Initiation of Clockwise Rotation and Eastward Transport of Southeastern Tibet Inferred from Deflected Fault Traces and GPS Observations

Evidence from much of eastern Asia suggests that long after India's collision with Eurasia at ca. 50 Ma, the style of deformation changed since 10–15 Ma. Although the thick crust beneath the Tibetan Plateau resulted from Late Cretaceous and Cenozoic horizontal shortening both before and since India collided with Eurasia (e.g., Kapp and DeCelles, 2019), both fault plane solutions of earthquakes and GPS measurements show that normal faulting, E-W extension, and crustal thinning characterize the latest active deformation within the plateau (e.g., Armijo et al., 1986; Elliott et al., 2010; Gan et al., 2007; Ge et al., 2015; Li et al., 2020; Molnar and Chen, 1983; Molnar and Lyon-Caen, 1989; Molnar and Tapponnier, 1978; Ni and York, 1978; Shen et al., 2005; Wang et al., 2001; Zhang et al., 2004). Detailed studies of individual grabens throughout Tibet suggest that the widespread normal faulting began since ca. 8–15 Ma (e.g., Blisniuk et al., 2001; Edwards and Harrison, 1997; Garzione et al., 2003; Harrison et al., 1992, 1995; Kali et al., 2010; Lee et al., 2011; McCallister et al., 2014; Murphy et al., 2002; Pan and Kidd 1992; Sanchez et al., 2013; Saylor et al., 2010; Styron et al., 2013; Woodruff et al., 2013).

In the context of active large-scale deformation of eastern Asia, the E-W extension across the Tibetan Plateau is reflected by GPS data showing more rapid eastward movement of eastern than western Tibet relative to the stable Eurasia (see sub-map in Fig. 1). This extension is especially clear when velocities are plotted in a “Tibetan Plateau-fixed” reference frame obtained by minimizing the net movement of sites within the plateau (Gan et al., 2007) (Fig. 1). East of the highest part of the plateau, GPS velocities show a clockwise rotation and eastward transport of the southeastern plateau, around the Eastern Himalayan Syntaxis (e.g., Gan et al., 2007; Li et al., 2020; Wang et al., 2001; Zhang et al., 2004). The smoothly varying velocity field southeast of Tibet raises the question of when did that pattern of deformation begin and did it occur concurrently with the onset of normal faulting in central and southern Tibet.

To address the timing of clockwise rotation about the eastern syntaxis, we exploit current rates of crustal deformation based on GPS velocities, and we use fault traces as passive markers that are being distorted by that deformation. In particular, we rely on the warping of the Red River fault (Schoenbohm et al., 2006) and the displacement and shear of the Yalong-Yulong thrust belt, which forms the southwestward continuation of the Longmen Shan thrust belt (e.g., Burchfiel and Chen., 2012; Liu-Zeng et al., 2008; Wang et al., 1998; Xu et al., 2003) (Fig. 2). We assume that the current horizontal deformation rate approximates the long-term rate. We contend that both theory and observations justify such extrapolations.

In geodynamics, the equation of equilibrium, or Stokes's equation, balances gradients of components of stress with the body force due to gravity. Forces that induce deformation of parcels of lithosphere are matched exactly by viscous forces that resist deformation, for the acceleration, and therefore the total force at any point, is zero. Consequently, changes in rates of deformation can arise only because of changes in boundary conditions or by alterations of rheological parameters, like viscosity, that relate stress to strain rate. As rheological properties are affected most strongly by temperature, and characteristic time constants for diffusion of heat through the lithosphere are tens of millions of years, such changes ought to have little effect on rates of deformation on the million-year time scale. Changes in lateral boundary conditions can occur on shorter time scales, such as when a new plate boundary forms, but such events are rare. For convergence between India and Eurasia, which is directly related to our study area, the rate and direction have changed little since ca. 10 Ma or more (e.g., DeMets et al., 2020; Molnar and Stock, 2009).

Changes in bottom boundaries, such as by the removal of mantle lithosphere, seem also to be possible in the few-million-year time scale (e.g., Bird and Baumgardner, 1981; Houseman et al., 1981), but such events also seem to be rare. Evidence of such a process is sparse, and we are unaware of evidence suggesting that removal of mantle lithosphere has occurred more than once in the typical tens-of-millions-of-year lifetimes of major mountain belts. Thus, in regions that have undergone deformation for periods as long as tens of millions of years, changes in rates of large-scale deformation that exceed several tens of percent ought to be rare, and occur only once, or at most twice, in such periods.

At the largest scale, numerous studies of high-resolution histories of relative plate motion show only small, <20% changes in relative velocities over ~10 m.y. periods (e.g., DeMets and Merkouriev, 2016; DeMets et al., 2005, 2015a, 2015b, 2020; Iaffaldano, 2014; Iaffaldano et al., 2014; Merkouriev and DeMets, 2006, 2008, 2014). In a noteworthy exception, convergence between the Nazca and South America plates seems to have changed by ~30% over a few-million-year period (DeMets and Merkouriev, 2019; Garzione et al., 2006; Iaffaldano et al., 2006), when the Central Andes seem to have risen 2–3 km in a comparably short period, and when mantle lithosphere seems to have been removed from beneath the region (e.g., Garzione et al., 2006, 2008; Gubbels et al., 1993; Kennan et al., 1997; Schildgen et al., 2007). The assumption of constant rates of relative plate motion, in fact, enabled Wilson (1993) to test refinements of the geomagnetic time scale, and Krijgsman et al. (1999) to refine that time scale further.

We recognize that changes in rates of deformation must necessarily occur to build mountain ranges and high plateaus, but we assume that in periods spanning tens of millions of years, horizontal deformation rates, such as the rotational deformation about the syntaxis, can occur at essentially constant rates over such periods.

The Ailao Shan shear zone, a long-standing, deeply exhumed fault zone, underwent several hundred kilometers of Oligo-Miocene left-lateral shear (Harrison et al., 1996; Leloup et al., 2001; Searle et al., 2010; Tapponnier et al., 1986; Wang and Burchfiel, 1997). Since 10–14 Ma, however, the Red River fault, which follows approximately the northeast margin of the Ailao Shan shear zone (Fig. 2), has undergone right-lateral displacement of 25 km (Replumaz et al., 2001), ~40 km (Schoenbohm et al., 2006), or 20–54 km (Wang et al., 1998). This reversal in regional deformation implies other changes in relative motions of surrounding domains. Important for us here, the Red River fault is not straight, but shows a marked left-lateral deflection of ~60 km across a region 100–200 km in width in the middle segment where the Xianshuihe-Xiaojiang fault approaches it (e.g., Schoenbohm et al., 2006; Wang et al., 1998) (Fig. 2). So, although active as a right-lateral strike-slip fault, the trace of Red River fault serves as a passive marker, deflected by shear of an internally deforming region around the Eastern Himalayan Syntaxis (Schoenbohm et al., 2006).

The Longmen Shan thrust belt lies within the Longmen Shan, the mountain range that separates the eastern Tibetan Plateau from the Sichuan Basin. We do not refer to the active thrust fault at that margin (e.g., Hubbard and Shaw, 2009; Zhang, 2013), but rather to a thrust belt that apparently began to develop in Late Triassic, and that accommodated significant crustal shortening during the Indosinian Orogeny in Mesozoic time (e.g., Burchfiel et al., 1995; Chen and Wilson 1996; Li et al., 2003; Searle et al., 2016; Wallis et al., 2003; Xue et al., 2017; Yan et al., 2011). Most agree with Burchfiel et al. (1995) that much of the crustal thickening in this region occurred in Mesozoic time. The Late Triassic Longmen Shan thrust belt, including deformed and metamorphosed surrounding rock, has been cut and offset ~60 km by the NNW-trending Xianshuihe fault (Burchfiel et al., 1995; Wang et al., 1998; Wang et al., 2009) (Fig. 2). As shown by Liu-Zeng et al. (2008) and Xu et al. (2003), the southwestward continuation of the Longmen Shan thrust belt, the Yalong-Yulong thrust belt, has been displaced ~100 km by southeastward crustal extrusion, and forms an arcuate trace with a length of ~200 km (Fig. 2).

As both the Red River fault and the Yalong-Yulong-Longmen Shan thrust belt existed before the late Miocene, their fault traces form regional strain markers of the southeastern plateau margin that have been deflected. Their deflections record shear that has accumulated by clockwise rotation that initiated since those features formed. We can estimate the date of the initiation of clockwise rotation by retro-deforming the deflected fault traces using current deformation rates from GPS velocities, and assuming that their original shapes of the traces were overall straight.

About 85% of the GPS velocities in Figure 1 are from published solutions of two Chinese national scientific projects, Crustal Movement Observation Network of China (CMONOC-I) and Tectonic and Environmental Observation Network of Mainland China (CMONOC-II) (Li et al., 2012). The detailed GPS observation methods and data processing strategies were introduced by Gan et al. (2007) and Liang et al. (2013). In addition to the GPS velocity data set of 423 stations from CMONOC, we merged some other published GPS velocities (Table 1) to enhance the density and coverage of GPS stations.

Although the CMONOC velocities and those of other data sets commonly are given in Eurasia-fixed reference frames, those frames differ slightly from each other. As each of the other data sets shared some stations with the CMONOC data set, we used stations common to the CMONOC data set as “links” to transform all the other velocities into the same reference frame as that for CMONOC by using rigid-body rotations with appropriate angular velocities. After the reference frame transformations, the maximum differences of north and east components of the velocities for the same stations in different data sets are 2.1 and 1.6 mm/yr, respectively; these values are within two standard deviations of the velocity components. The final velocities of the common stations are the weighted average of the values from all the data sets in the same Eurasia-fixed reference frame.

Two major earthquakes, 2008 Wenchuan Mw 7.9 (e.g., Zhang, 2013; Zhang et al., 2010) and 2013 Lu Shan Ms 7.0 earthquakes (Xu et al., 2013), occurred in the area of this study during the GPS observation span. In order to avoid the contamination by transient displacements associated with these two events, we excluded the GPS data observed after the earthquakes for those stations located within ~500 km of the epicentral areas of those earthquakes.

The combined velocities of 501 GPS stations in the Eurasia-fixed reference frame demonstrate the well-known crustal movement pattern (see sub-map in Fig. 1), but to highlight the clockwise rotation around the southeastern plateau margin relative to its stable neighbor block to the east, we transformed the velocities into a reference frame fixed to the South China Block, by minimizing the root mean square velocities of 83 stations located in that block (Figs. 3A). Not only is the rotation about the syntaxis clear, but it also is confined to the region west of the South China Block bounded by the Xianshuihe-Xiaojiang fault zone on the north and east, and therefore includes a major component of left-lateral shear along that boundary (Figs. 2 and 3A).

The surface traces of the Red River fault and Yalong-Yulong-Longmen Shan thrust belt (hereafter “fault traces”) are deflected left-laterally by the clockwise rotation around the eastern syntaxis. Moreover, marked gradients in GPS velocities characterize the segments of the fault traces where they are deflected. In order to estimate the initiation age of the clockwise rotation and the deflections of the fault traces using GPS velocity field, we adopted the following strategy:

• (1) By extending the undeflected segments of the fault traces, we defined two directions, N138°E and N51°E, which are approximately parallel to the traces in the regions where present-day rotation is modest (Figs. 3A).

• (2) The exact positions of the initial fault traces are unknown, but they can be represented approximately by linear extensions of the undeflected segments. By representing the current fault traces as a series of points, we further defined a series of expected corresponding points on the extensions of the undeflected segments (Figs. 3B). Using the positions of these points, we first obtained appropriate velocities of these points (shown as the blue vectors in Figs. 3B) by interpolation from the GPS velocity field using the interpolation algorithm of splines in tension (τ = 0.95) (Wessel and Bercovici, 1998; Gan et al., 2007). We then averaged the velocities of each pair of corresponding points. The averaged velocities (shown as the red vectors in Figs. 3B) were appointed to represent the long-term velocities of the fault traces.

• (3) We assume that all points on the original fault traces have moved with current average velocities since the traces were deflected. We assume that the initial shapes of the fault traces were straight, and we let points on the current fault traces move backward (Figs. 3C) until they recover a shape that is closest to a straight line (Fig. 4).

With the above strategy, we estimated similar ages of initiation of 9.6 Ma and 10.7 Ma from the Red River fault and Yalong-Yulong-Longmen Shan thrust belt, respectively, and therefore 10.1 Ma.

As we know, the overall shape of the initial fault trace need not be straight, although in theory and practice, it is more common for thrust and strike-slip faults to evolve into overall linear features. Hence, the estimation based on the ideal linear assumption could include deviations, which sometimes are not insignificant. We estimated the 1σ uncertainty of the initiation age with two approaches. First, the two independently deduced initiation ages were quite consistent, with a difference of ~1.1 m.y., which provided us with some clues to assess the level of uncertainty. Second, we empirically assumed that the best fitting line is not significantly different from the other good fitting lines when differences in correlation coefficients between them are less than ~1% of the correlation coefficient difference between the current and original fault traces in the inversion by least-squares linear fitting (see sub-maps in Fig. 4). This approach yielded 1σ uncertainties of ±1.3 Ma and ±1.2 Ma for the Red River fault and Yalong-Yulong-Longmen Shan thrust belt, respectively. Based on the above empirical assessment, we set ±1.5 Ma as a conservative estimate of the 1σ uncertainty, and therefore 10.1 ±1.5 Ma for the estimate of initiation age. Obviously, readers could multiply this uncertainty by a factor greater than one, if they doubted the approach. It should also be noted that the estimate in our method is affected mainly by the overall linearity of the fault traces, and not the local twists and turns in the traces. Figure 4 shows that the inferred original fault traces locally twist and turn and do not form ideal straight lines.

Numerous paleomagnetic studies show clockwise declination anomalies of rock from eastern Tibet, including the region northeast of the syntaxis to western Yunnan, southeast of Tibet. Thus, they indicate clockwise rotation of such rock. Two processes have contributed to these rotations. First, the present-day GPS velocity field (Figs. 3A) includes rotation about a vertical axis. The current pattern of flow around the eastern syntaxis includes clockwise rotation, but the amount is small. For a maximum rate of ~10 mm/yr about the syntaxis and ~500 km from it, displacement of ~100 km since 10 Ma, yields a rotation of only arctan(100/500) ≈ 11°. Second, and more importantly, as India penetrated into Eurasia, both it and the region to its north moved northward with respect to South China, and induced right-lateral shear and clockwise rotation of eastern Tibet (e.g., Cobbold and Davy, 1988; Davy and Cobbold, 1988; England and Molnar, 1990; England and Houseman, 1986). Blocks embedded in a zone of simple shear of width, L, and across which opposite sides are displaced Δu should undergo rotation, R, such that arctan (Δu/2L) < R < arctan (Δu/L), that is dependent on the shape of the object and boundary conditions (Lamb, 1987; McKenzie and Jackson, 1983). With a width L ≈ 500 km for eastern Tibet, and values of Δu between 200 km in the north and 1000 km in the south, we might expect R ~10°–25° in the north and R ~45°–65° in the south. Although the contributions of both processes to clockwise rotation are apparent, separating them is tricky, and the role of the first of these processes cannot be assessed well.

Despite scatter, amounts of clockwise rotation inferred from paleomagnetic declinations increase southward, as expected. In the northern region near latitudes of ~35°, where though present-day left-lateral slip on east-west planes and therefore counterclockwise rotation characterizes current deformation (Duvall and Clark, 2010), total amounts of clockwise rotation are ~15°–25° (Dupont-Nivet et al., 2004, 2008; Fang et al., 2003; Liu et al., 2010; Wang et al., 2011; Yan et al., 2006). Near the latitude of the syntaxis, most estimates of clockwise rotation are 25°–50° (e.g., Huang et al., 1992; Kornfeld et al., 2014; Otofuji et al., 1990; Todrani et al., 2020; Zhang et al., 2020). Farther south in Yunnan, yet larger amounts of clockwise rotation exceeding 50° (Chen et al., 1995; Huang and Opdyke, 1993; Li et al., 2018; Tong et al., 2016) and reaching 90° (Sato et al., 2001) have been reported, but many sites show smaller amounts of rotation. As important as the southward increase in amounts of rotation is the scatter in such data. In a particularly thorough study of the region just south of the Litang fault (Fig. 2), Todrani et al. (2020) reported a range from 26° ± 12° of counterclockwise rotation to as much as 37° ± 8° clockwise rotation, with three sites with as much as 60° to 86° clockwise rotations. Similarly, although large rotations have been recorded from parts of Yunnan, smaller amounts have also been measured: 36.3° ± 13.6° (Sato et al., 1999), 26° ± 17° (Funahara et al., 1992), 25° ± 16° (Funahara et al., 1993), 12.1° ± 10.0° (Otofuji et al., 1998), ~4° ± 8° (Yoshioka et al., 2003), essentially no rotation at a site next to the Red River fault (Huang and Opdyke, 1993) and even counterclockwise 8° ± 3° (Li et al., 2015) albeit from a site near the Xiaojiang fault. These rotations are given relative to Eurasia, but a better reference frame might be South China. Although paleomagnetic studies of its western edge show essentially no rotation (e.g., Huang and Opdyke, 1992; Otofuji et al., 1990), several sites surrounding the Sichuan Basin suggest a small clockwise rotation of 5–10° (Tong et al., 2020). Thus, values given above might be reduced by that amount.

As noted above, the current pattern of flow around the syntaxis contributes to the clockwise rotation but the estimated amount of ~11° is small compared with paleomagnetic declination anomalies. Moreover, most of the largest values (50°–90°) come from south of the region currently undergoing that rotation around the syntaxis. Otofuji et al. (2010) suggested that these regions with large paleomagnetic declinations underwent rotation when they lay near the eastern syntaxis, and subsequently were displaced southeastward, where GPS data show that only slow rotation occurs today. In a review of paleomagnetic data from Yunnan and surrounding regions, Tong et al. (2021) concluded that rotation began since ca. 25.7 ± 2.5 Ma at a higher rate than occurs today, despite scatter in amounts of rotation. At that time, the syntaxis would have lain ~1000–1500 km south of its current position with respect to South China, and hence west of this region where large rotations have accumulated. We conclude that the paleomagnetic declinations, despite scatter, have resulted largely from right-lateral shear of eastern Tibet and not from the ongoing flow around the syntaxis.

The estimated initiation age, 10.1 ± 1.5 Ma, of clockwise rotation of the southeastern plateau margin is approximately synchronous with a suite of tectonic phenomena, both near and far, within the scope of the wide Eurasia/Indian collision zone (Fig. 5), which collectively indicate that a major tectonic event happened since the middle Miocene.

The region undergoing clockwise rotation around the Eastern Himalayan Syntaxis is bounded on the north by the Ganzi-Yushu-Xianshuihe fault, which crops out as a ductile shear zone that consists of mylonite with mylonitized and migmatitized rock (Xu et al., 2007; Wang et al., 1998) (Fig. 2). The Xianshuihe segment offsets a huge Precambrian metamorphic complex left-laterally for 90–100 km. Also, a granodiorite body, with a length parallel to the Xianshuihe fault for 80–90 km and a width of only 7–20 km, was emplaced along the fault near the offset of the belt (Xu et al., 2007). From micro-structure and geochemical studies, Xu et al. (2007) suggested that this linear granodiorite was emplaced during the initiation of left-lateral displacement on the Xianshuihe fault, and thus, the age of emplacement dates the onset of slip on the fault. Roger et al. (1995) obtained U-Th and Rb-Sr ages of 10–12 Ma, and Xu et al. (2007) reported a Rb-Sr age of 9.9–11.6 Ma, which are consistent with the inference that clockwise rotation began at 10.1 ± 1.5 Ma, though Searle et al. (2016) suggested that slip might have begun at only 5 Ma.

Along the whole Ganzi-Yushu-Xianshuihe fault zone, the Ganzi-Yushu segment has the simplest geometrical structure (Fig. 2). Field investigations show that this segment has a total left-lateral offset of 75–80 km based on a dislocated Triassic granitic pluton (Wang et al., 1998). Wang et al. (2009) claimed that fission track ages decreased toward the fault zone, and they assumed that the oldest (12.8 ± 1.4 Ma) marked the date when the fault became active. In fact, by relying on GPS velocities, inferred late Quaternary offsets of moraines and other features, and a total offset of precisely dated igneous rock, Chevalier et al. (2017) inferred that the slip rate on the Ganzi-Xianshuihe fault (Fig. 1) has been constant since slip began at, or shortly after, ca. 12.6 Ma. Thus, one could derive a long-term average slip rate of 5.9–6.3 mm/yr based on the date and total offset. Multiple estimates of short-term average slip rates for late Quaternary or Holocene periods from the classic methods of dating offset landforms like terraces and alluvial fans have yielded a slip rate of ~7 mm/yr (Chevalier et al., 2017; Li et al., 1995; Zhou et al., 1996). In addition, GPS observations give a present-day slip rate of 6.6 ± 1.5 mm/yr (Wang et al., 2013). If the slip rate had been constant at 6.6 or 7.0 mm/yr, then 75–80 km of offset would imply an onset date of 10.7–12.1 Ma for the Ganzi segment.

Thus, these onset dates of the Ganzi-Yushu-Xianshuihe fault are indistinguishable from 10.1 ± 1.5 Ma, and the fault became active synchronously with the eastward transport and clockwise rotation.

The Red River fault is a first-order tectonic structure in the southeast margin of the Tibetan Plateau; it separates the South China Block to the northeast and the Indochina block to the southwest (Fig. 2). Numerous studies determined that the Red River fault experienced a reversal of slip from left- to right-lateral sometime between 16 Ma and 5.5 Ma (Allen et al., 1984; Wang et al., 1998; Replumaz et al., 2001; Schoenbohm et al., 2006; Zhu et al., 2009). Recently, based on the seismic analysis and stratigraphic information from confidential exploration wells in the Gulf of Tonkin, where the offshore segment of the Red River fault lies, Fyhn and Phach (2015) suggested that the reversal of slip may have commenced at ca. 10–8 Ma. In addition, Wang et al. (2016), using new apatite (U-Th)/He data together with existing thermochronological data, backed by stratigraphic, structural, and geomorphologic observations, inferred that the Ailao Shan-Red River shear zone experienced a phase of rapid cooling and exhumation starting near 14–10 Ma, and then, since 10 Ma a reversal from left-lateral slip on the Ailao Shan shear zone to right-lateral slip on the Red River fault. Again, these tectonic phenomena coincide with our estimate of the initiation of clockwise rotation of the southeastern Tibet.

In eastern Tibet east of the eastern syntaxis of the Himalaya, age-elevation transects of low-temperature thermochronometers suggest slow cooling between 100 Ma and 10–20 Ma and a change to rapid river incision between ca. 8 Ma and 13 Ma (Clark et al., 2005; Godard et al., 2009, 2010; Ouimet et al., 2010; Zhang et al., 2016). Thus, these data can be interpreted to indicate surface uplift, presumably in response to crustal thickening, concurrent with the onset of rotation in this region.

Since the initial recognition of active N-S–trending rifts in southern Tibet (Armijo et al., 1986; Molnar and Tapponnier, 1978; Ni and York, 1978; Tapponnier and Molnar, 1977), numerous studies have constrained the timing of the rift-related normal faulting and E-W crustal extension, as the onset of extension was argued to represent the time when the plateau began to collapse after it reached its highest elevation (England and Houseman 1989; Harrison et al., 1992; Molnar et al., 1993). Reported dates of onsets of normal faulting and crustal thinning range from 15 Ma to 8 Ma (e.g., Blisniuk et al., 2001; Dewane et al., 2006; Edwards and Harrison, 1997; Garzione et al., 2003; Kali et al., 2010; Lee et al., 2011; McCallister et al., 2014; Murphy et al., 2002; Sanchez et al., 2013; Saylor et al., 2010; Thiede et al., 2006; Wu et al., 2008) (Fig. 5). The relatively wide spread of ages, and the apparent absence of strict simultaneity with the onset of rotation about the eastern syntaxis may be due in part to inaccuracies in ages. More important, however, if development of normal faulting and crustal thinning did result from removal of mantle lithosphere, we would not expect such a process to have occurred at the same time across the whole of Tibet. Moreover, only after the entire plateau was undergoing extension and crustal thinning, would the effects of its increase in gravitational potential energy affect surrounding regions.

In northeastern Tibet, early Cenozoic deformation was accommodated primarily by slip on WNW-trending thrust faults, which indicates that during the early stages of orogenesis, the orientation of thrust faulting was approximately parallel to the orientation of plate convergence. Nevertheless, some studies reveal a middle-late Miocene change in the kinematic style of plateau growth, from long-standing NNE-SSW contraction that mimicked the orientation of plate convergence to the initiation of new structures accommodating E-W contraction (e.g., Lease et al., 2011; Zheng et al., 2006). Lease et al. (2011) reported apatite (U-Th)/He and apatite fission track ages from Laji-Jishi Shan in the northeastern Tibetan Plateau (Fig. 5), from which they inferred that accelerated growth of the WNW-trending eastern Laji Shan began ca. 22 Ma. Rapid growth of the adjacent, but north-trending Jishi Shan, however, did not commence until ca. 13 Ma. Thus, the primary orientation of contraction in the composite Laji-Jishi Shan changed >45° with the initiation of E-W shortening of the Jishi Shan. Furthermore, paleoclimate records from sediment in Neogene basins on either side of the Jishi Shan show the development of a rain shadow due to surface uplift of the range by ca. 11 Ma, but not before ca. 16 Ma (Hough et al., 2011). Zheng et al. (2006), using apatite fission track thermochronology, found that late Cenozoic rapid cooling occurred at ca. 8 Ma in the Liupan Shan region (Fig. 5), and the north-trending thrust faults there began to accommodate lateral motion at that time. This inception of E-W contraction is only slightly later than the eastward transport and clockwise rotation in southeastern Tibet, and reflects a change in the kinematic style of plateau uplift and lateral growth. Lateral, outward growth of the margin of the plateau should begin later than crustal shortening that is now within the high terrain.

North of the Tibetan Plateau, the current rate of convergence across the western Tien Shan is 20 ± 2 mm/yr (Zubovich et al., 2010) and the estimated shortening across the belt is ~200 km (Avouac et al., 1993). Assuming a 20% error in the amount of shortening and a constant convergence rate suggests an initiation of crustal shortening at 10 ± 2 Ma (e.g., Abdrakhmatov et al., 1996; Zubovich et al., 2010). Numerous studies of thermochronology, which date the onset of rapid cooling (Bullen et al., 2001, 2003; Macaulay et al., 2013), and of magnetostratigraphy, which date abrupt increases in sedimentation within a few millions years of 10 Ma (Charreau et al., 2005, 2008, 2009; Huang et al., 2006, 2010; Ji et al., 2008; Qiao et al., 2016; Sun et al., 2004, 2009), concur with an abrupt increase in the erosion rate of the Tien Shan, presumably associated with rapid surface uplift, at ca. 10 Ma, though abundant evidence suggests that some relief existed well before ca. 10 Ma.

where g is gravity, ρ_c and ρ_m are the densities of crust and mantle, respectively, L is the thickness of the thin viscous sheet, and therefore the lithosphere, and V is a characteristic horizontal component of velocity, like the rate at which the sheet is indented (England and McKenzie, 1982). One may think of Ar as scaling resistance to crustal thickening imposed by gravity to resistance by viscous processes; if Ar→0, gravity offers little resistance to crustal thickening, and the sheet is relatively strong, whereas when Ar is large, crust cannot thicken much because it tends to flow away from regions of thickened crust. The Argand number is sometimes called the feta-brie number, for feta cheese corresponds to small Ar, and brie to large Ar.

The reliance on just two dimensionless numbers allows their ranges of values to be explored, and for their roles in large scale deformation to be revealed and assessed (e.g., England and Houseman, 1986; Houseman and England, 1986). Comparisons of large-scale deformation from essentially all regions of large-scale deformation show patterns of deformation that can be matched by treating the continental lithosphere as a thin viscous sheet: eastern Asia (England and Houseman, 1986; England and Molnar, 1997; Flesch et al., 2001; Holt and Haines, 1993; Holt et al., 1991); Iran (Walters et al., 2017); Anatolia (England et al., 2016), and western North America (Bahadori et al., 2018; Flesch et al., 2000; Reitman and Molnar, 2021; Whitehouse et al., 2005). Moreover, strain-rate fields measured at the surface using GPS match those implied by seismic anisotropy in the underlying mantle, which suggests vertically coherent deformation through the lithosphere (e.g., Chang et al., 2015; Davis et al., 1997; Flesch et al., 2005; Holt, 2000).

The simplicity of the thin viscous sheet as a model for continental deformation allows it to reveal how changes in basic conditions or boundary conditions affect deformation. For example, as England and Houseman (1989) argued, a decrease in V, such as in the rate of indentation of a rigid object, like India, into a thin viscous sheet like the rest of Eurasia, would manifest itself as an increase in Ar, and hence could affect the pattern of deformation. Because of 1/n^th power of V in the denominator of Ar, however, a major effect in the deformation field requires a huge change in V. A change in the viscosity coefficient, B, should have a bigger effect than comparable change in V, but insofar as diffusion of heat has the primary effect on viscosity coefficient, such changes will require tens of millions of years to have an effect. Moreover, a concurrent change in B over an area the size of eastern Asia is implausible.

Both n and Ar are material parameters of the thin viscous sheet, and therefore changes in them, such as by changing the thickness of the lithosphere, L, or its average viscosity coefficient, B, strictly apply to the entire sheet, so that, in general, the equations solved do not depend separately on the ingredients of Ar in Equation (2). Lateral variations in Ar, however, offer the possibility of including more realistic models with the inevitable trade-offs with non-uniqueness and potential loss of simple understanding (e.g., England and Molnar, 1997; Flesch et al., 2001). As England and Houseman (1989) showed, the simplest process that could affect the pattern of deformation in a thin viscous sheet like the eastern Asian lithosphere and give rise to crustal thinning within the Tibetan Plateau would be to raise the surface of the plateau. Doing so would give the elevated region additional gravitational potential energy, which England and Houseman (1989) implemented by locally increasing Ar. They showed that other possible processes, like weakening the part of the sheet occupied by the plateau would have an opposite effect to what is observed. Where weaker, the sheet would shorten and thicken, rather than extend. They argued, and in our view convincingly, that the only way to induce a switch from thickening to thinning would be to raise the surface, and give the plateau additional gravitational potential energy, and greater outward force-per unit length from the plateau. We recognize that others have offered different explanations for the flow of material around the syntaxis, including channel flow within the crust (e.g., Clark and Royden, 2000; Royden et al., 1997; Schoenbohm et al., 2006) and gravity currents (Copley and McKenzie, 2007), and that some question the removal of mantle lithosphere beneath Tibet (e.g., McKenzie and Priestley, 2008). Nevertheless, we consider removal of mantle lithosphere to be the simplest process that can accomplish this increase in gravitational potential energy and the change in tectonic style.

The traces of two major fault systems in southwest China, the Red River fault and Yalong-Yulong-Longmen Shan thrust belt, are currently being distorted by crustal movements in the area southeast of the Tibetan Plateau. GPS measurements show that region currently undergoes widespread, apparently continuous deformation dominated by virtually concentric clockwise rotation about the Eastern Himalaya Syntaxis and associated left-lateral shear of the region (Figs. 1 and 3). By treating current rates of crustal deformation as approximating long-term average rates, we use that velocity field to restore offsets of these fault traces, and suggest that such rotation and shear initiated at 10.1 ± 1.5 Ma.

Significant change in styles and rates of deformation elsewhere in Tibet and its surroundings occurred at approximately the same time. For example, slip on the Ganzi-Yushu-Xianshuihe fault in eastern Tibet apparently began at that time. Slip on the Red River fault seems to have reversed, from left- to right-lateral, at that time. Several rivers began to incise rapidly into southeastern Tibet between 9 Ma and 13 Ma, presumably in response to a rapid increase in mean elevations of that region. Thus, the eastward transport and clockwise rotation seems to be one of several changes in style of deformation across the region.

Approximately concurrently with these changes in eastern Tibet, between 8 Ma and 15 Ma N-S–trending rifts within the Tibetan Plateau, on which E-W extension occurs by normal faulting, developed (Fig. 5). The change from roughly N-S crustal shortening and thickening within the plateau to extension and crustal thinning is arguably the biggest change in style of deformation of Asia since India collided with Eurasia. Also, between 8 Ma and 15 Ma, the orientation of contraction in northeastern Tibet changed from NE-SW—approximately parallel to convergence between India and Eurasia—to E-W, which is perpendicular to the local orientation of the margin of the plateau. Moreover, erosion of the Tien Shan north of Tibet accelerated near 10 Ma, again presumably in response to accelerations of N-S crustal shorting and crustal thickening and the resulting surface uplift (Fig. 5). All of these changes can be seen as a logical response to an abrupt rise of the interior of the Tibetan Plateau at 10–15 Ma. Following England and Houseman's (1989) logic, we concur that mantle lithosphere was removed from beneath the plateau, the surface rose, and the plateau gained gravitational potential energy per unit area. Hence, the elevated plateau applied an increased horizontal compressive force per unit length to its surroundings. The increased gravitational potential energy of that newly elevated terrain then powered outward growth of it, including extension and crustal thinning within the elevated region and enhanced crustal thickening of the surrounding regions, including regions as far away as the Tien Shan. The eastward movement of eastern Tibet and the clockwise rotation of that material around the Eastern Himalayan Syntaxis would be local manifestations of this larger geodynamic event. Of course, the surface across a region more than 1000 km in dimension is not likely to have risen simultaneous, let alone instantaneously. Thus, the difference between surrounding regions undergoing a change in tectonic style and rate at ca. 10 Ma, and normal faulting scattered across Tibet starting between ca. 16 Ma and 8 Ma ought not to be a surprise. Convective removal of some or all of Tibet's mantle lithosphere offers the simplest underlying mechanism not just for causing normal faulting within the Tibetan Plateau, but also for affecting the surrounding areas as described above.

The GPS velocity data set and the solution from this study, which are listed in Tables S2 and S8 of the supporting information¹, are under way to archive in the Data Sharing Infrastructure of Earth System Science (http://www.geodata.cn/) for free public access. Now we temporarily upload a copy of our data as supporting information for review purposes.

We thank all the Chinese colleagues who worked in the Crustal Movement Observation Network of China project and collected the GPS data used in this study. We are grateful to Roger Bilham, Huiping Zhang, and two anonymous reviewers for their thoughtful comments, which resulted in considerable improvement of this manuscript. The main figures in the manuscript were drawn with the help of the Generic Mapping Tools. This work was supported by the National Natural Science Foundation of China (41490615), the Special Research Project for Earthquake Science of China (2015419024), and the National Key Research and Development Plan of China (2018YFC1503304).