Extension Discrepancy of the Hyper‐Thinned Continental Crust in the Baiyun Rift, Northern Margin of the South China Sea

At rifted continental margins, it has been widely reported that the amount of extension measured from faulting is less than the one observed from crustal and lithospheric thinning, referred to as “extension discrepancy.” Here, we use observations from high‐resolution seismic reflection data over the Baiyun Rift to explore the discrepancy between upper and whole crustal thinning factors when the crust of the Northern South China Sea margin thinned from 30 to <10 km. We first restore the rift system of the Baiyun Rift in the absence of post‐rift sediments and water loading. Subsequently, we calculate thinning factors based on fault geometries and crustal thickness ratios to compare the deformation of the upper and whole crust. Our results show (1) in the eastern and western basins, upper crustal thinning factors exceed that of whole crust, showing an inverse discrepancy and (2) upper crustal faulting was sufficient to explain the whole crustal thinning in a narrow area of the basin center, indicating no extension discrepancy. The inverse discrepancy in the western basin results from relatively weak ductile shearing in the lower crust, while in the eastern basin, it is caused by intense sequential detachment faulting in the upper crust. Moreover, lower crust exhumation may trigger magma underplating, which further decreases whole crustal thinning factors. Overall, the hyper‐thinning process of the continental crust beneath the Baiyun Rift is dominated by sequential detachment faulting and is characterized by upper crustal thinning factors equaling or significantly exceeding whole crustal thinning factors.

been associated with exhumation of lower crust into hyper-extension area or extensive detachment faulting in the upper crust (Brune et al., 2014;Huismans & Beaumont, 2014;Whitney et al., 2013).
Extension discrepancy had also been observed in the SCS area, including the Pearl River Mouth basin (PRMB) (Bai et al., 2019;Clift et al., 2002;Tsai et al., 2004), the Qiongdongnan basin (QDNB) (Tong et al., 2009;, the Northwest Palawan margin (Franke et al., 2011), and the offshore Vietnam . Most of these results agree with the model of depth-dependent thinning model that the extension of the brittle crust is much smaller than the whole crust, with the exception of the western QDNB which exhibits inverse extension discrepancy .
Baiyun Rift is located in the southern part of the PRMB, where the continental crust thinned sharply from 25 to 7 km over a ∼50 km distance (Clift et al., 2002;Huang et al., 2005;Sun et al., 2010). We designate crustal thicknesses <10 km as hyper-thinned crust, after Ranero and Pérez-Gussinyé (2010). Attempts had been made to quantify the extension amount of the crust beneath the Baiyun Rift, which showed a typical extension discrepancy characterized by increasing thinning factors with depth (Bai et al., 2014;Chen, 2014;Clift et al., 2002;Hu et al., 2009;Liao et al., 2011;Lu et al., 2017;Zhang et al., 2014). Note that previous studies were (1) only carried out in 2D seismic sections across the central Baiyun Rift (e.g., 2002DSRP, Figure 1), (2) based on a classic rifted basin model that the rifting was dominated by high-angle normal faulting, and (3) under an assumption that the observed faulting is close to the total amount of brittle extension. However, newly acquired high-resolution multi-channel seismic data indicate that in the Baiyun Rift was controlled by a combination of high-angle and low-angle faults Yang et al., 2018;. Misinterpretation of the low-angle faulting leads to critical uncertainties of the stretching factors; therefore, the stretching factor of the Baiyun Rift needs to be reconsidered.
Based on the geological interpretation of the 3D seismic data, we restore the initial geometry of the rift system in the Baiyun Rift. Then we quantify the hyper-thinning process of continental crust by comparing the thinning factors that derived from faulting and crustal thickness ratios. We focus on the observational constraints on rift systems of the sedimentary basin and crust rather than modeling the entire lithosphere. This study aims to document: (1) what, if any, is the mechanism of the extension discrepancy; and (2) how brittle and ductile layers of the continental crust interact before the continental crust reaches final breakup.

Geological and Tectonic Setting
The SCS is located at the convergence of the Eurasian Plate, the Indo-Australian Plate, and the Pacific Plate ( Figure 1). With the subduction of the Paleo-Pacific Plate to the Eurasian Plate, the earliest rifting in the South China occurred from the Late Cretaceous to the Early Paleocene, which is related to the slab retreat of the Pacific Plate (Ren et al., 2002;Ru & Pigott, 1986;Yin, 2010). The Mesozoic pre-rift sedimentation and magmatism associated with the mainland created a series of fore-arc basins filled with volcanic and metamorphic clasts in the south of the PRMB (Lester et al., 2014;Li et al., 2012Zhang et al., 2005;Zhang & Wang, 2007;Zhou et al., 1995).
Episodic rifting that resulted in the formation of the Northern South China Sea (NSCS) margin occurred since Paleogene and mostly ceased with the opening of the oceanic basin during Early Oligocene (Figure 2) (Briais et al., 1993;Clift & Lin, 2001;Ding et al., 2020;Flower et al., 1992;Larsen et al., 2018;Ren et al., 2018;Taylor & Hayes, 1983). The stratigraphic record suggests the NSCS experienced three stages of rifting. (1) From the Paleocene to the Early Eocene, initial extension widely occurred and resulted in the deposition of 1,000 m thick lacustrine sediments bounded by high-angled faults. (2) During the Mid to Late Eocene, the rifting direction changed from NE to EW due to the reorientation of the Pacific Plate movement and the collision between the Eurasian-Indian Ocean plates (Briais et al., 1993;Ren, 2018). These events led to an intense extension in the continental margin and a widely development of isolated rift basins across the region (Cullen, 2014). Hyper-extended area with a < 10 km continental crust, including the Baiyun Rift, was dominated by a detachment fault system with large horizontal displacements (>10 km) and low dips (<30°) Yang et al., 2018;. (3) The extension continued until Early Oligocene and led to the final breakup of the continental lithosphere (Briais et al., 1993;Li et al., 2015;Sun et al., 2018). Corresponding to these extensional events, regional unconformities in the NSCS include top basement (TB), syn-detachment (SD) horizon, post-detachment (PD) horizon, and the breakup unconformity (BRU) (Ren et al., 2015). TB and PD mark the onset and cessation of detachment faulting in the NSCS, respectively, and BRU marks the initiation of the seafloor spreading.
The initial crustal thickness of the Cenozoic NSCS had been determined by multiple studies: 30 km based on the crustal thickness of the South China block (Zhou et al., 2012), 30-35 km based on seismic refraction modeling (Hu et al., 2009;Li et al., 2006;Nissen et al., 1995), 33 km in the eastern NSCS while 30.5 km in the western NSCS based on thermal-subsidence modeling (Dong et al., 2020). The episodic Cenozoic rifting formed an uneven thickness of continental crust wedging oceanwards across the NSCS. On the continental ZHAO ET AL.  Wang et al. (2006); ESP-E and ESP-W, Nissen et al. (1995); OBS2006-3, Wei et al. (2011); OBS1993, Yan et al. (2001);2002DSRP, Huang et al. (2005; OBS2006-1, Ding et al. (2012), and OBH1996, Qiu et al. (2001). (b) Isopach map of Eocene strata of the Baiyun Rift constrained by the 3D seismic data range. ①-④ are four sets of basement fault zones in the Baiyun Rift. Sections A-A′ to G-G′, with the exception of C-C′, are extracted from a 3D seismic survey. Section C-C′ is extracted from the 2D seismic profile 2002DSRP.  (Dong et al., 2008;Pang et al., 2007;, as well as the key tectonic events related to the South China Sea based on IODP expeditions 349/367/368/368X (Li et al., 2015;Sun et al., 2018). BRU, breakup unconformity; IODP, International Ocean Discovery Program; PD, post-detachment horizon; SD, syn-detachment horizon; TB, top basement horizon. Note that the horizons TB and BRU may have various ages in different sedimentary basins of the NSCS.
shelf, the present-day thickness of the crust remains 25-20 km, while for the areas further south, the crustal thickness ranges from 20 to 5 km (Chen, 2014;Hsu et al., 2004;Tong et al., 2009;Zhang et al., 2008). In the northeastern NSCS, the crust thins rapidly from 25 to ∼4 km (Lester et al., 2014). Various degrees of thinning suggest the lithospheric deformational architecture and stretching pattern differ remarkably across the NSCS.
The NSCS was previously believed to be a magma-poor margin since no Seaward Dipping Reflectors, which are formed by the stacking of basaltic flows and sediments during the breakup, have been drilled, dredged or observed in the NSCS (Brune et al., 2016;Franke et al., 2014). However, recent studies based on the drilling results of the International Ocean Discovery Program (IODP) Expeditions 367/368/368X suggested that a short-period magmatic event is associated with the final breakup of the continental lithosphere and the formation of the oceanic basin (Ding et al., 2020;Larsen et al., 2018;Sun et al., 2018). Before the continental crust reaches its final breakup, lithospheric hyper-thinning exhibits ductile deformation features (Clift, 2015;Sun et al., 2008;Zhang et al., 2019Zhang et al., , 2020. In addition, in the middle to northeastern area of the NSCS, a high velocity layer (HVL) is widely distributed beneath the lower crust, such as the Dongsha Islands and the Taixinan basin ( Figure 1) (Nissen et al., 1995;Wang et al., 2006;Yan et al., 2001;Zhao et al., 2010). This HVL is believed to be related to a magmatic mafic underplating (Ding et al., 2012;Nissen et al., 1995;Wei et al., 2011;Yan et al., 2001). Hence, the NSCS does not resemble a magma-poor archetype (Ding et al., 2020). The complexity of structures suggests that a hybrid of end-member models is needed to account for the regional tectonic evolution (Cullen et al., 2010).
Baiyun Rift is located in a transition area where the Moho depth changes from 29 to 19.5 km (Huang et al., 2005;Liu et al., 2007;Sun et al., 2008), and the eastern Baiyun Rift is developed above a "hot crust" with the development of HVL (Figure 1). In the Cenozoic, ultra-thick sediments (>10 km) were deposited on top of the hyper-thinned (7 km) crust Yan et al., 2006). Dominated by both high-angled and low-angled normal faulting , the Eocene strata consist of a lower succession of lacustrine sandstone interbedded thin mudstone and an upper succession of lacustrine mudstone as much as 120 m in thickness ( Figure 2). During Oligocene, most faulting stopped and the Baiyun Rift entered into a post-rift stage. The post-rift strata are mainly abyssal deposits, comprised of mudstone and progressively decreasing sandstone. Certain amounts of magmas intruded into the post-rift strata  and deformed early stage rifting patterns. In light of the intensive tectonic-magmatic interaction, related mechanisms of the basin formation include lower crustal flow (Clift et al., 2002;Ding et al., 2009;Huang et al., 2005;Morley & Westaway, 2006), long-lived large-scale faulting (Huang et al., 2005), or mantle upwelling along the pre-existing weak zone .

Data
This study uses high-resolution 3D seismic data that were acquired and processed by the Chinese National Offshore Oil Company (CNOOC). The 3D seismic data were obtained using 3 km long streamers with 240 channels, which allows for a bin size of 25 × 12.5 m, and a vertical seismic resolution of 31 m in the syn-rift strata. Seismic sections cover the entire Baiyun Rift over an area of 11,420 km 2 , and the sections are imaged to a depth of 8 s two-way travel time (TWT), showing an entire stratigraphy of the Baiyun Rift. We extract seven key seismic sections from the 3D seismic volumes in this study. Uninterpreted versions of the seismic sections have been shown in Supplementary Figures S4-S10. Based on the structural framework of the Baiyun Rift built by the previous work , this study further investigates the characteristics of the rifting system in the Cenozoic infillings.
Also, we re-interpret a 2D seismic profile, DSRP2002 (i.e., section C-C′ in Figure 3b), which was acquired in 2002 (Huang et al., 2005). This deep reflection seismic profile reaches a depth of 12 s TWT, which allows imaging the Moho reflections. Previous interpretation of MCS profiles (Gao et al., 2015;Huang et al., 2005;Yang et al., 2018), wide-angle seismic data modeling (Lester et al., 2014;Wei et al., 2011;Yan et al., 2001) (Figure 1), and gravity data analysis (Sandwell et al., 2014) have also provided constraints in determining the structures of the crystalline basement and the Moho beneath the Baiyun Rift.

Terminology
With the application of the McKenzie (1978) model for the lithospheric extension, stretching factors have been defined from changes in crustal thickness (Davis & Kusznir, 2004) and fault geometry (Le Pichon & Sibuet, 1981;Ziegler, 1983). In this paper, we apply both methods based on thickness changes and fault geometry to quantify the upper crustal thinning (i.e., brittle extension), and the method of thickness changes to quantify the whole crustal thinning. Hence, we utilize thinning factors as a proxy for the amount of brittle extension and crust thinning (Table 1).
In the NSCS, high-resolution seismic data allow for the identification of the transition between the brittle and ductile crust (Ding et al., 2020;Gao et al., 2015;Savva et al., 2014). In the Baiyun Rift, the main basement faults dipping <30° usually cut through the upper crust and sole out at a similar depth of 7-8 s TWT in the depocenters, such as f A-1 , f C-1 , f E-9 , and f G-7 (Figure 3). Below this depth is a series of high-amplitude, sub-horizontal reflections. We infer these reflections correspond to the brittle-ductile transition at a depth of 7-8 s TWT. Hence, this thin transition zone where the large-offset basement faults abruptly terminated is defined as the boundary between the upper crust (brittle crust) and lower crust (ductile crust) (ULB).
To be noted, some parts of the ULB cannot be observed. In the area adjacent to the uplifts, such as distances of <20 km in sections A-A′ and B-B′, the discontinuous ULB is due to the limited seismic data coverage (Figure 3a), while in the southern Baiyun Rift the fragmentary ULB is due to magmatic overprints, as shown in section D-D'. We aligned these fragmentary ULB by interpolation.
As the thickness of the brittle crust and whole crust become available, direct estimates for crustal thinning, as initially proposed by Davis and Kusznir (2004), have been used to calculate extension in the brittle crust and the whole crust. Besides, we also use an alternative method based on the normal faulting (Le Pichon & Sibuet, 1981), to compare the extension discrepancy between the brittle crust and the whole crust. To represent graphically, γ (γ = 1-1/β) (Hellinger & Sclater, 1983) is applied to describe the variations between no crustal thinning (γ = 0) and a complete lithospheric breakup (γ = 1). Here, we use γ uc and γ c as the crustal thinning factor that is related to the crustal thinning, and γ f as the thinning factor that is derived from normal faulting. Since the thinning process of the brittle crust is conducted through tectonic faulting, theoretically, the fault-derived thinning factor γ f is equal to the thinning factor of the brittle crust γ uc .

Time-Depth Conversion
A time-depth conversion of the structural framework is carried out based on the interval velocity of the structural sequence (Table 2), after Huang et al. (2005). The lithologies and the physical parameters of the stratigraphic sequences are shown in Table 2. For the mixed lithologies, a weighted sum of the physical parameters is performed. The velocity of the crust is based on the seismic modeling of the OBS1993 (Yan et al., 2001), which intersects with sections E-E′ and F-F' (Figure 1).
The Moho depth in the Baiyun Rift has been determined through P-wave velocity modeling (e.g., Huang et al., 2005), and joint inversion of the seismic and gravity data (Gozzard et al., 2018;Li et al., 2019;Ren et al., 2018). Specially, the time-migrated section C-C′ shows clear Moho reflections with continuous and high amplitude features at a depth of ∼10 s (Figure 3b). We re-process the Moho depth through a joint inversion of gravity (Sandwell et al., 2014) and seismic data, and project the results on the depth sections. The depth of the basement and the crustal thickness revealed by the depth sections are consistent with previous ZHAO ET AL. γ c Thinning factor of whole crust quantifies extension amount of whole crust based on thickness change Equation 3 γ f Fault-derived thinning factor quantifies extension amount of brittle crust based on fault geometry Equations 4-8 studies within an error range of 10% (Huang et al., 2005;Lester et al., 2014;Wei et al., 2011;Yan et al., 2001;Zhang et al., 2008).

Rift Structure Restoration
In the Baiyun Rift, the development of the low angle normal faults (LANFs) strongly deformed the syn-rift strata and crustal structures. Hence, the original geometry of the rift system is crucial for extension quantification. In this study, we apply backstripping to restore the syn-rift sequence in the absence of post-rift sediments and water loading (Watts & Ryan, 1976). In addition, the fault blocks are corrected for rotation until the syn-depositional strata are horizontal. The restored rift system is used for estimating the thinning factor based on fault geometry without the influence of footwall denudation and fault rotation.
We use software Move (by Midland Valley) for the reconstruction of the Baiyun Rift. By following the principle of areal balancing, a basin evolution reconstruction allows estimating the initial thickness of brittle crust before rifting (Dahlstrom, 1969). The principle states in a 2D section perpendicular to the rift trend, the area of the brittle crust remains unchanged before and after rifting/thinning deformation (Kusznir et al., 1995;Roberts et al., 1998). After measuring the length of the top basement (TB) and the area of the brittle crust along the selected seismic sections (Figure 1), the initial thickness of the brittle crust can be estimated as where: Z uc , initial thickness of pre-rift brittle crust; A, present-day area of the brittle crust; and L TB , length of the top-basement horizon.
ZHAO ET AL.
10.1029/2020TC006547 8 of 28  We ignore the lateral velocity changes within the same sequence due to the short section. Lithology percentage is based on previous studies that summarized from 41 wells in the PRMB (Dong et al., 2008) and the unpublished data from 2 deepwater wells acquired by CNOOC. Abbreviations: BRU, breakup unconformity; CNOOC, Chinese National Offshore Oil Company; PD, post-detachment; PRMB, Pearl River Mouth basin; SD, syn-detachment; TB, top basement; ULB, boundary of upper and lower crust.

Table 2 Sequences of Interval Velocity for Converting Time-Migrated Section Into the Depth domain
Multiple studies suggested thickness of pre-rift brittle crust ranges from 9.14 to 12 km (Chen, 2014;Dong et al., 2020;Zhang et al., 2008). A comparison between Z uc and the suggested thickness of 12 km provides further evidence of pre-rift inheritance of the basement crust.

Crustal Extension Quantification
(1) Crustal thinning factor γ uc and γ c The classic method to define the crustal stretching factor is by comparing the thickness of the lithosphere before and after rifting (Davis & Kusznir, 2004). We apply this terminology to quantify the deformation within the brittle crust and the whole crust.
To represent graphically, we use thinning factor (γ = 1−1/β) to present the crustal extension. Hence, the thinning factors of the brittle crust (γ uc ) and the whole crust (γ c ) can be calculated as where: T 0 , thickness of the pre-rift brittle crust; T i , thickness of the post-rift brittle crust; Z 0 , thickness of the pre-rift crust; and Z i , thickness of the post-rift crust.
Previous results in the NSCS determined the pre-rift crust thickness ranges from 30 to 35 km (Li et al., 2006) and an average proportion of brittle crust is 33% (Dong et al., 2020). We compute the thinning factors with Z 0 = 30-35 km and T 0 = 12-15 km for the error range analysis, and choose the nominal values of Z 0 = 30 km (Chen, 2014) and T 0 = 12 km slightly higher than 33% (Dong et al., 2020) but still on the lower range of values explored in this area (Bai et al., 2019). We emphasize that the choice of pre-rift values within the specified ranges does not change the main conclusion of this study (see sensitivity analysis in Supplementary Figure S2 for an exploration of pre-rift crustal thicknesses).
In the eastern Baiyun Rift, we calculate γ uc ' and γ c ' for additional discussions. As observed in the seismic sections, the brittle crust had been thickened by magma addition and embrittlement. For comparison, γ uc is calculated after removing the area of new brittle crust, while γ uc ' is calculated without removing them. Besides, pre-Cenozoic remnants can be observed in the basement with uncertain amounts. We calculate γ c ' under an extreme assumption that pre-Cenozoic materials accounts for 50% of the crystalline basement. Therefore, in the eastern Baiyun Rift, the actual thinning factor of the whole crust should range from γ c to γ c '.
(2) Fault-derived thinning factor γ f We use the basin modeling software Stretch (by Badley Geoscience) to calculate the thinning factor γ f . The extension of the brittle (upper) crust is accommodated by fault-related stretching (Davis & Kusznir, 2004;Kusznir et al., 1987;Weissel & Karner, 1989). Assume faults are planar, the extension amount of a single where: v is the vertical throw of the fault, θ is the fault dip, l is the horizontal heave of the fault, φ is the bedding rotation.
To compare brittle extension with similar estimates for the whole crust, extension from a single fault is first represented by a continuous function and is mapped to a profile of stretching factor β fi as A cosine-squared function is used to calculate the fault-derived stretching factor β fi as where: x, horizontal coordinate; x 0i , fault location coordinate; W, distributed pure-shear width, which is suggested to be ∼100 km in the NSCS (Davis & Kusznir, 2004); and β 0 is determined numerically such that the distributed extension given by Equation 5 is identical to that measured discretely on faults given by Equation 4.
The stretching factor associated with all faults is the product of β fi (x): where: N is the number of the faults in the profile.
To represent graphically: where: α, a scale factor to incorporate uncertainties in sub-resolution faulting. In main texts, we present the results using α = 1.
(3) Sub-resolution faulting examination Previous studies proposed that failure in recognizing the sub-resolution faults in the seismic profiles would result in a 25%-60% mismatch of total extension in the crust (Clift & Sun, 2006;Marrett & Allmendinger, 1992;Walsh et al., 1991). It leads to an underestimation of the brittle extension derived from fault geometry. Therefore, for uncertainty estimation of γ f , we consider a possible range of scale factors α in Equation 8, from 1 (no underestimation) to 1.4 (40% of underestimation, after Walsh et al., 1991) (Supplementary Figure S2).
(4) Evolution of brittle extension In the Baiyun Rift, brittle extension of the crust is largely accommodated by the basement faulting. Hence, the evolution of the fault heaves allows restoring the brittle extension process of the crust.
In this study, we propose a scalar value (η) to quantify the brittle extension for a given evolution stage. It demonstrates the proportion of the extension in the episodic rifting. The proportion can be derived by a comparison of the section length after extension (L) and before extension (L−ΣL fi ) where: L, section length at the end of rifting stage; and L fi , heave of any active fault during this rifting stage. For the present-day section, η = 100%.
Thereby, the brittle extension amount in each rifting stage (Γ f ) can be obtained through the fault-derived thinning factor multiply by the scalar value

Rift System in the Baiyun Rift
This study focuses on the fault pattern of the Baiyun Rift and the primary sequence that records the hyper-thinning process of the continental crust in the Cenozoic. Following the fault strike directions, we present four selected seismic sections to document the variation of basement fault pattern from west to east (Figure 3), as well as the structural evolution process (Figure 4).

Structural Framework
Four stratigraphic units can be identified with distinct reflection features and internal structures. They are initial rift unit, intense rift unit, transition unit, and post-rift unit (Figures 2 and 3). The horizons that bound ZHAO ET AL.
(1) Initial rift unit (RE I ) The inception of the Baiyun Rift started at (or perhaps preceded) the Early Eocene while the NSCS underwent uniform stretching. In the seismic sections, a clear boundary at the top of the acoustic basement (TB) represents the base of the Cenozoic rift infill. TB acts as a major unconformity characterized by high amplitude, continuous reflections ( Figure 3). Above TB, basement faults developed widely and offset the horizon syn-detachment (SD). The sequence between TB and SD formed isolated half-graben scattered in t the Baiyun Rift ( Figure 4).
(2) Intense rift unit (RE II ) Deposition above SD extended both southwards (e.g., section G-G′) and northwards (e.g., section E-E′), covering initial-formed half-graben ( Figure 4). After the rotation correction for the fault blocks, the dips of the main basement faults were <30° during the Mid Eocene, that is, LANFs. The wedge-shaped geometry between SD and post-detachment (PD) indicates syn-rifting strata thickening to the LANFs. Significant displacements of these LANFs (∼5 km in section E-E′) suggest an intensive extension during this stage.
(3) Transition unit Extension was significantly reduced after the formation of the PD, indicated by limited displacement of the basin-boundary faults in the west, such as the f A-4 in section A-A' (Figure 3a). The PD corresponds to the cessation of the extension of the Baiyun Rift ( Figure 4). Afterward, the extension massively weakened and migrated to the south. As observed in the Liwan sag to the south of the Baiyun Rift, during the hyper-extension, deposition and related faults migrated in sequence to the future oceanic domain. Few faults remain active until the formation of H 23.8 (Zhang et al., 2019).
Above the PD, a distinct seismic horizon can be tracked across the entire continental margin of the NSCS. The PD shows medium-high amplitude, patchy-to-continuous reflections, and it has either erosional truncation below or onlap or downlap above the horizon (Ding et al., 2020;Gao et al., 2015;Yang et al., 2018;). This regional unconformity marks the final breakup of the continental lithosphere of the NSCS, which is referred to as the breakup unconformity (BRU) (Briais et al., 1993;Gao et al., 2015;Lei et al., 2018;Li et al., 2014;Ren et al., 2018;Sun et al., 2014). The BRU postdates the PD, suggesting that the cessation of the tectonic movement in the Baiyun Rift predates the lithospheric breakup of the SCS.
(4) Post-rift unit The unit is bounded by the BRU and the seafloor, deposited since the Late Oligocene until present (Figure 4). This unit covers the entire margin of the NSCS, including those uplifts. A disk-shaped geometry in the seismic sections with the depocenters away from the basement faults indicates a thermal subsidence stage. Magmatic sills can be observed at a depth of 3 s TWT above the dome-shaped structure (section E-E′, Figure 3c). Limited amount of small faults (green faults in Figure 3) develop in the post-rift unit only affecting sediments, which suggests the extension has mostly ceased.

Basement Faults
Generally, the hyper-extended structure of the Baiyun Rift is characterized by highly asymmetric half-graben dominated by continentward-dipping faults. Section A-A′ crosses the western Baiyun Rift (Figure 3a). The geological interpretation shows that the rift system consists of a series of well-developed continentward-dipping faults. Wedge-shaped syn-tectonic deposition thickens to the major basement faults, including f A-1 , f A-4, and f A-8 . Toward the southwest, TB is getting shallower and stepping up to the Yunkai Uplift.
Section C-C′ runs across the center of the Baiyun Rift (Figure 3b). The basement faults have lower dipping angles with more significant horizontal displacement than those in section A-A'. The major fault, f C-1 , extends to the NW with an average dipping angle of 10° and soles out at a depth of ∼8 s TWT, that is, ULB. In the footwall of the f C-1 , the Moho reflection shallows up from ∼10 to ∼9 s TWT. Within the basement, f C-1 accommodates an accumulation displacement of 20 km at the horizon TB. Over 5 km thick syn-rift sediments were deposited in the hanging wall of the f C-1 , forming the largest depocenter of the Baiyun Rift. The brittle crust has been dramatically thinned to <10 km at the depocenter of f C-1 . At the distance of 55-75 km, small faults only affect sediments attach onto the major fault plane of the f C-1 . Therefore, in the central Baiyun Rift, the development of the LANFs is closely related to the crustal hyper-thinning.
In section E-E′, the f E-9 , as a major fault, extends a long distance from ∼50 to ∼85 km and penetrates into the basement at ∼12 s TWT, indicating extremely strong extension (Figures 3c and 3d). In the hanging wall of the f E-9 , small basement faults are well developed with similar tendency and offset the sediments between SD and PD.
The basement faults in section G-G′ shows a similar pattern as in section E-E'. The major fault f G-7 penetrates into the basement at ∼15 s TWT before joining the ULB to the NW, and flattens SE-wards along the TB. Noteworthy, in the hanging wall of f G-8, the transition unit (at the distances of 83-100 km) dips to NW, while the below initial and intense rift units (at the distances of 70-83 km) dips opposite, indicating the f G-8 has experienced episodic normal faulting characterized by a clockwise rotation. Furthermore, the initial rift unit between TB and SD in the hanging wall of f G-8 (with breakaway points b/b') is absent in the half-graben controlled by f G-7 (with breakaway points a/a'), suggesting the rifting system in the eastern Baiyun Rift includes at least two phases of faulting ( Figure 4c): (1) fault f G-8 (with breakaway points b/b') initiated during the Early Eocene and initial rift unit was deposited; (2) after that the f G-7 became active and resulted in the deposition of transition unit and passive rotation of f G-8. After the Late Eocene most faults became inactive.
In map view, alignments of major basement faults form four fault zones, numbered as ①-④. Fault ① only develops in the southern margin. The hanging wall of the fault zone ① had been strongly deformed by fault zone ②, as shown by f E-10 ( Figure 3c). Fault zone ② is a basin-scale boundary fault in E-W orientation with well-developed small normal faults in the hanging wall. For example, in section E-E′, the f E-9 and numerous small faults between 60 and 85 km (f E-1 to f E-8 ) constitute a comb shape divergence upward. Similar fault patterns have also been observed in other sections (e.g., section D-D′ in the Supplementary Figure S7). This comb structure might suggest that fault zone ② was formed by a sequential faulting (see Section 6.3 for a further discussion). Fault zone ③ is in a NWW-SEE orientation and separates the Baiyun Rift from the Yunkai Uplift in the southeast, as shown by f A-1 (Figure 3a). Fault ④ is located in the northeast of the Baiyun Rift with limited lateral extensibility. Typical pattern of fault ④ is shown in f G-7 (Figure 3d).
Overall, the faults become younger sequentially and nucleate in the hanging walls of the previous faults. As shown in section E-E′, sequential faults become superimposed and form asymmetric rift structures. The faulting pattern shows similarity to the one in Iberia-Newfoundland margins (Ranero & Pérez-Gussinyé, 2010).

Crustal Structures
In general, most of the major basement faults root at a semi-horizontal decollement between a depth of 7-8 s TWT in the time-migrated sections (Figure 3). This semi-horizontal decollement can be tracked across the entire Baiyun Rift with high-amplitude and continuous reflections. According to the P-wave velocity modeling of the OBS1993 (Figure 1), the ULB is located at a depth of 15-12 km in the central Baiyun Rift (Yan et al., 2001), which is consistent with the semi-horizontal decollement in the adjacent seismic sections (e.g., section E-E′). Therefore, we interpret this regional decollement in the seismic sections as the transition between the brittle and ductile crust, that is, ULB.
(1) Western Baiyun Rift In the western Baiyun Rift, between ULB and TB, several sets of NE-dipping, high-amplitude reflections extending from basement faults terminate at different depths in the upper crust (Figures 5a and 5b). In the lower crust, the ULB truncates multiple, subparallel seismic reflections dipping SW, as shown in sections A-A′ and B-B' (green dotted lines in Figures 3a, 5a, and 5b). These lower crustal reflections (LCRs) are several kilometers thick, mirroring to the fault reflections in the upper crust. Since neither magmatic intrusions or HVL has been observed in the western Baiyun Rift, these LCRs are unlikely to be magmatic origin. Similar reflections were observed in the West Iberia margin, which were interpreted as a ductile shear deformation (Pérez-Gussinyé et al., 2003). Analogously, we propose these LCRs are shear zones related to the hyper-extension of the crust. See Section 6.2 for a detailed discussion.
(2) Eastern Baiyun Rift Distinctively, in the eastern Baiyun Rift, a dome-shaped structure with chaotic internal reflections can be observed in the footwalls of LANFs (Figures 3c, 3d, 5c, and 5d). The LANFs extend parallelly to the northern flank of the dome, and the base of the dome is as deep as the current ULB. These contact relationships suggest the dome-shaped structure is brittle, and the regional decollement is the current ULB in the present-day section (base of the dome structure). Notably, the dome-shaped structure is close to the presence of magmatic intrusions and HVL (Figure 3c), which indicates the doming is related to magmatism.
In the footwall of the fault f G-7 in section G-G′, a set of high-amplitude, parallel reflections dipping to the SE lies above the southern flank of the dome-shaped structure (purple dash lines, Figures 3d and 5d). Considering the south of the PRMB is developed above a Mesozoic volcanic arc , we interpret these reflections as the pre-rift sequence associated with Mesozoic sedimentation and magmatism. A parallel contacting relationship between the southern flank of the dome, pre-Cenozoic strata, and the hanging wall of the f G-8 suggests that the passive rotation of the strata and the doming were synchronic (Figure 4c). The presence of the pre-Cenozoic sequence in the basement results in an uncertainty in calculating the thinning factor. See Section 6.3 for a further discussion. (

3) ULB and Moho depth
Across the entire Baiyun Rift, the ULB remains consistent at a depth of 7-8 s TWT. In the depth sections, the ULB in sections A-A′ shallows to the depocenters from ∼18 to ∼13 km when the basement deepens from ∼3 to ∼9 km, which induced the thinning of the brittle crust in the western Baiyun Rift (Figures 3a  and 3d). Likewise, in the central Baiyun Rift, the brittle crust is dramatically thinned. For example, in the depth section C-C′, a decrease in the ULB from 15 to 12 km is coincident with an increase in basement ZHAO ET AL.
Clear Moho reflections can be observed in the time-migrated section C-C′ with continuous and high amplitude features at a depth of 9-10 s TWT (Figure 3b). In the depth section C-C′, the Moho shallows from 30 km at the rift shoulders to 18 km below the depocenter. By projecting the Moho depth from previous studies, in the western and eastern areas of the Baiyun Rift, the Moho shows a gentle uplifting from 30 to 25 km (Figures 3a and 3d).

Results of Thinning Factors
(1) Comparisons between γ uc and γ c , γ f and γ c Figure 6 shows the results of the fault-derived thinning factor γ f , the thinning factors of the upper crust γ uc , and thinning factor of the whole crust γ c . In the central Baiyun Rift, as shown in sections C-C′ and D-D′ at the distance of <80 km, thinning factors have similar values and increase synchronically from the rift shoulders to the depocenters (Figure 6b). The result of γ f ≈ γ c with this region indicates no evident discrepancy between brittle extension and whole crustal extension in the central Baiyun Rift However, in the southern Baiyun Rift where the distances along sections C-C′ and D-D′ are >80 km, the whole crustal thinning factor γ c remains larger than the γ uc and γ f . The minimum γ c is > 0.4, while γ uc ranges between 0.38 and 0.28, and γ f drops from 0.5 to less than 0.1 (Figure 6b). Therefore, the southern Baiyun Rift shows a typical extension discrepancy that whole crustal extension exceeds the brittle crustal extension.
ZHAO ET AL.

10.1029/2020TC006547
15 of 28 Figure 6. Comparisons between whole crustal thinning factor (γ c ), upper crustal thinning factor (γ uc ), and fault-derived thinning factor (γ f ) along the depth sections. Blue and orange shaded regions represent γ uc > γ c and γ f > γ c , indicating inverse discrepancy. Pie graphs demonstrate the proportion of the extension in each structural evolution stages (η). The light brown category in the pie graph has the largest proportion shows the most intense extension happened in the RE II when low-angled faults were active. γ uc ' is estimated without removing the area of new brittle crust. γ c ' is the thinning factor of the whole crust under an assumption that 50% of the basement is pre-Cenozoic remnants. The actual thinning factor of the crust ranges from γ c to γ c '.
In the western Baiyun Rift, the extension is characterized by a remarkable difference between γ f and γ c (Figure 6a). In sections A-A′ and B-B′, the maximum γ f (0.6) and maximum γ c (0.45) occur at the depocenters of the LANFs, which suggests the low-angled faulting dominated the crustal thinning process. In section A-A′, the area with γ f > γ c (Orange shaded region in Figure 6a) suggests that the fault-related extension is greater than the whole crustal extension, that is, inverse extension discrepancy (e.g., Reston, 2007). In section B-B′, maximum γ f = maximum γ c = 0.5. After the error correction due to the 40% underestimation of the sub-seismic faulting (e.g., Clift et al., 2002), the actual upper crustal thinning factors would be larger than the whole crustal thinning factors (see uncertainty analysis in Supplementary Figure S2).
The thinning factor of the upper crust, γ uc , reaches a maximum of 0.65 in the depocenter and changes more sharply than γ f and γ c (sections A-A′ and B-B′ in Figure 5). Generally, the γ uc in the depocenters is larger than γ c (Blue shaded regions in Figure 5a), which agrees with the γ f > γ c , indicating an typical inverse extension discrepancy in the western Baiyun Rift.
Similar to the west, the eastern Baiyun Rift also shows an inverse extension discrepancy. In the depocenters of sections E-E′ and G-G′, both γ f and γ uc exceed the γ c (Orange and blue shaded regions in Figure 6c). In the Southern Uplift (i.e., distances of >80 km in section E-E′ and 60 km in section G-G′), γ f remains larger than or equal to γ c , while γ uc drops rapidly to lower than γ c . We believe that low values of γ uc in the Southern Uplift are due to the thickening of the brittle crust. The γ f > γ c across sections E-E′ and G-G′ indicates an inverse discrepancy across the eastern Baiyun Rift.
The seismic interpretation (see Section 4.3) and previous studies  show that the pre-rift basement of the Baiyun Rift may contain Mesozoic strata. Thus, we further calculated the thinning factor of the whole crust considering the existence of pre-Cenozoic strata in the basement (γ c '). Based on the distribution of the high amplitude reflections parallel to TB, we mapped the maximum distribution of the pre-Cenozoic remnants. Result shows the remnants are distributed below the depocenters and the southern uplifts in the areas from section E-E′ to the further east (Figures 3c and 3d). Then, we assumed a situation that the maximum volume of pre-Cenozoic materials accounts for 50% of the pre-rift basement. Based on the extreme distribution, a thinning factor of the whole crust, γ c ', is obtained. Under this assumption, the maximum γ c ' is equal to the maximum γ uc (Figure 6c), which would imply no extension discrepancy between the brittle crust and the whole crust.
However, these pre-Cenozoic layered seismic reflections are less than 50% of the pre-rift basement (Figures 3c and 3d). Therefore, the actual thinning factor of the whole crust should range from γ c to γ c ', which is likely remain smaller than the brittle extension and γ uc , in sections E-E′ and G-G'. This result further supports the observation of an inverse extension discrepancy in the eastern Baiyun Rift.
(2) Comparison between γ uc and γ f Along all sections, the γ uc and γ f synchronically increase from the rift shoulders to the depocenters and reach their maxima near the main LANFs. For the parts without clear ULB due to magmatism, such as section D-D′, the interpolated γ uc is also synchronous to γ c and γ f . Within the depocenters of the Baiyun Rift, the γ uc remains larger than the γ f . As representatively displayed in section C-C′, the γ uc reaches ∼0.8 at the distance of 65 km while the maximum γ f is 0.6 ( Figure 6b). After a sub-resolution faulting correction (α ranging from 1 to 1.4 in Equation 8; see Supplementary Figure S2), the maximum γ f ranges from 0.6 to 0.75, which remains smaller than the maximum γ uc . This observation indicates that in the depocenters of the basin, (1) the normal faulting is not sufficient to explain the brittle extension of the crust, and (2) the brittle extension was mainly accommodated by the LANFs.
In the western and eastern Baiyun Rift, γ uc changes more sharply than the one in the basin center. As observed in the seismic sections in the eastern Baiyun Rift (Figures 3c and 3d), new brittle crust is created in the footwall of the LANFs during the hyper-stretching. If the γ uc is not corrected for the effects of the new brittle crust (i.e., γ uc '), the resulting values will be much smaller than γ f along the sections. The areas with γ uc ' < γ f correspond to the dome-shaped structure with magma additions in the seismic sections E-E′ and G-G' (Figure 6c), which further indicate the brittle crust had been thickened instead of being thinned in the eastern Baiyun Rift.
Near the Yunkai Uplift, γ uc < γ f is related to an underestimation of the γ uc due to pre-rift basement (see Section 6.1 for a detailed discussion). To be noted, in the Panyu Uplift (at the distances of <20 km along sections A-A′ and B-B′), γ f > γ uc is obtained where no seismic data cover. Therefore, in the northwest of the Baiyun Rift, the missing of the ULB causes an uncertainty of the γ uc (see uncertainty analysis in Supplementary Figure S3).

Evolution of Brittle Extension
Pie graphs in Figure 6 show the proportion of the brittle extension in the crust in different structural evolution stages. The light brown category in the pie graph (RE II ) takes up the largest proportion, which shows that the crustal extension was mainly accommodated by detachment faulting before Late Eocene. During the transition stage, the brittle extension was significantly decreased (Pink category, Tran.). After the new oceanic crust of the SCS was formed, most of the extension ceased; only limited faulting occurred in the western Baiyun Rift (Light yellow category, PR).
A map view of the thinning factor derived from normal faulting (γ f ) shows the evolution of the brittle extension (Figure 7). is located in the hanging wall of the fault zone ②, which suggests the fault zone ② became the dominant rifted-control faults (Figure 7b). During the Late Eocene, the Baiyun Rift transited into the thermal subsidence stage (Figures 2 and 4). In Figures 7b and 7c, γ f in the east of the Baiyun Rift increases from 0.3 to 0.5, which suggests the extensional strain migrated from the center to the east. Since the Early Oligocene, the brittle extension in the Baiyun Rift showed little change (Figures 7c-7d). Arrows mark the results of the most conservative interpretation with minimum brittle extension and maximum whole crustal thinning. The blue arrows are estimated when T 0 is at its lower bound (12 km) and Z 0 is at its higher bound (35 km). The orange arrows are estimated when no underestimation of fault-derived thinning factor and Z 0 is 35 km. Values along the diagonal line indicate no extension discrepancy (Driscoll & Karner, 1998). Values in the top-left indicate a greater deformation of the brittle crust than the whole crust, that is, inverse discrepancy (Reston & Mcdermott, 2014). Values in the bottom-right suggests a typical extension discrepancy of brittle extension far less than whole crustal thinning (Driscoll & Karner, 1998). Maps compare the differences between γ f and γ c (b), γ uc and γ c (c), γ uc and γ f (d).

Spatial Distribution of the Extension Discrepancy
Extension discrepancy with γ f −γ c < 0 is concentrated in the center of the Baiyun Rift (b). The contrary observation of γ f −γ c > 0 and γ uc −γ c > 0, that is, inverse discrepancy, takes up the most area of the basin (b)-(c). Widely distribution of γ uc −γ f > 0 suggests that in the Baiyun Rift, without sub-resolution faulting correction, fault-derived thinning factor cannot adequately describe the brittle extension of the crust (d).
different pre-rift thickness (see methodology Section 3.5 for details). Solid dots mark the thinning factors with nominal values of pre-rift upper crustal thickness (T 0 ) = 12 km and pre-rift whole crustal thickness (Z 0 ) = 30 km. Most results of the γ f and γ uc are approximately along, or scatter on two sides of the diagonal line.
We also consider the most conservative interpretation with minimum brittle extension and maximum whole crustal thinning, marked by horizontal arrows in Figure 8a. The blue arrows are estimated when T 0 is at its lower bound (12 km) and Z 0 is at its higher bound (35 km). The orange arrows are estimated when no underestimation of sub-resolution faulting and Z 0 is 35 km. On the other hand, vertical bars show a more liberal interpretation with the maximum brittle extension and the minimum whole crustal thinning. Orange vertical bars are estimated with Z 0 = 30 km and 0%-40% of sub-resolution faulting underestimation. Blue vertical bars are estimated with T 0 = 12-15 km and Z 0 = 30 km. Including more conservative and bolder scenarios, results with error ranges suggest that both extension discrepancy and inverse discrepancy developed in the Baiyun Rift.
(1) Observation I: γ f > γ c Figure 8b, warm colors indicating γ f > γ c are generally distributed in the eastern and western Baiyun Rift. The maximum value of the γ f −γ c is located between the fault zones ② and ④ suggesting the maximum inverse discrepancy occurred in the eastern Baiyun Rift. On the contrary, the negative values, that is, extension discrepancy, are mainly distributed in a narrow area of the central Baiyun Rift as well as the Southern Uplift. Therefore, the normal faulting in the upper crust is sufficient to explain the whole crustal thinning in the most areas of the Baiyun Rift, except for the narrow area in the center.
(2) Observation II: γ uc > γ c The comparison between γ uc and γ c in Figure 8c suggests a similar inverse discrepancy pattern as Figure 8b but with higher magnitudes. Most area of the Baiyun Rift shows a greater thinning of the brittle crust than the whole crust with γ uc > γ c , especially in the hanging walls of the dominant LANFs with higher values of γ uc −γ c . However, near the adjacent uplifts, the thinning factor of the whole crust exceeds the one of the brittle crust (γ c > γ uc ). This extension discrepancy corresponds to a small amount of γ uc /γ c distributed below the diagonal line in Figure 8a. Based on the seismic interpretation, relatively weak upper crustal thinning might relate to the participation of magma additions in the rift shoulders, resulting in low values of γ uc.
(3) Observation III: γ f ≠ γ uc The values of γ uc > γ f cover the most areas of the Baiyun Rift (warm colors in Figure 8d). Two significantly differences between γ uc and γ f (γ uc −γ f > 0.2) are located in the northeast near the dominant LANFs zones ①, ②, and ④. As shown in Figure 4, the evolution of the LANFs involves early stage high-angled faulting, which cannot be fully restored in the modeling. Therefore, the warm colors in Figure 8d suggest where multiple phases faulting occurred, the fault-derived thinning factor based on the fault geometry is highly underestimated.
Results of γ uc < γ f are mainly located in the uplifts outside of the Baiyun Rift, which is due to the limitation of the seismic data coverage. Within the basin, γ uc < γ f can be only observed in the hanging wall of the fault system ②, eastern Baiyun Rift. These negative values of γ uc −γ f correspond to the dome-structures in seismic section (e.g., section G-G′, Figure 3d). Therefore, the result of γ uc << γ f in the eastern Baiyun Rift is a result of thickened brittle crust.

Possible Causes for Extension Discrepancy
Potential models interpreting the extension discrepancy (γ uc < γ c ) include (1) asymmetrical simple shear model (Wernicke, 1995); and (2) lateral lower crust flowing (McKenzie et al., 2000;McKenzie & Jackson, 2002). However, the former simple shear model does not agree with the observation in the Baiyun Rift where both extension discrepancy and inverse discrepancy are observed. As for the lower crustal flowing model, even it could produce crustal thickness variation, the ductile materials usually flow along the extension direction (McKenzie & Jackson, 2002). In the NSCS, a few of numerical simulations did propose that the lower crustal flowing occurred across the entire margin of the NSCS along the NW-SE extension direction (Clift, 2015;Dong et al., 2020). However, this study indicates an inhomogeneous thinning of the crust in an NE-SW direction, which is perpendicular to the extension direction ( Figure 8). Hence, the lateral lower crustal flow cannot primarily cause the localized discrepancy in the Baiyun Rift. We propose alternative explanation for the observed extension discrepancy in the Baiyun Rift. .

West-East Variations in the Pre-Rift Basement
In the Baiyun Rift, the pre-rift basement was very heterogeneous due to Paleo-Pacific subduction in Mesozoic Sun et al., 2008;Zhao et al., 2020;Zhou et al., 2005). A series of lithospheric-sale Mesozoic faults were proposed to separate the Mesozoic arc from the forearc-related areas with evident magnetic anomalies and segmented the NSCS margin . One of the Mesozoic fault, F2, also referred to as Baiyun-Liwan Fault Zone (Zhao et al., 2020), is a ∼220 km-long strikeslip fault across the central Baiyun Rift (Figure 1). Pre-rift inheritance in the basement results in a rapid change of the lower crust behavior over a short-distance of 100 km from west to east of the Baiyun Rift.
Multiple studies suggested the thickness of the pre-rift brittle crust (Z uc0 ) ranges from 9.14 to 12 km (Chen, 2014;Dong et al., 2020;Zhang et al., 2008). Under an assumption of areal balancing (Julivert & Arboleya, 1986), the thickness of the pre-rift brittle crust in the basin center is 11.8 km (section B-B′, Table 3), which is consistent previous studies. However, the thickness of the pre-rift brittle crust in sections G-G' (15.8 km) and E-E' (13.9 km) (Table 3), which are 31.6% and 15.7% larger than the average number of 12 km, respectively. Therefore, the structure restoration supports the pre-rift structure of the brittle crust in the Baiyun Rift are highly inhomogeneous.
Pre-rift inheritance increases the uncertainties in the extension discrepancy between the brittle crust and the whole crust. As shown in Figures 3c and 3d, under an extreme assumption with a maximum volume of 50% for the pre-Cenozoic strata in the crust, the brittle extension is still sufficient to explain the whole crust thinning. It further supports the observation of inverse discrepancy that brittle extension exceeds the whole crust thinning in the Baiyun Rift. Therefore, the inheritance from pre-rift inheritance cannot solely explain the inverse discrepancy in the Baiyun Rift.
Besides, the west-east variations in crustal structure extends further south to the continental-oceanic transition (COT) of the NSCS. Ding et al. (2020) found that there is a lateral variability of the COT in an eastwest direction. Separated by a boundary extending from the central Baiyun Rift, the eastern COT shows hyper-thinned crust covered by possible volcanoclastic materials, while the western COT shows no magmatism. This work agrees with our results that during the hyper-thinning of the crust, the eastern Baiyun Rift is developed above a "hot crust." Hot crust was proposed and demonstrated by field outcrops in Basin and Range Province in the western North America (Buck, 1991;Wernicke, 1995) and physical experiments (Brun et al., 1994). Hot crust model indicate hyper-extension in the upper crust lead to the exhumation of the lower crust by the development of rolling-hinge detachment (Brun et al., 1994;Whitney et al., 2013), which can be analog to the observations in the eastern Baiyun Rift.

Ductile Shearing Deformation in the Western Baiyun Rift
As shown in Section 4.3, the crustal structure in the west and east of the Baiyun Rift shows distinct differences. In the western Baiyun Rift, characteristically, the LCRs are identified near the brittle-ductile transition.
In the SCS, conjugate LCRs have been reported in the lowest crust of the oceanic basin, and interpreted as shear zones associated with the flowing mantle (Ding et al., 2018). However, the LCRs in the Baiyun Rift is in the middle-lower crust, notably abundant beneath the LANFs reflections. Furthermore, these LCRs seem symmetric to the continentward-dipping faults (Figures 5a and 5b). Such geometry has been reported in the other rifted continental margin, such as the west of Iberia (Pérez-Gussinyé, 2013; Pérez-Gussinyé ZHAO ET AL. Maximum γuc > Maximum γf suggests the thinning factor derived from fault geometry cannot adequately describe the brittle extension of the crust. Thicknesses of pre-rift brittle crust in the east (sections E-E′ and G-G′) are far greater than the one in other areas (e.g., section B-B′), suggesting the brittle crust had been thickened.  (Clerc et al., 2015), Gulf of Mexico (Pindell et al., 2014), and Northern North Sea rift (Fazlikhani et al., 2017), where they indicate a deformation dominated by a simple shear toward the continent.
Therefore, in the western Baiyun Rift with limited magmatism, these widely distributed LCRs were result from a ductile shearing. The inverse discrepancy of the crustal extension is likely to be achieved by intense tectonic faulting in the upper crust, and a relatively weak ductile shearing in the lower crust.

Lower Crustal Exhumation in the Eastern Baiyun Rift
As mentioned in the Observations I and II, the maximum inverse discrepancy occurred near the fault zones ② and ④, eastern Baiyun Rift. Comparing to the seismic interpretation ( Figure 3d) and the isopach map of Eocene strata (Figure 1), the area of maximum inverse discrepancy corresponds to the exhumed crust without syn-rift deposition.
One crucial evidence for the lower crustal exhumation is the intense magmatism in the eastern Baiyun Rift. Magmatic sills have been observed in the seismic section ( Figure 3c). Previous deep seismic refraction data show an HVL develops in the central and eastern Baiyun Rift (Figures 1 and 3) (Wan et al., 2017;Wang et al., 2006;Yan et al., 2001;Zhao et al., 2010). Although the origin and nature of the HVL are still under discussion, in the NSCS, the formation of the HVL is more suggestive of a magmatic underplating mechanism (Nissen et al., 1995;Wei et al., 2011;Yan et al., 2001). The magmatic underplating fits well with the high thermal environment in the eastern Baiyun Rift (Li et al., 2019;Shi et al., 2017). Moreover, recent results proved the occurrence of a short-period magmatism in the latest syn-rifting stage due to a weak lower crustal strength (Ding et al., 2020;Larsen et al., 2018;Zhang et al., 2019). The thermal activity and rheology in term favors the exhumation of the lower crust (Andrés-Martínez et al., 2019).
In addition, as proposed by the numerical models, the thinning of the rifted margin can be performed by either mantle exhumation or mantle removal, lower crustal exhumation or lower crustal removal (Huismans & Beaumont, 2008. For the wide type of the continental margin such as the NSCS, the crust is usually characterized by weak lower crust, magmatic underplating and intrusion during the extension, which corresponds to an abrupt transition from a continental margin to the new creation of the oceanic crust without the exhumation of the mantle (Ros et al., 2017). This model agrees with the IODP results of rapid transition from continental thinning to final breakup . These results further support the lower crustal exhumation occurred during the extension of the Baiyun Rift.
Thus, we believe the lower crustal exhumation caused inverse extension discrepancy in the eastern Baiyun Rift. As shown in the structural restoration (Figures 4 and 9), the sequential development of the f G-7 and f G-8 hyper-stretched the brittle crust. Subsequently, due to isostasy, magmas migrated upward and formed the dome-shaped structure at the footwall of the f G-7 . Meanwhile, the hanging wall of the f G-8 rotated clockwise synchronously with the magma upwelling ( Figure 9c). As the magmas cooled, the original ULB had no longer behave as a conservative boundary since the ductile materials become brittle (Perez-gussinye & Reston, 2001). New brittle crust and new ULB formed below the old ULB ( Figure 9d). Consequently, the total area of the crust would increase (Figures 3 and 4), which accordingly results in an underestimation of the whole crustal thinning factor γ c . Therefore, an inverse discrepancy of γ c < γ uc is observed.

Hyper-Thinning Process of the Continental Crust
The hyper-thinning process of the crust in the Baiyun Rift is mainly contributed by sequential basement faulting. Fault zone ① activated first (Figure 10a). It caused a limited extension focused on the shallow layer of the crust. Then, fault zone ② became active with low angle. The average dipping angle between TB and PD is 12°. During this stage, extension generally focused on the center of the basin (Figure 10b). With continuous extension, fault zone ③ became more intensive in the west (Figure 10c). Likewise, in the east, fault zone ④ with more massive displacements initiated, resulting in passive rotation of the fault zone ② (Figure 10c). The upper crust was hyper-thinned while the lower crust passively upwelled. In the northeast area of the Baiyun Rift, the lower crust had been exhumed at the footwall of the fault zone ④ (Figure 10d).
The crustal structure in the western and eastern Baiyun Rift not only demonstrates the spatial differences; more importantly, they represent a temporal evolution of a crustal thinning process.
The western Baiyun Rift demonstrates the initial stretching mode, which is characterized by a stair-step fault pattern with multiple deposition centers ( Figure 10e). Basement faults root at a strong ductile shear zone, which separates the upper and lower crust. The extension of the upper crust was dominated by intense faulting, while the extension of the lower crust was characterized by weak ductile shearing, which results in a stronger extension in the brittle crust than in the whole crust.
The thinning mode is characterized by a hyper-thinned whole crust with the thickness less than 10 km in the central Baiyun Rift (Figure 10f). LANFs with large horizontal displacements accommodated a significant extension in the brittle crust. A high-velocity lower crust related to the magmatic intrusion shows that magmatism could be involved in the crustal thinning. The combination of an intense brittle extension and the ductile deformation results in the upper crust of the Baiyun Rift thinned uniformly with the lower crust ( Figure 3b).
The exhumation mode is revealed by the eastern Baiyun Rift (Figure 10g). The sequential development of the fault zones ② and ④, hyper-stretched the upper crust. Due to isostasy, magma upwelled along the footwall of the LANFs. Where the lower crust exhumed, the igneous rocks became brittle and formed the dome-shaped structure locally. Therefore, the hyper-thinning of the continental crust in the Baiyun Rift is largely accommodated by detachment faulting.
ZHAO ET AL.

Conclusions
Based on the geological interpretation and restoration of high-resolution seismic sections extracted from a 3D seismic survey in the Baiyun Rift, four sets of the basement fault zones in the Baiyun Rift have been identified. These fault zones were initiated earlier in the central south and migrated to the NE and NW progressively. In the eastern Baiyun Rift, sequential faulting led to detachment faulting associated with a lower crustal exhumation.
We calculate the thinning factor derived from normal faulting (γ f ), the thinning factors of the brittle crust (γ uc ), and the whole crust (γ c ). Unlike the classic "extension discrepancy" phenomenon in rifted continental margin, most areas of the Baiyun Rift are characterized by "inverse extension discrepancy," indicated by a higher thinning factor of the brittle crust than the whole crust, especially in the eastern and western depocenters.
We propose that the inverse discrepancies in the west and the east are due to different mechanisms. In the western Baiyun Rift, the ductile shear zone separates the deformation in the upper crust and lower crust, which results in a decoupled thinning process characterized by a greater extension in the upper crust than in the whole crust. However, in the eastern Baiyun Rift, sequential detachment faulting caused the exhumation of the lower crust, leading to two separate processes. First, due to isostasy, magma from ductile crust passively upwelled and became brittle. The magma embrittlement would thicken the upper crust and cause an underestimation of the upper crustal thinning factor. Thus, we estimate an upper bound of upper crustal thinning factor by removing the area of new brittle crust and set the lower bound as thinning factors without any corrections. Second, the lower crustal exhumation may trigger decompression melting in the ZHAO ET AL.
10.1029/2020TC006547 23 of 28 Figure 10. Models of sequential faulting leading to the hyper-thinned crust accommodated by basin-controlling fault zones (a-d) Block diagrams display a perspective view from the NE of the 3D volume, without sediments to expose the top of the acoustic basement (e-g) Models for temporal and spatial evolution of the hyper-thinning process based on observations from seismic sections in the Baiyun Rift. Stretching mode (e) is characterized by listric faulting, a differential subsidence of half-graben, and a major ductile shear zone exemplified by section A-A′, western Baiyun Rift. Thinning mode (f) is characterized by the maximum thinning of the crust and the presence of the magmatic addition in the lowest crust. The exhumation mode (g) is well documented by section G-G′, eastern Baiyun Rift. This phase is distinguished by the exhumation and embrittlement of the lower crust from less than 5 km depth along a downwardconcave detachment fault ④.
mantle. Underplating of the lower crust by the mantle magma can further decrease the estimated thinning factor of the whole crust, although we are not able to quantify the extent (if any) of the underplating magma. Therefore, in the eastern Baiyun Rift, these effects lead to an inverse discrepancy characterized by an upper crustal thinning factor equaling (lower bound) or significantly exceeding (upper bound) the whole crustal thinning factor.
Overall, quantification of the spatial variations in crustal thinning factors were made possible by using a 3D seismic survey, which allowed for observations of crustal hyper-thinning throughout the Baiyun Rift. Through the addition of spatial dimensions, perpendicular to the primary extension direction of NE-SW, this work reveals the difference in inverse discrepancy mechanisms is likely attributable to pre-rift crustal structure inheritance and thermal environment. Future work in quantification of crustal deformation in other hyper-thinning domains of rifted margins will likely benefit from similar 3D seismic data.

Data Availability Statement
Data are available through .