Channelized lithospheric erosion shapes the western North American craton margin Authors: Xiaotao Yang, Lijun Liu, Zebin Cao Published: 2026 ABSTRACT AbstractModification of craton margins influences topographic evolution, magmatism, and mineralization, though the scales, geometries, and driving mechanisms remain debated. Previous studies of the western North American craton margin focused primarily on the Colorado Plateau, invoking hydration by the subducting slab as the controlling factor. Here we construct a high-resolution 3-D shear-wave velocity model of this margin using full-wave ambient noise tomography. Our model reveals a distinct inward retreat of the high-velocity cratonic keel with depth along the entire margin, not limited to the edge of the Colorado Plateau, as previously thought. Two laterally confined (150-200 km wide) low-velocity zones, reflecting channelized lithospheric weakening, extend 500-600 km into the craton, reaching the Black Hills in two opposite directions. Their spatial correlation with strong mantle flows suggests that this flow can erode the craton itself. Kimberlite and basalt geochemistry data suggest that this entire margin undergoes progressive and laterally heterogeneous lithospheric erosion and may eventually delaminate, facilitating progressive inward and lateral lithospheric thinning. This progressive erosion process drives the migration of volcanism and regional topographic uplifts. Our findings provide insights into the detailed characteristics and mechanisms of lithospheric modification at craton margins, shaping the long-term evolution of cratonic lithosphere.Subject terms: Geophysics, Seismology Introduction Cratons are relatively stable continental cores, characterized by old1, thick2, strong3, and seismically fast4 lithosphere. Here we define the “North American craton” in the broad seismological sense of the stable continental interior east of the Cordilleran orogen, encompassing both Archean nuclei (e.g., the Wyoming craton) and Proterozoic provinces (e.g., Yavapai-Mazatzal), rather than a single Archean craton (Fig. 1a). Craton margins are in direct contact with tectonically active orogenic provinces, with sharp lithospheric thickness changes, making them more susceptible to deformation. This deformation along the margin potentially localizes the formation of base metals5–9 and diamonds10,11, making them important locations for mineral exploration. However, the detailed characteristics (e.g., scales, geometry, and distribution) and mechanism of such deformation and its influence on magmatism and surface topography, although important for the long-term evolution of cratons and mineral exploration, remain poorly constrained. In this study, we examine the structural modification of the western margin of the North American cratonic lithosphere east of the Phanerozoic accreted terranes (Fig. 1a) through full-wave ambient noise tomography, complemented by geological analyses and geochemistry of volcanic rocks. Our study area covers both the craton margin domains (Rocky Mountains, Colorado Plateau, Wyoming Basin, and Great Plains) and the Phanerozoic orogenic provinces (Snake River Plain, Basin and Range, and Rio Grande Rift) to the west and southwest (Fig. 1a). Tectonic settings and seismic stations. a Major tectonic domains and age of basaltic volcanoes. The black outlines show major physiographic domains west of the Rocky Mountain Front, digitized after ref. 23. The inset map shows the study area (green box) and the outline of the region with the identified Precambrian basement as in ref. 102). The colored dots are the Cenozoic basaltic volcanoes since 20 Ma60. The red arrows highlight major younging directions toward the craton. The background image shows the thermal lithosphere-asthenosphere boundary (LAB) defined as the 1300 °C adiabat21,69. Only LAB depths of 50 km and deeper are plotted. RM Rocky Mountains, SRP Snake River Plain, WB Wyoming Basin, NBR/CBR/SBR Northern/Central/Southern Basin and Range, RGR Rio Grande Rift, SRM Southern Rocky Mountains, YS Yellowstone. b Seismic stations (black triangles) used in this study. The red outlines show major physiographic domains west of the Rocky Mountain Front, digitized after ref. 23. The dashed blue line delineates the Rocky Mountain Front. The red star magenta square and start mark the locations of Yellowstone and the Black Hills, respectively. The western North American craton margin has undergone multiple tectonic events since the late Cretaceous. The Laramide Orogeny, which started at around 90 Ma and reached the Black Hills at around 60 Ma12, caused basement-cored uplifts at the Wyoming Basin as well as the initial rise of the Colorado Plateau during the early Cenozoic. However, the subsequent uplift history of the Colorado Plateau remains elusive, with the timing, amplitude, and mechanism of the uplift being heavily debated13–18. Since about 30 Ma19,20, the Colorado Plateau was partially separated from the North American cratonic interior to the east by the extension along the north-south trending Rio Grande Rift. Magmatic activity has been migrating toward the interior of the craton along multiple paths since the early Miocene (Fig. 1a). The lithosphere around this craton margin thickens from about 50–100 km below the northern Rocky Mountains and the central Colorado Plateau to >200 km below most of the Great Plains to the east21 (Fig. 1a). In contrast, the lithosphere below the Phanerozoic accreted terranes and the Rio Grande Rift is <50 km21 (Fig. 1a), forming a prominent lithospheric step across the craton margin2,21–23. The strong variation in lithospheric thickness at this location can focus edge-driven convection and stress localization23–29, facilitating intraplate crustal deformation and seismicity in the western United States23. Previous studies of lithospheric modification along the western North American craton margin primarily focused on the edges of the Colorado Plateau and the Rio Grande Rift13,26,30–32. Seismic tomographic studies over the past two decades that cover the western U.S.31,33–41 have provided valuable structural constraints on the first-order velocity contrast between the orogenic provinces and the Great Plains, approximately across the Rocky Mountain Front. However, the detailed along-strike and depth-dependent variation of this velocity contrast, which is essential for understanding the mechanism, mechanical process, and scale of craton modification along the margin, is yet poorly resolved. It remains unclear how the craton margin evolved to its present state, how and at what scale it has been modified by past tectonic events and mantle flows, and how it influenced volcanic activity and surface tectonics over time. In this study, we develop a 3-D shear-wave velocity model of the western North American lithosphere using full-wave tomography with Rayleigh wave empirical Green’s functions (15–100 s) from ambient noise cross-correlations. The model images pronounced along-strike variations at the transition from the Colorado Plateau to the adjacent cratonic interior, including a broadly weakened mantle lithosphere that follows the western craton margin and extends locally into the Great Plains. Together with comparisons to prior seismic models and independent geologic and geodynamic constraints, these observations are consistent with a view in which the craton margin has been progressively modified since the Cenozoic, with deformation and weakening focused along specific segments of the margin. Our results provide a regionally coherent seismic framework for evaluating how mantle flow and surface processes interact to reshape cratonic lithosphere over long timescales. Results To refine the lithospheric structure and characterize its deformation along the western North American craton margin, we constructed a 3-D shear-wave velocity model of the crust and upper mantle across the western U.S. We utilized a finite-frequency full-wave ambient noise tomography method with continuous waveforms from 1088 permanent and temporary seismic stations from 32 networks (Fig. 1b and Supplementary Tab. 1). See “Methods” for a detailed description of the tomographic steps and parameters. See Supplementary Figs. 1 and 2 for examples of Rayleigh wave empirical Green’s functions and the data coverage. We selected the preferred velocity model after four iterations when major seismic velocity features became stable (see Supplementary Fig. 3 for the comparison of models from different iterations). The phase delays reduce with each iteration of our inversion, with the most significant reduction after iterations 1 and 2. After four iterations, the improvement in fit between iterations is statistically negligible. With this final model (after iteration 4), the average phase delays are close to and median statistically indistinguishable from zeros (Supplementary Figs. 4 and 5). The improvement in the velocity model is also evidenced by the comparison of synthetic and observed Green’s functions (Fig. 2), with significantly better matches in phase arrivals and waveforms with the synthetics computed using the final model than those computed using the initial model. Since the tomographic imaging method exploited in this study uses only Rayleigh waves, the resultant shear-wave velocity model reflects the velocity structure of vertically polarized shear waves. Although both azimuthal and radial anisotropy may be present in this region [e.g., refs. 42–46], imaging the first-order isotropic shear-wave velocity structure remains a standard and useful step for robustly characterizing the dominant 3-D heterogeneity. The velocity model presented here (Figs. 3–5 and Supplementary Figs. 6) serves as a critical foundation for future work to investigate anisotropic structures as a second-order parameter. The best resolved depth range is from 20 km to about 170 km (Supplementary Figs. 7 and 8; see “Methods” for detailed model resolution analysis). The following paragraphs summarize key shear-wave velocity features within the mantle lithosphere along the craton margin. Shear-wave velocity contours (4.5 km/s) from 74 to 159 km. a Map-view of the velocity contours color-coded by depth. The magenta lines show the locations of the vertical profiles as in Fig. 5. Only contour segments with enclosed surface areas of ≥5000 km2 are plotted. b 3-D perspective view of the velocity contours. Improvements in waveform matches at 30–60 s between IU.TUC and other stations. The blue dotted wiggles are synthetics computed using the reference velocity model. The red solid wiggles are those computed using the final velocity model. The map on the right shows the distribution of all stations used in this study (black open triangles) and those selected in the example (blue filled triangles). For visibility, we only plot the results for stations within 450–600 km with a signal-to-noise ratio (SNR) above 7. The SNR of the observed Green’s functions is computed as the ratio between the maximum absolute amplitude of the signal (plotted segment) and the standard deviation of the absolute amplitudes of the entire trace. The average SNR of the negative and positive lags is used in selecting the traces. Depth slices of the shear-wave velocity model. a–c Shear-wave velocities at 74, 97, and 140 km depths. Major structural domains are the same as in Fig. 1. The white lines show the velocity contours at 4.5 km/s. The L1 and L2 in (c) label the two elongated low-velocity zones. Vertical cross-sections of the shear-wave velocity model. a–d Cross-sections from 20 to 170 km. The solid and dashed black lines are the Moho36 and negative velocity discontinuity71 from receiver function analyses. The dotted black line is the thermal lithosphere–asthenosphere boundary (LAB) defined as the 1300 °C adiabat21,69. See Fig. 4a for profile locations. The line segment with a diamond marks the LAB depth range from kimberlite thermobarometric studies for a nearby group of kimberlite samples (see Fig. 6 for sample locations). The shear-wave velocity model shows a prominent contrast below about 70 km across the Rocky Mountain Front, with velocities increasing from the Phanerozoic accreted terranes toward the cratonic interior (Figs. 3–5). At 74 km depth, an SSW-NNE trending high-velocity anomaly is imaged below the north-central Colorado Plateau and the central Wyoming Basin, connected with the broader high-velocity region below the Great Plains. While previously imaged, the reported scale, geometry, and amplitude of this feature vary significantly across different models13,34,36,38,41,47–50. By about 140 km, this high-velocity anomaly retreats to only below the Wyoming Basin region. Meanwhile, from approximately 70 to 140 km, the western limit of the high-velocity (≥4.5 km/s) volumes below the Great Plains retreats inward relative to the shallower margin (Figs. 3b, c and 5a, b). This inward retreat of the high-velocity front extends across the entire margin with various magnitudes. The largest retreats occur locally where the 4.5 km/s velocity contour follows the low-velocity lineaments at depth oblique to the margin. The average eastward shift of these features is more modest. The inward retreat of the high-velocity lithosphere with depth corresponds to a prominent reduction in velocity below 110 km (Figs. 3c and 5). These low velocities are scattered below the Snake River Plain, the Basin and Range, the Colorado Plateau, and the Rio Grande Rift. They are up to 8% lower than the mantle lithosphere velocity below the Great Plains (Supplementary Fig. 6b, c) but are not as slow as the low-velocity anomaly (up to 14% lower) below the Yellowstone region. Two elongated low-velocity zones below the Great Plains (L1 and L2 anomalies in Figs. 3 and 5) extend toward the cratonic interior for as much as 500–600 km beyond the lithospheric margin, as measured down the axes of the surface lineaments approximately following the Rocky Mountain Front. The L1 zone extends southeastward from the Rock Mountain Front to the north of the Black Hills (Figs. 3c and 5a). The L2 zone extends north-northeastward from the eastern edge of the Rio Grande Rift to the south of the Black Hills (Figs. 3c and 5b). These two low-velocity anomalies are laterally localized within 150–200 km zones (Fig. 5c, d), with a velocity reduction of approximately 6–8% compared to the surrounding Great Plains (Supplementary Fig. 6b, c). Our model can sufficiently resolve the scales, geometry, and amplitudes of the described low-velocity anomalies (Supplementary Figs. 7–11). The depth-dependent inward retreat and the detailed geometry and extent of the two low-velocity channels were ambiguous in previous velocity models34–41 that we compared with (Supplementary Figs. 12–14; see “Methods” for a detailed comparison). Discussion Our velocity model reveals two elongated low-velocity zones (L1 and L2) along the western North American craton margin, extending inward into the cratonic mantle lithosphere. Vertically, these anomalies occupy a “sandwiched" position between the mid-lithosphere discontinuity (MLD), an internal velocity contrast within cratonic lithosphere, and the deeper cratonic lithosphere-asthenosphere boundary (LAB), which is the base of the mechanically rigid lid. This geometry favors localized degradation at the base of the cratonic lid, potentially via metasomatism, melt-assisted weakening, or shear localization, rather than a simple asthenospheric intrusion. While not uniquely diagnostic, azimuthal anisotropy provides critical context: the P-wave fast directions at 100–220 km remain broadly coherent across the Great Plains and dominantly parallel/subparallel to the Rocky Mountain Front48 (Fig. 6a). The lack of notable localized rotation tied to the mapped L1/L2 trends suggests that while anisotropy does not require strong channel-parallel flow confined to L1/L2, it also does not exclude broader sub-lithospheric contributions. In the following sections, we evaluate the physical mechanisms behind these velocity reductions, characterize the nature of marginal craton erosion, and discuss the implications for long-term cratonic stability. Progressive erosion of the western North American craton margin. a Simplified shear-wave velocity map at 140 km overlain by the present-day mantle flow at the same depth (blue arrows) from ref. 23. Only mantle flows ≥0.1 cm/year are plotted, highlighting the concentration of flow at lithospheric steps. Locations of the kimberlites and lamprophyre samples are from refs. 61,62, with the age (in Ma) and source depth (in km) shown from reference61. The double-headed arrows show the overall P-wave fast directions at 100–220 km48. The colored dots are the basalt groups in (b) and (c), numbered consistently. b Basalt groups with decreasing minimum basalt-derived lithospheric thickness60 since 20 Ma. We grouped the samples east of longitude −115° and south of latitude 42° into 0.5° by 0.5° spatial grids, followed by a grid search to identify the groups with decreasing minimum LAB depths. After linear regression, we excluded groups/grids with R2 < 0.85 and slope < 0.1. c Slope of the linear regression for each basalt group in b as a proxy for the rate of lithospheric thinning over time. We use the slopes of 2 and 10 km/Myr to divide the results into categories with high, moderate, and low thinning rates. Margin-wide differential lithospheric weakening The shear-wave velocity within the L1 and L2 low-velocity zones is significantly lower (about 4.3 km/s or 6–8% reduction) than the surrounding high-velocity cratonic mantle lithosphere within the Great Plains region (>4.5 km/s; Figs. 3 and 5; Supplementary Fig. 6). Multiple factors can contribute to shear-wave velocity reductions within the mantle lithosphere. Pure compositional variations in mantle peridotites, such as a decrease in magnesium number, typically contribute <1% to shear-wave velocity reductions51 and are thus insufficient to explain the high-amplitude anomalies observed here. Conversely, thermal anomalies exert a primary control on seismic velocities and lithospheric strength52. While purely anharmonic (elastic) models51 require excessive temperature increases (>400 °C) to achieve an 8% reduction at 110–170 km (average of 140 km), the incorporation of temperature-dependent anelasticity and grain-boundary relaxation53–55 reconciles these observations with more moderate thermal anomalies of approximately 220–270 °C, assuming a grain size of 5 mm (see “Methods”). The impact of these thermal anomalies is likely further amplified by metasomatism resulting from the infiltration of hydrothermal fluids or melts56–58. Although the direct seismic impact of hydrous minerals like amphibole or chlorite is limited by their phase stability at depths exceeding 100 km56, the introduction of even trace amounts of volatiles can significantly enhance anelastic relaxation and lower the solidus. In this framework, volatile-induced metasomatism amplifies the temperature sensitivity of the shear-wave velocity, allowing for the observed 6–8% velocity drops. This coupled thermo-chemical weakening decreases the structural integrity59 and increases the density57 of the lithospheric mantle, potentially localized along the margins of the North American Craton, thereby facilitating the potential of lithospheric delamination 3,13. While inherited lithospheric variations are possible, the confined low-velocity zones we observe are more consistent with later-stage modifications. Their depth-dependent geometry and spatial correlation with mantle flow and geochemical anomalies suggest active processes rather than static, inherited structures that caused the imaged variations in seismic velocity structures. In particular, as detailed in discussions below, the low-velocity zones align with regions of strong mantle flow inferred from geodynamic simulations23 and with geochemical signatures from basaltic volcanism60 and kimberlites61,62 (Fig. 6). These correlations suggest that the depth-dependent, confined seismic velocity anomalies along the western North American craton margin are more likely linked to ongoing or recent lithosphere-asthenosphere interactions, rather than ancient tectonic imprints. Geometry and spatial extent of craton margin retreat The cumulative effect of the thermo-chemical weakening discussed above is reflected in the large-scale architecture of the craton margin, which exhibits a pronounced and systematic inward migration with increasing depth. The inward retreat of high-velocity mantle keels, shown as the lateral offset between the 4.5 km/s contours at depths of 74 km and 140 km, respectively, delineates a widespread depth-dependent lithospheric modification along the entire western North American margin (Fig. 6a). The width of this modified lithospheric zone increases with depth (Fig. 3c). It also shows a large variability along the margin (Fig. 6a). The farthest inward modification is observed below the Colorado Plateau-Wyoming Basin region, with an inward retreat of over 600 km from the central Colorado Plateau to the Rocky Mountain Front. The L1 and L2 low-velocity zones correspond to a retreat of approximately 500–600 km into the Great Plains cratonic lithosphere toward the Black Hills. Unlike the southern and southeastern Colorado Plateau, where the entire mantle lithosphere is missing13, the margin-wide differential reduction in seismic velocity suggests a relatively mild ongoing or recent modification of the mantle lithosphere, not yet being removed and replaced by asthenospheric materials. The along-strike heterogeneity of the lithospheric modification (Fig. 6a) suggests strong spatial variability of the driving forces, which could deviate from the overall margin tectonics and plate motion patterns. The western North American craton margin, with flat subduction of the Farallon slab and edge-driven mantle convection, provides viable conditions for lithospheric modification through thermal erosion, melt infiltration, and metasomatism. As the thickness of the lithosphere increases from the orogenic margin to the craton, edge-driven mantle convection is expected24–26 (Fig. 6a). The intruding hot mantle flow would produce asthenospheric melts25,63,64, thermally eroding the thicker cratonic mantle lithosphere and leading to metasomatism. On the other hand, the cratonic mantle lithosphere was in direct contact with the subducting Farallon slab during the Laramide Orogeny30. The flat subduction of the Farallon slab may have released excess water and other volatiles to hydrate the overriding lithosphere9,13,25,30,31,59, thus metasomatizing the western margin of the North American craton. Furthermore, strong mantle flow and small-scale convection can concentrate at the corners of lithospheric blocks with lateral steps of lithospheric thicknesses, such as the regions between the Colorado Plateau-Wyoming Basin block and the Great Plains23,25 (Fig. 6a). Time-dependent mantle flow simulations suggest that the concentration of strong flows at these two lithospheric corners was fully developed at around 10 Ma, with hot asthenospheric materials slowly migrating inward64,65. These mantle flows would cause stress localization that enhances the mechanical weakening of the surrounding lithosphere. This localized weakening could be further facilitated by hydration of the lithosphere29. Indeed, these corners defined by lithospheric thickness steps also spatially coincide with the approximate locations of the two low-velocity zones with the highest degree of inward retreat (Fig. 6a). Furthermore, compression during the Laramide Orogeny31,66, as well as the later extensional phases that formed the basin and range province31 and the Rio Grande Rift19,20,67, would all have contributed to some degree to the lithospheric deformation along the western North American craton margin. It is worth noting that the mantle flow simulations shown in Fig. 6a23 used a composite LAB model from LITHO1.02 and body-wave tomography68, interpreting continuous fast anomalies down to about 250 km as the thick continental lithosphere. The LAB depths in Figs. 1a and 5 were derived from thermal modeling based on seismic velocity data21,69, which yield relatively shallow estimates beneath the Colorado Plateau. Despite differences in absolute LAB depths in LAB models [e.g., refs. 2,21–23,70,71], they all consistently identify a lithospheric step around the edge of the Colorado Plateau and approximately below the Rocky Mountain Front. This structural contrast around the Colorado Plateau is sufficient to deflect mantle flows, as demonstrated in the simulation23. The magnitude and spatial extent of mantle flow deflection around the plateau margin may vary depending on the LAB models. While hydration from subduction has been invoked as the primary mechanism for craton margin weakening13,26,72, craton margin modification may persist even without active subduction. The locations and geometry of the channelized erosion toward the Black Hills spatially coincide with strong mantle flows (Fig. 6a). As discussed earlier, the directions of these localized mantle flows are controlled primarily by the heterogeneous lithosphere thickness along the strike23, independent of the overall direction of subduction (Fig. 6a). Therefore, the long-term lithospheric weakening at craton margins and its channelization may extend much further inward from the current convergent plate boundary. Nonetheless, the hydration with fluids from subduction would accelerate the weakening of the cratonic lithosphere29. Progressive lithospheric thinning over time Kimberlites and basalt geochemistry data suggest progressive metasomatism and thinning of the cratonic lithosphere over time along the western North American craton margin. Thermobarometric data from kimberlites61,62 suggest a thicker paleo-lithosphere than the modern lithosphere along the craton margin (Fig. 6a). The source depths of the kimberlites and lamprophyre samples were about 150–170 km around the northern Rocky Mountain Front during 47–52 Ma, 175–190 km in the central Rocky Mountain Front near the north end of the Rio Grande Rift during 186–614 Ma, and 130 km below the central Colorado Plateau during 20–30 Ma61,62 (Fig. 6a). These depths are consistently larger than their present values of 150 km, 140–150 km, and 50–70 km in the nearby regions, respectively, indicating thinning of the metasomatized lithosphere over time. Along the Rio Grande Rift, Cenozoic lithospheric extension during rifting could have also contributed to the thinning of the lithosphere. To better quantify lithosphere thinning, we further analyzed the paleo-lithospheric thicknesses from basalt geochemistry60 (Fig. 6b, c). The result suggests spatially variable thinning rates of the lithosphere since the early Miocene (Fig. 6b), implying progressive and differential lithospheric modification over time and space. From the database, we conducted a grid search in 0.5° by 0.5° spatial grids from lower-left to upper-right to identify basalt groups with decreasing minimum lithospheric thicknesses. We categorized the sample groups by the lithospheric thinning rate after applying a linear regression. After excluding groups with R2 < 0.85 and slope <0.1 (equivalent to lithosphere thinning rate of <0.1 km/Myr) after linear regression, we identified 12 basalt groups (Fig. 6b). The lithosphere thinning rate ranges from 0.6 to 65.8 km/Myr. The basalt groups can be categorized into three major clusters, with high (≥10 km/Myr; groups 1–3 and 12), moderate (between 1 and 10 km/Myr; groups 4–8 and 10–11), and low (≤1 km/Myr; group 9) thinning rates (Fig. 6c). The high-rate groups 1–3 are located at the southwestern edge of the Colorado Plateau, where a surface lift of approximately 1 km has occurred since 5 Ma73, consistent with the removal of 50 km metasomatized mantle that has an excess density of 2%74. Group 12 has the highest thinning rate of roughly 65.8 km/Myr. It is located in the Raton–Clayton volcanic field in northeast New Mexico, where an active surface uplift of up to 114 m/Myr has been documented75. Groups with moderate and low lithospheric thinning rates are located near the southern and southeastern edge of the Colorado Plateau and along the Rio Grande Rift (Fig. 6a, c). The locations of these basalt groups with lithospheric thinning also coincide with the occurrence of relatively strong mantle flow (e.g., >2 cm/year) toward the cratonic lithosphere (Fig. 6a). Intruding hot mantle flow can locally increase the degree of metasomatism, leading to strong lithospheric modification. In addition, these locations may experience stress localization due to the concentration of basal traction imposed by the mantle flow22,23, ultimately leading to lithospheric delamination and rapid surface uplift. Notably, the lithospheric thinning rate is not constant through time for each group. Local and regional tectonic activities, temporal variation of mantle flows [e.g., refs. 64,65], heterogeneity of material properties, and the geometry and thickness contrast at the lithospheric step may all contribute to the temporal variation in the lithospheric thinning rate. The exact mechanism of the temporal variation is unclear and is beyond the scope of this work. Implications for the evolution of cratonic lithosphere The widespread reduction in seismic shear-wave velocities along the western North American craton margin and the low velocities within the elongated zones below the Great Plains suggest differential lithospheric modification. Combined with kimberlite data61,62, basalt geochemistry60, and numerical simulations from previous studies23,25,26, the inferred lithospheric modification may reflect an ongoing process of progressive erosion inward of the cratonic lithosphere. This progressive and differential inward lithospheric modification may be a common phenomenon at craton margins and play an important role in the long-term evolution of cratonic lithosphere. Building upon the observations along the western North American craton margin, we summarize and generalize the dynamic evolution of the lithosphere at craton margins in three stages (Fig. 7), including initial weakening (Stage 1), channelized progression (Stage 2), and lithospheric thinning (Stage 3). These stages describe the evolution of representative structural characteristics of a continuous process. Schematic illustration of channelized progressive erosion of craton margins. a Initial weakening due to the infiltration of fluids and partial melts from the underlying asthenosphere (Stage 1). b Channelized progression of lithospheric weakening with increased metasomatism (Stage 2). The structural characteristics of this stage are based on and generalized from the velocity cross-sections in Fig. 5. c Lithospheric thinning with the foundering of the weakened mantle lithosphere (Stage 3). Mantle flow patterns are based on refs. 23,25,26,42,48,76. The left panels highlight the processes perpendicular to the margin, while the right panels highlight the along-strike variation of these processes with channelized erosion. UML upper mantle lithosphere, MLD mid-lithosphere discontinuity, LML lower mantle lithosphere, LAB lithosphere–asthenosphere boundary. Stage 1: Initial weakening with infiltration of hydrothermal fluids and partial melts (Fig. 7a). The step in lithospheric thickness, which increases from the orogenic provinces to the craton, facilitates small-scale edge-driven mantle convection along the margin23–25. The mantle flow brings hot asthenospheric materials toward the craton. These materials, together with fluids and partial melts from dehydration of the subduction slab13,26,72, weaken the mantle lithosphere by heating and metasomatism. Strong laterally variable asthenospheric flow toward the margin may impose spatially variable stress and lead to channelized weakening of the cratonic mantle lithosphere. This process gradually erodes the cratonic mantle keel and induces magmatic activity along the margin. This lithospheric weakening may occur within laterally confined zones. While subduction plays an important role in providing fluids for metasomatizing the western North American craton margin, it may not be required for generalized lithospheric erosion at all craton margins. However, subduction would accelerate the weakening process. At this stage, the LAB is relatively sharp, with a prominent contrast in seismic velocities across it. Stage 2: Channelized progression of lithospheric modification along laterally localized zones (Fig. 7b). The focused weakening in Stage 1 establishes strong lateral strength contrasts along the margin. As the lithospheric modification process continues and progressively migrates in narrow channels inward and laterally, the cratonic mantle keel will eventually be completely metasomatized. Meanwhile, with continued inward lithospheric erosion at the margin, the magmatic activity also migrates toward the cratonic interior. The lithosphere below the orogenic provinces thins due to sustained thermal erosion, accompanied by continued surface uplift and stronger edge-driven mantle flows24,76,77. The decreased seismic velocity of the metasomatized cratonic mantle keel56 would smooth the velocity contrast across the LAB, resulting in a more transitional velocity boundary. The imaged margin-wide inward retreat of the high-velocity front and the elongated low-velocity zones (L1 and L2 in Figs. 3–5) at the western North American craton margin are consistent with the structural modification in Stage 2. Stage 3: Lithospheric thinning due to delamination or gradual erosion of the modified cratonic mantle keel (Fig. 7c). The basal traction toward the craton imposed by the mantle flow23 would facilitate the delamination or erosion of the modified mantle keel, causing lithospheric thinning. Consequently, the LAB becomes much shallower as asthenospheric material replaces what was previously the lower part of the mantle lithosphere. The compression at the convergent margin31 may also contribute to triggering the potential delamination1 and surface uplift78. Following thinning of the cratonic lithosphere, the front of small-scale mantle convection at the lithospheric edge would also migrate inward toward the new step. The magmatic activity continues to migrate inward. Migration of convective mantle flows, combined with asthenospheric upwelling due to delamination, would lead to long-wavelength surface uplift of the craton76,79,80. The location of the lamprophyre sample in the center of the Colorado plateau also coincides with the 4.5 km/s velocity contour at 74 km depth. This suggests that the lithosphere to the south and southwest of this sample has been thinned to <74 km thick from >130 km at 20–30 Ma (Fig. 6a). This is consistent with the structural characterization of Stage 3. The proposed three-stage model represents a conceptual framework describing the progressive lithospheric modification encroaching craton margins within laterally confined zones or channels. Given the continuity of the proposed evolution stages, the exact time span of each stage is not constrained by the current data. It likely varies significantly from place to place, as shown by the basalt data with spatially and temporally varying lithospheric thinning rates (Fig. 6). Future studies with geodynamic modeling of these stages and a more comprehensive dataset of basalt geochemistry and kimberlites would help to further evaluate the proposed lithospheric erosion model. At a specific craton margin, these three stages may coexist along different segments. For instance, along the western North American craton margin, Stage 3 (delamination and lithospheric thinning) is primarily localized below the southern Colorado Plateau and eastern edge of the Rio Grande Rift. Other regions along the margin remain in earlier stages of progressive weakening. Given the observed along-strike variability, it is plausible that some regions may not evolve to full delamination (Stage 3). With the inward and lateral progression of the lithospheric weakening and delamination, the mantle lithosphere along the entire margin would be gradually or episodically removed. With time, the lithospheric step would become sufficiently gentle so that strong small-scale convective mantle flows are absent. Our model provides insights into the scale, geometry, and process of the long-term evolution of the continental lithosphere and its dynamic interaction with the underlying mantle. Future numerical simulations are necessary to quantify the mechanical process and the criteria for the continuation and termination of this cyclic margin evolution model. Methods Extraction of Rayleigh wave empirical Green’s functions We downloaded continuous vertical component seismic waveforms from January 2007 to December 2012 recorded by 1088 seismic stations within the study area from the EarthScope Data Management Center (Fig. 1b). These stations are from 32 permanent and temporary seismic networks. See Supplementary Table 1 for detailed information on all seismic networks used in this study and associated references. We downloaded the continuous waveforms as 12-h segments and removed the instrument responses after resampling to 1 Hz. To compute the cross-correlations, we sliced the 12-h waveform record into 6-h short-term segments with a two-hour overlap between segments. The short-term segments were tapered using a Tukey window with a cosine fraction of 0.05, followed by demeaning and detrending procedures. To remove transient signals with abnormally large amplitudes, we calculated the ratio between the maximum absolute amplitude of each 6-h segment and the standard deviation of the absolute amplitudes of all segments. We discarded waveform segments with the above ratio greater than 10. We computed the waveform cross-correlations in the frequency domain using the SeisGo Python toolbox81,82. Before cross-correlating the waveforms, we normalized the spectrum of the short-term waveform segments with a frequency-domain moving window average method. We extracted the Rayleigh-wave empirical Green’s functions (EGFs) as the negative time derivatives of the waveform cross-correlations in the time domain83,84. We then used the Robust stacking method81,85 to stack the short-term EGFs within 30 days to enhance coherent signals. To be consistent with the synthetic waveforms computed in the full-wave tomography workflow, we convolved the EGFs with a Gaussian wavelet with a standard deviation of 2 s and a shift of 6 s. Full-wave ambient noise tomography The full-wave ambient noise tomography workflow includes three major steps83,84: finite-difference simulation of Rayleigh waves, measuring Rayleigh wave phase delays between synthetics and observed EGFs, and least squares inversion for P and S wave perturbations to update the reference model based on 3-D finite-frequency sensitivity kernels. In the following paragraphs, we briefly describe these steps and the corresponding parameters. Step-1: finite-difference wave simulation. We simulated wave propagation in a 3-D spherical Earth using the nonstaggered-grid, finite-difference method86. The simulation used a Gaussian source time function with a standard deviation of 2 s and a time shift of 6 s, the same as the Gaussian shaping wavelet convolved with the EGFs. Our simulation grid size is 0.05 degrees laterally in the latitudinal direction from 31.2° to 50.5° and 0.0764° in the longitudinal direction from −118° to −98°. The vertical grid size increases with depth from 1.83 km at the surface to about 11 km at 1005 km depth, which is the bottom of the simulation grid. There were 260, 384, and 136 simulation grid points in the longitudinal, latitudinal, and vertical directions. There were 12 absorbing boundary grid points at each side of the model, except for the free surface. The simulation time step was 0.125 s, saved every 8 steps or 1 s. The 3-D strain tensors from the simulation were saved every 4, 4, and 2 steps in the three spatial directions for the calculation of the sensitivity kernels and the inversion. Considering the minimal influence of starting models87, we used the AK135 1-D model88 as our initial model for the first iteration. We chose this global 1-D model to remove the potential influence of lateral variation in the starting model on lateral velocity heterogeneities in the resultant model. The wave simulation did not account for attenuation or anelasticity. We conducted a synthetic test to evaluate the effect of attenuation on Rayleigh wave phase velocity dispersions (Supplementary Fig. 15). The total shear-wave velocity reduction in the mantle is a combination of anharmonic (purely elastic) and anelastic effects. While the anharmonic velocity, Vanh, accounts for the temperature and pressure dependence of the elastic moduli at an infinite frequency, the observed seismic velocity, V(ω), at a given frequency ω is lower due to physical dispersion. Following the framework of refs. 53 and 89, the reduction in the observed phase velocity at a certain frequency (dV(ω)) relative to the purely elastic velocity is expressed as: where α represents the frequency dependence of the attenuation Q−1. In our test, we used α=0.3, following experimental constraints on mantle olivine53. Supplementary Fig. 15 compares the velocity reduction for stable continents (Q=20090), active continents (Q = 11190), and a high-attenuation anomaly (Q = 60). The results show that while anelasticity contributes to a velocity reduction of about 0.3–1.6% across our period band, this effect is co-located with the thermal anomaly and/or metasomatism, which further suppresses seismic velocities51,53,56. Therefore, the presence of localized low-velocity features L1 and L2 (with 6–8% velocity reduction relative to the more stable craton) is a robust feature. Step-2: measuring Rayleigh wave phase delays. We measured the difference in Rayleigh wave arrival times between the observed EGFs and the synthetics through cross-correlations. We searched for the time delay maximizing the correlation coefficient between the two waveforms. We measured the phase delays separately for the negative and positive lags and then averaged them as the final phase delay. We used five overlapping period bands, including 15–30, 20–40, 30–60, 40–75, and 60–100 s. We selected phase delay measurements for the tomographic inversion step using the following criteria: (1) For the first iteration, we manually removed negative phase delays <−8 s that may suffer from cycle skipping, showing decreasing absolute values with distance. The minimum signal-to-noise ratios for the five periods are 7, 9, 11, 11, and 11. The minimum correlation coefficient between the EGFs and the synthetics is 0.75; (2) for iterations 2 through 4, we increased the minimum correlation coefficient to 0.8 while keeping the same signal-to-noise ratio threshold. We used different thresholds to balance the number of measurements of each period while keeping the total dataset manageable with the available computational resources. The number of phase delay measurements for the four iterations is 286,489, 304,735, 313,742, and 314,579. Supplementary Fig. 2 shows the distribution of ray path counts within each 0.5-degree grid using the phase delay measurements for the final (fourth) iteration. Most of the study area is sampled by more than 100 ray paths. Step-3: inversion for velocity perturbations with 3-D finite-frequency sensitivity kernels. We calculated the three-dimensional, finite-frequency sensitivity kernels of Rayleigh waves at the above five period bands for all phase delay measurements that passed the quality control thresholds. We considered the influence of both P- and S-wave velocities on the propagation of Rayleigh waves91. The influence of P-wave velocities on Rayleigh wave propagation is limited to the top 10–15 km83,91,92. The inclusion of P-wave velocities in calculating the sensitivity kernels helps reduce the influence of P-wave velocity uncertainties on deeper shear-wave structures83,92. This would lead to more accurate measurements of phase arrivals and more robust solutions for shear wave velocity perturbations during the inversion. Following refs. 83 and 84, the phase delay, δt, caused by velocity perturbations of the P (Δmα) and S (Δmβ) can be expressed as: where κα(mr, x) and κβ(mr, x) are the 3-D finite-frequency structural sensitivity kernels of Rayleigh waves to P- and S-wave velocities, respectively, mr is the velocity at location x used in computing the synthetic waveforms, and dV is the volumetric increment for the integral. We calculated the sensitivity kernels with the strain-Green-tensor-based scattering-integral method93. The use of 3-D finite-frequency sensitivity kernels allows the resolution of structures smaller than the dominant wavelength, as demonstrated by previous studies83,94,95 With Eq. (2), we built the inversion as solving for the 3-D velocity perturbations, m, in where G is the matrix containing the structural sensitivity kernels and d is the measured phase delays at multiple frequencies. We solved for velocity perturbations using a damped least-squares scheme83,84,96 to match the phase delay measurements, using the conjugate gradient method to search for the optimal solution. Density was not directly solved in the inversion. Instead, we estimated the density using P-wave velocities based on the empirical relationships from rock physics experiments97. We chose a damping and smoothing coefficient of 8 for all iterations after comparing the solutions with a series of damping and smoothing coefficients, including 2, 4, 8, 12, and 16. We carried out tradeoff analysis (L-curves) between model norm and variance reduction (see Supplementary Fig. 16 for the L-curves for the first iteration) and visual inspection of the inverted velocity perturbations (Supplementary Fig. 17). The selected damping and smoothing parameters are approximately at the elbow of the L-curves. The comparison of inverted perturbations with different smoothing and damping coefficients demonstrates that the variation of damping and smoothing coefficients primarily affects the amplitudes and smoothness of the velocity features. The existence and overall geometry of the major seismic velocity anomalies analyzed in our model are largely consistent (Supplementary Fig. 17). In the least-squares inversion, we accounted for the uncertainties in the phase delays using the standard error of the mean phase delay measurements from the 30-day sub-stacks, combined with the simulation errors. See ref. 98 for a detailed description of the calculation of these uncertainties. The full-wave ambient noise tomography method is an iterative process containing the above key steps. The velocity model used in wave simulation was updated after each iteration using the velocity perturbations from the inversion (Step-3 above). The final preferred velocity model was selected after four iterations when the changes in the velocity model between iterations became negligible and major seismic velocity features become stable (Supplementary Fig. 3). The Rayleigh wave phase delays measured with the synthetics computed using the final velocity model show significant improvements compared to those computed using the initial AK135 velocity model (Supplementary Figs. 4 and 5). The final phase delays are concentrated statistically much closer to zero (Supplementary Fig. 5). The largest reduction in phase delays occurs in the first 1–2 iterations, while later iterations yield smaller but still measurable improvements (e.g., Supplementary Fig. 5), refining shorter-wavelength details. This is consistent with the convergence of key model features shown in Supplementary Fig. 3. Following common practice in iterative tomography, we carried out a small number of additional iterations after the main misfit reduction to confirm stability and avoid stopping on a transient solution. Resolution analysis We tested the model resolution in the horizontal and vertical directions with checkerboard tests (Supplementary Figs. 7 and 8). We built the input checkerboard models with varying cell sizes in the lateral and vertical directions. The velocity anomalies of adjacent cells alternate between −10% and 10%. We computed the synthetic phase delays using the same sensitivity kernels as for the tomographic inversion of the real data used for the final iteration. We then inverted the noise-free synthetic phase delays to recover the velocity perturbations, following the same procedure as for the real data. We tested adding 10% random noise to the synthetic phase delays in the checkerboard tests (Supplementary Fig. 7e–h). The differences in recovered velocity anomalies are negligible (Supplementary Fig. 7i–l), mostly below 0.4% at small (<50–100 km) spatial scales. Considering the smaller synthetic phase delays for the checkerboard models, we use two as the smoothing and damping coefficient in the least squares inversion. Considering the depth of the major velocity features discussed in this paper, our resolution analysis below is focused on the mantle lithosphere. The checkerboard tests show that we can recover shear-wave velocity perturbations of >60% recovery of the input perturbations at horizontal scales down to 110 km above the depth of about <74 km, 154 km down to 109 km, and 198 km at greater depths (Supplementary Fig. 7). The major seismic features (L1 and L2) discussed in this study are all larger than these resolvable scales at the corresponding depths. In the vertical direction, the checkerboard test results suggest a resolvable vertical scale of about 40 km with >60% recovery of the perturbation amplitudes down to a depth of around 160 km (Supplementary Fig. 8). At greater depths, the amplitude recovery of an anomaly on a scale of 40–50 km is approximately 20–30% in the center of the study area. To evaluate the resolution of the variation of the mantle lithospheric velocity along the margin, we conducted additional model recovery tests (Supplementary Figs. 9–11). After building the input models, we followed the same procedure as for the checkerboard resolution tests. We tested the resolution of the high-velocity mantle lithosphere with a vertical edge (Supplementary Fig. 9). The results show that we can recover the vertical edge with ≥60% amplitude recovery. We can also recover the mantle lithosphere with a retreating edge with over 60% amplitude recovery (Supplementary Fig. 10). The input model in Supplementary Fig. 11 simulates the narrow (about 200 km) low-velocity zones in the real velocity model (Fig. 3–5). The results show that both the geometry and velocity contrasts can be well resolved. It is worth noting that although small-scale variations of the detailed geometry of L1 and L2 may fall below the resolvable scale, the existence and orientations of these two low-velocity zones are robust. Comparison with previous velocity models Seismic tomography over the past two decades has produced multiple seismic velocity models at different scales covering the western U.S. We compared our velocity model with eight previously published and digitally available shear-wave velocity models34–41 at depths around 74, 97, and 140 km (Supplementary Figs. 12–14). Except for the GyPSuM model35, all other velocity models in this comparison used the EarthScope Transportable Array stations within the study area. As a reference, we plotted the velocity contour at 4.5 km/s from our velocity model on the velocity models at the corresponding target depth. We show the 74-km velocity contour on all depth plots to illustrate the inward retreat of the high-velocity margin. All of the examined velocity models show a distinct eastward increase in shear-wave velocities from the orogenic margin to the Great Plains, with significant variations in the scale, geometry, and amplitude of velocity anomalies (Supplementary Figs. 12–14). The location and geometry of the boundary between the low- and high-velocity anomalies (i.e., transition from negative to positive anomalies) vary notably across models along the strike at all plotted depths. At around 74 km, the localized high-velocity anomaly below the Colorado Plateau and the Wyoming Basin is imaged by all examined models, except for the two global models35,40 (Supplementary Fig. 12a, g). The outline of this high-velocity anomaly from the model by ref. 41 (Supplementary Fig. 12h) best matches the 4.5 km/s velocity contour extracted from the model by this study (Supplementary Fig. 12i). However, the inward retreat of the high-velocity margin at greater depths is very ambiguous in previous models (Supplementary Figs. 13 and 14). For example, at 140 km depth, the margin of the high-velocity anomalies in most of the velocity models (Supplementary Fig. 14a, c–g) is generally to the west and southwest of the 4.5 km/s velocity contour from our model. The remaining two models (Supplementary Fig. 14b, h) show a similar location of the transition from low- to high-velocity anomalies, though the geometry of this transitional boundary is much smoother than that from our velocity model (Supplementary Fig. 14i). In summary, while sharing the existence of major velocity features, our velocity model reveals (1) a distinct depth-dependent inward retreat of the high-velocity features below the Great Plains, (2) detailed along-strike variation of this retreat, exemplified as two narrow low-velocity channels below the Great Plains. The better-resolved details on the geometry, scale, and extent of the inward low-velocity anomalies and their depth dependency are critical for understanding the associated tectonic and mechanical processes. Thermal interpretation of seismic velocity reductions To estimate the temperature anomalies (ΔT) from the observed shear-wave velocity reductions (ΔVs), we employed a modeling framework that combined anharmonic (elastic) and anelastic effects. The model first calculated the anharmonic contribution using the temperature derivative of the elastic modulus54,55. Supplementary Table 2 summarizes the key parameters and related references used in this model. Following the mineral physics framework of ref. 51, we estimated the elastic sensitivity constant from a volume-weighted average of the primary mantle phases listed in their Table A1 of ref. 54 (60% olivine, 25% orthopyroxene, and 15% clinopyroxene). Using the laboratory-derived shear modulus (μ) and its temperature derivative (∂μ/∂T), the elastic sensitivity is calculated as: where μavg ≈ 79.5 GPa, ∂μ/∂T ≈ − 14.4 MPa/K, and the thermal expansion coefficient αavg ≈ 3 × 10−5 K−1. The resulting value is rounded to 1.1 × 10−4 K−1, consistent with standard benchmarks for a dry, pyrolitic upper mantle51. The purely elastic velocity reduction can then be computed as: To incorporate grain-boundary relaxation and pre-melting (near solidus) effects, we applied the anelastic scaling framework of ref. 54. This model amplifies the elastic sensitivity using a non-linear multiplier or scaling factor, M, computed as: where M¯ is the average multiplier between the cratonic baseline temperature (Tbase = 1100 °C 99) and the perturbed temperature (T = Tbase + ΔT). The multiplier, M, captures the enhancement of seismic sensitivity as mantle temperature approaches the solidus (Tsol). We assumed an effective volatile-bearing solidus of Tsol = 1450 °C at 140 km depth with 0.05 bulk wt% water100. The baseline multiplier with the reference grain size, dref, can be computed as: where T represents the absolute temperature, d denotes the assumed mantle grain size, dref is the reference grain size (5 mm), and p = 0.25 provided the grain-size scaling exponent53,55. Equation (7) is to approximate the pre-melt model fit in ref. 54 for the dataset by ref. 101. The scaling factor of 150 is an empirical constant that defines the temperature width of the anelastic transition, aligning with the observed normalized activation temperature at 0.94 in Table 4 of ref. 54. It is calibrated to match laboratory observations of seismic velocity reduction in the pre-melting regime54. This factor ensures that the transition from purely elastic behavior to grain-boundary sliding and pre-melt attenuation matches the observed sensitivities in the upper mantle, as in ref. 54. The offset (1.5) and scaling parameter (3.5) are based on the laboratory measurements by ref. 54. Specifically, our maximum multiplier of 5 for the reference grain size at the solidus is consistent with their derived pre-melting sensitivity parameter (γ = 5 in their Table 4). This model operates within the high-normalized-frequency regime, where the high-frequency relaxation peak is broad and is relatively insensitive to specific frequencies. Because the 15–100 s period range of our data closely matches the 20–160 s range of the dataset by ref. 101 used in the original calibration54, the numerical constants are effectively applicable to our target depth. Our modeling results illustrate the non-linear relationship between seismic velocity and the underlying thermal state (Supplementary Fig. 18). While elastic sensitivity remains constant, anelastic effects significantly amplify shear-wave velocity reduction as the temperature approaches the solidus (Supplementary Fig. 18a). This amplification is sensitive to grain size, with smaller grains further enhancing the velocity drop, requiring lower temperature anomaly (Supplementary Fig. 18b). For a representative mantle grain size of 5 mm, the observed 6–8% velocity reduction corresponds to a temperature anomaly of approximately 220–270 °C relative to the cratonic baseline. These results place the target region at an absolute temperature of 1320–1370 °C, indicating a mantle state that is significantly warmed but remains just below the volatile-bearing solidus. Supplementary information supplementary information Transparent Peer Review file Acknowledgements This work is supported by NSFC grant 92355302 and Strategy Priority Research Program (Category B) of the Chinese Academy of Sciences XDB0710000 to L. Liu and Purdue University startup funding and NSF award EAR-2319445 to X. Yang. The authors thank the PIs and field teams for the EarthScope Transportable Array and other flexible arrays within the study area that contributed data to this study. The authors thank Ryan Porter for sharing the thermal LAB data. This work benefited from the discussion with Joseph Byrnes. The computational resources used in this study are provided through the Purdue University Rosen Center for Advanced Computing. Author contributions X. Yang contributed to the conceptualization, seismic imaging, formal analysis, interpretation, and writing of the original paper. L. Liu contributed to the conceptualization, interpretation of the results, and the reviewing and editing of the paper. Z. Cao contributed to the reviewing and editing of the paper. Peer review Peer review information Nature Communications thanks Eva Golos, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available. Funding L. Liu is supported by NSFC grant 92355302 and Strategy Priority Research Program (Category B) of the Chinese Academy of Sciences XDB0710000. X. Yang is supported by the startup funding of Purdue University and the NSF award EAR-2319445. Data availability The continuous seismic waveforms are available through the EarthScope Data Management Center (https://ds.iris.edu/ds/nodes/dmc/), operated by the EarthScope Consortium that is funded through the Seismological Facility for the Advancement of Geoscience (SAGE) Award of the National Science Foundation under Cooperative Agreement EAR-1724509. All seismic networks used are referenced in Supplementary Table 1. The velocity model is available on Zenodo (10.5281/zenodo.17478703) and the EarthScope Earth Model Collaboration (10.17611/dp/emc.2026). Code availability The open-source Python toolbox SeisGo for extracting empirical Green’s functions from ambient noise is available on Zenodo (10.5281/zenodo.5873724). The Python scripts for waveform downloading and ambient noise cross-correlations are archived on Zenodo (10.5281/zenodo.17478703). The full-wave ambient noise tomography codes are available from the authors upon request. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Xiaotao Yang, Email: xtyang@purdue.edu. Lijun Liu, Email: ljliu@mail.iggcas.ac.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-73462-w. References Associated Data Supplementary Materials supplementary information Transparent Peer Review file Data Availability Statement The continuous seismic waveforms are available through the EarthScope Data Management Center (https://ds.iris.edu/ds/nodes/dmc/), operated by the EarthScope Consortium that is funded through the Seismological Facility for the Advancement of Geoscience (SAGE) Award of the National Science Foundation under Cooperative Agreement EAR-1724509. All seismic networks used are referenced in Supplementary Table 1. The velocity model is available on Zenodo (10.5281/zenodo.17478703) and the EarthScope Earth Model Collaboration (10.17611/dp/emc.2026). The open-source Python toolbox SeisGo for extracting empirical Green’s functions from ambient noise is available on Zenodo (10.5281/zenodo.5873724). The Python scripts for waveform downloading and ambient noise cross-correlations are archived on Zenodo (10.5281/zenodo.17478703). The full-wave ambient noise tomography codes are available from the authors upon request.