The Tibetan Plateau has a long history of deformation within the last 50 million years e.g.. The reliability of seismic anisotropy measurements is a challenging issue as it is essential to identify the tectonics, coupling–decoupling of the crust-lithospheric mantle, multi-layered anisotropic modelling, and active seismicity in relation to the type of deformation and possible flow patterns, which are still a matter of debate in understanding the formation process and future challenges of this active region.

Lattice-preferred orientation (LPO) of olivine mineral in the mantle as a result of plate interactions is controlled by various geodynamic processes and is considered to be the main cause of the shear wave splitting observations on the teleseismic S and SKS waves. Deformation in the upper mantle generally takes place through two processes: diffusion and dislocation creep under favourable conditions. The dislocation creep process, which is the creeping motion of crystal dislocation, is considered to be the leading cause of mantle anisotropy . It can be caused by either high-stress conditions or large grain size or both, but the nonlinear increase in the strain rate is independent of the grain size . This type of deformation is expected to occur at a depth range of less than 400 km e.g. where olivine is the most common mineral, and hence LPO development and observed anisotropy mainly represents the upper 400 km of the mantle .

Several observations on seismic anisotropy have greatly contributed to elucidating these deformation patterns in relation to the past and present geodynamic activity of the region. Generally speaking, SKS splitting analyses are the most diagnostic, quick, and well-established way of detection and quantification of seismic anisotropy. The SKS phase does not propagate as an S wave in the liquid outer core and refracts from a P wave into an SV (radially polarized) wave when entering the receiver-side mantle. Hence a recorded SKS phase at the surface is not influenced by the source-side anisotropy. The main disadvantage of using the SKS phase in splitting measurements is that finding good-quality observations is restricted by several parameters, i.e. epicentral distance and propagation direction of the event, and therefore the measurements need to be supplemented with other phases (e.g. ScS and direct S) that can provide better azimuthal coverage. However, employing such additional phases may introduce contamination due to the source-side anisotropy. Splitting of shear waves is similar to the birefringence phenomena in optics. Shear waves split into fast and slow components when they pass through an anisotropic medium. In such a situation, we obtain particle motion (e.g. elliptical, cruciform) with different shapes that depend on the anisotropy along the ray path. If the anisotropy is the only cause of splitting, then the observed shear wave (fast or slow) can be rotated in such a way that two very similar phases are seen, apart from scaling and a time delay between them . Resultant splitting parameters, ϕ and δt, indicate the rotation angle in relation to the flow direction and shearing or extension along the ray path under the assumption of LPO in the upper mantle (a.k.a. fast polarization direction or FPD), and to the strength and thickness of the anisotropic layer (a.k.a. delay time), respectively. Splitting measurements from the Himalaya–Tibet collision zone have long been explained by the presence of a single homogeneous layer with a horizontal axis of symmetry e.g..

The use of direct S waves of earthquakes at teleseismic distances (30–90∘) can provide complementary splitting measurements to SKS measurements as this helps in establishing a more complete database of anisotropy that will be inferred from good-quality S wave signals from an enhanced azimuthal distribution relative to SKS splitting only. This is crucial for the Indian subcontinent where SKS measurements are skewed towards eastern azimuths, and very few SKS measurements have been obtained due to temporary deployments see. However, the major problem in including direct S waves in splitting measurements is the contamination of the S wave signals due to the influence of anisotropic structures existing within the source-side region. have recently shown that this problem can be overcome using an array-based approach, known as the reference station technique (RST). The method assumes an identical source-side anisotropy effect at two closely located stations (reference and target stations) with small differences in epicentral distances. In this case, optimum splitting parameters can be estimated by searching for receiver-side correction parameters for the target station that result in maximum similarity to the S wave signal corrected for previously known receiver-side anisotropy beneath the reference station in a grid search scheme. Signals used for that comparison are those of the reference station previously corrected for known reference receiver-side anisotropy and of the target station whose receiver-side splitting parameters are desired to be estimated. In this technique, we utilize seismic anisotropy parameters, which were previously inferred from the SKS measurements by as the reference knowledge of the receiver-side anisotropy beneath the reference station. The RST has been successfully tested through both synthetic and observed data collected along the northeastern and southwestern parts of the Tibetan Plateau and the Hellenic Trench in the eastern Mediterranean e.g.. The present study focuses on the southeastern part of Tibet near Namcha Barwa (Fig. 1). The study region is located between and around the Indus–Tsangpo suture zone (ITSZ) and Bangong–Nuijiang suture zone (BNSZ). Our major motivation is to calculate S-wave-derived seismic anisotropic parameters that may have a potential link to tectonic setting and deformation history with the help of a correlative analysis of resultant anisotropy observations with absolute plate motion (APM) directions, GPS measurements, and the structural and topographic features. Our results contradict previous interpretations of an isotropic Indian lithospheric mantle and add new constraints in understanding the types of deformation and their causes in the region.

The formation of the Tibetan Plateau and the Himalayan mountain belt is due to collision and post-collision processes of the Indian and Eurasian plates starting at around 50 Ma . Underthrusting of the Indian lithosphere beneath the Eurasian lithosphere has been proposed to be the main reason for the formation of the Himalayan and Karakorum ranges along with the formation of the central Tibetan region . The underlying reason for the development of the northern and eastern Tibetan Plateau, however, remains enigmatic . discuss the development of the northern Tibetan lithosphere as an accreted one. argue that the Tibetan Plateau evolved due to the subduction of the Indian lithosphere beneath Eurasia, which is also responsible for the thickening of the Tibetan crust and afterwards the extrusion of the Tibetan lithosphere towards the east.

Various models have been developed regarding the deformation of Tibet , but no single model can explain all of it. The debate regarding the crust and mantle deformation patterns and ongoing geodynamics has not been settled. Recent work by and suggests a model involving multistage subduction of the Tethys oceanic plate and the Indian Plate below the Eurasian Plate resulting in a highly heterogeneous and anisotropic lithosphere. The Tibetan and Himalayan region is mainly dominated by thrust and strike–slip faulting. Suture zones are extended in the E–W direction and take a sharp turn around the eastern Himalayan syntaxis (EHS; Fig. 1). Strike–slip faulting becomes more dominant to the east of the EHS. Figure 1 shows that the eastern portion of the subducting Indian Plate is found adjacent to the EHS where the structural and topographical features take a sharp trend from nearly W–E striking to N–S striking.

In this study, we examine a total of 5285 waveforms with the direct S waves extracted from 161 teleseismic events with magnitudes ≥ 5.5 within an epicentral distance range from 30 to 90∘. The teleseismic events used in this study are recorded at 47 seismic stations of the XE network, which operated between 2003 and 2004 . Of those 47 stations, we selected for use as reference stations only those 36 seismic stations where we have knowledge of seismic anisotropy inferred from SKS splitting measurements performed by . Prior to the data analysis, we remove the instrument response from the original seismograms to overcome biases that can depend on the potential use of different stations (at reference and target sites). At the stage of the preprocessing, a band-pass filter between 0.03 and 0.2 Hz is performed to enhance S signals and resample the seismograms at 20 samples per second to avoid aliased signals and to reconstruct the waveforms in the appropriate frequency range. Signals with possible contamination with other phases, such as ScS, SKKS, and SKS, are omitted from the analysis. We select only those waveforms, which have ≥ 2.5 signal-to-noise ratio (SNR) on the transverse and radial components for further analysis. The selection of the waveforms is achieved by performing a manual visual inspection that allowed only 40 % of the direct S waveforms. We begin data analysis by determining station pairs over the entire area. We form the station pairs by selecting the same earthquake event recorded at both (reference and target) stations. and successfully applied the RST to regional arrays with an interstation distance less than 300 km. By taking 300 km interstation spacing as the limit in a similar fashion, we have formed 22 816 station pairs with four horizontal components available at reference and target stations out of 35 649 possible station pairs; these are based on 161 teleseismic events prior to the application of the technique. To minimize the effects of the coda waves and converted phases, we use a 45 s time window starting 15 s before the theoretical onset of the direct S waves on the basis of the IASP91 1-D radial earth velocity model of . This excludes the undesired effect of crustal S multiples in the thick Tibetan crust.

The approach used in the present study avoids the source-side anisotropy by minimizing the misfit function between the corrected seismic waveforms at the reference and target stations. At the first stage, an inverse splitting operator depending on a backward angular rotation with two horizontal components, a time shift, and the reversal of the back angular rotation are employed to correct the reference station for known receiver-side anisotropy (generally inferred from SKS splitting analyses) when estimating the direct-S-derived individual splitting parameter . Following the correction of the reference station, S signals are corrected for splitting parameters in a grid search manner at the target stations. Corrected S wave signals at reference and target stations are compared to each other for each pair of splitting parameters. Such comparison also allows for time shifts and amplitude corrections to account for the lateral heterogeneities and differences in site response between these stations by optimizing the time shift (Δt) and amplitude factor (a). First, we assign splitting parameters that minimized the misfit function simply representing the difference between the corrected reference and target station traces as optimum splitting parameters for the receiver-side beneath the target station at a given station pair. Later, taking the average of all optimum splitting parameters estimated at station pairs related to a given target station are considered representative of a given event. In the end, station-pair averaged splitting parameters are averaged over all events to estimate the final splitting parameters at each given target station.

Here we should note that our approach initially depends on the knowledge of receiver-side seismic anisotropy that can be most likely inferred from directionally averaged SKS splitting parameters at a given reference station. Conventional SKS splitting measurements are performed under the assumption of a single-layer anisotropic structure with a horizontal axis of symmetry. However, beneath the regions with complex anisotropic structures, for instance, in the case of a well-developed continental lithosphere with a dipping axis of symmetry (i.e. stable cratonic regions, see ) or an existing double-layer anisotropy , significant directional variation of apparent splitting parameters will likely be expected. In such regions, the average value of splitting parameters as reference knowledge of seismic anisotropy cannot be representative of the events from different directions, thus making resultant average S-derived splitting parameters misleading in our method since complicated anisotropic structures likely introduce a similar influence on both the SKS phase and direct S waves. However due to the fact that our approach certainly provides more splitting observations from an increased amount of back-azimuths, these new directionally enhanced apparent S-derived splitting parameters help in resolving the actual orientation of the anisotropic structure by using more sophisticated modelling strategies, which is not within the scope of the present work.

The RST relies on two important underlying assumptions: (i) the ray path at two stations can be considered equivalent in the deeper mantle part and near the source-side region due to the fact that the distance between receiver and target stations is small (< 300 km) compared to the epicentral distance; (ii) waveform differences between the receiver and target stations are only due to differences in anisotropic structure after correcting any waveform differences in time and amplitude presumably due to the lateral heterogeneities and differences in site response between these stations. Any potential difference between the thickness of the crust and sedimentary layers will also cause the timing and amplitude of converted phase, but showed, numerically, its influence on expected splitting parameters would be negligible. During the application of the technique, we let ϕ and δt vary from 0 to 180∘ with an increment of 1∘ and from 0 to 3 s with an increment of 0.05 s, respectively. We perform an inverse F test error analysis for uncertainty estimates of obtained splitting parameters. In this process, we check the reliability of the individual splitting parameters by comparing variation in the residual energy distribution away from the minimum with the variation according to the preset confidence level of 95 %. At this stage, the number of degrees of freedom in the data and unknown model parameters becomes crucial. According to 1 degree of freedom is set to 1 s and considering two horizontal components, the number of degrees of freedom becomes 2 times the data length, which could be considered a typical value for teleseismic data. In our case, however, the number of unknown model parameters is four at the minimum point (ϕ and δt, isotropic delay, amplitude correction factor) or two at any given splitting parameters tried in the grid search, reducing the number of degrees of freedom by four or two, respectively. Estimating the number of degrees of freedom is a challenging task. For an appropriate uncertainty analysis, the assumption of band-limited Gaussian noise is required to be justified as reported by . Thus, taking a fixed value for the degrees of freedom as performed in this study will allow us to compare the reliability of different individual splitting estimates rather than the absolute value of the error bounds.

An example of the basic steps of the RST can be found in Fig. 3 for target stations ES01 and ES35, respectively. Figure 4 present the examples of the obtained splitting parameters at target stations ES19 and ES33, where null or no measurements are reported by . Null splitting may be observed for three reasons: (i) if the incoming polarization direction below an anisotropic layer is parallel to the fast or slow axis; (ii) if the region is isotropic in nature due to complex anisotropy e.g.; (iii) if the region itself is isotropic in nature. Following and , we use the F test with the null-split rejection criteria to be able to avoid the contamination of the null measurement with the good splitting measurements. In this process, we calculate the theoretical residual energy under the assumption of null measurements and compare this with the observed residual energy at the minimum. Figure 5 shows two examples of null-splitting measurements at target stations ES29 and ES37, respectively. To ensure the stability of the results, we perform a stepwise quality assessment criterion before calculating the average splitting parameters at each station. To achieve this aim, we considered only those waveform pairs that have (i) a normalized residual energy (ΔE) smaller than 0.5; (ii) an amplitude correction factor parameter (a) in between 0.4 and 0.6; and (iii) a 95 % confidence level for null-splitting rejection. We rejected the waveform pairs that have a ϕ error greater than 25∘ and a delay time error greater than half of the delay time itself. After these quality assessments, we are left with only 3231 waveform pairs. At this stage of the processing, we perform another visual inspection to enhance the quality of our estimates, yielding only 501 very high-quality waveform pairs. These final waveforms show clear splitting and are free of any distortions due to signal processing. The remaining 501 waveform pairs are extracted from only 25 teleseismic events (Fig. 6) and are used to calculate the average splitting parameter at each station. The list of these 25 teleseismic events is provided in Table 1. We apply the Von Mises approach to calculate the circular mean at each target stations for ϕ and an arithmetic mean is used for δt.

Our splitting observations using direct S waves add new constraints in understanding the deformation pattern and its causes in the southeastern Tibetan region near Namcha Barwa. We list the main concluding remarks from the present study as follows:

The observed splitting analysis suggests a highly deformed crust and lithospheric mantle.