Spatially variable extension in southern Tibet based on GPS measurements



[1] We use Global Positioning System (GPS) data and kinematic block models to study the present-day deformation of southern Tibet. GPS data from 33 sites in southern Tibet and Nepal surveyed between 1991 and 2000 reveal 13 ± 2 mm/yr of N110°E extension between Lhasa and Shiquanhe (80°E–91°E), of which 9.7 ± 3.0 mm/yr represents permanent extension of the Tibetan crust. The remaining ∼3 mm/yr results from elastic deformation from the locked, curving Main Himalayan Thrust fault. This extension is spatially nonuniform. One half to two thirds of the permanent extension is concentrated across the Yadong-Gulu rift, with an opening rate of 6.5 ± 1.5 mm/yr; most of the remainder occurs on or near the Thakkola graben, with little extension across the rifts between them. Differential velocities of sites north and south of the Yarlung-Zangbo suture in western Tibet imply that the suture or an adjacent structure may be active as a strike-slip fault. A numerical model suggests that right-lateral strike slip on the Yarlung-Zangbo suture may extend from the Karakorum Fault Zone in the west at least to the Yadong-Gulu rift in the east with ∼3 mm/yr slip rate, accommodating part of the eastward extrusion of Tibet. The convergence directions inferred from GPS are consistent with slip vectors of earthquakes; however, the rate of slip of India beneath Tibet along the Himalaya is lower than those previously estimated. We estimate a slip rate of 12.4 ± 0.4 mm/yr between longitudes 83°E and 88°E and 17 ± 1 and 19 ± 1 mm/yr in the western and eastern Himalaya, respectively. The variable slip rate correlates with the variable extension rate in southern Tibet, and we suggest that it results from variation in the deformation rate of the overriding Tibetan crust. We infer that the slower convergence rate in the central Himalaya is significant.

1. Introduction

[2] The Tibetan Plateau lies in the collision zone between the Indian and Eurasian plates (Figure 1) and is a complex region with broadly distributed seismicity and many active faults [Molnar and Lyon-Caen, 1989; Rothery and Drury, 1984; Armijo et al., 1986, 1989]. The motion of India relative to Eurasia is now constrained by geologic data (NUVEL-1A) [DeMets et al., 1990, 1994] and geodetic data [e.g., Freymueller et al., 1996; Larson et al., 1999; Paul et al., 2001; Sella et al., 2002], but Tibet is also moving significantly relative to Eurasia and undergoes significant internal deformation [Armijo et al., 1986, 1989; Molnar and Tapponnier, 1975; Harrison et al., 1992; Molnar, 1984, 1990; Molnar and Chen, 1983; Rothery and Drury, 1984; Molnar and Lyon-Caen, 1989; Yin et al., 1999; Chen et al., 2004].

Figure 1.

Shaded relief map showing the GPS sites used in this study. Solid circles indicate campaign sites; solid triangles indicate continuous sites. Earthquake focal mechanisms are from Harvard CMT catalogue (1976–2003, Ms > 5.0) and Molnar and Lyon-Caen [1989] (1962–1986). Stars indicate historical earthquakes with no focal mechanisms available (1700–1976, Ms > 7.0, from NEIC catalogue).

[3] According to the NUVEL-1A model [DeMets et al., 1990, 1994], the NNE movement of India toward Asia is approximately 50 mm/yr. More recently, space geodetic data show a lower rate of 36–37 mm/yr between India and Eurasia [e.g., Sella et al., 2002; Wang et al., 2001]. One third to one half of the India-Eurasia convergence occurs across the Himalaya [Bilham et al., 1997], and the rest is distributed over a region extending north ∼1500 km to Siberia. It has long been known that convergence across the Himalaya is accompanied by arc-parallel extension within southernmost Tibet [e.g., Armijo et al., 1986; Molnar and Lyon-Caen, 1989], but the rate of extension and its variability along the arc have not been measured precisely.

[4] In this paper, we use Global Positioning System (GPS) data to study the convergence across the Himalaya and the internal deformation of southern Tibet. Eighteen GPS sites in southern Tibet surveyed between 1991 and 2001 [Wang et al., 2001] show that the rate of roughly east-west (N110°E) extension across southern Tibet is spatially variable, and most of the extension is concentrated in two places, the Yadong-Gulu rift and the Thakkola graben. GPS sites in southwestern Tibet show strong evidence for right-lateral strike-slip motion along the Yarlung-Zangbo suture (YZS). We present a model to explain these observations and highlight the relationship between the convergence across the Himalaya and deformation within Tibet.

2. Tectonic Background

[5] Field investigations [Tapponnier et al., 1981b; Armijo et al., 1986, 1989; Burchfiel et al., 1991; Mercier et al., 1987], Landsat image interpretation [Molnar and Tapponnier, 1978; Rothery and Drury, 1984] and earthquake fault plane solutions [Molnar and Tapponnier, 1978; Molnar and Chen, 1982; Molnar and Lyon-Caen, 1989] demonstrate that north-south trending normal faulting dominates the present tectonics of southern Tibet. Seven rift systems with a regular pattern, a few hundred kilometers long and about 200 km apart, extend north from the Himalaya and can be identified mostly south of latitude 32°N [Armijo et al., 1986]. On average, the normal faults trend N5.9°E ± 7.5° and dip between 45° and 60°. These rift systems cut across the YZS, which broadly coincides at the surface with the Yarlung-Zangbo river. Only two of the rifts, the Yadong-Gulu rift at ∼90°E and the Thakkola graben at ∼83°E, cut across the Himalayan range. Geologic reconstructions suggest that these two rifts have undergone more total extension than the others [see Replumaz and Tapponnier, 2003 and references therein].

[6] A broad zone of en échelon right-lateral strike-slip faults at about latitude 32°N, termed the Karakorum-Jiali Fault Zone (KJFZ) by Armijo et al. [1986], separates the well-developed normal faulting in southern Tibet from more diffuse extension in central and northern Tibet. These faults trend 120–140°. Armijo et al. [1989] proposed that the Tibetan Plateau north of the KJFZ and south of the left-lateral Altyn Tagh fault is relatively undeforming and is extruding rapidly to the east. In contrast, based on interpretation of lineaments in Landsat images, Rothery and Drury [1984] suggested that E–W extension, strike-slip, and normal faulting extend throughout the Tibetan Plateau, and that the N–S shortening within the plateau could be about one third of the total convergence between India and Eurasia. GPS results from sites across the Tibetan Plateau show broadly distributed contraction across the plateau, oriented ∼N30°E, as well as orthogonal extension [Wang et al., 2001]. Chen et al. [2004] presented a model for these results that shows that strain between the major strike-slip faults is as significant as slip on the faults, and suggested that a large component of the rapid eastward extrusion of Tibet is caused by distributed deformation within the plateau, not only slip on the major faults.

2.1. Yarlung-Zangbo Suture

[7] The YZS in southern Tibet is the Tertiary suture that marks the boundary between rocks of Eurasian and Indian origin. The early convergence between India and Eurasia took place along the YZS, and the suture closed during the Late Cretaceous to Eocene following the closure of the neo-Tethys [Burg and Chen, 1984; Gansser, 1980; Molnar and Tapponnier, 1975; Ni and Barazangi, 1984; Tapponnier et al., 1981a]. Presently, the Indian plate underthrusts rocks of originally Indian origin and dips beneath the southern Tibetan Plateau [Seeber et al., 1981; Molnar and Chen, 1982; Barazangi and Ni, 1982; Hauck et al., 1998].

[8] Whether the suture remained active during the late stages of the Indo-Asia collision, and whether it is active today are both subject to debate. The linearity of the YZS and the steeply dipping faults accompanying it [e.g., Tapponnier et al., 1981a] indicate that the suture is a steeply dipping zone [Burg and Chen, 1984] and that postcollision strike-slip movement along the suture may have been large [Tapponnier et al., 1981a; Burg and Chen, 1984; Tapponnier et al., 1986]. Tapponnier et al. [1986] suggested the YZS was active as a strike-slip shear zone with many kilometers of right-lateral offset in the Oligo-Miocene. The suture is not marked by continuous young fault traces except in the northwest where the suture appears to be a continuation (to about 86°E) of the active Karakorum Fault [Tapponnier et al., 1981a; Rothery and Drury, 1984; Tapponnier et al., 1986; Lacassin et al., 2004].

[9] The connection between the Karakorum fault and the YZS has been subject to recent debate. Murphy et al. [2000, 2002] argued that the Karakorum fault did not extend east of Mount Kailash (81°E), instead terminating in the Gurla Mandhata detachment. However, on the basis of field and radiometric data, Ratschbacher et al. [1994] argued that strike-slip faults along the YZS near Xigatse (near 89°E) may have been active recently. Tapponnier et al. [2001] report that field evidence in several locations supports this conclusion. Lacassin et al. [2004] showed that the Karakorum fault continues east of 81°E as a dextral, transpressive flower structure reactivating the YZS.

2.2. Yadong-Gulu Rift

[10] The Yadong-Gulu rift extends from south to north across the Himalaya, the Yarlung-Zangbo suture, the Lhasa block, and the Nyainqentanglha range [Armijo et al., 1986]. The overall trend of the rift is about N30°E (Figure 1).

[11] Armijo et al. [1986] divided the Yadong-Gulu rift into three subsections: (1) south of the Zangbo suture where normal faults dominate and dip to the west; (2) between the suture and Nyainqentanglha range, where the rift is nearly symmetrical with faults dipping east and west; and (3) along the Nyainqentanglha range, where major faults dip to the east and have a large component of left-lateral movement (Figure 1). Armijo et al. [1986] estimated an average rate across the Yadong-Gulu rift of 1.4 ± 0.8 mm/yr, but noted that the throw rate at the northernmost part of the rift appeared to be much higher, perhaps as high as 15 mm/yr as the fault approaches the Beng Co strike-slip fault.

[12] Several ductile high strain shear zones have been identified in the southern and northern Yadong-Gulu rift [Pan and Kidd, 1992; Harrison et al., 1995; Kidd et al., 1995; Burchfiel et al., 1991]. The locations and the shear sense of these localized high strain zones may be associated with the opening of the Yadong-Gulu rift [Cogan et al., 1998]. Cogan et al. [1998] suggested that the opening of the rift is being accommodated at depth by flow in the middle crust, perhaps a consequence of the middle crust being partially molten as suggested by INDEPTH data [Nelson et al., 1996].

2.3. Extension Rate Across Southern Tibet

[13] Seismicity [Molnar and Tapponnier, 1978; Ni and Barazangi, 1984; Molnar and Lyon-Caen, 1989] and active north-south trending graben systems [Armijo et al., 1986] show that significant east-west extension occurs in the Himalaya and southern Tibet. The rate of extension across southern Tibet was estimated as 5–10 mm/yr by Baranowski et al. [1984] and 18 ± 9 mm/yr in ∼N115°E by Molnar and Lyon-Caen [1989] from seismic moment release in earthquakes between 1962 and 1986. Armijo et al. [1986] estimated the total rate of extension across southern Tibet by extrapolating measured slip rates from the Yadong-Gulu rift, assuming spatially uniform extension, and obtained an estimate of 10 ± 5 mm/yr. Since the westernmost graben is near 80°E, their estimate of 10 ± 5 mm/yr ignores extension farther west, which was included in the estimate of Molnar and Lyon-Caen [1989]. Cogan et al. [1998] suggested that the total E–W extension across the Tibetan Plateau could be much greater than Armijo et al. [1986] estimated if ductile detachment faults observed along the Yadong-Gulu are indicative of rapid extension.

[14] All of the previously estimated rates are highly uncertain, due to weak age control and the short seismic record. More recently, based on six years of GPS data in Nepal and southern Tibet, Larson et al. [1999] observed east-west extension across southern Tibet at a rate of 11 ± 3 mm/yr between northwestern Nepal and Lhasa. However, given the available site spacing, Larson et al. [1999] could not determine whether the extension was uniform or variable in space nor distinguish clearly between elastic deformation due to the locked Main Himalayan Thrust and extension due to faulting within the Tibetan crust.

2.4. Convergence Rate Between Tibet and India

[15] Early estimates of the long-term convergence rate across the Himalaya were in the range of 10–25 mm/yr from geologic field observations [Armijo et al., 1986], seismic moment release rate for major thrust events along the Himalayan arc [Molnar and Deng, 1984], inversion of profiles of uplift rates [Molnar, 1987], and changes in the age of basal sediments in the Ganga basin over the last 10–20 Myr [Lyon-Caen and Molnar, 1985]. However, these early estimates were based on some assumed features, such as assumed dates for the geologic estimate and assumed rupture zones for the seismic estimate, and thus yielded large uncertainties.

[16] More recently, precise GPS geodesy suggested a convergence rate across the Himalaya of 14–18 mm/yr [Bilham et al., 1997; Larson et al., 1999; Paul et al., 2001; Banerjee and Bürgmann, 2002]. These geodetic convergence rates were derived from sites close to the Himalayan arc, and use an elastic faulting model [Savage, 1983] to estimate the total convergence rate, as all sites lie within the zone of elastic deformation associated with the locked thrust. These geodetic convergence rates are thus sensitive to a trade-off between the fault geometry and slip rate.

[17] On the basis of uplift profiles derived from abandoned river terraces, Lavé and Avouac [2000] derived a slightly higher rate (21 ± 1.5 mm/yr) of underthrusting beneath the Siwalik Hills at the Main Frontal Thrust of the Himalaya. As with the geodetic estimates, this shortening rate is dependent on certain assumptions. In this case, the shortening rate is dependent on the assumptions that terrace treads represent isochrons, and that long-term uplift results from fault-bend folding. A less precise, but similar rate (20 ± 5 mm/yr) was determined from conservation of mass constraints, that require no assumptions about fault geometry. Cattin and Avouac [2000] developed a two-dimensional finite element model that showed that the geodetic [Larson et al., 1999] and geologic data are internally consistent.

[18] The GPS sites used by Larson et al. [1999] and all subsequent studies using the same data set were in Nepal or in southern Tibet within ∼70 km of the Himalaya. In this paper, new data from southern Tibet extend data coverage farther from the Himalayan arc, which will strengthen convergence rate estimates, and allow us to model not only the Himalayan thrust, but also the internal deformation of southern Tibet. We find that significant modifications to Larson et al.'s [1999] models are required by these new data, and that the three GPS velocities from southern Tibet used in these previous studies are significantly biased. We also show that the convergence rate in central Himalaya is significantly lower than previous geodetic estimates [Bilham et al., 1997; Larson et al., 1999] and that this rate varies along the Himalayan arc.

3. Measurement and Analysis

3.1. GPS Observations

[19] A total of 18 GPS sites in southern Tibet lie within a roughly N110°E transect about 1400 km long and 660 km wide, extending from east of Lhasa west almost to the Karakorum fault (Figure 1). Data were collected between 1991 and 2000 through the collaborative efforts of researchers from the United States and China. Most measurements were in campaign style, with most surveys lasting two or more days at each site. Fifteen sites in Nepal from Larson et al. [1999] are also included in this study, and 1998–2000 observations at Nepalese sites AIRP, NAGA, and NEPA in Nepal are used here, which were not available to Larson et al. [1999]. In addition, new observations at two sites in southern Tibet used by Larson et al. [1999], RONG and TING, produce much more precise and quite different site velocities. All of this data was included in the compilation of Wang et al. [2001], and velocities used here are only slightly different than presented in their paper. We use a slightly improved version of the velocity field of Wang et al. [2001] but only discuss the 33 sites in Nepal and southern Tibet in this paper (Table 1). Site velocities are provided in Table 2. Velocities for additional sites to the north of the study area are presented by Chen et al. [2004].

Table 1. GPS Sites Observed in Southern Tibeta
SiteIDLatitude, degLongitude, degHeight, mObservationAgency
  • a

    CU, University of Colorado; UAF-Xi'an, University of Alaska Fairbanks and Chang'an University, China; SSB, State Seismological Bureau of China; IGS, International Geodetic Service; UAF-WTUSM, Wuhan Technical University of Surveying and Mapping (now Wuhan University).

YadongWT1527.4988.9129281997.462000.42WTUSM, SSB
Table 2. Site Velocities in ITRF97, Eurasian, and Indian Framesa
SiteVelocities in ITRF97CorrelationEurasia-FixedIndia-Fixed
  • a

    Velocity in mm/yr. Eurasian and Indian frames are defined by REVEL [Sella et al., 2002]. EN, EU, and NU are correlation coefficients between east and north, east and vertical, and north and vertical, respectively.

AIRP38.2 ± 2.929.1 ± 1.55.9 ± 8.40.046−0.239−0.1569.7 ± 2.931.1 ± 1.50.072−2.6 ± 3.4−2.4 ± 1.8−0.261
BALA45.6 ± 0.813.7 ± 0.43.6 ± 1.80.052−0.023−0.19317.2 ± 0.917.1 ± 0.60.2863.9 ± 2.2−17.1 ± 1.8−0.874
BHAR38.9 ± 2.431.6 ± 1.28.9 ± 6.70.081−0.237−0.18310.4 ± 2.433.3 ± 1.30.116−1.7 ± 3.00.1 ± 1.5−0.293
BIRA41.1 ± 2.832.9 ± 1.48.2 ± 7.80.079−0.316−0.06512.7 ± 2.835.3 ± 1.40.107−0.5 ± 3.21.6 ± 1.9−0.286
DAGZ46.2 ± 0.611.7 ± 0.4−7.0 ± 1.40.044−0.073−0.13017.8 ± 0.815.3 ± 0.50.3594.4 ± 2.1−19.0 ± 1.8−0.908
GNGB56.1 ± 2.50.8 ± 1.47.6 ± 6.40.0400.083−0.13427.7 ± 2.54.9 ± 1.50.07313.8 ± 3.2−29.6 ± 2.5−0.489
GGAR49.5 ± 2.311.2 ± 1.3−0.7 ± 5.50.0420.016−0.12221.1 ± 2.414.7 ± 1.30.0807.7 ± 3.1−19.6 ± 2.2−0.484
GUCO43.1 ± 2.519.4 ± 1.47.8 ± 5.70.080−0.028−0.13614.5 ± 2.521.6 ± 1.40.1102.3 ± 3.2−12.0 ± 1.8−0.335
SHOT41.3 ± 2.517.7 ± 1.411.5 ± 5.60.0650.029−0.12112.7 ± 2.519.8 ± 1.40.0960.9 ± 3.2−13.7 ± 1.7−0.341
JANK39.9 ± 4.432.1 ± 1.911.5 ± 11.60.087−0.336−0.08011.4 ± 4.434.2 ± 2.00.100−1.3 ± 4.70.7 ± 2.2−0.100
JIAN42.2 ± 0.818.0 ± 0.52.1 ± 1.90.056−0.028−0.15313.7 ± 0.921.1 ± 0.60.2730.6 ± 2.1−13.0 ± 1.6−0.849
JIRI36.0 ± 3.225.6 ± 1.63.2 ± 9.40.151−0.292−0.2527.5 ± 3.227.8 ± 1.60.170−5.0 ± 3.7−5.8 ± 1.9−0.174
JOMO37.7 ± 1.522.6 ± 0.8−2.0 ± 4.10.040−0.152−0.2299.1 ± 1.524.2 ± 0.90.124−2.4 ± 2.5−9.0 ± 1.1−0.524
LHAS46.7 ± 0.213.2 ± 0.12.0 ± 0.40.0250.104−0.09618.3 ± 0.516.7 ± 0.40.7024.9 ± 2.0−17.6 ± 1.8−0.964
LAZE42.6 ± 1.120.6 ± 0.6−0.8 ± 2.30.0100.034−0.15914.1 ± 1.223.2 ± 0.70.1601.6 ± 2.3−10.6 ± 1.4−0.770
MAHE33.6 ± 2.232.5 ± 1.1−4.9 ± 6.00.078−0.102−0.1995.0 ± 2.333.2 ± 1.10.116−5.5 ± 3.00.8 ± 1.2−0.126
NAGA37.5 ± 0.529.7 ± 0.25.2 ± 1.10.022−0.022−0.0679.0 ± 0.631.7 ± 0.40.484−3.3 ± 1.9−1.7 ± 1.1−0.903
NEPA37.9 ± 1.133.0 ± 0.62.2 ± 2.90.044−0.124−0.1679.4 ± 1.234.0 ± 0.70.175−1.8 ± 2.21.3 ± 0.8−0.505
WT1138.0 ± 3.022.3 ± 1.51.2 ± 6.6−0.0610.051−0.1379.5 ± 3.024.4 ± 1.6−0.034−2.8 ± 3.6−9.1 ± 1.9−0.342
WT1640.4 ± 1.419.8 ± 0.90.4 ± 3.8−0.0630.131−0.24211.9 ± 1.522.0 ± 1.00.028−0.4 ± 2.4−11.6 ± 1.4−0.620
POKH37.1 ± 1.928.8 ± 1.08.8 ± 5.10.074−0.217−0.1828.5 ± 2.030.5 ± 1.10.124−3.3 ± 2.7−2.7 ± 1.3−0.381
RANJ37.3 ± 2.529.5 ± 1.35.8 ± 6.20.063−0.133−0.1828.7 ± 2.530.8 ± 1.30.095−2.7 ± 3.1−2.1 ± 1.5−0.216
RONG37.7 ± 1.322.4 ± 0.76.7 ± 2.90.0890.016−0.2089.2 ± 1.324.8 ± 0.80.196−3.4 ± 2.3−8.9 ± 1.4−0.673
WT1233.9 ± 1.020.7 ± 0.6−9.2 ± 2.4−0.0690.106−0.2165.3 ± 1.122.7 ± 0.70.094−6.4 ± 2.3−10.7 ± 1.2−0.747
SHIQ30.9 ± 1.214.9 ± 0.6−2.1 ± 2.5−0.0640.126−0.1502.3 ± 1.315.5 ± 0.70.055−6.9 ± 2.8−16.9 ± 0.7−0.390
SIMA39.2 ± 2.233.5 ± 1.0−0.6 ± 5.30.055−0.134−0.06910.7 ± 2.235.4 ± 1.10.103−1.6 ± 2.82.0 ± 1.4−0.389
SIMI34.3 ± 1.619.4 ± 0.911.2 ± 4.60.050−0.108−0.2845.7 ± 1.720.5 ± 1.00.114−4.9 ± 2.7−12.3 ± 1.1−0.365
SURK34.5 ± 1.829.1 ± 1.06.8 ± 5.10.044−0.156−0.1985.9 ± 1.930.1 ± 1.00.098−5.1 ± 2.7−2.6 ± 1.1−0.296
TANS37.4 ± 2.428.4 ± 1.317.6 ± 6.30.062−0.172−0.1688.9 ± 2.529.9 ± 1.30.095−2.9 ± 3.1−3.2 ± 1.5−0.258
TCOQ34.9 ± 2.416.7 ± 1.4−0.1 ± 5.60.0880.015−0.1396.3 ± 2.418.6 ± 1.40.118−4.9 ± 3.3−14.8 ± 1.7−0.338
TING35.5 ± 1.420.8 ± 0.72.2 ± 3.10.092−0.029−0.1436.9 ± 1.523.3 ± 0.80.184−5.6 ± 2.4−10.5 ± 1.5−0.648
XIGA41.2 ± 0.919.4 ± 0.5−0.8 ± 2.00.062−0.080−0.15412.8 ± 1.022.2 ± 0.60.255−0.1 ± 2.2−11.7 ± 1.5−0.832
WT1540.5 ± 2.122.6 ± 1.25.8 ± 5.7−0.0770.170−0.20412.1 ± 2.125.5 ± 1.3−0.025−1.2 ± 2.8−8.5 ± 1.9−0.521

[20] Updates and changes have been made in the velocity solution of Wang et al. [2001], but differences between the velocities presented by Wang et al. [2001] and in this paper are generally less than 1 mm/yr. About 60% of the solutions used by Wang et al. [2001] were improved, mostly by adding data from continuous sites surrounding the area of interest (no data were added from China or Nepal). The improvements in the solutions result from an improved determination of the reference frame, and from improved quality control. We identified and corrected a few antenna height errors, fixed a reference clock problem that biased a 10–20 daily solutions, and excluded a few outlier measurements.

3.2. GPS Data Analysis

[21] The GPS data in this paper were analyzed using the GIPSY/OASIS II software (release 6) developed at the Jet Propulsion Laboratory (JPL) [Zumberge et al., 1997; Gregorius, 1996]. Raw GPS data collected in the field were analyzed in 24-hour daily solutions along with regional and global continuous sites. Two types of GPS solutions were used in this study. For data prior to 1996, global GPS solutions were estimated using all data from Tibet, along with a well-distributed set of global sites. Satellite orbits were estimated in these solutions. For the more recent data, we used regional solutions by combining campaign data with a regional set of continuous sites and used fixed orbits obtained from JPL's submission to the International GPS Service (IGS). For more details, refer to Larson et al. [1997] and Freymueller et al. [1999, 2000].

[22] Each daily solution was then transformed into the International Terrestrial Reference Frame 1997 (ITRF97). Finally, the individual daily GPS solutions were combined together to determine site velocities. The coordinates in the ITRF97 reference frame and their covariance matrix from the daily GPS solutions were used in a standard weighted least squares solution to estimate site positions at epoch 1995.0 and site velocities.

[23] For ease of interpretation, we express our velocities either relative to Xigatse (XIGA), or to the Indian plate, by subtracting the motion of XIGA or the Indian plate relative to the ITRF97 reference frame (Figure 2). We use the REVEL model [Sella et al., 2002] to define an India-fixed reference frame. This is a different and more robust approach than was possible with previous studies. For example, compared to the velocities in an India-fixed frame as estimated by Larson et al. [1999], our velocities are slightly slower. The uncertainty in the REVEL model, based on propagation of the full REVEL covariance matrix, is included in the uncertainties of velocities of sites relative to the Indian plate. For many sites, the uncertainty in the Indian frame is the dominant uncertainty in the velocity relative to the Indian plate, so further refinement of the India-fixed frame would be beneficial.

Figure 2.

GPS velocities 1991–2001 for sites in southern Tibet and Nepal, with 95% confidence ellipses. Active faults are shown by solid lines, modified from Tapponnier et al. [2001] and Taylor et al. [2003]. (a) Velocities relative to Xigatse (XIGA). Xigatse is shown as a solid diamond. (b) Velocities relative to the Indian plate [Sella et al., 2002]. Uncertainties in Figure 2b include the uncertainty in the motion of the Indian plate relative to ITRF97.

3.3. Differences From Previous Work

[24] Compared to Larson et al. [1999], there are significant differences in the velocities for three sites within our study area, sites RONG, TING and LHAS in southern Tibet. All three had limited data in the work by Larson et al. [1999]. Figure 3 compares the two velocity fields by constructing a profile using site velocities relative to site NAGA (Kathmandu). Because of reference frame differences, this is simpler than rotating Larson et al.'s [1999] data into an India-fixed frame. Clearly, the velocities at sites in Nepal are consistent within uncertainties, but velocities of the sites in southern Tibet are significantly different, ∼6 mm/yr lower than that estimated by Larson et al. [1999]. TING and RONG are critical sites because both constrain the total convergence rate between India and Tibet [Larson et al., 1999].

Figure 3.

N12°E velocities relative to NAGA as a function of distance from NAGA along a N102°E profile. Error bars represent one standard deviation. Circles indicate velocities from this study; dots indicate velocities from Larson et al. [1999].

[25] Figure 4 shows the time series of north components for NAGA and TING in the ITRF97 frame. NAGA was first surveyed in 1991, and then again in 1992, and 1995 in campaign mode. After 1996, NAGA was observed as a continuous site (Figure 4a). Larson et al. [1999] included data from NAGA measured between 1991 and 1997, while we used data between 1991 and 2000. TING is a campaign site, first surveyed in 1991. We include all available data for TING between 1991 and 1998 (Figure 4b), but only the 1991 and 1995 data were available to Larson et al. [1999] (they did not have access to data from the 1994 survey, which was carried out by a Chinese collaborator). The large scatter in the 1991 estimates of the site position results from the limited global tracking network available at that time and consequently poor reference frame definition. We found that the uncertainty in the determination of the reference frame transformation for the 1991 data should be more than 10 cm due to the limited global tracking network. However, the GIPSY software transformation program does not account for this uncertainty. Thus the site positions for the 1991 data could be biased and the uncertainties were underestimated. We removed these data from our solution. For the remainder of the data used in this study, the uncertainty in the reference frame definition is small.

Figure 4.

Time series of north components of sites (a) NAGA, and (b) TING. Solid line indicates best fit for the complete time series; dashed line in Figure 4a marks the end of the temporal span of Larson et al. [1999]. The dotted line in Figure 4b is best line fit when excluding 1991 survey, while the dashed line is best line fit based on the data available to Larson et al. [1999].

[26] A single biased site position could bias the site velocity unless we have a large enough number of the site occupations. For example, the velocity at NAGA does not change whether or not we use the 1991 survey because there are many other data spanning a long survey period. However, this is not the case at TING. Excluding either the 1991 or 1998 data for TING would significantly change its velocity (Figure 4b). We observe a similar pattern with RONG and LHAS. In our velocity solution, we exclude the 1991 solutions for the reasons described above. The velocities for TING, RONG, and LHAS estimated by Larson et al. [1999] are all biased toward faster convergence across the Himalaya, and these velocities have been used in several subsequent studies [e.g., Cattin and Avouac, 2000]. By eliminating these biases, our new velocity field is a significant improvement over past published versions.

4. Active Deformation in Southern Tibet

4.1. East-West Extension

[27] Geologic and seismic evidence suggest that the orientation of extension in southern Tibet is ∼N110°E [Armijo et al., 1986; Molnar and Lyon-Caen, 1989]. We show horizontal velocity components perpendicular and parallel to a N110°E profile through XIGA as a function of distance along the profile in Figure 5. Only sites on the Tibetan Plateau are shown in Figure 5.

Figure 5.

(a) Velocity relative to XIGA in the N110°E direction as a function of distance along a N110°E profile. (b) Velocity relative to XIGA in the N20°E direction as a function of distance along a N110°E profile. Error bars represent one standard deviation. Circles indicate sites north of the YZS; dots indicate sites south of the YZS; triangles indicate projected locations of seven rift systems.

[28] In southern Tibet and northern Nepal, the total rate of extension in the N110°E direction between Lhasa (LHAS) and Simikot (SIMI) is about 13 ± 2 mm/yr over about 1000 km, or about 0.013 μstrain/yr. This is consistent with the 11 ± 3 mm/yr of extension rate observed by Larson et al. [1999] using a subset of our full data set. The total extension rate between 80°E and 93°E (SHIQ and GNGB) in the N110°E direction is about 26 ± 3 mm/yr. The extension rate (N110°E) between SIMI in northwestern Nepal (82°E) and SHIQ is insignificant (1 ± 2 mm/yr), although these sites lie on opposite sides of the Karakorum fault and thus should measure the difference between the Karakorum fault slip rate and the extension rate of the Tibetan crust south of the Karakorum fault. About 13 mm/yr of extension occurs between Lhasa and Gongbu, which are only about 300 km apart. Clearly, the extension rate is not uniform along the length of the Himalayan arc.

[29] The site velocities over the entire area cannot be explained by a simple combination of uniform N110°E extension and strain resulting from uniform coupling along the Main Himalayan Thrust [Larson et al., 1999; Bilham et al., 1997]. Along the profile, the velocities in the N110°E direction (relative to Lhasa) of sites Jiangzi (JIAN) and Xigatse (200–300 km west of Lhasa) are 5.5 ± 1.2 mm/yr and 6.7 ± 0.8 mm/yr, respectively, implying that the extension rate across the Yadong-Gulu rift is ∼5–7 mm/yr, while those of Tingri (TING), Rongbuk (RONG), and Jomosom (JOMO) (500–800 km west of Lhasa) average 10 ± 2 mm/yr. Thus about one half of the east-west extension rate between 80°E and 90°E may be concentrated across the Yadong-Gulu rift. The extension rate increases to the east of Lhasa, although this observation is based on only one site (GNGB, Figures 2 and 5).

[30] About 350 km west of Xigatse, near 86°E, sites SHOT north of the Yarlung-Zangbo suture and WT12 south of the suture have significantly different velocities (about 7 ± 3 mm/yr) in the N110°E direction (Figure 5). This can also be seen clearly in the velocities relative to Xigatse (Figure 2a). These sites are only 50 km apart, and we interpret this difference as being due to the presence of an active strike-slip fault between them. Recently published work [Lacassin et al., 2004] supports our interpretation, which was made independently. Figures 2 and 5 show that sites on the north of the YZS show significantly slower westward velocities (relative to XIGA) than sites farther to the south. This shows that not only is the extension rate nonuniform across all of southern Tibet, but also that the crust south of the Yarlung Zangbo suture moves westward relative to that north of the suture.

4.2. Convergence Between Tibet and India

[31] Figure 5b shows velocity components in the N20°E direction along the N110°E profile (orthogonal to Figure 5a). For most sites in southern Tibet between the Yadong-Gulu rift and Thakkola graben system, the southward motion relative to XIGA is not significant, and slower than those of the sites east of the Yadong-Gulu rift or west of Thakkola graben.

[32] The convergence across the Himalaya is better illustrated by examining the N12°E component of site velocities relative to India (Figure 6), the approximate convergence direction between India and Eurasia [Chen et al., 2004; Sella et al., 2002]. Sites across all of southern Nepal do not move significantly relative to India. However, sites across southern Tibet show a range of velocities relative to India that varies systematically with longitude. East of Kathmandu, Nepal, we observe 11.7 ± 1.5 mm/yr of N12°E convergence between India and sites in southern Tibet (e.g., TING and XIGA) (Figures 2b and 6). This is significantly lower than 18 ± 2 mm/yr estimated by Larson et al. [1999] based on more limited data from the same sites. We see the same situation in western Nepal; SIMI is moving south at 13 ± 1 mm/yr, while sites in southwestern Nepal (MAHE and NEPA) are moving negligibly relative to India. The rate of shortening within the network in western Nepal (between SIMI, MAHE, and NEPA) is 13.4 ± 0.9 mm/yr. Farther to the west, and also east of the Yadong-Gulu rift, we do not have data near the Himalayan arc, but the contraction rate across these segments must be ∼18 mm/yr based on SHIQ and LHAS (Figure 6), assuming that India is rigid.

Figure 6.

Velocities relative to the Indian plate in the N12°E direction as a function of distance from NAGA along a N102°E profile. Error bars represent one standard deviation. Circles indicate northern sites (in Tibet); dots indicate sites south of the MHT; stars indicate sites in between. The region of the Nepal GPS network used by Larson et al. [1999] is shaded. Zero velocity is shown as dashed line.

[33] If we postulate that this southward motion is mostly caused by the steady slipping portion of the thrust system below the Himalaya, the different southward motion at sites of southern Tibet could imply variation in the slip rate of India beneath Tibet along the Himalayan arc. However, to confirm this we need to use a model that incorporates elastic deformation, since the contraction within the network should always be less than or equal to the total convergence rate.

5. Modeling

[34] Strain resulting from the locked Main Himalayan Thrust (MHT) was modeled by Bilham et al. [1997] and Larson et al. [1999] as uniform slip on rectangular dislocation planes embedded in a homogeneous, isotropic, and elastic half-space. Their models fit the data from Nepal quite well but do not explain the observations at sites within Tibet. Figure 7 shows the residuals of the observed and predicted velocity field by Larson et al. [1999], shown relative to Xigatse. The residuals of sites within Tibet show that this model fails to explain the east-west extension we observe in southern Tibet. The extensional residual across the Himalaya indicates that the model overestimates the contraction between India and southern Tibet by ∼4 mm/yr, because the slip rate of India beneath Tibet is too high (see section 4.2). The sites in Tibet systematically move more slowly relative to India than predicted by this model. Clearly, a revised and more complex model is required.

Figure 7.

Residual velocity field (relative to XIGA) in Nepal and southern Tibet after removing the model of Larson et al. [1999]. XIGA is shown as a solid square. The surface projections of the downdip ends of the MHT fault planes are shown as black lines.

5.1. Dislocation and Block Model

[35] We use a block modeling approach to explain the GPS velocities in terms of the motions of rigid crustal blocks and elastic deformation caused by locked faults at the block boundaries. There is a long history of block models in geodetic literature, a recent example being the work by Meade et al. [2002], who developed a model for the Marmara Sea region of Turkey. Our approach is quite similar to theirs, although developed independently.

[36] Our block model is by necessity highly simplified, consisting of five blocks within southern Tibet plus rigid India (Figure 8). The blocks within Tibet are divided by the major faults identified by previous work, and where significant relative motions are clear in our results: the Yadong-Gulu rift, Thakkola graben, Karakorum fault, and Yarlung-Zangbo suture. We assume that the latter two faults connect near Lake Manasarovar, consistent with Lacassin et al. [2004]. We neglect the other five north-south trending rift systems in southern Tibet, because they appear to have much lower slip rates and inclusion of these additional faults would increase the degrees of freedom of our model without significantly improving the fit to the data. The model is bounded on the north by the KJFZ. Site velocities are modeled as a sum of rigid block motions and elastic deformation (modeled via elastic dislocations with virtual backslip, as by Savage [1983] and many other studies) resulting from the block-bounding faults being locked. The elastic component is most important for sites very close to the block-bounding faults, but most of our sites are several tens of kilometers or more from the block-bounding faults. The dipping MHT produces significant strain over a broad area, so it is critical that the model correctly incorporate the elastic deformation from this fault. However, for the other faults we found that we could assume either that they were creeping to the surface or fully locked to ∼15 km, with little to no change in the model results, because their slip rates are slow and our sites are too far from the block boundaries. For simplicity, we assume here that the block-bounding faults within Tibet are creeping.

Figure 8.

Map outlining the block boundaries in this study. The Himalayan arc is modeled as three faults. The horizontal locations of the downdip end of the locked fault planes are shown, with diamonds at the midpoints. Sites in different blocks are shown with different symbols. The Thakkola graben and Yadong-Gulu rift are assumed to extend across the Himalaya to the surface trace of the MHT.

[37] For the same reason, even though the boundaries are quite simplified relative to the true faults, our model results are not impacted because we do not assume the sense of slip on any boundary. For example, although the Thakkola graben does not continue north of the YZS as a continuous feature, there are normal faults to the north of the YZS and slip on those faults are treated as a northward continuation of the Thakkola graben in the model. A corollary of the lack of sensitivity to the details of the block boundaries within Tibet is that the identification of the block boundaries with specific faults is in most cases interpretive, based on prior geologic work. The block boundaries may not be single fault, and our estimated slip rates should be considered as the sum of slip on all faults that accommodate the relative block motion.

[38] Strain resulting from the locked thrust system below the Himalaya is modeled by rectangular dislocation planes embedded in a homogeneous, isotropic, and elastic half-space [Okada, 1985]. Compared to Larson et al. [1999], we increased the number of dislocations along the Himalaya to better model the geometry of the MHT. We divided the eastern fault of Larson et al. [1999] into two dislocations with different orientations, with the boundary at the Yadong-Gulu rift. Both strike-slip and dip-slip motion are allowed on the MHT. Our approximation of the MHT as a planar fault is oversimplified [Cattin and Avouac, 2000], but the detailed geometry of the locked portion has little impact on the interseismic deformation. The location and depth of the downdip end of the locked zone are what control the interseismic deformation to first order [Freymueller et al., 2000; Vergne et al., 2001].

[39] All slip rates on the dislocations are calculated using the velocities of the blocks as the fundamental parameters. A site velocity can be calculated by v = vblockequation imageimage where vblock is the velocity of the block the site lies upon, and image is the elastic contribution from the ith locked dislocation, and N is the number of the dislocations. As mentioned before, except for the MHT we neglect the elastic deformation because the fault slip rates within southern Tibet are low and the sites far from the faults. We use all data shown in Figure 2, except for the isolated site Gongbu (GNGB) ∼250 km east of Lhasa. Because this site moves eastward so rapidly relative to Lhasa (13 ± 3 mm/yr), its velocity could be fit only by adding another block with only that one site on it. We have no simple explanation for this velocity, but there are two independent sites in the same town that give compatible velocities, so it cannot be explained away as an unstable site.

5.2. Inversion Method

[40] We inverted the GPS velocities to estimate the relative block motions, including the convergence rate across the Himalaya, plus the geometry of the MHT. Table 3 summarizes the block boundaries. For each section along the MHT, we estimated the fault length, width, depth, dip, strike, location (x and y) of the middle point of upper tip of the locked portion, plus the strike-slip and dip-slip rates determined from the relative block motions. Any model that estimates fault geometries from geodetic data is a nonlinear inversion problem (estimating only slip rates with a fixed geometry is a linear problem). We use the random cost algorithm [Berg, 1993], a computationally efficient method of inverting source geometry using geodetic data [e.g., Murray et al., 1996; Freymueller et al., 1999; Cervelli et al., 2001] to estimate the MHT fault geometry. We then fixed the optimal MHT fault geometry estimated from the random cost algorithm, and investigated the block motions using linear least squares techniques.

Table 3. Model Parametersa
FaultLength, kmDepth, kmMidpointbDip, degStrike, degDip Slip,c mm/yrStrike Slip,d mm/yrOpening,e mm/yr
  • a

    Results in parentheses are from the alternative model, see text for detail.

  • b

    Surface projections of midpoints of down tip of locked part.

  • c

    Negative values indicate thrusting.

  • d

    Positive values indicate right-lateral motion; negative values indicate left-lateral motion.

  • e

    Positive values indicate opening; negative values indicate shortening.

Karakorum489.215.032.379.8−90.0140.0-3.6 ± 1.0−0.7 ± 1.0
       -(3.7 ± 1.0)(−0.5 ± 1.0)
YZS813.415.029.585.5−90.0106.0-2.8 ± 0.80.0 ± 0.3
       -(2.9 ± 0.8)(−0.1 ± 0.3)
North Yadong357.215.029.690.3−90.030.0-2.6 ± 0.66.6 ± 2.2
       -(2.3 ± 0.6)(4.3 ± 2.0)
South Yadong141.215.027.689.0−90.030.0-2.0 ± 0.69.3 ± 2.2
       -(1.7 ± 0.6)(7.1 ± 2.0)
North Thakkola319.515.031.583.5−90.0−1.0-−5.9 ± 1.46.1 ± 1.9
       -(−5.6 ± 1.2)(5.0 ± 1.7)
South Thakkola207.415.029.183.6−90.0−1.0-−3.5 ± 1.45.2 ± 1.9
       -(−3.3 ± 1.2)(4.2 ± 1.7)
West Himalaya800.018.330.079.3−9.5115.0−17.0 ± 0.90.8 ± 1.9-
       (−16.4 ± 0.8)(1.7 ± 1.8)-
Central Himalaya485.014.328.085.5−2.5102.2−12.2 ± 0.41.2 ± 0.8-
       (−12.2 ± 0.4)(1.4 ± 0.8)-
East Himalaya660.020.327.492.5−3.183.9−19.0 ± 1.23.7 ± 1.2-
       (−17.5 ± 1.0)(2.2 ± 1.0)-

[41] We took this approach because the geometry of parts of the MHT is not well constrained. Specifically, the width of the locked zone on the MHT is unknown except in central Nepal, where it is constrained by past studies and by leveling data [Jackson and Bilham, 1994]. Because we lack data from GPS sites close to the arc in western Nepal and east of Nepal, we cannot determine the width of the locked zone and the dip angle there only from the GPS data. We chose an a priori model using the western fault of Larson et al.'s [1999] model as the western Himalaya segment, and we defined the center and eastern Himalaya segments to be consistent with the approximately small circle shape of the India-Eurasia collision boundary inferred from seismicity, topography, stress state, and geodetic data [Bendick and Bilham, 2001]. We constrained the model fault to remain close to the mapped location of the small circle shape of the MHT and the dip angle in the eastern dislocation to be similar to that in central Nepal. Like Larson et al.'s [1999], we find a steeper dip angle in western Nepal compared to eastern Nepal, consistent with earthquake focal mechanisms [Molnar and Tapponnier, 1978; Molnar and Chen, 1982; Molnar and Lyon-Caen, 1989].

5.3. Results

[42] The estimated parameters for the optimal model are summarized in Table 3. Figure 9 shows the observed and model predicted horizontal velocities relative to the rigid Indian plate, as well as the horizontal location of the downdip edge of the MHT fault planes. The modeled velocity field is matched quite well, and yields a weighted misfit per degree of freedom of 1.12, indicating an excellent fit to the data. Uncertainties in this section are based on the block model and have been scaled to reflect the misfit.

Figure 9.

Observed horizontal velocities (solid vectors) relative to the Indian plate and predicted horizontal velocities (open vectors) from our preferred model. The 95% confidence ellipses are shown. The locations of the downdip ends of the MHT fault planes are shown as in Figure 8. Large grey vectors indicate predicted block velocities relative to Indian plate. Active faults are shown as solid lines; block boundaries are dashed lines.

[43] The Himalayan arc is modeled by three dislocations with different orientations. The estimated strikes of the western, central, and eastern faults are 115°, 102°, and 84°, respectively. The geometries (dip, width of locked zones) of the fault tips in the western and eastern Himalaya are poorly determined, since we do not have data near the arc. In contrast, the geometry of the central fault is very well determined, and we find that in the central Himalaya, the Indian plate is locked northward to 80 km north of Kathmandu (NAGA), with a shallow dip of ∼2.5°. The width of the locked zone here is similar to the ∼100 km determined for the NW Himalaya [Banerjee and Bürgmann, 2002]. Uncertainty in the dip angle is of the order of a few degrees. Compared to the models for eastern Nepal discussed by Vergne et al. [2001], our estimated dip angle is shallower, locked zone slightly wider, and the slip rate lower.

[44] Bilham et al. [1997] suggested a deeper locking depth in western Nepal than in eastern Nepal, with dislocations dipping at ∼4° in the seismogenic portion, and increasing smoothly in dip to ∼9° northward at a deeper level. Larson et al.'s [1999] two-fault model confirmed that. Our models favor a similar locking depth of 18–20 km along the Himalayan arc, with a steeper dip angle of ∼9° in western Himalaya, and a shallower dip of ∼3° in the central and eastern Himalaya.

[45] Our model suggests slip rates of 17 ± 1 and 19 ± 1 mm/yr for India beneath Tibet in the western and eastern Himalaya, respectively. These are consistent with previous estimates from GPS observations based on more limited data [Bilham et al., 1997; Larson et al., 1999]. However, along the central Himalaya, between the longitudes of 83°E and 88°E, we find a lower slip rate of 12.2 ± 0.4 mm/yr. We will discuss this in more detail in the following section. Our models predict right-lateral strike-slip motion of 3 ± 1 mm/yr on or close to the YZS, extending from the Karakorum fault in the west at least to the Yadong-Gulu rift in the east.

[46] Our model predicts right-lateral strike-slip motion of 4 ± 1 mm/yr on the Karakorum fault, similar to the recent estimate by Brown et al. [2002] based on cosmic ray exposure dating of debris flows and moraines, but significantly lower than the ∼30 mm/yr estimated by Liu [1993]. Lacassin et al. [2004] recently estimated an average slip rate for the Karakorum fault over the last 23–34 Myr of 10 ± 3 mm/yr. Our estimate results from the block model, and is constrained only by the (slow) relative motion of sites SIMI and SHIQ, and given a lack of data (or access to data) from the Himalaya west of Simikot, we assumed that there was no arc-parallel extension as is seen farther east. This assumption may be wrong, and if there is significant arc-parallel extension in this part of the Himalaya our Karakorum fault slip rate would be underestimated by roughly the magnitude of that extension. Banerjee and Bürgmann [2002] estimated a Karakorum fault slip rate of 11 ± 4 mm/yr, while GPS data from Ladakh, southwest or Shiquanhe suggest a lower rate (R. Bendick, personal communication, 2003). Our data are sufficient to demonstrate that a high slip rate, such as that proposed by Liu [1993], is untenable. If this part of the Karakorum fault had a high slip rate, SHIQ would move rapidly relative to SIMI. However, our data are not sufficient to distinguish between the rates proposed by Brown et al. [1993] and Banerjee and Bürgmann [2002]. A better estimate of the Karakorum fault slip rate will require the precise integration of GPS data from India and China, a difficult proposition given the policies of those countries regarding GPS position data.

6. Discussion

6.1. Convergence Across the Himalaya

[47] We estimated that the rates of slip of India beneath Eurasia in the western Himalaya between 76°E and 83°E and the eastern Himalaya between 89°E and 96°E are comparable to the previous estimates based on more limited GPS data [Bilham et al., 1997; Larson et al., 1999], but in the central Himalaya between 83°E and 88°E, our estimated rate of 12.2 ± 0.4 mm/yr is significantly lower than previously estimated. For comparison, Figure 10 shows the GPS site velocities (relative to NAGA) in the fault perpendicular (N12°E) component plotted against the distance from NAGA, as well as the predictions of our model and that of Larson et al. [1999]. The models presented by Vergne et al. [2001] and Galahaut and Chander [1997] are very similar to that of Larson et al. [1999]. Our model is quite different because of the improved (and much slower) velocities of RONG and TING, as well as the addition of more data from Tibet. Velocities relative to India are also slightly slower than previously estimated, due to the improved definition of an Indian reference frame (this does not affect the relative velocities shown in Figure 10).

Figure 10.

Velocities (relative to NAGA) in the fault perpendicular (strike 102°) direction of sites between 80°E and 90°E with respect to the distance from NAGA along N102°E. Error bars represent one standard deviation. Solid line is predicted by our model; dashed line is predicted by Larson et al. [1999]. Circles indicate velocities in this study; stars indicate velocities from Larson et al. [1999] for sites RONG and TING.

[48] To test whether this low convergence rate might apply to the entire arc, we tried an alternative model in which we added pseudo-observations to constrain the slip rates a priori along the eastern and western Himalaya to be 13 ± 2 mm/yr, comparable with that in the central Himalaya. The total sum of squared errors increases from 60 to 70 when these pseudo-observations are added, but even so the convergence rate to the east and west remains substantially higher than in the central region. Table 3 also gives the slip rate on each dislocation predicted by this alternative model. We infer that the variation in convergence rate is significant and required by the data.

[49] Our estimated convergence directions along the western, central, and eastern Himalaya are N25°E, N12°E, and N12°W, respectively, approximately normal to the local strike of the Himalayan arc, and consistent with the orientation of slip vectors of large earthquakes [Molnar and Lyon-Caen, 1989]. This is in accord with the conclusion of Armijo et al. [1986] connecting the rifting in southern Tibet with the divergence of Himalayan earthquake slip vectors (Figure 11). Paul et al. [2001] and Banerjee and Bürgmann [2002] showed that this same conclusion applies to the NW Himalaya as well, and it is likely that it applies to the entire Himalayan arc. We will discuss this in more detail in section 6.4.

Figure 11.

Azimuths of convergence vectors estimated in this study and slip vectors of earthquakes in the Himalaya [Molnar and Lyon-Caen, 1989] plotted as a function of the longitudes. Error bars represent one standard deviation. Circles indicate earthquake slip vectors; dots indicate model convergent vectors; stars indicate P axes of earthquakes in the western Himalaya.

[50] A low convergence rate across the Himalaya in east central Nepal conflicts with the shortening rate across the Siwalik Hills estimated by Lavé and Avouac [2000], who estimated 21 ± 1.5 mm/yr shortening from uplifted Holocene river terrace profiles and an assumption of fault-bed folding. They also determined a weaker constraint of 20 ± 5 mm/yr from conservation of mass, which makes no assumption about the mechanism of folding but still assumes that the terrace treads represent isochrons. It is not clear how to reconcile these observations, although both convergence rates are dependent on the assumptions of the models used. The geologically inferred rate might be in error if fault-bend folding was not the correct folding model, or if the terrace treads were not isochrons. The geodetically inferred rate might be in error if time-dependent earthquake cycle effects caused strain from the locked MHT to extend much farther north than predicted by an elastic dislocation model. All of these model assumptions should be subject to further scrutiny and testing. Postseismic deformation is not likely to cause a decrease in the present contraction rate across the Nepal Himalaya at present, because postseismic deformation will in general increase the contraction rate across the region near the downdip end of the locked zone (see, for example, the Alaska postseismic models of Freymueller et al. [2000]). If the downdip end of the locked MHT was significantly mislocated, it would need to be significantly deeper and farther north than we estimate to support a higher convergence rate.

6.2. Yadong-Gulu Rift

[51] Our estimate of the opening rate for the Yadong-Gulu rift is 6.6 ± 2.2 mm/yr, much higher than the average rate of 1.4 ± 0.8 mm/yr estimated by Armijo et al. [1986]. However, Armijo et al. [1986] also noted higher throw rates on the northernmost Gulu rift near Damxung, with estimates ranging from ∼6 to as high as ∼15 mm/yr. The relationship of extension rate to throw rate depends on the dip of the master fault, which is ∼45–60°, giving an extension rate ∼50–70% of the throw rate for the likely range of dip angles. The individual graben and half graben valleys that compose the Yadong-Gulu rift are generally a few to ∼15 km wide. The Yadong-Gulu rift could have been opening at the current rate for no more than about 3 Myr. Harrison et al. [1995] and Edwards and Harrison [1997] suggested that the onset of opening along the Yadong-Gulu rift was at circa 8 Ma. If both this date and the GPS observations are correct, then the opening rate must have increased dramatically within the last 3 Myr.

[52] The presence of partially molten rocks and the locally extreme heat flow near the Yadong-Gulu rift [Francheteau et al., 1984] suggest that the crust may be weak along the rift, allowing rapid localized strain. Given the nonuniform spatial extension rate in southern Tibet we observe here, it is questionable whether the crust in the Yadong-Gulu rift is typical of southern Tibet. The concentration of strain at the Yadong-Gulu rift compared to other rifts calls into question whether inferences about the crust there (for example, from INDEPTH) can be extrapolated easily to the rest of southern Tibet.

6.3. Yarlung-Zangbo Suture

[53] The Yarlung-Zangbo suture is the principal boundary along which early convergence between India and Eurasia took place [Molnar and Tapponnier, 1975; Gansser, 1980; Tapponnier et al., 1981a; Burg and Chen, 1984]. We find evidence for an active strike-slip fault along or near the YZS in the western Tibet. We interpret strike-slip faulting on or near the suture as a continuation of the active Karakorum fault farther west. In our numerical model, we assume this explicitly through our block boundaries. Because of the limited number of GPS sites, we cannot say whether the suture itself has been reactivated as a strike-slip fault, but the active fault must be within ∼10–20 km of the suture. Although developed independently, this is in agreement with the recent work of Lacassin et al. [2004]. Our results cast doubt on models that call for the Karakorum fault to end at 81°E. We thus identify the suture itself as the active fault in the discussion that follows.

[54] Examination of the velocity field suggests that right-lateral strike-slip motion continues on the YZS as far east as the Yadong-Gulu rift, as noted by Ratschbacher et al. [1994] based on the field observations and radiometric data. Tapponnier et al. [2001] draws the same conclusion. East of the Yadong-Gulu rift, we have no data from south of the suture, so we cannot tell if strike-slip motion continues farther east. In the block model we obtain a right-lateral slip rate of 2.6 ± 0.7 mm/yr.

6.4. Nonuniform East-West Extension in Southern Tibet

[55] Table 4 summarizes the block velocities relative to the Indian plate. Figure 12 shows the block velocities relative to both the Indian plate and relative to the block north of the YZS between the Thakkola graben and the Yadong-Gulu rift. East-west extension in southern Tibet is revealed in the N110°E component of the block motions. The block model allows us to separate permanent deformation of Tibet from elastic strain due to the curving and locked MHT, which is presumably recovered in great earthquakes. Although simplified, our block model is sufficient to explain the first-order features of the GPS data, and allows us to relate the GPS velocities to motions on the major fault zones of southern Tibet, while also removing the large elastic deformation associated with the MHT. The remaining block motions are presumed to represent permanent deformation of overriding southern Tibet.

Figure 12.

Estimated block velocities relative to the Indian plate (solid vectors) and relative to the block north of the YZS between the Yadong-Gulu rift and Thakkola graben (open vectors). Block boundaries are shown as dashed lines.

Table 4. Estimated Block Motionsa
  • a

    All velocities in mm/yr relative to the Indian plate. YZS, Yarlung-Zangbo Suture.

South Karakorum−6.5 ± 2.0−15.8 ± 0.6
North Karakorum−4.8 ± 2.6−18.9 ± 0.8
South YZS−1.3 ± 0.7−12.1 ± 0.5
North YZS1.3 ± 1.1−12.9 ± 0.7
East Yadong Rift5.7 ± 1.3−18.9 ± 1.0

[56] The block west of the Thakkola graben and north of the Karakorum fault moves at 2.0 ± 2.5 mm/yr in the N110°E direction relative to India. The N110°E component of the block east of the Yadong-Gulu rift is 11.7 ± 1.5 mm/yr. Thus the total permanent extension in the N110°E direction between longitudes of 78°E and 91°E is 9.7 ± 3.0 mm/yr. The larger rate of 13 ± 1 mm/yr observed between SHIQ and Lhasa includes a component due to elastic strain from the locked MHT. That is, about 10–25% of the observed extension rate reflects elastic strain from the curving, locked MHT.

[57] The block east of the Yadong-Gulu rift moves at ∼7 mm/yr in the N140°E direction relative to the block north of the YZS and west of the rift, implying half of the relative motion between the longitudes of 78°E and 91°E north of the YZS is concentrated across the Yadong-Gulu rift. However, south of the YZS, about two thirds of the extension occurs on the rift, as a consequence of the termination of strike-slip motion (in our model) on the suture at that point. The higher rate there is completely dependent on that assumption. Direct measurements of the opening rate across the rift on both sides of the YZS can test where the strike-slip motion terminates.

[58] About one half to two thirds of the total permanent extension between SHIQ and LHAS is concentrated along the Yadong-Gulu rift, with most of the remainder inferred to occur on or near the Thakkola graben. Thus extension cannot be uniform across all seven rifts in southern Tibet. Having rate data from only one rift, Armijo et al. [1986] extrapolated assuming uniform extension, and obtained a total extension rate of 10 ± 5.6 mm/yr across all seven rifts. Although our data clearly show that the present extension is spatially nonuniform, the total rate of permanent extension across the region is quite similar to their estimate (albeit now much more precise).

[59] Given the current spacing of GPS sites, we cannot rule out a total of ∼1–2 mm/yr extension across the rifts in central southern Tibet. These faults are active [Armijo et al., 1986], but apparently have much lower present-day extension rates than the Yadong-Gulu rift and Thakkola graben. We find 5–6 mm/yr extension across the Thakkola graben. Recall that in our block model, this boundary represents the sum of the Thakkola graben and nearby faults, but it is reasonable to assign the majority of this slip to the Thakkola graben for two reasons. First, the Thakkola graben is the only rift other than the Yadong-Gulu rift to cut through the High Himalaya. Second, geologic reconstructions for southern Tibet suggest that the Yadong-Gulu rift and Thakkola graben both have greater total extension than the other rifts [Replumaz and Tapponnier, 2003, and references therein].

[60] We propose that the geometries of the two most active rifts in southern Tibet, the Thakkola graben and the Yadong-Gulu rift, and the concentration of extension in these two rifts, cause the observed variations in the MHT convergence rates (Figure 13). The Yadong-Gulu rift trends N30°E, and the Thakkola graben strikes roughly north-south [Armijo et al., 1986; Mercier et al., 1987]. Thus the convergence direction roughly bisects the orientations of the two rifts, and each one strikes about ∼15° away from the convergence direction (Figure 13). Opening on these rifts would cause the block between them to move northward relative to its neighbors. This would cause the average convergence rate along the central segment to be slower than for the western and eastern segments.

Figure 13.

Cartoon of the kinematics of southern Tibet. Opening on the Yadong-Gulu rift and Thakkola graben results in slower convergence rates in the central Himalaya. The convergence direction in central segment of the Himalayan arc roughly bisects the orientations of the two rifts, and each rift strikes about ∼15° away from the convergence direction.

[61] The Yarlung-Zangbo suture also plays a role in the spatially nonuniform extension, with a right-lateral strike-slip rate of ∼3 mm/yr. The area south of the YZS (including Nepal) moves to the west relative to Xigatse (Figure 2). McCaffrey and Nabelek [1998] suggested that the extension in southern Tibet ought to be interpreted in terms of slip partitioning of oblique subduction, with the KJFZ serving as the strike-slip component of the partitioned deformation. We believe that the slip partitioning model is more appropriately applied to explain the right-lateral strike-slip motion on the suture than on the KJFZ, which is located much farther to the north. The model of McCaffrey and Nabelek [1998] predicts that the roughly arc-parallel extension in southern Tibet, as a consequence of the frictional partitioning of slip, should not be greater than the convergence rate between India and Tibet. The total extension rate observed across southern Tibet is too high in relation with the convergence rate for slip partitioning to be the sole driver of extension in southern Tibet.

[62] Gongbu (GNGB), 250 km east of Lhasa, moves at 13 ± 3 mm/yr relative to Lhasa in the N110°E direction, suggesting further extension is concentrated in between. However, the velocity at GNGB cannot be explained easily by extending our block model farther east, unless we simply add another block east of Lhasa with only one site on it. Figure 5 shows that the average extension rate changes dramatically at the Yadong-Gulu rift, and an intriguing possibility is that the extension rate may be higher across the entire easternmost part of the Himalayan arc. Geological evidence for well-developed rift systems east of Lhasa is lacking, so the interpretation of the motion of this site is not obvious, although the existence of two sites with consistent velocities means it cannot simply be rejected as a biased measurement. More GPS data between Gongbu and Lhasa will help improve our knowledge of east-west extension east of Lhasa in southern Tibet.

6.5. Motions at the North Edge of the Model

[63] Because the block boundaries we have assumed are simplified, it is instructive to consider what style of faulting might be optimal to accommodate the relative motion between the blocks north of the suture at their northern boundaries, given reasonable assumptions of the motion of central Tibet. Chen et al. [2004] documented significant extension of Tibet north of the KJFZ, and used a deforming block model to explain the large-scale deformation of Tibet. That model suggests that the extension associated with the Yadong-Gulu rift extends significantly north of the KJFZ, and ∼7 mm/yr of right-lateral strike-slip motion is concentrated across the KJFZ. In addition, the deformation of the entire Tibetan Plateau includes a component of distributed pure shear, with the axis of maximum compression oriented N32°E.

[64] The southeastward relative motion between the block east of the Yadong-Gulu rift and the block between the Yadong-Gulu rift and Thakkola graben north of the YZS could be accommodated by pure right-lateral slip on faults striking N140°E. This orientation is strikingly similar to that of the Jiali fault and other faults of the KJFZ [Armijo et al., 1986]. This suggests an alternative interpretation of the role of the Jiali fault and the KJFZ. These faults could have formed to accommodate motion similar to the block motion we infer. The relative motion between the block north of the Karakorum fault and west of the Thakkola graben and the central block north of the YZS could be accommodated by SW-NE trending left-lateral strike-slip faults (Figure 12), such as the fault mapped in this area by Yin et al. [1999].

7. Conclusion

[65] GPS velocities in southern Tibet show that 9.7 ± 3.0 mm/yr of permanent extension between 78°E and 92°E, roughly in the N110°E direction, occurs between Lhasa and Shiquanhe. Over this same distance, an additional ∼3 mm/yr of elastic deformation in the N110°E direction results from the curvature of the locked MHT. The rate of extension is not uniform across southern Tibet, with extension concentrated across the Yadong-Gulu rift and the Thakkola graben system. The extension rate increases dramatically east of Lhasa, where no well-developed rift systems have been identified.

[66] Opening of 6.6 ± 2.2 mm/yr along the Yadong-Gulu rift accounts for about one half to two thirds of the total permanent extension between Lhasa and Shiquanhe. A relatively undeforming block lies to the west of this rift and east of the Thakkola graben. This comprises the five remaining rift systems of southern Tibet, which extend at a total rate no larger than 1–2 mm/yr. Right-lateral strike-slip motion of 3 ± 1 mm/yr on or near the YZS, extending from the Karakorum fault on the west to the Yadong-Gulu rift on the east contributes to the east-west extension in southern Tibet.

[67] Arc-normal convergence directions along the Himalaya are in the directions of N25°E, N12°E, and N12°W on the western, central, and eastern segments of the Himalaya, respectively, consistent with slip vectors of moderate earthquakes along the Himalaya. We find slip rates of India beneath southern Tibet to be 17 ± 1, 12.2 ± 0.4, and 19 ± 1 mm/yr on the western, central, and eastern segments, respectively. The lower rate in the central Himalaya segment is significant, and the variation in slip rate results from the geometries of the two most active rift systems, the Yadong-Gulu rift and Thakkola graben, in southern Tibet and the concentrated extension rates there.


[68] We thank Wang Wenying, Ju Tianyi, Su Shengrui, You Xinzhao, Qiao Xuejun, Hilary Fletcher, Fu Zhongtan, Ciren Wangdui, Rebecca Bendick, and many others for help in the Tibet field campaigns. We thank Paul Tapponnier and an anonymous reviewer for constructive and helpful reviews. This research was supported by NSF grants EAR 9521922, EAR 9725563, and EAR 0125995.