan-Ganzi crust from the north (Figures 2f, 9a, and 9b). The loss of a lesser quantity of Lhasa-block lower crustal material is also suggested by the velocity structure as described above.
 Ductile lateral flow (Figure 9) [e.g., Clark and Royden, 2000] provides a reasonable and likely explanation for the loss of this lower crust, especially since the regionally smooth elevation of the plateau and results from gravity inversion suggest the presence of a ductile compensating layer at depth [Fielding et al., 1994; Jin et al., 1994]. In addition, shear-wave splitting results show a pronounced E-W fast direction in the Qiangtang block [Huang et al., 2000; McNamara et al., 1994; Herquel et al., 1995], indicating strong E-W directed creep. Because the continental margin to the east is essentially a free edge (the Pacific subduction zones), it can be assumed that any flow would be to the east, an assumption supported by the gentle topographic slopes of the southeastern and northeastern plateau margins relative to the steep margins to the north, south, and west [e.g., Clark and Royden, 2000]. Huang et al.  suggest that this flow takes place in the lower crust as well as in the upper mantle, on the basis of the spatially rapid transition from a null splitting result to large (up to 2.25 s) delay times at about 32°N. Lower crustal flow is also consistent with laminar reflectivity imaged at that depth by Zhao et al. . The extrusion of Tibetan lower crustal material to the east suggests that areas to the east of the Tibetan Plateau should be expected to show an abnormally great thickness of higher-velocity lower crustal material, a prediction that can easily be tested as new seismic data become available.
 An explanation for the low viscosity needed for ductile flow is provided by the fluid accumulations (likely melt) postulated to exist within the crust of eastern Tibet [e.g., Wei et al., 2001] as described previously. The less viscous (likely warmer) material under eastern Tibet would facilitate channel flow (in a region where lateral extrusion is already more likely owing to its proximity to the eastern plateau margin), while the more rigid crust beneath central Tibet would be more conducive to underthrusting. Thus we propose an E-W gradient (Figure 9c), with lower viscosity and greater rapidity of channel flow in the east and greater rigidity and volume of underthrusting in the west.
 The underthrusting of Lhasa-block lower crustal material beneath the Qiangtang block in the Late Cretaceous–early Tertiary formed the Qiangtang Anticlinorium, one aspect of major ongoing regional compression [Yin and Harrison, 2000]. Then, facilitated by warmer (lower viscosity) material and the proximal free surface (eastern plateau margin), the lower crustal material beneath the eastern plateau began ductile flow to the east, driven by the plateau topography. Thinner lower crust beneath the eastern Qiangtang (relative to the western Qiangtang block) provides an explanation for the eastward plunge of the Qiangtang Anticlinorium shown by Yin and Harrison  (Figures 1 and 9). Our model for plateau development (Figures 2 and 9) combines the advantages of many previously proposed mechanisms, and agrees with available data. In particular, our model achieves the final velocity structure determined from INDEPTH III seismic data while incorporating the Indian plate underthrusting and lower crustal flow strongly suggested by various data sets.