Introduction

Unlike the “ideal” scenario of plume-oceanic plate interaction where the surface develops a long-wavelength topographical swell1 (Fig. 1a), the rise of a plume toward continental lithosphere with a complex and layered rheological structure could cause “repetitive” topography variation across a broadly uplifted region2. Depending on the stress state of the lithosphere, ripple-shaped or Basin-and-Range-like topography variations (Fig. 1b) could potentially develop3. A previous numerical study further raised the possibility that the ponding of plume materials beneath the Mantle Transition Zone (MTZ) could induce broad surface subsidence (Fig. 1c)4. Therefore, the topographical response to deep-seated plume activities may potentially be diverse and might vary between different tectonic environments.

Fig. 1: Previous models of plume-induced topography response.
Fig. 1: Previous models of plume-induced topography response.
Full size image

a Dome-like long-wavelength uplift1. b Basin-and-Range-like short-wavelength topography response2,3. c Plume-ponding induced subsidence due to mantle phase transitions4.

Recent seismic tomography models and geological studies potentially reveal an additional type of topographical response to mantle-plume-related activity, which features anomalous subsidence in elongated regions of thin lithosphere. As shown in Fig. 2, extended but narrow (tens of to a few hundred kilometers wide) subsidence has developed in the vicinity of Yellowstone-Snake-River Plain (YSRP)5,6,7 and the South China Sea (SCS)8. Notably, these subsidence centers appear to lie above seismically imaged lower-mantle upwelling that may be impeded by cold anomalies9,10,11 (Fig. 2b, e), i.e., subducted/delaminated blocks not currently connected to a subducting/delaminating lithosphere10,11,12 (here, we will call these “remnant blocks”). In both scenarios, geological and geophysical observations suggest the presence of local ~1–2 km anomalous subsidence that cannot be easily explained by the regional tectonic evolution5,6,7,13.

Fig. 2: Localized anomalously low topography above the mantle domains with seismically imaged plume-remnant block interactions.
Fig. 2: Localized anomalously low topography above the mantle domains with seismically imaged plume-remnant block interactions.
Full size image

a Geologic sketch map of the western USA7. Contours of slow and fast seismic anomalies (≥0.6% P wave velocity change) are based on Zhao et al. (2024). HLP— High Lava Plains. b P wave tomography results along the transect A-A’ in a11. A similar scenario was also suggested to exist beneath this region back to ~30–20 Ma33. Data for this panel can be obtained from ref. 85. c Geologic sketch map of the South China Sea (SCS) region. The brown line demonstrates the region of anomalous subsidence center8. d Comparison between the SCS sea floor depth and the theoretical predictions for the subsidence of young oceanic lithosphere89,90. Sampling locations of U1431E and U1433B are shown in (c). e P wave tomography result along the transect B-B’ in c10. The cited 3-D tomographic models are available from ref. 86.

The observed extra surface subsidence has been traditionally believed to reflect the dynamic effects of nearby oceanic subduction14. However, because the dynamic effect of subduction tends to affect a broad region (1000 s of km)15, it cannot easily explain the intense subsidence in the ~100–200 km banded region of the western USA and the SCS7,8. In the YSRP, the depressed topography was suggested to be due to the existence of infiltrated mafic melts at the middle-upper crust depths, assuming that the YSRP’s crustal and lithospheric thicknesses are roughly equal to the surroundings16. However, more recent studies reveal that the YSRP crust is ~10–15 km thicker than the crust in the adjacent Northern Basin and Range (NBR) (Fig. S1)17. Even considering its basalt-enriched composition, the predicted elevation of the YSRP is ~1 km higher than observed18. In addition, anomalous subsidence could also be caused by the localized crustal thinning due to the lower crustal flow during the plume’s melt infiltration5. However, subsidence has been inferred to have happened nearly simultaneously throughout the YSRP and/or NBR region (Fig. 2a)7,19; In both YSRP and SCS, the Moho beneath the subsidence center is not shallower than beneath its surroundings20,21. Therefore, other deep-seated processes, like the previously discussed plume-upwelling-related activities (Fig. 1b, c) nowadays near both regions (Fig. 2b–e), could have played a role in causing these anomalous subsidence events.

As noted above, the topographic responses predicted by existing models (Fig. 1) do not appear to easily match the scenario of localized subsidence (Fig. 2a, c). In addition, as the temporal effect of a plume-remnant block interaction (Fig. 2b, e) was not considered, the specific mechanisms operating in this scenario remain to be analyzed.

Here, using numerical modeling, we study the surface topographic responses to interactions between lower mantle upwelling and remnant blocks. Our modeling results reveal that lower mantle upwelling can temporarily induce broad surface subsidence, whose magnitude and duration can be extended due to the existence of sinking remnant blocks above the upwelling’s path.

Results

We design three types of numerical models to explore the effects of plume-like mantle upwelling: 1) an initial hot mantle portion in the lower mantle only to trigger plume-like upwelling22 (Fig. S3); 2) a remnant block only in the upper mantle; and 3) both initial structures (Fig. 10 in the methods and S4 in the supplements). For each type, we summarize 623, 120, and 120 test results, respectively, to show the model’s sensitivity to the critical variables considered here.

Topographic response to lower mantle upwelling

In general, most models with a deep mantle upwelling predict dome-like surface uplift before the upwelling reaches the lithosphere-asthenosphere boundary (LAB) (Figs. S5b–S5c). When the plume materials reach the LAB, they impose divergent horizontal drag and cause short-wave-length topography variations in the uplifted region (Figs. S5d–S5f vs. 1b)2.

However, in some instances, the models predict surface subsidence with horizontal stretching before the upwelling reaches the LAB. Run 4 represents a typical model exhibiting this behavior (Fig. 3). In this model, lower mantle thermal anomaly drives mantle upwelling and convection. The surface experiences ~1 km topography variation within a 1500 km wide horizontal range at ~2.5–3 Myr (Fig. 3b, c), consistent with previous studies (Fig. 1c). By further analyzing the deviatoric stress within the lithosphere, we find that a brief, but prominent extensional period happens just before the “plume” arrives at the LAB, which explains the crustal thinning and surface subsidence (Fig. 3g–i). This extensional episode is mainly due to the basal drag of divergent mantle flow caused by the lower mantle upwelling (Fig. 3c). However, when the plume meets the LAB, it uplifts the LAB (Fig. 3g) and gradually reverses the topography (Fig. 3d–e, i). Towards the end of the model run, the topography changes back to a quasi-flat state (Fig. 3f).

Fig. 3: Result of a typical plume upwelling model (Run 4) (Fig. 4).
Fig. 3: Result of a typical plume upwelling model (Run 4) (Fig. 4).
Full size image

a–f Snapshots of model evolution. Gray arrows demonstrate the convection velocity. Astheno. – Asthenosphere. g Evolution of the horizontal deviatoric stress within the lithosphere. The deviatoric stress with the largest magnitude within the lithosphere is sampled every 5 km horizontally to reveal the dominant lithospheric deformation state. The range of this plot is between 1000 and 3000 km along the horizontal axis. h Comparisons of the crustal and lithospheric thicknesses between the model center (horizontal axis: 1750–2250 km) and its left flank (horizontal axis: 1250–1750 km). The lithospheric thickness increases briefly after ~3 Myr because when sampling the lithosphere thickness, both the lithospheric mantle and the melted asthenosphere (inset in (e) are treated as lithospheric materials. i Comparisons of the elevation between the model center (horizontal axis: 1750–2250 km) and its left flank (horizontal axis: 1250–1750 km).

More sensitivity tests for the models with lower mantle upwelling

Since the surface response to hot lower-mantle upwelling is affected by the thickness of the overriding plate and the phase transitions in the MTZ, in the models with lower mantle upwelling only, we mainly test seven variables to explore the full spectrum of possible dynamic regions, including the thickness of overriding plate (Fig. 5), the Clapeyron slope (γ410 and γ660, Fig. 4)23 and the density jump ratio (Δρ410 and Δρ660, Fig. 4) at the 410 km24,25 and 660 km26,27,28, and the belt width of the two mantle phase transitions (Fig. S7)29. We summarize model results in the form of checkbox diagrams to demonstrate the value of surface uplift or subsidence (Figs. 4–5, S7). Through the sensitivity test, we find that 1) surface subsidence mainly happens when Δρ660 is larger than 7%26,27,28, and γ660 is >-2.8 (Figs. 3–4); 2) the magnitude of topography uplift/subsidence increases with the decrease of the lithosphere thickness (Fig. 5a-d) while the wavelength of topographical variation does not change obviously, which is consistent with the previous 3D modeling results30 (Fig. 5e); and 3) the transition belt widths do not affect the modeling results, except when they are very narrow (i.e., 1 km) (Fig. S7).

Fig. 4: Sensitivity test results regarding the mantle phase transformations (model setup in Fig. S3).
Fig. 4: Sensitivity test results regarding the mantle phase transformations (model setup in Fig. S3).
Full size image

Here, we mainly consider the effects of the 410 and 660 km phase transformations. The belt widths of both mantle transitions are 25 km. γ410 and Δρ410- the Clapeyron slope and the complete density-jump ratio of the 410 km phase transformations. γ660 and Δρ660- the Clapeyron slope and the complete density-jump ratio of the 660 km phase transformations. R xxx indicates the model number. The specific number shows the most significant topography change (uplift/subsidence) between 1500 and 2500 km along the horizontal axis at 3.5 Myr. The result at 3.5 Myr was chosen because significant topography changes (start to) primarily happen at this time.

Fig. 5: Sensitivity test results regarding the lithospheric thickness (model setup in Fig. S3).
Fig. 5: Sensitivity test results regarding the lithospheric thickness (model setup in Fig. S3).
Full size image

a–d Summary of topographical variations. e Comparison of the topography at 3.5 Myr in four representative models. In the presented models, the belt widths of both mantle transitions are 25 km. γ410 and Δρ410 (5%)- the Clapeyron slope and the complete density-jump ratio of the 410 km phase transformations. γ660 and Δρ660 (9.6%)- the Clapeyron slope and the complete density-jump ratio of the 660 km phase transformations. Other plotting habits are the same as in Fig. 4.

Topographic response to a sinking remnant block

We also quantify the surface response to a remnant block in the deep upper mantle. In all models, the sinking cold block always causes surface subsidence (Fig. S8)15 with all tested values of initial length, thermal age (for calculating the thermal structure based on the half-space cooling model31), and the duration of the block’s stay within the mantle (e.g., Fig. S9). However, the subsidence magnitude is small (~200 m) (Fig. S8), not enough to explain the observed anomalous subsidence (Fig. 2).

Topographic response to upwelling-remnant block interactions

In the third model type, we further quantify the surface response to upwelling-remnant block interactions (Figs. 2 vs. 6; Figs. S6, S10). Because the number of free parameters is large, we only limitedly tested the effects of the scale, thermal structure, and compositional density of the remnant block and the temperature and scale of the lower mantle thermal anomaly (Figs. S6, S10).

Fig. 6: Result of a typical upwelling-remnant block interaction model (Run 760, Fig. S10).
Fig. 6: Result of a typical upwelling-remnant block interaction model (Run 760, Fig. S10).
Full size image

a–f Snapshots of model evolution. The remnant block is 400 km wide, and its crustal age and heating time are 80 Myr and 20 Myr, respectively. The insect in (d) demonstrates the thinning of the upper crust and thickening of the lower crust towards the subsidence center. g Evolution of the horizontal deviatoric stress within the lithosphere. The range of this plot is between 1000 and 3000 km along the horizontal axis. h Comparisons of the upper crust and lithosphere thickness between the model center and its left flank. i Comparisons of the total crustal thickness between the model center and its left flank. Other plotting habits are the same as in Fig. 3.

As shown in Fig. 6, the combined effects of mantle-upwelling-induced stretching and remnant-block sinking can induce >2 km topographic variations, with the subsidence center at a depth deeper than -1 km (Fig. 6a, b). This enhanced subsidence compared to the Type-2 models (Fig. S8) is because the lower mantle upwelling in this model accelerates the upper mantle convection velocity. Thereafter, the reduced viscosity of the upper mantle (the strain rate weakening effects, Equation S4) allows the remnant block to sink rapidly to cause a more prominent surface subsidence.

While the remnant block is cold and strong, it can divert the path of upwelling (Figs. 6c–d and 7 vs. Fig. 2b)11. Depending on the scale of the remnant block, the latter can either be pushed to be nearly vertical (Fig. 6e) or only mildly tilted by the lower mantle upwelling (Fig. 7c). Then, the upwelling reaches the LAB and uplifts the left shoulder of the subsidence region (Fig. 6d–g). Subsequently, the left shoulder keeps rising, while to the right, because this side is less affected by the upwelling, the topography remains low (Fig. 6e, f). As a result, the subsidence center shrinks to a ~ 200 km wide V-shape at the bottom (Fig. 6f).

Fig. 7: Result of typical plume-remnant block interaction models (a–c: Run 821; d–f: Run 813).
Fig. 7: Result of typical plume-remnant block interaction models (a–c: Run 821; d–f: Run 813).
Full size image

a–f Snapshots of model evolution. The remnant block is 1000 km wide; the seafloor age is 60 and 20 Myr in Run 821 and 813, respectively, and the heating time is 20 Myr (Fig. S10). Other plotting habits are the same as in Fig. 3.

Unlike the scenario shown in Fig. 3, the dynamic topography caused by the remnant block’s sinking, which induces downwelling in the upper mantle, enhances the initial subsidence magnitude (Fig. 6b). Later, the upper crust of the subsidence center experiences quick thinning after ~3 Myr (Fig. 6h). At the same time, the total crustal thickness gradually increases (Fig. 6d, i) as weak lower crustal materials flow from the compressive left shoulder to the extensive subsidence center (Fig. 6g)32. Because the lithosphere at the subsidence center is (comparatively) thinned more significantly than the crust (Fig. 6h) and is no longer affected by the mantle upwelling, its topography remains low during the following evolution of the model (Fig. 6f, i).

Instead, in models where the remnant block is large and weak (Fig. 7f), the plume can penetrate through it7,33. In this kind of model, the topography of the subsidence center can be partially reversed.

Discussion

Potential scope of model application

Based on previous studies, a lower-mantle plume tends to develop a cylindrical geometry during upwelling, especially within a homogeneous background mantle34,35. In 2D models, we need to assume that the variations of composition, velocity, and other physical properties are minor or smooth in the direction perpendicular to the model slice. In other words, an appropriate plume example for 2D models to simulate should have a prolonged ellipsoid-like shape. Therefore, simplified models like Run 4 (Fig. 3) may mainly fit the conditions where the mantle plume derives from the edge of the thermally and/or chemically abnormal piles at the core-mantle boundary36. However, if the overriding lithosphere is in an extensional state, the expansion of plume materials beneath the LAB would tend to develop in the direction perpendicular to the stretching3, more closely mimicking a 2D scenario. Furthermore, in situations where a plume meets remnant blocks (Fig. 6) or even penetrates through them (Fig. 7f), the induced mantle flow could lose its cylindrical pattern (Fig. 8b)10,11,37 again more closely approximating a 2D scenario. The existence of remnant blocks on the path of lower mantle upwelling could help focus the surface subsidence within a ~ 200 km range (Figs. 6h and 7c) and extend the subsidence’s duration and magnitude (Figs. 6 vs. 3). Here we call this the ‘lens effect’ of a remnant block.

Fig. 8: Potential 3D surface expressions of lower mantle upwellings.
Fig. 8: Potential 3D surface expressions of lower mantle upwellings.
Full size image

a Ripple-shaped topography variation when the lithosphere was previously in a neutral state3. b Valley-like short-wave-length surface depression due to the interaction with remnant blocks (this study).

Compared to the predicted long-wavelength surface subsidence (Fig. 1c) and the ripple-shaped (Fig. 8a) or Basin-and-Range-like (Fig. 1b) topographical variations in previous studies, the model of plume-remnant block interactions may better explain the anomalous subsidence in an elongated region of thin lithosphere (Figs. 2 and 6), which cannot be easily explained by variations in crustal properties6,13. Our numerical model predicts a localized subsidence center with asymmetrical surface topography on its flanks (Fig. 6), which closely matches the observed surface topography in the YSRP (Fig. 2a) and the SCS (Fig. 2c). Compared to the pure plume case (Fig. 3), the plume-remnant block interaction model shows a long-lasting subsidence center that does not fade away when the plume reaches the LAB (cf. Figs. 3f and 6f). This is because, in the latter scenario, the lithosphere below the subsidence center is ~25 km thinner than the flanks at the late stage of the model (Fig. 6h) and is not reversed by the rising plume materials (Fig. 6f). The model of plume-remnant block interactions also predicts thickening of the total crust with time (Fig. 6i) due to the flow of the weak lower crustal materials from the compressive region of the plume-lithosphere interaction towards the stretching center (Fig. 6g)32,38 (while it is still not enough to cancel out the effect of lithospheric thickness variation upon subsidence, Fig. 6f). The accumulation of the weak lower crustal materials may help explain the aseismic and unfaulted features of the YSRP39; the juxtaposition of the thinned lithosphere and thick crust also well aligns with the those beneath the western USA (Fig. S1). However, as the largest predicted surface subsidence is ~1.2 km, the up to 2 km anomalous subsidence in the YSRP and the SCS could also be partially augmented by the presence of mafic intrusions within the upper-middle crust16,18 or the locally increased density of the shallow upper mantle13.

The sequence of surface subsidence, plume-related magmatism, and the evolution of crustal and lithospheric thickness appear key to distinguishing the effects of different models. For example, in the regions where surface subsidence happened earlier than plume-related magmatism and depressed topography remained after the magmatism (YSRP-NBR in Fig. 2a6,7), the model of plume-remnant block interactions (Fig. 8b) might be the only appropriate model to explain the observed anomalous subsidence. In the scenario of YSRP-NBR, the initial subsidence happens around 23 Ma (Fig. 2a) before the nearby slab rollback event, and the plume-related magmatism or crust extension7,40,41; and the distribution of faults and magmatism here cannot be explained by the regional boundary forces42. Based on our model, the broad subsidence at ~23 Ma in this region could be due to the divergent mantle flow when the plume was still in the lower mantle and not affected by remnant blocks (Figs. 3c, 6b, c). Then, the interaction between the plume and remnant blocks could have deflected the mantle flow and caused the extensive lithospheric stretching and magmatism in the YSRP region at ~17 Ma7,43,44 (Fig. 6f), supporting the previously suggested role of mantle flow upon overriding lithospheric stretching7,45. As previous studies suggested a deep origin below the lithosphere for the abnormal subsidence of the SCS13, the surface response to the divergent mantle flow driven by the plume in the lower mantle could play a role (Figs. 3c vs. 2e). However, since the SCS is below sea level, a detailed comparison between the model and existing sparse observations is more difficult.

Model-implied mantle transition zone condition

In the literature, the Clapeyron slope γ660 has been proposed to vary in a broad range between −0.4 to −4.0 MPa/K, while the range of γ410 is better constrained to lie within 1.5—3.5 MPa/K23. Δρ410 has been inferred to lie between 1%25 and 7%24, mainly around 5%25, while Δρ660 has been inferred to lie between 5.1% and 10.2%26,27,28. Based on these studies, we designed the above sensitivity tests (Fig. 4). Our modeling results demonstrate that lower-mantle-upwelling-induced surface subsidence mainly happens when Δρ660 is ≥7% and γ660 is >-2.8 MPa/K (Fig. 4). According to the previous lab experiments, the required large Δρ660 can happen during conditions of oceanic subduction26,28. As the MTZ in regions with long-term oceanic subduction is enriched in water46, the belt width of the 660-km phase transition could be increased26,29, and in this situation, γ660 would become >-2.8 MPa/K26,28,29. Therefore, the model-implied mantle conditions, which favor anomalous subsidence, appear to be broadly linked to subduction zones, i.e., to occur in the regions that motivated this study (Fig. 2).

Other candidates for plume-induced anomalous subsidence

Regional seismic tomography studies reveal the juxtaposition of slow and fast anomalies at the MTZ base beneath Northeast China, similar to those observed in the western USA and SCS (Fig. 9a, b)47. Because it has experienced long-term Pacific subduction, the fast anomalies within this MTZ have been generally interpreted to be fragments of stagnant slabs47,48, while some of them could also have been from the formerly delaminated cratonic root49,50,51,52. On the other hand, results of receiver function studies indicate that the MTZ thickness is thinner in regions above the slow anomalies compared to their surroundings53, and thus, slow anomalies below the fast could represent hot and ascending materials54. Global seismic tomography results further reveal a broad distribution of the lower mantle slow anomalies in adjacent regions, with roots extending to deeper than 1000 km and connecting to a potential mantle plume beneath central Mongolia (~45–55 °N, 100–110 °E) (Fig. S11)55,56. Therefore, interactions between lower mantle upwelling and remnant blocks appear likely to be active beneath northeastern Asia. If our model predictions were robust, anomalous subsidence should have developed in this region, too.

Fig. 9: Additional anomalous subsidence examples.
Fig. 9: Additional anomalous subsidence examples.
Full size image

a Sketched geological map of Northeast Asia57,59. BBB-Bohai Bay Basin. b P wave tomography results along the transect in a47. c Average S-wave tomography results at 750 km depth. We used the 15 S-wave tomography models without restricted depths in the Submachine to calculate the average87,88. The average results of S- and P wave tomography models at different depths are shown in Fig. S11. d–g Tectonic subduction results in the BBB57 and Songliao Basin59. See locations in (a).

Interestingly, the northern Bohai Bay Basin (BBB) above the region of the potential upwelling-remnant block interactions (Fig. 9a–c) recorded a post-extension anomalous subsidence event that accelerated at ~10 Ma (Fig. 9d, e)57 when the present-day stagnant slab had roughly reached this region58. The extra subsidence amplitude was up to ~1 km (Fig. 9d, e), which is a typical value predicted by our numerical models (Figs. 6 and 7c). Similarly, a post-extension anomalous subsidence event also occurred in the Songliao Basin (SLB) at ~105 Ma, with an amplitude of 700 m—~1 km (Fig. 9d, e)59. However, currently, the basin is not underlain by prominent lower mantle slow anomalies (Fig. 9c). Given that the northeast Aisa drafted southwards since ~105 Ma60, if we reconstruct the geographical location of the SLB back to that time, it appears to have lain above another branch of slow anomalies connecting to the potential central Mongolia plume (Figs. 9c and S11)55,56. Geological studies also suggest the occurrence of the Cretaceous slab stagnation or lithospheric delamination in this region, like that within the present-day MTZ (Fig. 2b)61,62. Therefore, the juxtaposition of anomalous subsidence, lower mantle upwelling, and remnant blocks also seems to have existed below the SLB (Fig. 8). In addition, even though the surrounding region experienced a broad uplift in the Cenozoic (e.g., Hangay dome to the west and Greater Xing’an range to the east)63,64, the elongated Mesozoic rifting basins above the potential plume in eastern Mongolia65 remain depressed (Fig. 9a). Given that geophysical and geochemical studies suggest that present-day stagnated slabs extend(ed) to central Asia47,66, surface subsidence induced by the plume-remnant block interaction is likely ongoing there too. However, the lack of quantitative constraints for the subsidence history of the eastern Mongolian basins precludes a further analysis like that for the above examples. In summary, the shallow responses to deep plume-remnant block interactions could also have contributed to the Cretaceous and Cenozoic localized anomalous subsidence events broadly distributed on the thin lithosphere of northeastern Asia.

Model limitations awaiting future investigation

In some of the models we present here, we assume that a lower mantle upwelling encounters a previously subducted/delaminated block that is not currently connected to an actively subducting/delaminating plate (Fig. 8b). Our idealized model setups are mainly designed to explore the potential topographical response to interactions between deep mantle upwelling and remnant blocks, without accounting for additional interfering factors such as nearby active subduction or the stability of the lower mantle upwelling.

In regions like the western United States and northeastern China, previous studies have shown that high-wave-speed anomalies within the mantle transition zone are typically associated with actively subducting plates67,68. If this observation is consistently accurate across these regions, the presence of a subducting plate would overshadow the lower mantle upwelling, leading to effects on upper mantle convection that differ from our model predictions (Fig. 6)69. However, recent localized seismological studies indicate that some, if not all, high-wave-speed anomalies in these areas are separated from the subducting plate by a seismic gap70 or a low-wave-speed ‘hole’ or channels11,54. Given that lithospheric delamination events could have happened in these regions, some high-wave-speed anomalies can be delaminated cratonic roots instead49,50,52. Indeed, the subducting plate itself adjacent to the target anomalies seems to have locally penetrated into the lower mantle independently12,71, particularly beneath regions experiencing anomalous subsidence, where some lower mantle upwelling may occur (Figs. 8 vs. 1 & 9). From this perspective, the simultaneous occurrence of lower mantle upwelling, remnant blocks, and localized anomalous subsidence could be dynamically connected by our proposed scenario (Fig. 8b). Nonetheless, we emphasize that our model cannot be generalized to other areas within these regions that lie outside the seismic gap. Further numerical modeling incorporating more complex initial geometries than those used here would be necessary.

Furthermore, our idealized model presupposes there exists an active lower upper mantle upwelling with a path in the lower mantle that remains relatively unchanged, similar to previous studies22. In nature, such vertical upwellings may be more transient. Nearby subducted slabs may induce changes in the lower mantle convection field and/or the position of thermal or chemical piles at the core-mantle boundary, potentially distorting the plume path over time36,72. This represents another potential complexity that our current modeling approach cannot readily address. However, given that our modeling timeframe, particularly for the occurrence of abnormal subsidence, spans only several million years (Figs. 6, 7), the simplified model setup here may still be valid for a preliminary exploration of surface responses to deep mantle processes (Fig. 8b).

In summary, we studied the potential effects of lower-mantle upwelling-remnant block interactions on inducing anomalous surface subsidence. Numerical models demonstrate that lower mantle upwelling can temporarily induce broad surface subsidence with a ~ 1500 km wide range. The existence of sinking remnant blocks above the upwelling’s path can extend the magnitude and duration of surface subsidence and focus the subsidence center to a ~ 200 km wide elongated region; the ‘lens effect’ of remnant blocks on underlying plume upwelling (Fig. 8b). This study further implies that in regions without other robust geological or geophysical evidence, the coexistence of anomalous surface subsidence, lower-mantle slow seismic anomalies, and fast seismic anomalies above the slow can probably be a diagnostic feature for a lower mantle plume.

Methods

We use a finite element code written in MATLAB to conduct these 2D numerical experiments38,73,74,75,76. This code uses well-vectorized block matrix-assembly techniques popularized by MILAMIN77. It includes a free surface on its top surface (a surface with no shear or normal stress)78 and can simulate mantle melting behavior that depends on the temperature, pressure, and melting degree (Fig. S2, Equations S10–S12)49,50. Because the largest density change in current models is <10%, the mantle should be well approximated by the Boussinesq approximation75. Therefore, we assume an incompressible viscoplastic rheology (Fig. 10c, e) that is governed by the conservation of mass, moment, and energy (Equations S1–S7). The phase transitions of mantle materials (olivine-to-wadsleyite and ringwoodite-to-bridgmanite mainly) are calculated from Equation S8–S9 (Fig. 10d)79,80. More details on our numerical method can be found in the supplements.

Fig. 10: Model setup (also see Figs. S3–S4).
Fig. 10: Model setup (also see Figs. S3–S4).
Full size image

a Initial and boundary conditions. In all models, the 2D numerical box is 4000 km wide × 2000 km deep. The finest resolution is ~5 km, where the mesh node is within 25 km from the nearest material interface, and it gradually increases to ~25 km, where the node is >120 km away from the closest material interface. The lithospheric thickness varies between ~60 and 120 km (see Fig. 5), which is composed of the upper and lower crust and lithospheric mantle. b Initial temperature structures. The transect locations are in a, with the corresponding line styles. The lithosphere portion follows a 1-D steady-state conductive thermal profile. The surface and bottom temperatures remain at 0 and 1350 °C. c 1-D Viscosity structure at 0.5 Myr. The transect locations are in a, with the corresponding line styles. d 2-D density structure at 0.5 Myr. e 2- D viscosity structure at 0.5 Myr.

Model setup

In all runs, we utilized a free surface boundary condition78 as most of the models would be applied to stretched continental margins; in submarine regions of the marginal sea (e.g., SCS), the topographic relief should be increased to reflect the extra load associated with water infill. Other boundary sections are set to be free slip. The overriding lithosphere has varied thicknesses (except for tests in Fig. 5 and S10, the lithosphere is ~65 km thick) and contains a 30 km thick crust (20 km lower crust + 10 km upper crust) (Fig. 10a), and its initial temperature structure is controlled by 1-D steady-state conductive thermal profile and the mantle solidus (Figs. 10b and S2)31,81. In most models with lower mantle upwelling, the geometry of the hot anomaly in the lower mantle comprises a 400 km× 200 km square and a 400 km× 200 km equilateral triangle (Fig. 10b), which could cause a ~ 1000 km wide and ~1 km surface uplift after the plume contacts with the lithosphere (Fig. S5), mimicking the scale of hotspot swells82,83; the effect of the hot anomaly scale is also tested (Fig. S6). The hot center (orange color in Fig. 10a) is initially 1650 °C84 (except for the tests in Fig. S6), and the temperature gradually reduces to 1350 °C towards the edge (within the gray region in Fig. 10a). The temperature and material structures are reinforced every time step within the 400 km× 200 km square region to approximate the continuous heat supply from the core-mantle boundary (e.g., Shi and Morgan, 2024). In models with the remnant block, the block initially lies above the lower mantle (600 km deep), and its initial temperature follows the half-space cooling model (Fig. 10b)31. We further assume that the remnant block has stayed within the mantle for a varied period and, thus, is warmer than the prediction of the half-space cooling model (Fig. 10a, b). Besides, the surface and bottom temperatures remain at 0 and 1350 °C, respectively. Key model parameters are shown in Table S1.