Main

Cratons, known as relatively stable cores of continents, have traditionally been regarded as the most long-lived regions on Earth due to their thick lithospheric roots1,2,3. However, different studies have raised doubts about the long-term stability of cratons and have suggested the possibility of craton deformation, a process referred to as decratonization4,5,6,7,8.

While mantle plumes can induce lithospheric thinning through thermal erosion7,9, oceanic subduction is increasingly recognized as an important driver of craton deformation4,5,6,8,10,11,12,13. Two primary models related to subduction-induced decratonization have been proposed: the flat-slab and slab rollback models. The flat-slab subduction model was initially devised to explain the deformation of the Wyoming Craton as a result of the shallow-angle subduction of the Farallon slab beneath North America. However, the Wyoming Craton’s ultimate decratonization was attributed to the impingement of the Yellowstone plume upon the craton’s base9. The slab rollback subduction model6 is often invoked to explain extensional rifting associated with decratonization but encounters challenges in explaining deeper intracratonic deformation, especially contractional deformation within a craton. Consequently, a fundamental question arises: What triggers decratonization?

The North China Craton (NCC), a prominent example of extensive decratonization14, experienced plate subduction in the western Pacific to the east in the late Mesozoic and Cenozoic10 (Fig. 1a and Extended Data Fig. 1a). In this study, we integrate deformable plate tectonic and mantle flow models to investigate plate subduction and deep mantle processes and their connections to craton deformation and topographic evolution to address the question of what initiated decratonization in the context of slab subduction.

Fig. 1: Tectonics and reconstructions of the deforming plate in Northeast Asia and the western Pacific.
figure 1

a, Tectonics of Northeast Asia10,21. The light-pink and light-green colours are depictions of arcs, relict arcs and ocean island basalts in the Philippine Sea plate. The sky-blue lines outline the present-day coastlines. The coloured lines marked pre-J, J, K and T represent pre-Jurassic, Cretaceous and Tertiary sutures or subduction zones, respectively10. The red lines denote faults, rifts and seafloor spreading centres. b–g, Reconstruction of the northeast Asian deformation and the western Pacific flat slab and its rollback subduction at 168 Ma (b), 137 Ma (c), 120 Ma (d), 90 Ma (e), 80 Ma (f) and 67 Ma (g) in the anchored Eurasia plate reference system. The dilatation strain rate is shown by the colour coding (red indicates extension, and blue indicates compression). Principal components of strains accumulated during 200–168 Ma (b), 200–137 Ma (c), 136–120 Ma (d), 136–90 Ma (e), 89–80 Ma (f) and 79–67 Ma (g), computed via strain markers. The transparent white segments depict flat-slab fragments subducting beneath the eastern margin of the NCC, with the thick white lines representing their leading edges.

Full size image

Tracing subduction processes through deformation

To explore the link between deformation and plate subduction in the western Pacific, we employed global models developed by Müller (2016) (ref. 15) (model 1), Müller (2019) (ref. 16) (model 2) and our combination of models from Müller (2019) and Merdith (2021) (ref. 17) (model 3) (Methods and Extended Data Fig. 2) to reconstruct the tectonic history of East Asia since 250 Ma or 230 Ma. Our global reconstruction models incorporated diffuse deformation in the NCC and adjacent regions since 200 Ma (Methods and Fig. 1b–g). These reconstructions reveal NE- or NNE-striking thrust and transpressional faulting during the Jurassic and Early Cretaceous, with an initial basement-involved thrusting front migrating westwards from the Korean Peninsula at approximately 180 Ma to Chengde and Xiahuayuan–Xuanhua at 152 or 158 Ma (Fig. 1b,c and Extended Data Fig. 3). Additionally, the tectonic subsidence curves (which include both isostatic subsidence induced by lithospheric deformation and dynamic subsidence due to deep mantle flow) in the Yanshan thrust-top basins show initial rapid flexural subsidence migrating from the Chaoyang and Jingxi Basins in the east at approximately 172–173 Ma to the Chengde and Xuanhua Basins in the west at approximately 147 Ma (Fig. 2).

Fig. 2: Airy-backstripped tectonic subsidence curves in the thrust-top and rift basins of the NCC.
figure 2

The horizontal axis shows the basin location (top) and the subsidence scale for each curve (bottom); the vertical axis shows ages. The width of the curves represents the uncertainty in the palaeobathymetry. The arrows on the dashed light-green and light-orange lines indicate the migration directions of initial rapid subsidence centres in thrust-top basins and rift basins, respectively, moving towards the west and east. The basin location is shown in Extended Data Fig. 1.

Full size image

From 136 Ma onwards, several phases of extension with NW‒SE or N‒S orientations and eastwards migration of synrifting episodes occurred (Figs. 1d,e and 2). However, this extensional retreat experienced an interval of compression (inversion) from ca. 93 to 85 Ma (ref. 18) in the Laiyang Basin and from 89 to 80 Ma (ref. 19) in the Songliao Basin (Fig. 1f). In response to these multiple extensional events, rapid initial rift subsidence shifted from Yanshan at approximately 130 Ma to the Laiyang Basin in the east at approximately 121 Ma, with a second phase of rapid subsidence starting at approximately 85 or 79 Ma (Figs. 1g and 2 and Supplementary Fig. 1b,c). Moreover, the spatiotemporal distribution of magmatism4 shows a two-stage migration pattern relative to the palaeo-Pacific trench, with a younging trend from east to west for Jurassic and earliest Cretaceous igneous rocks before 137 Ma and the opposite trend towards the trench for Cretaceous igneous rocks (Supplementary Fig. 2).

The kinematics of plate tectonics, deformation and surface topography are strong indicators of coupling between plates and the mantle10,20,21,22. All the deformation patterns in the NCC that are characterized by NE- or NNE-oriented deformation centres originating from the eastern ocean–continent convergent margin indicate that the deformation of the NCC was probably related primarily to the subduction of oceanic plates from the Pacific rather than mantle plumes within the craton9,23. The basement-involved thrusting and associated flexure, occurring between 180 Ma and 137 Ma and continuously shifting inland, suggest a connection to flat-slab or gentle-slab subduction24,25 (Fig. 1b,c). Importantly, the inversion of structures and the rapid subsidence mechanism at approximately 137 Ma may indicate a transition from flat-slab subduction, with its leading edge shifting inland between 180 Ma and 136 Ma, to rapid seaward retreat to the east from 136 Ma to 90 Ma (ref. 10) (Fig. 1d,e). Structural inversion along the northeastern Asian margin from 89 to 80 Ma might have been induced by the return of advancing subduction with initial slab rupture or terrane accretion26, followed by rifting and subsequent trench retreat to the present day21 (Fig. 1f,g).

Validation of mantle flow model success

To investigate the impact of flat-slab and rollback subduction on craton stability, we created more than 26 global mantle flow models. We used these models to explore the effects of three different reconstructed tectonic histories (models 1–3) and variations in the physical properties of the mantle, including the ability of the subducting slab to reach the lower mantle (Methods and Extended Data Table 1). These variations were chosen on the basis of their demonstrated influence on lithospheric deformation20,21,22.

We evaluated the flow models (case 1-1, case 2-1 and the case 3 series) by comparing their predicted present-day structures with high-resolution P-wave seismic tomography (MIT-P08 (ref. 27) and GAP-P4 (ref. 28)) along four NW‒SE profiles across Northeast Asia (Methods and Extended Data Figs. 4 and 5). The case 3 series showed good agreement with tomographic images, achieving average accuracy rates—representing the ratios between seismic tomography observations and model predictions—ranging from 74.0% to 80.3% for MIT-P08 and from 84.4% to 87.4% for GAP-P4 (Supplementary Table 1). Specifically, cases 3-2, 3-3 and 3-4 not only accurately reproduced a 4,000-km-wide horizontal slab within the mantle transition zone (660 km and 410 km deep), primarily reflecting Cenozoic plate motion, but also captured eastwards-dipping structures with positive P-wave velocity anomalies located below 2,000 km to the east and shallowing to the west in the lower mantle, which are closely linked to the Mesozoic tectonic evolution29,30,31 (Extended Data Fig. 5). These models exhibited high average prediction accuracies of 78–80% for MIT-P08 and 86% for GAP-P4. A comparison of the predicted architectures with those of the seismic tomography (GAP-P4 and MIT-P08) revealed that case 3-2, but not case 3-1, demonstrated better fits and higher accuracies than cases 1-1 and 2-1, achieving accuracies of 80% for MIT-P08 and 86% for GAP-P4 (Supplementary Table 1). Overall, case 3-2 effectively matched the slab structures observed in the MIT-P08 and GAP-P4 tomographic models (Extended Data Fig. 4).

The mantle flow model computed from the case 1, 2 and 3 series generated dynamic topography32,33 that was compared against present-day residual topographies created by removing isostatic and thermal oceanic lithosphere effects34. A root-mean-square comparison of the predicted present-day dynamic topography against residual topographies in the western Pacific, Philippine Sea, Japan Sea and South China Sea indicated that cases 3-2 and 3-4 exhibited significantly fewer differences and greater percentages of agreement within one standard deviation than did the other cases (Methods and Supplementary Figs. 3 and 4).

By comparing the dynamic topography at specific times predicted by five models with the observed cumulative residual topography in the Ordos Basin, Xuanhua Basin and Jingxi Basin, we identified consistent trends and amplitudes between dynamic and residual topographies across all the models (Methods and Extended Data Fig. 6). For the Ordos Basin, the average residual topography was backstripped, ranging from approximately −337 to −445 m at 153 Ma and subsequently decreasing to −543 m at 136 Ma (Supplementary Fig. 5 and Extended Data Fig. 6a–d); these values are cumulative starting at 174 Ma with respect to the base level of the basin. In the Xuanhua Basin and Jingxi Basin, the average residual topographies were −462 m and −512 m at 136 Ma, respectively, which are cumulative since 160 Ma and 175 Ma, with respect to the base levels of the basins (Extended Data Fig. 6e, f).

Notably, the comparison between the evolutions of the observed residual topography and the model-predicted dynamic topography at different observation points from west to east in the NCC indicated that, while the dynamic topography predictions from cases 1-1 and 2-1 for the period of 135–60 Ma showed poor agreement with the observed residual topography, the predictions from cases 3-1, 3-2 and 3-4 for the period of 174–0 Ma demonstrated stronger matches with the residual topography, with case 3-2 showing the best alignment (Methods, Extended Data Fig. 7 and Supplementary Fig. 1). Essentially, these scenarios revealed dynamically driven subsidence and uplift in the NCC core to the west of the Tan–Lu Fault (Extended Data Fig. 7a–d), coinciding with dynamically driven uplift and subsidence to the east (Extended Data Fig. 7e), at approximately 140–120 Ma and 120–90 Ma, respectively (Fig. 2). Consequently, we successfully developed a new mantle flow model, case 3-2, incorporating a flat slab and its rollback subduction, which aligns with surface geological evolution and the present-day mantle slab structure.

Shaping large mantle wedge through flat-slab rollback

Interestingly, the successfully validated mantle-flow model, case 3-2, combined with our globally reconstructed tectonic history (Extended Data Table 1), can effectively describe the space–time characteristics and topographic response of mantle slab subduction over time (Fig. 3 and Supplementary Video 1). Commencing at approximately 180 Ma, the Izanagi plate transitioned from early intraoceanic subduction to initial flat-slab subduction beneath the NCC, which was driven by the relatively swift eastwards motion of the Eurasian plate and the integration of inferred exotic terrane accretion10,26,35. The sinking slab tore away from the subducting flat slab. The leading edge of the flat slab advanced westwards, steeply subducted and partly traversed the 660 km mantle phase transition boundary. By 137 Ma, the curved leading edge of the flat slab reached its westernmost front, corresponding to the eastern Taihang Mountains on the Earth’s surface (Fig. 3a,b and Supplementary Video 1).

Fig. 3: Evolution of slab subduction.
figure 3

a,b, Flat-slab subduction at 163 Ma (a) and 138 Ma (b). c,d, Flat-slab rollback and Cretaceous LMW formation at 105 Ma (c) and 90 Ma (d). e,f, Subducted horizontal slab descending at 80 Ma (e) and 68 Ma (f). Flat-slab rollback processes and dynamic topographic evolution modelled by case 3-2 are shown in vertical profile c (bottom row) and map view (top row) for mutual identification. The horizontal dashed line is the phase transition and viscosity jump at a depth of 660 km. For each profile, slabs are represented by a blue colour delineated by a non-dimensional temperature of <0.45, and mantle flow velocities are shown with black vectors. IZA, Izanagi plate; EUR, Eurasian plate; PAC, Pacific plate. The thin white lines on the map view represent the continental boundaries16. The present-day location for profile c is shown in Fig. 1a.

Full size image

Starting at approximately 139 Ma, the eastwards motion of the overlying NCC slowed and gradually shifted southwards. Additionally, the downward traction induced by the sinking slab intensified. These plate–mantle dynamics may have triggered rollback of the slab’s leading edge beneath the NCC, commencing at approximately 136 Ma (ref. 10). The persistent slab rollback, accompanied by a viscosity jump and a phase change across the 660 km discontinuity, facilitated the horizontal subduction of the Izanagi slab within the mantle transition zone ahead of the leading edge. This process led to the formation of a notable mantle wedge, referred to as the large mantle wedge (LMW), between 130 and 90 Ma (Fig. 3c,d and Supplementary Video 1). This LMW formed through the coalescence of the horizontal slab within the transition zone and the steep frontal slab in the upper mantle36. Mg isotope analysis of volcanic rocks from eastern China indicates the presence of the LMW, with recycled carbonates from the subducted slab into the upper mantle, which formed carbonated peridotite that underwent initial melting at depths exceeding 360 km above the mantle transition zone (the subducted horizontal slab)37.

During the interval from 89 to 80 Ma, the complete rollback–subduction of the flat slab resulted in a tectonic inversion event18, during which the slab experienced tearing while retaining some of the original structural attributes of the LMW (Fig. 3e and Supplementary Video 1). However, between 80 and 55 Ma, the palaeo-Pacific slab descended into the lower mantle37, causing the LMW to disappear with renewed slab rollback (Fig. 3f). Throughout this period, the initially subducted horizontal slab gradually descended into the lower mantle across the 660 km phase transition boundary beneath the NCC. Hence, the emergence and collapse of the LMW beneath the NCC were intricately linked to flat-slab subduction and rollback.

The dynamic topography calculated by the mantle flow model in case 3-2 closely matches the residual topography, confirming the plausibility of the dynamic topographic outcomes. During the interval from the Early Jurassic (178 Ma) to the Early Cretaceous (137 Ma) (Extended Data Fig. 8a–c), dynamic subsidence exhibited a distinct NE‒SW pattern. Subsidence centres exceeding −600 m, aligned parallel to the leading edge of the flat-slab subduction zone, migrated inland from the eastern margin of the NCC to the Taihang Mountains (Fig. 3a,b). This dynamic subsidence resulted primarily from the downwards pull and mantle flow driven by flat-slab subduction, which greatly contributed to the subsidence of thrust-top basins during the Jurassic to Early Cretaceous. After 136 Ma, the eastwards rollback of the flat slab led to a synchronous eastwards shift in the dynamic subsidence centres. This migration coincided with relative uplift in the NCC core west of the Tan–Lu Fault, terminating rifting and causing the absence of coeval strata in that area (Figs. 2 and 3c,d and Extended Data Figs. 7a–d and 8d,e). Notably, as the LMW formed from 100 Ma to 90 Ma, the width of the subsidence centres, defined by the −700 m subsidence contour, expanded westwards accordingly (Fig. 3d and Extended Data Fig. 8f). From 79 Ma to 55 Ma, as the Izanagi horizontal slab in the mantle transition zone gradually sank, the dynamic subsidence belt expanded (Fig. 3e,f and Extended Data Fig. 8g–i).

Linking the LMW to craton deformation

Our four-dimensional geodynamic model constrained by deforming plate reconstructions suggests that the formation process of the LMW induced by flat-slab subduction and rollback played an important role in craton deformation. Robust stress transmission and tectonic interplay between the oceanic flat slab and the overlying NCC resulted in inwards-shifting basement-involved thrusts and associated thrust-top basin formation with crustal thickening24,38 before the inception of the LMW. Due to this shortening deformation, the crust (or lithosphere) compression factor (β) decreased to a minimum of approximately 0.65, while the thickness increased by approximately 54% compared with its initial thickness, with more intense and longer-lasting deformation further east (Methods, Fig. 4a,b and Supplementary Fig. 7). This process was accompanied by westwards-younging magmatism from the craton margin to its interior4 and concurrent hydration and metasomatism of the subcratonic lithospheric mantle. This process occurred between 180 Ma and 137 Ma. Similarly, in the western United States, flat-slab subduction led to the formation of basement-involved thrusts and broken foreland basins11, accompanied by modification of the basal CLM12,39,40.

Fig. 4: Conceptual model illustrating LMW formation and its relationship with craton deformation.
figure 4

a,b, Flat-slab subduction at 163 Ma (a) and 138 Ma (b). c,d, Flat-slab rollback subduction at 120 Ma (c) and 90 Ma (d). We delineate the key components of kinematic vectors in this model: the overriding plate (O), the downgoing plate (D), the trench (T) and the leading edge of the flat slab (L). Despite the motion of the overriding plate, it remains anchored in the model and is marked by a dot. The velocity lines or marked dots on the plates and plate boundaries illustrate the following: the plate convergence rate between the motion vectors of the downgoing and overriding plates, represented by D; the overriding plate shortening and the trench advance in a and b; and the extension and trench retreat rate in c and d between the trench and overriding plate, represented by T; the subduction rate between the downgoing plate and the trench, represented by D minus T in a and b and D plus T in c and d; and the advance in a and b and the rollback rate in c and d between the leading edge of the flat slab and the overriding plate, represented by L. a and b represent inland flat-slab subduction, while c and d illustrate subsequent backwards flat-slab rollback. DT, dynamic topography; CC, continental crust; CLM, cratonic lithospheric mantle; OC, oceanic crust; OLM, oceanic lithospheric mantle.

Full size image

During LMW formation (Fig. 4c,d), the eastwards retreat of the western Pacific subduction zone and the rollback of its flat slab triggered progressively younger magmatism, episodic crustal extension and continental lithospheric thinning eastwards. Fascinatingly, the western boundary of rifting migrated synchronously with the eastwards rollback of the leading edge of the flat slab, shifting from the region west of the Taihang Mountains (136 Ma to 110 Ma) to the area east of the Tan–Lu Fault (110 Ma to 90 Ma) (Figs. 2 and 4c,d). Rift activity predominantly occurred throughout the NCC, situated above the rolling-back flat slab, while the tectonics in the region west of the LMW, situated in front of the leading edge of the rolling-back slab, remained relatively stable. This geological evidence strongly supports the concept that lithospheric rifting was a result of the interaction between the subducting flat slab and the overlying craton, with the rollback of the flat slab relative to the overlying NCC leading to its extensional deformation. This observation suggests that the extensional processes within the NCC intensified towards the east, coinciding with a corresponding increase in lithospheric thinning by a maximum of approximately 26% compared with its initial thickness, reaching a β value of approximately 1.35 (Supplementary Fig. 7c,d). At this stage of the LMW formation, intense metasomatism and partial melting may be induced at the interface between the oceanic slab and the asthenospheric mantle, along with heating and erosion at the base of the sub-CLM4,41,42, and even lithospheric removal in the lower cratonic lithosphere. These processes may be crucial for lithospheric thinning during tectonic extension4,41.

After the complete rollback of the subducted flat slab, the LMW transitioned into an avalanching stage (for example, ref. 43), descending through the transition zone into the lower mantle. Following approximately 10 million years of basin inversion and uplift on the east Asian continental margin, the western Izanagi plate underwent prolonged subduction rollback until 55 Ma. This event once again triggered rifting and lithospheric thinning to the east of the Tan–Lu Fault in the NCC. With the emergence of a new Cenozoic LMW21,44, the NCC to the east of the Taihang Mountains entered a renewed phase of intense back-arc extension and lithospheric thinning10,45,46. In this dynamic deep mantle setting, the lithosphere experienced initial compression and subsequent extensional thinning, ultimately leading to disintegration of the craton. We conclude that the formation and evolution of the LMW induced by flat-slab rollback subduction were the primary causes of craton destruction. Our results indicate that cratons located closer to subduction zones are more likely to destabilize during their lifespan than those situated in continental interiors. This study documents the intricate interactions among subduction, mantle processes and craton deformation, creating opportunities for broader investigations.

Methods

Deforming plate reconstruction

The global plate tectonic reconstructions (Extended Data Fig. 2) utilized in this study were built on the basis of previous models15,17. The method of linking rigid plates with deforming plates was developed as described in a previous study47. Reconstruction model 1, based on the global reconstruction by ref. 15, spanned from 230 Ma to the present. This reconstruction involved rigid plates and plate boundaries without explicit deformation modelling. The global absolute reference frame depended on the moving hotspot reference frame described in ref. 48, linked to a palaeomagnetic reference frame corrected for true polar wander for earlier times.

Reconstruction model 2, derived from the global deforming plate model of refs. 16,48, covered the period from 250 Ma to the present. The model incorporated the deformation model of ref. 10 for Northeast Asia. The absolute reference frame was established using the iterative optimization technique detailed by refs. 49,50, ensuring reasonable net lithospheric rotation and absolute trench migration, and was fit to available hotspot trails.

The newly updated reconstruction model 3, developed specifically for this study, extended the model of ref. 16 back to 410 Ma. This extension utilized the relative plate motion framework introduced by ref. 17 for periods older than 250 Ma. Based on geological evidence, we identified the Mongol–Okhotsk Ocean (MOO) as a distinct plate and reconstructed its closure along with the orogenic processes responsible for the formation of the Tuva–Mongol–Okhotsk orocline51. Furthermore, we identified the northwards or northeastwards motion of the Amur and North China Block during the Jurassic to Early Cretaceous52, which contrasts with the predominant westwards motion of the northeast Asian continent in reconstruction models 1 and 2 (Extended Data Fig. 2). Our updated reconstruction model depicts the NCC moving northwards or northeastwards and the scissor-like closure of the MOO, accompanied by eastwards trench retreat along the western Pacific from 200 to 130 Ma. In addition, new research35 has indicated that Izanagi plate subduction along the northeast Asian margin initiated at approximately 200–180 Ma. In the model before this period, we introduced an intraoceanic subduction zone in the northwestern Panthalassa to address uneven oceanic crust consumption, based on ref. 53.

To incorporate distributed deformation data for East Asia since 200 Ma, we utilized data, including information on faults and folds, structural sections, geological maps, timing and various other deformation-related elements. These data provide accurate kinematic parameters for geological features, such as points, lines and surfaces obtained from balanced structural profiles. Deformation zones were divided into smaller triangular topological grids using the Delaunay triangulation algorithm, enabling linear interpolation within each triangle to calculate the velocity and strain rate at any position within the deforming region (Fig. 1b–g).

In deforming regions, horizontal divergence or convergence results in vertically uniform thinning or thickening. The rate of thinning or thickening is directly proportional to the surface dilatation rate, which is computed from the divergence of the surface velocity field within the deforming region. Via calculation of the dilatation rate, stretching or compressional factors can be derived (Supplementary Fig. 7). This provides insights into variations in the thickness of the crust (or lithosphere) relative to the initial stages of deformation16.

Mantle flow modelling

We utilized the CitcomS package54 to compute time-dependent global mantle-flow models, which solve the thermal convection problem of an incompressible fluid within a spherical shell using the finite element method. The average lateral resolution of the surface in this spherical domain is approximately 50 km, with an average radial resolution of approximately 45 km. However, to enhance the resolution, mesh refinement techniques were applied, resulting in resolutions of less than 20 km near the surface and core–mantle boundary. These mantle flow models were constrained by the tectonic history extracted from the reconstruction of the deforming plate, which encompasses various boundary and initial conditions55. The temperature fields of the lithosphere and slabs above 350 km in the upper mantle were progressively assimilated using an ideal thermal boundary field, which was created in a manner similar to that used for the initial conditions, following the evolving tectonic reconstruction for each million-year period. For nodes without assimilation below a depth of 350 km, the mantle convection process is purely dynamic. We developed a series of global models, beginning at 200 Ma, 230 Ma, 250 Ma or 310 Ma, and the reconstructed tectonic history was assimilated into the flow.

For case 1, case 2 and case 3, we employed deforming plate reconstructions from previous global models, as in refs. 15,16, as well as our own newly reconstructed global plate motion model (Extended Data Fig. 2). While these models exhibit different plate kinematics, particularly in East Asia, we set a uniform approach for determining the subduction history, with consistent flat-slab subduction and rollback from 180 to 90 Ma and subduction modes involving a normal dip (45°) from 89 to 55 Ma (Fig. 1b–g). In case 3, the Pacific plate along the East Asian margin was subjected to high-angle subduction from 50 to 22 Ma, unlike in cases 1 and 2 (ref. 21). Additionally, in case 3, we employed a normal dip (45°) for the northwestern subduction zone from 250 to 130 Ma and a steep dip (50°) from 250 to 220 Ma, transitioning to a transform fault from 219 to 130 Ma for the southeastern subduction zone in the MOO. However, in cases 1 and 2, we maintained a normal dip (45°) for the subduction zones in the MOO, and in all the cases, a normal dip was employed for the Tethys Ocean. By integrating the velocity boundary conditions derived from the tectonic reconstruction of the deforming plate, a series of global flow models were established (case series 1, 2 and 3) utilizing diverse parameter settings for viscosity layering, the Clapeyron slope and constants (Extended Data Table 1).

Our initial approach was to use a reference case series (case 1; ref. 21). Our reference case was case 1-1 (ref. 21), as this case leads to a computed upper mantle structure and a horizontal slab in the mantle transition zone, which is in agreement with seismic images. We computed variations in viscosity layering and the Clapeyron slope (cases 1-1 to 1-6) under identical subduction boundary conditions to assess the effects of mantle physical properties and identify an optimized model with suitable parameters. This demonstrated the effective implementation of viscous layering within the lithosphere and mantle, encompassing various parameters (Extended Data Table 1). The case 2 and case 3 series were formulated for comparative analysis with the case 1 series. This comparison across different case series proved essential for evaluating the impact of accurate and comprehensive plate motion reconstruction in East Asia, the closure timelines of the MOO, and the subduction processes of plates along the eastern, southern and northern margins of the NCC.

Mantle convection and viscous drag exerted by a subducting slab play crucial roles in inducing lithospheric deformation, resulting in variations in surface topography, known as dynamic topography56,57,58. The computation of dynamic topography is a vital component of mantle flow modelling. We computed the dynamic topography driven by mantle flow and its temporal evolution using the mantle flow model described earlier following the methodologies outlined in refs. 22,56. By considering density data beneath depths of 350 km and integrating compositional and thermal field data, we derived the dynamic topography of the surface.

Differentiating the residual subsidence

The Jurassic to earliest Cretaceous Ordos Basin experienced widespread intracontinental subsidence with peripheral shortening deformation24 (Supplementary Fig. 6). We employed a flexural backstripping approach for stratigraphic records spanning from 174 to 153 Ma (or to 136 Ma) across three well sections perpendicular to the western margin of the Ordos Basin (Supplementary Fig. 5), as outlined in refs. 57,58. The backstripped results reveal distinct long-wavelength anomalous subsidence components, termed residual subsidence. Furthermore, we differentiated residual components from the total subsidence in thrust-top basins in Yanshan (Extended Data Fig. 6e,f).

In the context of rift phases, the observed tectonic subsidence encompasses both water-loaded isostatic subsidence and residual subsidence45. The calculation of water-loaded isostatic subsidence, attributed to continental thinning and thermal contraction, was conducted using the equations outlined by McKenzie (1978)59 and White (1994)60. A crucial parameter, the stretching factor, was derived from a balanced seismic profile or a reconstruction model of a deforming plate. We selected the Fuxin Basin and the Laiyang and Zhucheng Basins within the Jiaolai Basin to compute the water-loaded isostatic subsidence and residual subsidence (Supplementary Fig. 1), following the methodology described in ref. 21.

Geodynamic model validation

To understand mantle dynamic processes during craton evolution, accurately reconstructing the subduction process of oceanic plates and constructing a four-dimensional plate–mantle system model are essential. The present-day slab architecture in the mantle, as inferred from seismic tomography, records plate subduction processes integrated through time21. Validating a reasonable mantle flow model that effectively reproduces the explicit structure and evolution of residual slabs in the lower mantle while also revealing their characteristics and origin is of paramount importance. To ascertain the level of agreement between the predicted mantle structure in various modelled cases and the global seismic tomography observations from the MIT-P08 and GAP-P4 models along the East Asian margin, we conducted a quantitative analysis spanning profiles extending into the lower mantle and computed the prediction accuracies for all the profiles across the cases. The method for this quantitative analysis followed the approach introduced by ref. 21. This analysis aimed to identify the case in which the proposed model demonstrated superior performance in predicting the current mantle structure.

Surface observations of long-wavelength topography, also referred to as residual topography, play a crucial role in assessing the mantle flow history over time21,22,58. To accurately evaluate the four-dimensional dynamics of the plate–mantle system and verify the accuracy of plate subduction beneath cratons, we conducted three comparative analyses using the observed residual topography. First, we compared the predicted present-day dynamic topography with the observed present-day residual topography. We utilized a root-mean-square analysis method, as detailed in ref. 21, to assess the consistency and correlation between the predicted dynamic topography and residual topography, thus evaluating the precision of our flow models. Second, we compared the predicted dynamic topography at specific times derived from geodynamic models and the observed accumulated residual topography up to those times. This evaluation aimed to assess the accuracy of our models by examining the consistency in spatial distribution trends, amplitudes and wavelengths between them (in profiles or planes). The third method involved a quantitative comparison of evolving dynamic topography and residual topography differentiated from intracontinental stratigraphic records at different locations. These multiple quantitative comparative methods enabled us to evaluate the temporal evolution and consistency in the spatial distribution between the predicted and observed values, validating the effectiveness of the mantle flow model.

As the residual topography at the baseline of the stratigraphy represents the starting point for residual topography when the basin was initially filled with sediments and often remains unknown or is preserved within the underlying strata (Extended Data Fig. 9), the residual subsidence extracted from various methods generally exhibits an increasing trend during the initial evolutionary phase. Therefore, when contrasting evolving residual topography with dynamic topography, the following scenarios or approaches were considered. First, during periods characterized by continuous growth in the predicted dynamic subsidence, an initial subsidence value could be introduced into the residual subsidence calculation. This value was extrapolated from older stratigraphic profiles in neighbouring areas or from the same region (Extended Data Figs. 6e and 7b,d,e). Alternatively, this initial subsidence could be omitted, and only the evolutionary trends of residual subsidence could be compared. However, during the initial stages of evolution, the trends of residual and dynamic topographies typically did not closely align (Extended Data Fig. 7a,c). Second, if the predicted dynamic topography corresponded to a period of nonsubsidence, an evolutionary period of stabilized residual subsidence was recommended for comparison with the dynamic topography (Extended Data Fig. 7f).