Crustal structure of the East African Limpopo margin, a strike-slip rifted corridor along the continental Mozambique Coastal Plain and North Natal Valley

Abstract. Coincident wide-angle and multi-channel seismic data acquired within the scope of the PAMELA Moz3-5 project allow us to reconsider the formation mechanism of East African margins offshore of southern Mozambique. This study specifically focuses on the sedimentary and deep-crustal architecture of the Limpopo margin (LM) that fringes the eastern edge of the Mozambique’s Coastal Plain (MCP) and its offshore southern prolongation the North Natal Valley (NNV). It relies primarily on the MZ3 profile that runs obliquely from the northeastern NNV towards the Mozambique basin (MB) with additional inputs from a tectonostratigraphy analysis of industrial onshore–offshore seismic lines and nearby or crossing velocity models from companion studies. Over its entire N–S extension the LM appears segmented into (1) a western domain that shows the progressive eastward crustal thinning and termination of the MCP/NNV continental crust and its overlying pre-Neocomian volcano-sedimentary basement and (2) a central corridor of anomalous crust bounded to the east by the Mozambique fracture zone (MFZ) and the oceanic crust of the MB. A prominent basement high marks the boundary between these two domains. Its development was most probably controlled by a steep and deeply rooted fault, i.e., the Limpopo fault. We infer that strike-slip or slightly transtensional rifting occurred along the LM and was accommodated along this Limpopo fault. At depth we propose that ductile shearing was responsible for the thinning of the continental crust and an oceanward flow of lower crustal material. This process was accompanied by intense magmatism that extruded to form the volcanic basement and gave the corridor its peculiar structure and mixed nature. The whole region remained at a relative high level during the rifting period and a shallow marine environment dominated the pre-Neocomian period during the early phase of continent–ocean interaction. It is only some time after break-up in the MB and the initiation of the MFZ that decoupling occurred between the MCP/NNV and the corridor, allowing for the latter to subside and become covered by deep marine sediments. A scenario for the early evolution and formation of the LM is proposed taking into account both recent kinematic and geological constraints. It implies that no or little change in extensional direction occurred between the intra-continental rifting and subsequent phase of continent–ocean interaction.


). The amplitude contrast is created by a strong velocity jump of at least 1.5 km/s. Instruments located over the NNV and the LM west of the basement high indicates the presence of two layers within the acoustic basement. There is a first set of associated refracted/reflected phases (SV1) with apparent velocities lower than 5 km/s and a second set (SV2) with velocities higher than 5 km/s. (e.g. MZ3OBS20 on figure 3C). The first of these layers 175 when converted to two-way-time coincides with the deep layering previously described on MCS (SV1 layer on Figure 2b).
However, its base and the SV2 layer below are entirely constrained by wide-angle data. To the NE of the basement high, OBS evidence a clear change in the characteristics of the acoustic basement internal structure as already suggested from its facies on MCS. Here a single layer (SV1) continues while SV2 is absent below the basement high. Apparent velocities are between 5 and 5.5 km/s (e.g. MZ3OBS30 on figure 3D)  Triplication between crustal and mantle phases occurs typically at about 150 km offset. This offset decreases to less than 100 km for instruments located close to the NE end of the profile (e.g MZ3OBS31 on Figure 6) implying important crustal thinning.  Figure 6. Same as figure 4 for MZ3OBS31 A different pattern of crustal phases is indeed observed on OBS records from the NE extremity of the LM. Crustal refracted phases at short offset suggest a stronger gradient and an average apparent velocity around 6.5 km/s. The gradient reduces at greater offsets but apparent velocities do not exceed 7 km/s. Some reflected phases can be distinguished but overall there is no more evidence for intense crustal layering, the most prominent corresponding to the Moho. The transition to mantle phases occurs at a relatively short offset (about 50 km on MZ3OBS31; Figure 6)   From MZ3 OBS data, a total of 72750 arrival times were picked from the phases described above and summarized in Table   1. Travel-time uncertainty for each phase was automatically calculated based on traces signal to noise ratio. They range from 0.025 s for high ratio to 0.25 s for poor ratio.
We used the iterative procedure of two-dimensional forward ray-tracing followed by the damped least-squares travel-time 225 inversion of the RAYINVR software (Zelt and Smith, 1992). Our modeling proceeded following a top-to-down strategy of arrival times fitting of both wide-angle reflected and refracted phases identified on OBS data. Model interfaces were inserted when a velocity change was observed in apparent refracted velocity which mostly coincide with high amplitude reflective  Table 1 and Table 2 respectively. Logically the density of velocity and depth nodes as well as the number of reflective segments are higher in the central part of the model and decreases towards its edges and with depth ( Figure 9a). Indeed these regions of the model are less sampled by rays ( Figure 9b) usually from a single direction and data quality is either poorer or degrade naturally with increasing offsets.
Because the model parameterization have been adapted to these limitations MZ3 shows overall limited smearing (+/-3, Figure   245 9c) and very good resolution (above 0.9, Figure 12d). The quality of the modeling decreases at depth and towards model extremities. Resolution values remain, however, higher than 0.5 which are still considered acceptable. Greater smearing occurs essentially in the lower crust where only few refracted rays travel. Those have been difficult to identified on OBS records because many phases converge around to the triplication point with the PmP and Pn (see 'Wide-angle seismic data' section above).

Gravity modeling
We tested the gravimetric response of our final model against the measured (in green) and satellite-derived (in yellow and red) free-air gravity anomaly along the profile ( Figure 10). A 2-D model consisting of homogeneous density blocks was constructed from the MZ3 final velocity model by converting P-wave velocities to densities according to (Ludwig et al., 1970)  (colored) and depth nodes (squares). c) Smearing from Spread-Point Function (SPF) for velocity (colored) and depth nodes (squares). d) Resolution of velocity (colored) and depth nodes (squares). Zones that were not imaged are blanked.
based on the conversion of MZ3 velocity model reflects well the overall shape of the gravimetric anomaly all along the profile.
MZ3 do not generate a broad negative anomaly at the transition beneath the basement high of the LM but the change in crustal properties generates a clear jump from negative to positive values consistent we the observed gravity anomaly. Note that we had to lower the densities within the mantle layers from 3300 to 3150 kg/m3 in order to adjust the measured gravimetric anomaly

Uncertainties estimation
Velocity and depth uncertainties of MZ3 final velocity model were estimated using the Vmontecarlo code (Loureiro et al., 2016). Vmontecarlo explores the model space by generating random models. It evaluates the ability of each model to fit the observed data set and translate it to estimates of uncertainties given some quality thresholds.

270
For computational cost, the explored model space was reduced by limiting the number of parameters and fixing some bounds. We also allowed a maximum fluctuation of +/-0.50 km/s on velocity nodes while the maximum depth node variations were set to bands of 1, 2 and 3 km for the upper crustal, the 3 mid-crustal and the Moho interfaces respectively. We further limited 275 the search by generating a maximum of 50000 random models and imposed a scaling factor for velocity and depth bounds that, starting at 20% of their maximum value, progressively increased to 100% within the first half of the randomly generated models. Finally the total number of observed travel times was decimated to be less than 50000 picks to reduced ray tracing computing time.
Scores are calculated for each model according to a function that takes into account travel time fit and the ratio of traced rays It is interesting to note that depth uncertainties for a velocity of 7 km/s is usually high 290 over a couple of km to a maximum of 5 km in a few places. Up to 300 km model distance this can be explained by the very low velocity gradient required in mid/lower crust. In fact Vmontecarlo analysis further confirms that most of the basement is composed of material with velocities higher than 6.5 km/s even towards the NE end of the profile (distance greater than 320 km).
In order to visualize uncertainty estimates on profile a subset of possible alternative models were selected. These models 295 respect the following criteria we judge acceptable: a score over 75% of the preferred model's score, a chi 2 lower than 2, a RMS lower than 0.1 s and at least 80% of the rays traced by the preferred model. This subset represents 126 models that were combined to produce the minimum and maximum admissible velocity deviations maps shown in figure 11. Large uncertainties within the depth bounds allowed for interfaces are expected when large velocity contrast exist and should be ignored. Globally outside these hatched areas velocities can vary from +/-0.10 km/s to +/-0.15 km/s with rare excursions over +/-0.2 km/s. to constrain the ages of the oldest deposits which span the syn-rift and early post-rift period that are of interest for this study.
More precisely we extended to the LM previous seismo-stratigraphy studies produced for the NNV (Baby et al., 2018;Schnürle et al.;Verrier et al.). Profile A is the direct north-south prolongation along the LM of a profile presented by these authors. It connects to the Sunray 1 well (Salman and Abdula, 1995) located in the NW corner of the NNV (Figure 13). We further analyzed two profiles (B and C) than are normal to profile A and strike roughly E-W across the LM. Among them Profile B has an onshore portion that connects to Funhalouro-1 and Nhachengue-1 wells and Line SM-59 presented by Salman and Abdula (1995). Along all these lines, the horizons of late-Jurassic to Neocomian formations, namely Red Beds Fm and Maputo Fm (in green and yellow respectively on figure 12), can be confidently extended towards the LM from both the NNV (Sunray-1 well) and the MCP (Funhalouro-1 and Nhachengue-1 wells). As mentioned by Salman and Abdula (1995), these two formation unconformably cover the acoustic basement. Red Beds are only observed filling local fossil depressions on the MCP and   (Figure 8), it is characterized by a large velocity jump (1-1.5 km/s) and shows values above 4.5-5 km/s just below. It is relatively flat over the NNV at about 2 km depth and deepens to 3.5-4 km in the LM to finally reach 6 km passed 340 330 km model distance. Its highly perturbed topography evidences important deformations most probably controlled by highly dipping and near vertical faults. On each side of the ALR they bound a few horsts and grabens while over the LM they are highlighted by fan-like dipping reflectors and steep basement highs. The most prominent basement high, located at 310-340 km model distance, is clearly an isolated basement structure controlled on each side by steep faults. To its SW, the fault zone appears wider at the basement surface suggesting a deep fault rising as a flower system (Figure 2).

345
Internally the acoustic basement show chaotic and incoherent signal in some places (60-120 km, 300-420 km, Figure 2) but also clear basement-parallel or dipping reflectors of variable amplitude in others (-60 to -20 km; 120-260 km). This already suggests that the top of the acoustic basement is not the roof of the crystalline basement, at least up to the basement high at 330 km. Its internal structure is indeed only partially resolved by the MCS. OBS data further constrain a 3 to 4 km thick unit made of two layers (SV1 and SV2 layers in green on figure 2b). They are respectively 1-1.5 km and 1-2.5 km thick with contiguous with the northern NNV ( Figure 12). All together this reveals that the volcano-sedimentary unit is ubiquitous to our 365 study area and form a widespread basin that terminates oceanward as a prominent basement high.

Crustal nature and segmentation
Below the volcano-sedimentary basin, the top of the crystalline crust is higher at 4-5 km depth in the NNV than in the LM where it reaches 6-7 km depth on MZ3 (Figure 8) basement (from 6.2 to 6.4 km/s) and a progressive decrease at its base (from 7.3 to 7.0 km/s). The velocity gradient remains thus stable and low (Figure 14b) while its internal highly reflective character is still observed. Such characteristics are coherent with those described on velocity models produced along crossing MZ4 and MZ5 profiles (Watremez et al.) and further south along MZ1 profiles (Leprêtre et al., a). All together they emphasize an eastward prolongation and thinning of magmatically intruded continental crust beneath the LM (Figure 13). On the northeastern side of the basement high, the crystalline crust of the LM is only 10 km thick on MZ3 velocity model ( Figure 8). It is made of 3 layers with velocities between 6.5 km/s to 7.0 km/s top to bottom that show little internal reflectivity.
While the two deepest layers present a low gradient (0.025 km/s/km) similar to the adjacent continental crystalline basement, the upper layer has an intermediate velocity gradient (0.1 km/s/km) between those of the acoustic basement above and the deep crystalline basement below (Figure 14b). This anomalous trend is also observed on the 'transitional crust' identified on MZ4 400 and MZ5 (Watremez et al.) and east of the basement high on MZ1 velocity model (Leprêtre et al., a). Taken all together they reveal a 60-80 km wide, N-S trending corridor of anomalous crust isolated between the thinned MCP and NNV continental crust to the west and the MB oceanic crust to the east (Figure 13).
The basement high locates therefore an important segmentation in the crustal structure of the LM (Figure 13). Not only it coincides with the eastward termination of the volcano-sedimentary basin and a change in nature of the acoustic basement but 405 it also marks a profound modification of the crystalline crust. We can further notice that uppermost mantle velocities appears normal and stable at 8 km/s on each side of the basement high while a slight velocity decrease to 7.9 km/s (Figure 8), more pronounced on MZ4 and MZ5 velocity models (Watremez et al.) is present beneath the feature. This suggest that the basement high is the surface expression of a vertical frontier that is deeply rooted in the lithosphere. As such it is comparable to the MFZ that bounds the opposite side of the corridor and separates it from the oceanic crust of the MB. We will therefore refer to this 410 zone of strongly localized deformation as the Limpopo Fault (LF, Figure 13).
To summarize, the LM is a N-S elongated margin cut by two major fault zones that segment its crustal structure in a western continental crust attest of an intra-continental depositional process Moulin et al., 2020) rather than a rifted margin process (e.g. Cox, 1992;Klausen, 2009;Watkeys, 2002). Our tectono-stratigraphic analysis only place an upper bound 430 to the formation of the basin in pre-Neocomian. So far it has not been dated more precisely despite that its upper volcanic layer was reached by several wells (see section 4.1 and Schnürle et al.; Verrier et al.). We can thus speculate that it formed during the Karoo phase which affected the entire African continent in late Palaeozoic/early Triassic ( (Daly et al., 1991).
However, this would leave a large, 30 My, sedimentary gaps between extrusions of Karoo volcanics, which might compose the roof of the basin, and deposition of late-Jurassic to Neocomien Red Beds and Cretaceous Maputo formations. Therefore, 435 another hypothesis is that it formed in late-Jurassic either before or contemporaneously with the formation of the adjacent MB (Aslanian et al.). The high magmatic content of the basin as well as evidence of strongly intruded continental crust suggest that magmatism may have overloaded the crust and created the necessary vertical subsidence Moulin et al., 2020;Tozer et al., 2017).
Our study also emphasizes the eastward prolongation and termination of the volcano-sedimentary basin along the LM where 440 it is strongly deformed. At depth, it is also the place where important crustal thinning and segmentation is evidenced ( Figure   13). This clearly indicates that the LM is the place where rifting localized to accommodate the opening of the MB rather than over then entire MCP/NNV. In previous scenarios based on 'tight' fit kinematic framework in which Antarctica partly overlaps Africa, rifting was either postulated to concentrate along the Lebombo monocline or beneath the MCP/NNV depending whether the area was interpreted as oceanic or a volcanic rifted margin (Cox, 1992;Klausen, 2009;Leinweber and Jokat, 2012;Mueller 445 and Jokat, 2019;Watkeys, 2002). Neither of these hypothesis are supported by our new observations.
Overall, this means that the MCP/NNV must be excluded from the Africa-Antarctica corridor (AAC in Leinweber and Jokat, 2012;Mueller and Jokat, 2019) and a 'looser' plate fit must be adopted in East-Gondwana kinematic reconstruction (Moulin et al., 2020;Thompson et al., 2019). Such framework excludes any initial rifting phase with normal ( E-W) or oblique (NW-SE) plates movement (e.g. Cox, 1992;Reeves et al., 2016) which, to the our opinion, has never been clearly evidenced or described.

450
Invoking indeed the orientation of magmatic dyke swarms to attest stress field direction (Mueller and Jokat, 2019;Reeves et al., 2016) is highly speculative as they may be strongly controlled by inherited lithospheric discontinuities (Jourdan et al., 2006).
Similarly, the presence of a wide crustal necking zone cannot solely justify normal or oblique rifting (e.g. Vormann et al., 2020). Indeed the LM itself shows such characteristics but given our alternative geodynamic framework it was affected instead by strike-slip or slightly trans-tensional rifting following a continuous N-S direction of plate motions during the opening of 455 the MB.  Moulin et al., 2015;Jolivet et al., 2015). It appears as a common process of rifted continental margin independently of their tectonic context. In fact it has been shown numerically that lower crust may flow across both divergent margin segments (Huismans and Beaumont, 2011) and strike-slip segments (Le Pourhiet et al., 2017;Reid, 1989). Usually it is 485 noticed that weak lower crust may favor such exhumation process. It was undoubtedly the case for the southern-Mozambique lithosphere which on the verge of its dislocation was affected by a large thermal anomaly (Karoo?) responsible for its anomalous magmatic content. The same anomaly might have favored coeval partial melting and extrusion of volcanic material to form the volcanic basement of the corridor.

Rift to drift evolution and vertical movements 490
In the alternative East-Gondwana 'fit' proposed by Thompson et al. (2019), the continental domain to the south-west of the Astrid Ridge on the Antarctica plate is the conjugate of the African corner made by the MCP and the Beira High ( Figure   151515a-b). Further south the Grunneghona craton is facing the NNV. The exact seaward limit of the Antarctica continental crust is inferred at the location of a strong positive free-air gravity anomaly (Mueller and Jokat, 2019;Scheinert et al., 2016) since transitional crust is observed further offshore on seismic data (Jokat et al., 2004). At this stage, the Antarctica plate 495 overlaps the corridor which will develop progressively during rifting ( Figure 15c) and before oceanic spreading in the MB starts from Chron M25 ( 155 Ma) onwards ( Figure 1515d). As discussed above we infer an intra-continental sag phase for the development of the volcano-sedimentary basin over the MCP and NNV (Figure 21b). Deep crustal magmatic intrusions led to the subsidence of the area without major plate movements (( Figure 15b; Aslanian et al.)). On the other hand rifting along the LM proceeded from N-S plate motion responsible for the opening of the MB (Figure 15c). The LM developed as a wide shear 500 zone along the eastern margin of the MCP and NNV. The LF acted at this time as a major strike-slip fault while easing the decoupling between upper and lower crust at depth. In the upper crust we infer that strain partitioning is responsible for shear folds, small pull-apart basins and grabens within the volcano-sedimentary basement. Deeper, extension was accommodated by lithospheric flow leading to lower crustal thinning of the eastern fringe of MCP/NNV continental crust and its oceanward exhumation. This was accompanied by the extrusion of a large amount of volcanics over the corridor. Further west, deep 505 seismic acquisition revealed a transitional domain capped by a thick volcanic layer (the Explora Wedge, e.g. Jokat et al., 2004) off the Antarctic margin which might attest of a equivalent process, thought along a divergent segment. On the African side, extension initially focused in the offshore Zambezi depression before migrating to the south of the Beira High, isolating this continental block (Mahanjane, 2012). The nature of the crust to the north of the Beira High is also deemed to be transitional with syn-rift magmatism reported (Mueller et al., 2016;Mueller and Jokat, 2019).

510
According to our stratigraphic analysis, scarce Red Beds continental sedimentation was followed by the widespread deposition of shallow marine Maputo Sands over both the MCP/NNV and LM. Only then, the eastern extremity of the corridor was covered by the first deep marine sediments. Lithospheric and/or asthenospheric flow may have sustained the area to a relative high level during continental rifting (Reid, 1989) allowing continental Red Beds deposits first, then slight subsidence responsible for the first shallow marine incursion. Minor vertical movements must have occurred along the LF only due to strike-slip 515 shearing. It is some times after break-up that larger differential vertical movement took place on either side of the LF between the continental domain and the corridor, hence accentuating the basement high and limiting the westward incursion of the first deep marine sedimentary horizons.
As in previous studies (see Basile, 2015, and references therein) we recognize that a stage of continent-ocean interaction followed continental rifting and its break-up. Analyzing the possible interplay between the corridor and the mid-oceanic ridge 520 requires, however, a careful analysis of later stratigraphic horizons across and all along the margin which is out of the scope of our study. Figure 15c