Introduction

In the Wilson Cycle, divergent plate boundaries are initiated by the rifting of continental lithosphere, leading to continental break up and the creation of oceanic crust along oceanic spreading centers1. Several competing geological and geodynamic factors control the breakup mechanism and determine rift configuration throughout the extension lifecycle2. However, a long-standing question remains on the rifting of the thick, old, cold continental lithosphere—a problem encapsulated by the tectonic force paradox3. The cratonic lithospheric strength far exceeds the available tectonic forces, implying that rifting requires other major driving mechanisms for weakening. Resolving this paradox will provide insight into the breakup of past supercontinents. Geodynamic models commonly employ thermal anomalies to promote strain localization by eroding the lithospheric mantle2. However, several active nonvolcanic (amagmatic, magma-poor) rifts with extensive crustal deformation show only minor lithospheric thinning4. Similarly, rifted margins show segments of voluminous volcanism and pervasive magmatic intrusions5 as well as segments with little or no volcanism6. Although inherited crustal and lithospheric structures may help to nucleate rift faulting, their influence on continued strain accommodation may not persist beyond the early stages of tectonic extension2. Thus, an important knowledge gap remains on how persistent tectonic strain is being accommodated along active continental rift zones where there is no evidence of volcanism.

The Cenozoic East African Rift System (EARS; Fig. 1a) is the largest active continental rift system on Earth. It is associated with the eastward deflection of rising hot materials from a mantle plume source that results in a magma-rich eastern branch and a magma-starved western branch7. Although magma-poor, the western branch has a few localized volcanic centers, among which are the Rungwe Volcanic Province just north of the Malawi Rift and the Virunga Volcanic Province in the Kivu Rift located further north (Fig. 1a). In between these volcanic areas is the broad nonvolcanic section of the rift branch hosting the NW-trending Tanganyika-Rukwa Rift Zone (Fig. 1b). Regional tomographic models show shear wave velocities indicative of unelevated mantle temperatures8. The rift zone is characterized by broad and elevated rift flanks, long border faults, and basins with tectonic extension that accommodate >7 km deep sedimentary basins of Mesozoic–Cenozoic age9. The rifts are surrounded by the Bangweulu and Tanzania cratons, with estimated lithospheric thicknesses up to 200 km4, and Proterozoic mobile belts with northwest-trending regional strain fabrics, which the rifts exploit during their development10. Seismic activity in the area is high, having hosted one of the largest African earthquakes in recent history, the 1910 Mw 7 Rukwa Rift earthquake event11. Within the southern Tanganyika Rift, earthquakes extend down to the lower crust and upper mantle (Fig. 1b, c). The rift hosts a narrow zone of increasing axial strain12 manifested by the collocation of clustered intra-rift faults featuring a prominent fault-bounded syn-rift wedge that increases in thickness with time (S1 to S3–6), at the deepest parts of the lake bed (e.g., white arrow in B–B’ in Fig. 1b inset).

Fig. 1: Study area maps of the East African Rift System.
figure 1

a Rift faults, earthquakes, and volcanic centers highlight the dichotomy of the western and eastern rift branches. b Tanganyika-Rukwa Rift Zone showing the locations of the seismometers and earthquakes used in the analysis20. Colored polygons define Archean cratons and Proterozoic mobile belts. Bottom left inset: seismic reflection profiles with interpretations12, showing that although both the border fault and rift axis are accommodating considerable tectonic strain, extension at the rift axis is increasing (bottom right inset). c Seismicity depth profile. Color-coded earthquakes in panels b and c represent events with reliable shear wave splitting measurements.

Full size image

The timing of Cenozoic rift initiation is not well-determined. Uranium-lead dating of carbonatitic tuffs in the Rungwe Volcanic Province suggests an upper bound of 25 Ma13, and detrital zircon dating of the Rukwa Lake sediments gives a lower bound of 8 Ma14. Geochemical analysis of nearby volcanic rocks indicates a plume source15. Until recently, geophysical monitoring has been too sparse to provide a detailed view of subsurface processes in the Tanganyika-Rukwa Rift Zone. The deployment of broadband seismographs has allowed some constraints on the crustal structure, which reveals that the crust has accommodated up to 20% extension and has anomalously high compressional-shear wave velocity ratios comparable to the magma-rich Main Ethiopian Rift in the eastern branch16,17. Observations of mantle anisotropy indicate a dominant rift-parallel (NW-trending) orientation, interpreted to be associated with a local deflection of the regional mantle flow by the surrounding cratons and oriented melt pockets in the upper mantle18. Combined with observations of high surface heat flow and hydrothermal activities19, these previous results hint at a subsurface magmatic influence.

We present crustal seismic anisotropy of the Tanganyika-Rukwa Rift Zone using a seismic waveform database and associated local earthquake catalog20 developed from the TANGA14 seismic array, which was deployed on the Ufipa Plateau between the Tanganyika and Rukwa Rift basins (Fig. 1b). Shear wave splitting analysis reveals fast polarization orientations of the shear waves and the corresponding strength of the anisotropy normalized over the entire ray path of the associated earthquake (Fig. 2). The strongest anisotropy is measured along ray paths that sample the middle-to-lower crust and are collocated with areas having high compressional-shear wave velocity ratios; thus, confirming the presence of oriented melt pockets, including volatiles, in the crust. These results and surface heat flow measurements explain the enigmatic topography (Fig. 3) of the rift flanks and indicate that lower crustal magmatism with no surface manifestation is responsible for the reduction in lithospheric strength that is driving rifting (Fig. 4).

Fig. 2: Crustal seismic anisotropy results compared with previous mantle anisotropy and surficial trends.
figure 2

a Map of shear wave splitting results plotted at the midpoint of their ray paths and colored using the shear wave anisotropy. b–f Rose orientation plots for the rift faults, pre-rift Precambrian metamorphic fabrics, mantle anisotropy18, megafractures, and crustal anisotropy. g Polar plot showing the anisotropy results as a function of the earthquake depth and northern direction of the polarization orientations and colored using the shear wave anisotropy.

Full size image
Fig. 3: Topographic anomalies.
figure 3

a Polygons labeled A, B, C highlight the three high-relief zones in the region. Other symbols indicate a compilation of surface heat flow measurements, sites with high compressional-shear wave velocity ratios, helium degassing, thermal springs, and volcanism. The black line indicates the profile below in (b). b Results of a flexural cantilever model9 compared with topography and the vertically averaged gravitational potential energy (GPE)50.

Full size image
Fig. 4: One-dimensional (1D) thermo-rheological model showing the isostatic contribution to the surface elevation and decreasing lithospheric strength due to evolving surface heat flow.
figure 4

a Predicted surface uplift from a combination of lower crustal dedensification and lithospheric mantle thinning. b Geothermal evolution based on the observed range of surface heat flow measurements. c Yield strength envelope corresponding to geotherms in (b). d Schematic summary of the proposed mechanism driving continued extension and transition to rift-axial strain localization along nonvolcanic active rifts.

Full size image

Results and discussion

Coupled mantle and crustal anisotropy

Seismic anisotropy is the directional dependence of seismic wave velocity polarization that provides information about rock fabric21. The most dominant form of anisotropy in the mantle, Lattice Preferred Orientation (LPO), is caused by the strain-induced alignment of olivine crystals22. In the crust, anisotropy is commonly attributed to Shape Preferred Orientation (SPO)23 caused by lithologic layering or oriented inclusions, such as water-filled fractures or aligned melt lenses. Analysis of SPO in the crust and LPO in the mantle can be used to explore deep crustal processes in the Tanganyika-Rukwa Rift Zone that subsurface imaging techniques cannot resolve due to the poor spatial density of existing geophysical datasets. These include the existing receiver function database, which provides estimates of the bulk compressional-shear velocity ratios and crustal thicknesses in the Ubendian Belt16 but whose poor depth resolution precludes discriminating possible mechanisms for variations in these properties.

We determine seismic anisotropy in the crust using shear wave splitting measurements from local earthquakes recorded by 13 broadband three-component seismometers deployed from June 2014 to September 201516. High-quality measurements are made using earthquakes with 0–35 km depths, a range that is entirely within the crust16. The splitting measurements show patterns of dominant rift-parallel northwest fast polarization axes that are subparallel to the active faults and pre-rift basement metamorphic shear zones (Fig. 2b, c, f). Less prevalent trends have subparallel orientations to the megafractures mapped on granitic terranes within the Bangweulu Craton (Fig. 2e). The strength of anisotropy is largest for events occurring in the lower crust (Fig. 2g) and propagation paths that sample the region beneath the southern Tanganyika Rift with high compressional-shear wave velocity ratios (Figs. 1b and 2a). Shallower events that sample this same region have weaker seismic anisotropy (Fig. 2a). As the strength of anisotropy dramatically increases with hypocentral depth, we hypothesize that the anisotropy is associated with SPO caused by melt pocket alignment, restricted to the lower crust, and is also the source of the high compressional-shear wave velocity ratios beneath the seismic stations. The weaker anisotropy from shallower events may be due to basement metamorphic fabrics or fluid-filled fractures.

Shear wave splitting from teleseismic SKS waves18, which sample the whole lithosphere and asthenosphere, has two dominant fast-axis orientations: northwest and northeast (Fig. 2d). The northwest trend is interpreted to result from mantle flow that deflected around the Bangweulu Craton into the Tanganyika-Rukwa Rift Zone, inducing vertical flow that generates oriented melt pockets in the upper mantle18. The northeast trend is attributed to LPO from the regional plate motion18. The similarity of the orientation of the crustal and SKS fast axes (both about N30°W) indicates that melt-induced anisotropy extends from the upper mantle into the lower crust and provides strong evidence for melt emplacement in the lower crust and upper mantle beneath the rift axis. These crustal anisotropy observations can be explained by vertically aligned and horizontally oriented melt channels in the lower crust along the rift axis24.

Above the zone of strong crustal anisotropy in the southern Tanganyika Rift, tectonic extension is accommodated by both the border and intra-rift faults12, and increasingly on the intra-rift faults (S3–6 in Fig. 1b insets). This pattern of rift faulting is anomalous for early-stage magma-poor rifts since strain localizes at the rift axis during the later phases of extension25,26,27. While unusual, rifting in the region may have evolved beyond the earliest phases, such that the rift axis is approaching the onset of strain localization. Inherited orogenic shear zones commonly focus on early intra-rift faulting28; however, this is unlikely the case in the southern Tanganyika Rift as the rift is not within a mobile belt (Fig. 1b).

Thermally supported surface elevations

Although surface magmatism in the region is absent, other geological observations suggest the existence of deep thermal anomalies, including high surface heat flow, hydrothermal vents, and helium-degassing zones19,29. A major unexplained feature of the rift zone is its anomalously high topographic relief (high-relief zone ‘B’ in Fig. 3a). Similar high and broad regions of elevated topography occur north of the Tanganyika Rift (high-relief zone ‘A’) and south of the Rukwa Rift (high-relief zone ‘C’), which are associated with surface volcanism (Fig. 3a). The ~400 km wavelength of these features is too large to be associated with footwall uplift from normal faulting but may be too small to be a result of dynamic topography due to a regionally impinging mantle plume30. Flexural cantilever models fail to produce the elevation or wavelength of the observed topography (Fig. 3b) with elevation underprediction exceeding 1.5 km9 even when using grossly underestimated elastic thickness values31; altogether, indicating the influence of an external and relatively local control.

We explain these topographic anomalies by considering the isostatic effects of lower crustal melting, including volatiles, and elevated mantle temperatures32. The complex evolution of the rift zone, imprecise timing of rift initiation, and the absence of erupted volcanic materials for geochemical analysis to lend insight into the melt source preclude a sophisticated thermo-mechanical model. Therefore, we approximate the geothermal evolution model in the Tanganyika-Rukwa Rift zone (zone B in Fig. 3) using a set of steady-state geotherms corresponding to the approximate range of observed surface heat flow measurements (Fig. 4b), where the smallest and largest values correspond to the craton and most mature part of the rift zone respectively and do not consider any specific rheology as it remains mostly unknown. The depth of the 1300 °C isotherm, which is a proxy for the lithosphere-asthenosphere boundary, is generally consistent with regional values from previous studies4,33,34; a constant 40 km crustal thickness is assumed, close to the observed average. These geotherms are used to model the evolution of the lithospheric yield strength envelope (Fig. 4c) and the topographic uplift (Fig. 4a).

The maximum plausible melt and volatile fraction in the lower crust is 0.2, estimated using liquidus and solidus relations35 for common lower crustal mafic rocks like gabbro and the highest observed surface heat flow value of ~90 mW m−2 (Zone B in Fig. 3a). Were the bottom 20 km of the crust to have this melt fraction, isostasy would cause approximately 400 m of uplift. Thermal isostatic effects from lithospheric mantle thinning can account for the remaining 1100 m of the observed relief (Fig. 4a), even accounting for variable erosion12. These results suggest a decoupling in the amount of deformation in the area since the extension in the crust is considerably less than what is predicted in the mantle by the geotherms. Without explicit consideration of magmatism on the lithospheric strength envelope, the evolution in the strength profile following the geotherms shows a substantial decrease in the lithospheric strength to values within the range of tectonic forces (1–5 TN m−1).

Implications for amagmatic continental rifting

Our model has deep crustal melts and volatiles resulting from thermal perturbations from the mantle, promoting crustal weakening and tectonic deformation in the absence of surface volcanism. In our study, measurements of crustal anisotropy from seismic data show a strong dependence on lower crustal melt distribution, representing ‘blind melts’ that are yet to breach the surface, possibly due to the low permeability of existing deep fault networks20 (Fig. 4d). Along the western branch, previous geophysical surveys have lacked the spatial resolution to detect low-volume melts that are trapped within the deep crystalline crust and upper-mantle, leading to the supposition that they were not present8. However, recent geophysical models developed using sophisticated imaging algorithms show that the upper mantle beneath the western branch may not be as cold as previously suggested, with low shear wave velocity anomalies beneath the study area36. Our findings present compelling evidence that links lower crustal magmatism with its influence on thermal isostasy, lithospheric weakening, and strain accommodation in nonvolcanic rift settings. We argue that during nonvolcanic rifting, the persistence of tectonic extension and initiation of strain localization at the rift axis is assisted by deep crustal melts and volatiles (Fig. 4d).

Studies of the volcanic phases of large igneous provinces show that an enormous amount of degassing can occur thousands of years before the arrival of the magma at the surface37. Consequently, deep ‘blind melts’ in nonvolcanic active continental rifts have implications for climate change because they might act as a source of diffuse and yet undetected volatiles, e.g., carbon dioxide emissions. Overall, the findings in this study highlight that a better understanding of blind melting products beneath nonvolcanic rifts requires higher-resolution geophysical detection analyses38.

Methods

Crustal seismic anisotropy

We use the minimum-eigenvalue shear wave splitting technique with cluster analysis as implemented in the MFAST39 program to obtain the seismic anisotropy measurements. The method does not require knowledge of the initial polarization of the shear waves. A numerical grid search over a range of polarization orientations and delay times is used to determine the optimal parameters that linearize the particle motions on the horizontal components and unsplit the shear waves within a selected seismogram window. All shear wave splitting results are manually inspected to ensure that they meet high-quality selection criteria, including a minimum signal-to-noise ratio of 3, maximum delay time between the fast and slow shear wave of 0.3 s, and uncertainties in delay times and polarization orientations of less than 0.03 and 16°, respectively. Null split measurements with initial polarization orientations within 20° of the fast polarization orientations are discarded. Supplementary Fig. 1 shows an example of a high-quality shear wave splitting result. The complete list of the most reliable shear wave splitting results is listed in the Supplementary Data.

Data

We use the continuous seismic data from the TANGA14 array comprising 13 broadband seismometers and deployed for a 15-month period from June 2014 to September 2015. The anisotropy analysis was performed using the recently developed local earthquake catalog consisting of 2213 earthquakes20. S phase picking was performed manually on filtered traces, resulting in a total of 3121 picks from 1261 events. Roughly 3% (79) of the S picks led to reliable shear wave splitting measurements.

Ray tracing

To compute the S wave anisotropy (SWA) and obtain better estimates of the incident angles for further refinement of the shear wave splitting results, we compute raypaths in a 3D volume using a radially symmetric velocity model20 rather than using a straight-line approximation that would be suitable for a homogenous media. For each source, we compute the travel time field by solving the eikonal equation40:

$${\Vert \nabla {{{{{\rm{t}}}}}}({{{{{\bf{x}}}}}})\Vert }_{2}=s{({{{{{\bf{x}}}}}})}^{2},$$
(1)

where \(t({{{{{\bf{x}}}}}})\) is the space-dependent \({{{{{\bf{x}}}}}}\) travel time field, \({\Vert .\Vert }_{2}\) is the Euclidean norm, and s(x) is slowness defined as the inverse of velocity. For computational efficiency, we treat the seismometers as sources and compute only 13 travel time fields. We then simultaneously trace the raypaths between each receiver and any number of events using the gradient of the travel time field (Supplementary Fig. 2). The calculated incident angle is also corrected for topographic slope, which is nearly zero at most stations. We take caution to avoid the misinterpretation of the effect of S-to-P headwave conversions at the free surface as shear wave splitting by ensuring that the incident angle of the rays fall within the shear wave window at each station given as \({\sin }^{-1}\left(\frac{{v}_{s}}{{v}_{p}}\right)\), where \({v}_{s}\) and \({v}_{p}\), are the shear and compressional wave velocity at the receiver location41.

Anisotropic strength

We compute the S wave anisotropy (SWA), which normalizes the observed delay times \(\delta t\) using the average shear wave velocities \(v\) and distance \(d\) along each raypath, to better localize the source of the strongest anisotropic signal:

$${{{{{\rm{SWA}}}}}}=\left(\sqrt{4+{\left(\frac{2d}{v\delta t}\right)}^{2}}-\frac{2d}{v\delta t}\right)\times 100.$$
(2)

We focus on interpreting the S wave anisotropy (SWA) rather than the delay time between the polarized fast and slow shear waves due to the dependence of the latter on the ray path distance, which can lead to ambiguity in interpretation (Supplementary Figs. 3 and 4). SWA measures the fractional perturbation of the average S wave velocity along the ray path, which normalizes the effect of distance and the sampled velocity structure42. However, we note that the absolute value of SWA in Fig. 2 is not as important as the relative value since anisotropy is assumed to be accrued along the entire ray path. Thus, SWA values may be higher when normalized only over the shorter ray path segments within the true anisotropic domain or when straight-line ray paths are assumed.

Geothermal evolution

We use the range of surface heat flow measurements in the area (zone B in Fig. 3a) to compute lower-bound geotherms that may represent the geothermal evolution. Thus, the evolution is realized as a set of steady-state geotherms by solving the Poisson form of the temperature equation43:

$${\nabla }^{2}T=-\frac{H}{k},$$
(3)

where T (°C) is the temperature field, \(H\,\)(\(\mu\)W m−3) is the volumetric heat production, and k is the thermal conductivity. Using both the surface heat flow \({q}_{0}\,\)(mW m−3) and surface temperature \({T}_{0}\) as boundary conditions and integrating Eq. 3 twice yields:

$${q}_{0}={q}_{z}+{Hz},$$
(4)

and

$${T}_{z}={T}_{0}+\frac{{q}_{0}}{k}z-\frac{H}{2k}{z}^{2},$$
(5)

where \({k}\) (W m−1 K−1) is the thermal conductivity.

The surface heat flow in the upper crust is partitioned to be 40% from radiogenic sources and 60% from deeper mantle sources, with an exponentially decreasing volumetric heat production:

$$H\left(z\right)={H}_{0}\exp \left(-\frac{z}{D}\right),$$
(6)
$${H}_{0}=0.4\times \frac{{q}_{0}}{D},$$
(7)

where \(D\) is the radiogenic length taken to be 10 km, consistent with empirical studies around the world44 and \({H}_{0}\) is the surface radiogenic heat production. We allow the upper crustal volumetric heat production to decay until it matches the value in the lower crust or crosses the Conrad discontinuity. The convective geotherm in the asthenosphere is computed using a potential mantle temperature of 1300 °C and an adiabatic gradient of 0.4 K km−1.

Yield strength envelope

For each geotherm, we compute the associated yield strength envelope using brittle and ductile flow laws. The thickness of the upper and lower crust is kept constant at 20 km but the thicknesses of the lithospheric and asthenospheric mantle is allowed to vary depending on the lithosphere-asthenosphere boundary determined using the 1300 °C isotherm in the geotherms. For the brittle layer, we compute the differential stress \(\sigma\) as

$$\sigma =\left\{\begin{array}{c}0.85\times P,{{{{{\rm{\& }}}}}}P < 0.2\,{{{{{\rm{GPa}}}}}}\\ 0.5+0.6\times P,{{{{{\rm{\& }}}}}}{{Otherwise}}\end{array},\right.$$
(8)

where \(P\) is the lithostatic pressure. The diffusion creep follows the power law:

$$\dot{\varepsilon \,}=A{\sigma }^{n}{d}^{-m}{f}_{{H}_{2}O}^{r}\exp \left(-\frac{Q+{PV}}{{RT}}\right),$$
(9)

where \(\dot{\varepsilon }\) is the strain rate (1e−15 s−1)45, A is the pre-exponential material constant, n is the differential stress exponent equaling 1 for diffusion creep, \({f}_{{{{{\rm{H}}}}}_{2}O}\) is the water fugacity with exponent \(r\) both assumed to be 1, \(Q\) is the activation energy, V is the activation volume, and R is the molar gas constant (8.314 J Kmol−1). We then estimate the lithospheric strength by integrating the yield stress envelopes. Modeling parameters46,47,48 are listed in Supplementary Table 1. The values assumed for the modeling parameters, such as the strain rate, affect the absolute lithospheric strength; the point remains to emphasize the strength reduction with increasing surface heat flows.

Surface uplift modeling

The discrepancy of the topographic anomaly is modeled as a combined isostatic effect resulting from lower crustal dedensification due to partial melting and thermal expansion from lithospheric mantle thinning. Underplating is not considered because previous subsurface imaging results indicate thinner crust beneath areas with higher compressional-to-shear wave velocity ratios16. Uplift \({\delta h}^{1}\) resulting from dedensification is

$${\delta h}^{1}={z}_{l}\frac{\delta \rho }{{\rho }_{m}},$$
(10)

where \({z}_{l}\) is the lower crustal thickness assumed to be 20 km, \({\rho }_{m}\) is the mantle density, and \(\delta \rho\) is the density change computed as a function of melt and volatile fraction \(M\):

$$\delta \rho =M({\rho }_{{solid}}-{\rho }_{{molten}}),$$
(11)

and \({\rho }_{{{{{\rm{solid}}}}}}\) and \({\rho }_{{{{{\rm{molten}}}}}}\), are the densities of the solid and molten lower crust. The melt and volatile fraction in the lower crust is determined by

$$M=\frac{T-{T}_{{solidus}}}{{T}_{{liquidus}}-{T}_{{solidus}}},$$
(12)

using the temperature at 25 km and the following liquidus and solidus relations

$${T}_{{liquidus}}(K)=1423+0.105\times P,$$
(13)
$${T}_{{solidus}}(K)=\left\{\begin{array}{c}973-\frac{70,400}{P+354}+\frac{77,800,000}{{\left(P+354\right)}^{2}},{{{{{\rm{\& }}}}}}P < 1.6\,{{{{{\rm{GPa}}}}}}\\ 935+0.0035\times P+0.0000062{\times P}^{2},{{Otherwise}}\end{array},\right.$$
(14)

common for lower crustal rocks, such as Gabbro. Thermal isostatic contribution to the uplift is determined by integrating from the surface to the lithosphere-asthenosphere boundary over the difference of the initial geotherm corresponding to the lowest surface heat flow measurement \({q}_{1}\) and subsequent geotherms for higher surface heat flow values \({q}_{i}\),

$${{\delta h}_{i}}^{2}=\alpha {\int }_{\!\!\!0}^{{{LAB}}_{1}}{T}_{{q}_{i}}(z)-{T}_{{q}_{1}}(z){dz},$$
(15)

where \(\alpha\) is the linear thermal expansion coefficient (2e−5 K−1).