Numerical simulation of contemporary kinematics at the northeastern Tibetan Plateau and its implications for seismic hazard assessment

. The slip rates of active faults in the northeastern Tibetan Plateau (NETP) require clariﬁcation to understand the lateral expansion of the Tibetan Plateau and assess the seismic hazards in this region. To obtain the continuous slip rates of active faults at the NETP, we constructed a three-dimensional (3D) numerical geomechanics model that includes a complex 3D fault system. The model also accounts for the physical rock properties, gravity ﬁelds, fault friction coefﬁcients, initial stress, and boundary conditions. Following this, we present the long-term kinematics of NETP based on the horizontal and vertical velocities and fault slip rates acquired from the model. The fault kinematic characteristics indicate that the Laohushan, middle–southern Liupanshan, and Guguan–Baoji faults, as well as the junction area of the Maxianshan and Zhuanglanghe faults, are potential hazard areas for strong earthquakes. However, as these faults are currently in the stress accumulation stage, they are unlikely to cause


Introduction
The northeastern Tibetan Plateau (NETP) is the growth front of the Tibetan Plateau (TP). Modern geomorphology and tectonic features of the NETP are thought to be formed by the expansion of the TP toward its periphery, which has been ongoing since the initial collision of the Indian and Eurasian plates Patriat and Achache 1984;P. Zhang et al., , 2014. Having experienced strong Cenozoic deformation, crust in this area has developed a complex fault system with several large and deep faults, such as the Haiyuan fault (F1), West Ordos fault (F2), West Qinling fault (F3), and East Kunlun fault (F4), that divide the NETP into the Alxa, Ordos, Qilian, Qaidam, and Bayan Har blocks Fig. 1). These faults are characterized by intense tectonic movements and seismic activity (Zhang, 1999;Zheng et al., 2016b). At least five earthquakes with magnitudes of ≥ M 8, such as the 1654 M 8.0 Tianshui, 1739M 8.0 Pingluo, 1879M 8.0 Wudu, 1920M 8.0 Haiyuan, and 1927 Gulang earthquakes, have occurred in this area and caused huge losses of life and property (Fig. 1). Because the generation and magnitude of an earthquake are closely related to fault activity, long-term fault slip rate plays a key role in medium-and long-term seismic hazard assessment (Ding et al., 1993;Xu et al., 2018). For example, combined with coseismic displacements, long-term fault slip rates can be used to calculate the earthquake recurrence interval (Shen et al., 2009) and assess the magnitudes of potential earthquakes (Bai et al., 2018;Hergert and Heidbach, 2010). Moreover, the spatially continuous fault slip rates that NETP lacks can also be used to reconstruct the tectonic evolution of this area and provide important insights into the lateral expansion pattern and deformation mechanisms of the TP (Royden et al., 1997;Tapponnier et al., 1982;Zhang et al., 2004).
Although there has been extensive research on fault slip rates in the NETP using geologic C. Li et al., 2009;Matrau et al., 2019;Wang et al., 2021; or geodetic (Hao et al., 2021;X. Li et al., , 2021a approaches, a systematic mismatch is typically found between the slip rates calculated from geological records and those obtained from geodetic inversion, owing to the limited assumptions behind these two methods. For example, the geological slip rates only represent the activities of one fault branch measured in a fault zone, which always consists of several branches. They are usually lower than the geodetic slip rates as a whole if a rigid block assumption is adopted in the geodetic inversion process (Shen et al., 2009). However, several crustal deformation studies conducted in TP have demonstrated that the internal block deformation in the NETP cannot be ignored (Royden et al., 1997;Zhang et al., 2004;Y. Li et al., 2017. Numerical modeling provides a powerful tool for studying large-scale crustal kinematics (Hergert and Heidbach, 2010;, as well as a comprehensive three-dimensional (3D) view of fault activities with a spatially continuous distribution. High efficiency and accuracy have made numerical modeling a widespread technology in the geosciences, especially for the study of kinematics and dynamics of the NETP (Pang et al., 2019a, b;Sun et al., 2018Sun et al., , 2019Zhu et al., 2018;Xiao and He, 2015). However, these previous numerical models have been either two-dimensional (2D) or 3D with extremely simplified fault planes. To our knowledge, there is not currently a 3D geomechanical model that considers the complex 3D fault system in the NETP. Therefore, detailed kinematics of the crust and faults in the NETP remain unclear.
In this study, a comprehensive 3D geomechanical model of the NETP was constructed with detailed complex 3D fault geometries, heterogeneous rock properties, and reasonable initial crustal stress. After calibration with modelindependent observations, the results of the geomechanical model, such as the horizontal crustal velocities and spatially continuous slip rates of major faults, were presented. Based on these results, we summarized the long-term crustal deformation characteristics in the NETP. Finally, we assessed the seismic hazards of major faults in the study area and suggested that the Jinqiangshan-Maomaoshan fault has the potential for an M S 7.1-7.3 earthquake in the coming decades.
2 Model concept and input

Model geometry
The proposed 3D geomechanical model is a rectangular Cartesian 3D block with an E-W length of 654 km (101-108 • E), N-S length of 777 km (33-40 • N), and thickness of ∼ 80 km. The topography of the model surface is based on GTOPO30 elevation data, which have a resolution of 30 arcsec (approximately 900 m). The model consists of four layers: (from top to bottom) the upper crust, middle crust, lower crust, and upper mantle. The geometric data of the layer interfaces were derived from CRUST1.0 (Laske et al., 2013).
Based on their depth, the faults of the model can be categorized into lithospheric and crustal faults. Lithospheric faults (i.e., F1, F2, F3, and F4) cut through the Moho and reach the bottom of the model (Zhan et al., 2005;Zhao et al., 2015;Fig. 2a, b; Table 1). All other faults are crustal faults that terminate in the upper, middle, or lower crust (Yuan et al., 2002b(Yuan et al., , 2003Lease et al., 2012;Meng et al., 2012;Wu et al., 2020). The fault traces were modified from Xu et al. (2016), and the attitudes of these faults were summarized from previously published data, including surface fault surveys and deep seismic profiles (Table 1). In the NETP, many geophysical investigations found that low-velocity bodies exist in the middlelower crust and deformations in the upper crust are not coupled to those in the middle-lower crust (Bao et al., 2013;Wang et al., 2018;Ye et al., 2016). Therefore, we introduced a contact surface between the upper and middle crusts into the model and treat it as a decollement layer that makes the upper crust decouple from the middle crust and allows the upper crust to slide freely along the contact surface based on the stress conditions. The model was meshed using tetrahedron elements. The element size was 1-2 km near the faults and increased to ∼ 10 km at the model boundary. The model contained a total of 8 463 583 elements (Fig. 2a).

Rock properties
To obtain the time-independent contemporary background crustal stress of the NETP, we assumed that the rock rheology in the model was linear elasticity, consistent with the observations that the large-scale continental crust always exhibits elastic-brittle behavior rather than elastic-viscous behavior (Armijo et al., 2004;Hubert-Ferrari et al., 2003). The model was divided into three major tectonic units -the Alxa, Ordos, and TP block regions. The portion of the TP Block region located in the study area consisted of the Qilian, Qaidam, and Bayan Har blocks. The rock properties of each block included the elastic parameters (i.e., the Young's modulus, density, and Poisson's ratio) of the upper, middle, and lower crust, as well as those of the upper mantle (Fig. 2a). The den- sity and Poisson's ratio were derived from CRUST1.0 (Laske et al., 2013). The adopted Young's modulus was static and converted from the dynamic elastic modulus using the empirical equation of Brotons et al. (2016), which was calculated based on the P-wave velocities, S-wave velocities, and densities derived from CRUST1.0. The elastic parameters used in the model are listed in Table 2.

Friction coefficient
In our model, frictional faults obeyed the Mohr-Coulomb friction law: where σ s is the shear stress on the fault surface at time of rupture, C 0 is the cohesion, µ is the coefficient of friction, σ n   is the normal stress on the fault surface, P f is the pore pressure, and µ is the effective coefficient of friction of the fault surface when accounting for the pore pressure. The cohesion C 0 of the rocks was assumed to be negligible (Jamison et al., 1980). The friction coefficient of the fault surface is important to the kinematics of a fault, but it is complex and varies in time and space. The exact magnitude of the friction coefficient can be affected by various external factors, including fluid, temperature, stress states, and fault slip rate (Zhu et al., 2009). Therefore, it is challenging to obtain the precise friction coefficient of a fault. However, the results of many studies showed that large faults generally have low effective friction coefficients . For instance, the Haiyuan Fault has friction coefficients as low as 0.05 , and faults on the eastern margin of the TP have friction coefficients as low as 0.02 (H. Li et al., 2015;X. Li et al., 2021b). Simulations were performed with a series of friction coefficients (0-0.1), and the results were compared with GPS observations (Cianetti et al., 2001). The results showed that setting a friction coefficient of 0.02 for all faults yields the smallest fitting error (Fig. 3a). To minimize the fitting error, localized adjustments were made for the friction coefficients of faults F1 and F3, which are large strike-slip faults.
The final friction coefficients of F1, F3, and all other faults were 0.01, 0.1, and 0.02, respectively, with a fitting error of 0.0536.

Initial stress state
The initial crustal stress affects the state of stress acting on a fault, which further controls the kinematics of the fault via the Mohr-Coulomb friction law. Therefore, the selection of appropriate initial stress is important when studying fault slip rates based on geomechanical models. The initial stress state most commonly employed in previous numerical modeling studies of the NETP is the uniaxial strain reference state (Zhu et al., 2016), which is based on the boundary condition that no elongation occurs in the horizontal direction, and the strain only occurs in the vertical direction. In this stress state, the ratio (k) of mean horizontal stress to vertical stress is only dependent on the Poisson's ratio: where S H , S h , and S V are the maximum horizontal, minimum horizontal, and vertical stress, respectively, and v is Poisson's ratio. For the typical v-value of 0.25, Eq.
(2) gives k = 1/3, which implies that the vertical stress acting on the rock mass far exceeds the horizontal stress, and the crust is always in a normal faulting or extensional stress regime. However, this assumption contradicts the thrust and strike-slip stress regimes common in the crust. Furthermore, k values obtained globally from in situ measurements greatly exceed one-third, even in extensional tectonic environments . Based on a spherical shell model, Sheorey (1994) proposed a method for estimating the initial crustal stress, which accounts for the curvature of the Earth, the properties of crustal and mantle materials, temperature fields, and other thermally dependent properties, as shown in Eq. (3): where E is the Young's modulus (GPa) and z is the depth (m).
Because the k values predicted by this method are consistent with those obtained from the Continental Deep Drilling Program , Eq.
In this study, we used Eq.
(3) to calculate the initial stress state for our model. The exact procedure for obtaining the initial stress proposed by Sheorey (1994) has been described previously by Hergert (2009) in detail. Figure 3b shows that the modeled initial stress is in agreement with the theoretical result given by Eq. (3).

Kinematic boundary conditions
The lateral boundary conditions of our model were constrained by the GPS velocity field data of Wang and Shen (2020) and assumed to be constant along depth . The top surface of the model was configured as a free boundary, whereas the bottom surface slid freely in the horizontal direction, with a vertical velocity of 0. The detailed boundary conditions are shown in Fig. 4. For the calculation, we used the finite-element software Abaqus™ because of its powerful nonlinear processing capabilities.
The model time was set as 500 kyr, as required to generate a proper contemporary state of stress, and deformation was modeled until the accumulated displacements at the boundaries were propagated into the model.

Horizontal crustal velocities
The calculated horizontal crustal velocities in the study area are shown in Fig. 5. The western and central parts of the study area moved in the NE-E and near-E-W directions, respectively and the eastern part gradually changed in motion  toward the SW or SE-E directions. Therefore, the study area was characterized by clockwise crustal motions. The model showed that the rates in the southwestern part of the study area were high, whereas they were low in the northeast. The movement rates of the Alxa Block in the north and Ordos Block in the east were as low as ∼ 4-6 mm yr −1 . The Bayan Har Block in the south had the highest movement rate at 13-14 mm yr −1 . Both the Qaidam Block and the Qilian Block exhibited a higher movement rate in the west than that in the east, decreasing from 12 to 9 mm yr −1 and from 10 to 8 mm yr −1 , respectively. In addition, Fig. 5 also shows that large strike-slip faults accommodate velocity gradients between adjacent blocks. For example, the modeled velocity map can be divided into several parts by the F1, F3, and F4 block boundary faults, across which the movement rates of the crust change remarkably, especially at fault F1. The movement rate on the southern side of the Qilian Block was 9-11 mm yr −1 , whereas that on the northern side of the Alxa Block was 4-6 mm yr −1 . A different rate of ≥ 3 mm yr −1 between the two blocks was accommodated by F1.

Fault slip rates of the main faults
The dominant depth range of seismic activities in the study area is 5-15 km, with a few events at depths up to 20 km (L. Li et al., 2020;. As mentioned in the introduction, seismic activities are closely related to fault activities. To assess the seismic hazard reasonably on the faults in the NETP, we focused on the fault kinematics in the brittle crust and extracted the kinematic characteristics up to a depth of 20 km for the main faults from our 3D geomechan-

Slip rate of F1
From west to east, F1 consists of the Lenglongling, Jinqianghe, Maomaoshan, Laohushan, Haiyuan, Liupanshan, and Guguan-Baoji faults (Fig. 1). Two earthquakes with magnitudes M ≥ 8.0 have occurred along this fault: the 1920 M 8.0 Haiyuan Earthquake and 1927 M 8.0 Gulang Earthquake, which formed 220 and 120 km long surface rupture zones, respectively (Guo et al., 2020;Liu-Zeng et al., 2015;Zhang et al., 1987). The kinematic state of F1 was extracted from the 3D geomechanical model, as shown in Fig. 6. On the western segment of F1 (Lenglongling, Jinqianghe, and Maomaoshan faults), the slip rates increased from 2.0 to 4.0 mm yr −1 from west to east and continued to increase further east (Laohushan Fault and western segment of the Haiyuan Fault), reaching a maximum of 4.5 mm yr −1 . The slip rates of F1 decreased in the middle of the Haiyuan Fault. The eastern segment of the Haiyuan Fault had a slip rate of ∼ 3.5 mm yr −1 . The slip rate of the Liupanshan Fault decreased in the SE direction, with a minimum of ∼ 2.0 mm yr −1 ; on the Guguan-Baoji Fault, the slip rate was lower than 1 mm yr −1 (Fig. 6a). The slip rates obtained by our 3D geomechanical model for the Haiyuan Fault were much lower than earlier estimates (8.0-12 mm yr −1 ; Burchfiel et al., 1991;Lasserre et al., 1999;Zhang et al., 1988) but similar to more recent estimates (3.2-4.5 mm yr −1 ; Li et al., 2009;Matrau et al., 2019;Y. Li et al., 2017). The disagreements with earlier geological estimates may be related to the timescale and the extent to which faults have been studied (C. Li et al., 2009).
Although F1 is predominantly a left-lateral strike-slip fault, it also has a thrust component (Fig. 6b). The Lenglongling, Jinqianghe, Maomaoshan, middle Haiyuan, and Liupanshan faults displayed rakes ranging from 10 to 20 • , whereas the Guguan-Baoji Fault had a rake varying from 40 to 50 • , indicating that they are oblique thrust faults. The Laohushan and western and eastern Haiyuan faults had rakes below 10 • , showing that left-lateral strike-slip faulting is dominant at these faults. Figure 6c and d show the continuous slip rates of F1 along its strike and dip, respectively. The slip rates along the strike (Fig. 6c) were similar to the total slip rates (Fig. 6a). The Laohushan and Haiyuan faults had the highest slip rates on F1. Conversely, the Lenglongling, Jinqianghe, Maomaoshan, and Liupanshan faults had high dip-slip rates.

Slip rate of F2
The F2 fault consists of the Zhuozishan, Huanghe, Luoshan, and Yunwushan-Xiaoguanshan faults (Fig. 1). Figure 7 shows that right-lateral strike-slip faulting is prevalent across the entire F2 fault. However, the magnitude of the dip-slip component varied from one location to another. The Zhuozishan Fault is an oblique-slip reverse fault with slip rates ranging from 0.8 to 1.6 mm yr −1 . The Huanghe Fault had a slip rate of 1.6-2.6 mm yr −1 . Its northern segment is an oblique-slip normal fault, whereas its southern segment displayed almost no dip-slip component. The Luoshan and Yunwushan faults had slip rates ranging from 2.6 to 3.0 mm yr −1 and are dominated by right-lateral strike-slip faulting. The Xiaoguanshan Fault is an oblique-slip reverse fault with slip rates ranging from 3.0 (northern end) to 1.4 mm yr −1 (southern end).

Slip rate of F3
F3 includes the Daotanghe-Linxia and West Qinling faults (Fig. 1) and is primarily a left-lateral strike-slip fault, as shown in Fig. 8. The western Daotanghe-Linxia Fault is an oblique-slip reverse fault, with slip rates ranging from 0.8 to 1.8 mm yr −1 . The West Qinling Fault had slip rates varying from 1.8 to 2.8 mm yr −1 , but each segment of this fault exhibited slightly different kinematics. The West Qinling Fault is an oblique-slip reverse fault west of Zhangxian but is dominated by left-lateral strike-slip faulting in the Zhangxian-Tianshui-Baoji region. The West Qinling Fault becomes an oblique-slip normal fault only near Baoji. The kinematics of other faults in the study area are not described in this study. Their horizontal velocities are listed in Fig. 9. Figure 10a shows the distribution of vertical velocities in the study area. In our model, surface subsidence was observed only in the Yinchuan Basin west of Huanghe Fault, in the Or-dos basin in the east, and in the Sikouzi Basin in the southern Ningxia arc tectonic belt. Most of the subsidence rates ranged between 0 and 0.2 mm yr −1 , except in the center of the Yinchuan Basin, which exhibited subsidence rates varying from 0.2 to 0.4 mm yr −1 . Based on paleomagnetic studies, the Sikouzi Basin had a subsidence rate of 0.22 mm yr −1 during the Pliocene , whereas the subsidence rate of the Yinchuan Basin has been 0.32 mm yr −1  since the Middle Pleistocene (Ma et al., 2021). Both values are consistent with those derived from our simulation.

Vertical velocities
Most other parts of the study area exhibited uplift, albeit at low rates (generally less than 0.2 mm yr −1 ). Most areas with high uplift rates (0.8-1.0 mm yr −1 ) were in the Qilian Block, such as the Lenglinglong and Jinqianghe faults, the southern side of the middle Haiyuan Fault, the western side of the Liupanshan segment, and the southern side of the Lajishan Fault ( Fig. 10a). High vertical fault slip rates also appeared on the F1 and F3 thrust faults that form the north and south boundaries of the Qilian Block (Fig. 10b). These high vertical fault slip rates were thought to be caused by the compression from the NE expansion of the TP Block and SW subduction of the Alxa Block (Ye et al., 2015). Therefore, outward thrust stacking occurs on the southern and northern boundaries (F1 and F3) of the Qilian Block.

Comparison with previous results
The boundary conditions of our model were derived from the GPS data of Wang and Shen (2020). Figure 11 shows a comparison of the results of our model with the GPS data.
The modeled velocity field is consistent in both direction and magnitude with the GPS-derived velocity field. To further examine the fit between the model results and GPS data, we selected a NE-SW profile that crosses through the study area (Fig. 11, C-C ) and projected all GPS-observed values within 50 km of both sides of the profile. Figure 12 shows that the modeled and observed values on either end of the profile (i.e., locations close to the model's boundaries) are almost identical. Although there were differences between the modeled and GPS-observed values, the differences were within the margin of error for the GPS data. Therefore, based on comparisons between the modeled and GPS-observed values in the map and profile, our modeled kinematics agree well with the GPS data. Table 3 is a comparison of the modeled slip rates and model-independent geological slip rates compiled from previous studies. From the comparison, our modeled results were consistent with the slip rates obtained by geologic approaches. For example, the modeled horizontal slip rates on F1 (e.g., the Laohushan, Haiyuan, and Liupanshan faults) were similar to the geological slip rates obtained by previous pointwise measurements (Table 3). The modeled slip rates on the F2-3 Luoshan Fault (2.6-3.0 mm yr −1 ) agreed with the geological slip rate (2.2 mm yr −1 ). A good agreement between these two kinds of slip rates also existed on  (Table 3).

Tianzhu seismic gap
The M 8.0 Gulang earthquake occurred in 1927 along the northwestern segment of F1, whereas the M 8.0 Haiyuan earthquake occurred in 1920 on the Haiyuan Fault (Fig. 1). The Jinqianghe, Maomaoshan, and Laohushan faults, which are in the region between the locations of these two strong earthquakes and remained unruptured during both events, are collectively known as the Tianzhu seismic gap (TSG, Guo et al., 2019;Y. Li et al., 2016;Fig. 13a). Based on the simulation, the left-lateral strike-slip rates of the Jinqianghe, Maomaoshan, and Laohushan faults were 3.2-3.8, 3.8-4.0, and 4.0-4.2 mm yr −1 , respectively (Table 2, Fig. 13a). Therefore, all three faults have relatively high slip rates compared with the rest of the study area. Based on the slip rates and other fault data, we estimated the earthquake magnitude based on the energy accumulated during the time elapsed from the previous event (Purcaru et al., 1978) and recurrence intervals (Shen et al., 2009), as shown in Table 4. The Jinqianghe, Maomaoshan, and Laohushan faults can generate M S 7.1, 7.3, and 6.6 earthquakes, with recurrence intervals of 424, 571, and 1910 years, respectively. A total of 675 and 952 years have elapsed since the last rupture along the Jinqianghe and Maomaoshan faults (Gan et al., 2002;Table 4). Therefore, the likelihood of reactivation along these two faults is high. For the Laohushan Fault, the most recent earthquake with surface ruptures is thought to be the 1888 M 6.25 Jingtai earthquake. Because only 133 years have elapsed since this event, the seismic hazard of this fault in the near future is low. Although several researchers suggested that the TSG could be ruptured thoroughly by a M ≥ 8.0 earthquake with a recurrence interval of 1000 years (Chen,  Liu et al., 2018), further work is needed to verify this assumption, especially on the stress state of the faults. Nevertheless, our results together with the previous studies suggest that the seismic hazard caused by a large earthquake in the TSG is high in the next few decades, and careful earthquake monitoring should be conducted in this area.

Seismic gap at the southern Liupanshan and Guguan-Baoji faults
The Liupanshan and Guguan-Baoji faults are jointly affected by three interacting blocks with contrasting velocity fields (Fig. 13b). First, the SE movement of the Qilian Block horizontally compresses this region against the stable Ordos Block to the east and causes strong thrusting motion (Fig. 6a). Second, the Liupanshan Fault zone, which is adjacent to the SE end of the Haiyuan Fault, accommodates the shortening caused by the left-lateral strike-slip motion of the Haiyuan Fault. Third, our modeled results showed that the Yunwushan-Xiaoguanshan Fault has a significant rightlateral strike-slip component (Fig. 7), which contributes to the accumulation of right-lateral shear strain in the Liupanshan and Guguan-Baoji fault zones (Du et al., 2018). Based on the analysis of the velocity field in the region, the Liu-panshan and Guguan-Baoji faults are a prime location for elastic strain accumulation. The distribution of the velocities of the faults are also indicative of stress accumulation in this region (Fig. 7). The northern segment of the Liupanshan Fault had slip rates of 3.2-3.6 mm yr −1 , which dramatically decreased to 2.5 mm yr −1 in the middle and southern segments of the fault. In addition, the slip rate of the Guguan-Baoji Fault was only 0.7 mm yr −1 . The northern part of the Yunwushan-Xiaoguanshan Fault had slip rates of 2.8-3.0 mm yr −1 , which decreased to 1.5 mm yr −1 in the southern segment (Fig. 13b). These changes in the slip rate indicate that the middle-southern segments of the Liupanshan and Guguan-Baoji faults have high slip rate deficits, implying this region accumulates strain. However, in terms of seismic activity, only the northern segment of the Liupanshan Fault has a history of major earthquakes, including an M 7 earthquake in 1219, M 6.5 earthquake in 1306, and M 6.5 earthquake in 1921 (Fig. 13b). Earthquakes stronger than M 6.0 have not been recorded in the middle-southern parts of Liupanshan Fault. Their seismic activity mainly manifests as small and sparse earthquakes, and the most recent significant activation was recorded ∼ 570 years ago (Shi et al., 2014). Minor earthquakes related to the Guguan-Baoji Fault are scarce, and the only notable earthquakes that have occurred near this fault were an M 6.0 earthquake in 1704 and  GG-BJF 0.6-0.8 0.5-0.6 n/a n/a n/a F2-1 ZZSF -(1.2-1.4) 0.2-0.3 n/a n/a n/a F2 YWSF -(2.6-3.0) n/a n/a n/a / F2-5 XGSF -(1.4-3.0) 0-0.4 n/a n/a / F3-1 DTH-LXF 0.8-1.8 0.5-0.8 n/a n/a / F3-2 WQLF 2.0-3.0 0.1-0.9 2.5-2.9 n/a Chen et al. the 600 CE Qinlong M 6.75 earthquake (Shi et al., 2013). Approximately 1400 years have elapsed since the last M ∼ 7 earthquake. Based on the above analysis, we suggest that the southern segments of the Liupanshan and Guguan-Baoji faults are in prime locations for stress and strain accumulation and constitute a seismic gap at the end of the large-scale Haiyuan strikeslip fault zone. Because this region has a history of strong earthquakes, it is necessary to assess the seismic hazard and its urgency. Based on the fault slip rates obtained from our model and fault data in the literature, we estimated that the energy accumulated on the middle-southern Liupanshan and Guguan-Baoji faults during the elapsed time is sufficient to generate M S 7.2 and 7.1 earthquakes, with recurrence intervals of 1397 and 4365 years, respectively (Table 4). Obviously, the elapsed time on the middle-southern Liupanshan (570 years) and Guguan-Baoji faults (1400 years) (see the column "t" in Table 4) are much shorter than their typical recurrence intervals. Therefore, we inferred that the middlesouthern Liupanshan fault and the Guguan-Baoji fault are most likely in a state of stress accumulation, and the likelihood of a large earthquake on these fault segments in the next few decades is low.

Maxianshan-Zhanglanghe fault zone
The Maxianshan Fault near Lanzhou city is a large Holocene strike-slip fault with a thrust component and acts as an important earthquake-controlling fault that affects and constrains the seismicity of this region (Yuan et al., 2003). Different left-lateral strike-slip rates have been reported for this fault: 3.73 (Yuan et al., 2002a), 0.5-1.72 (Song et al., 2006), and 0.93 mm yr −1 (Liang et al., 2008). These discrepancies in fault slip rate may be attributed to the loess that covers the extension of the fault, which obscures the fault traces in many segments and makes it difficult to track its activity. V 1 is the modeled average slip rate of the fault in this study. V 2 is the aseismic creep rate of the fault (Y. . L 1 is the length of the fault (Xu et al., 2016). L 2 is the depth of the seismogenic structure, which refers to the locking depth (Y. . µ is the shear modulus of the rocks (Aki et al., 2002). t is the time that has elapsed since the most recent remarkable earthquake (Gan et al., 2002;Shi et al., 2013Shi et al., , 2014Wang and Liu, 2001). S is the largest maximum coseismic displacement, calculated using the method of Gan et al. (2002). M S is the earthquake magnitude, corresponding to the energy accumulated by the fault between recurrences (Purcaru et al., 1978). T is the recurrence interval of the fault, where T = S/ (V 1 − V 2 ) (Shen et al., 2009). The fault names are defined in Figs. 1 and 13. n/a: not applicable. Figure 11. Comparison of the modeled horizontal velocities and GPS velocities. The red arrows represent the modeled results, and the blue arrows are the GPS measurements (Wang and Shen, 2020). The dashed line is the location of the profile in Figs. 12 and 14.
The modeled results in this study indicated that the leftlateral strike-slip rates of the Maxianshan Fault range from 0.2 to 0.4 mm yr −1 , and the vertical slip rates vary from 0.1 to 0.4 mm yr −1 (Fig. 13c, Table 3). Therefore, the left-lateral strike-slip rates of the Maxianshan Fault may not be as large as previously thought, but they have a relatively large thrust component. The Zhanglanghe Fault is predominantly a rightlateral strike-slip fault with slip rates ranging from 0.6 to 0.8 mm yr −1 (Fig. 13c, Table 3). These slip rates are signifi- Figure 12. Comparison of the modeled horizontal velocities and GPS velocities along the C-C profile in Fig. 11. The red points are model data, and blue circles indicate GPS measurements. The fault names are defined in Fig. 1 (He et al., 1997;Fig. 13c). Given that the recurrence interval of this region is 2250-3590 years and the last event was only 896 years ago (Liang et al., 2008), the near-term risk of a large earthquake in this region is low. Therefore, we speculate that the junction of the Maxianshan and Zhuanglanghe faults is currently in a state of stress accumulation.

Isolated uplift areas and earthquakes
As mentioned above, we considered that earthquakes are less likely to occur on the Laohushan, Liupanshan, and Haiyuan faults in the short term owing to the earthquake recurrence cycle and the elapsed time from the previous earthquake. However, the Haiyuan, Liupanshan, Lajishan, and Daotanghe-Linxia faults are all located near the isolated rapid uplift areas of the Qilian block (Fig. 10a). Many studies have also found that low-velocity bodies are widely distributed in the middle-lower crust of the Qilian Block (Bao et al., 2013;Wang et al., 2018;Ye et al., 2016). The spatial coupling of active faults, isolated uplift areas, and low-velocity bodies is highly similar to the seismogenic conditions elaborated by the "seismic source cavity" model recently proposed by Zeng et al. (2021). That is, during the rapid uplift of the isolated areas (Fig. 10a), the low-velocity bodies in the middle-lower crust easily intruded into the weak space of the crust under differential pressure to form a seismic source cavity. If the isolated uplift areas continue to rise, the seismic source cavity may rise to the shallow part of the crust to intersect with brittle faults, causing strong earthquakes (Yang et al., 2009;Zeng et al., 2021). Therefore, in addition to the Jinqianghe and Maomaoshan faults mentioned above, the Haiyuan, Liupanshan, Lajishan, and the Daotanghe-Linxia faults also have favorable structural conditions for strong earthquakes, although a number of areas have not experienced events in recorded history.

Implication for deformation mechanism of NETP
The deformation of NETP is the result of the combined action of block rotation, faulting, and intra-block strain (Meade and Loveless, 2009). We analyzed four velocity profiles to compare the contributions of block rotation, faulting, and intra-block strain to determine the total deformation of NETP (Fig. 14). The rigid displacements caused by block rotation were calculated according to the Euler pole locations and rotation rates with respect to the Eurasian plate Y. Li et al., 2022), as shown in Fig. 14a-d. The velocity gradient caused by block rotation accounted for more than 80 % of that on the profiles. Obviously, the block rotation should be the primary mechanism for the deformation of the NETP, which is similar to southeastern Tibet (Z. . However, the intra-block strain of the Bayan Har and Qaidam blocks contributed approximately 4 and 3 mm yr −1 shortening in profiles of A-A and B-B ( Fig. 14a-b). The Qilian Block also exhibited a contribution of 2 mm yr −1 shortening in profile B-B but decreased to approximately 1 mm yr −1 in profile C-C (Fig. 14b-c). Therefore, the intra-block strain is still significant for regional de-formation. The boundary faults of the blocks, such as the East Kunlun, Haiyuan, and West Qinling faults, also play an important role in regulating the deformation differences between blocks.
The D-D profile shows that the tectonic deformations of the Yinchuan Basin structural belt slightly differ from those in other profiles. The NE expansion of the TP leads to near-N-S compression on the Yinchuan Basin (Yang, 2018), which causes it to move eastwards faster than the Alxa Block. This manifests as an eastward extension in the Yinchuan Basin. The crustal deformations caused by this process are accommodated by the right-lateral strike-slip of Huanghe Fault (Fig. 14d).

Conclusions
In this study, a detailed 3D geomechanical model of the NETP was constructed based on geophysical, geodetic, and geological data. This model accounted for 3D fault geometries, variational rock properties, a reasonable initial crustal stress, and gravity. Special attention has been given to the evaluation of fault friction coefficients and initial stress to ensure that the model is consistent with the geological conditions as much as possible. In addition, we extracted particular data from the model and obtained the horizontal and vertical crustal velocities of the study area and the horizontal and vertical slip rates of the major faults. The results are consistent with the conclusions obtained by geodesy, geology, paleomagnetism, etc.
Based on the analysis of the kinematics of major faults, we suggest that the Jinqianghe-Maomaoshan Fault will likely experience a M S 7.1-7.3 earthquake in the following decades owing to its relatively high slip rates with an elapsed time close to the recurrence interval. In contrast, the Laohushan and middle-southern Liupanshan faults, as well as the Guguan-Baoji and the junction of the Maxianshan and Zhuanglanghe faults, are thought to have a low seismic hazard in the near future owing to the very short elapsed time since the last significant event, although stress is easily accumulated in these areas. The model also provided information on the deformation mechanism of the NETP. Because of velocity differences between the opposing sides of the Haiyuan, West Qinling, and East Kunlun faults, as well as the relative stability of the Alxa and Ordos blocks, the NE expansion of the TP has caused the fault-separated Qilian, Qaidam, and Bayan Har blocks to extrude in the SEE direction and rotate in the clockwise direction. The block rotation is the primary mechanism for the deformation of the NETP even though intra-block strain and faulting are non-negligible.
Author contributions. LL and XL contributed to the model building. LL conducted the analysis, wrote the paper, and prepared the figures. FY, LP, and JT reviewed and edited the paper.