Analogue modelling of marginal flexure in Afar, East Africa: Implications for passive margin formation

The Afar region in East Africa is a key locality for studying continental break-up. Within Afar proper, passive margins are developing, of which the Southern Afar Margin (SAM) contains synthetic (basinward) faulting, whereas crustal flexure, antithetic faulting and marginal grabens occur along the Western Afar Margin (WAM). Numerous conflicting scenarios for the evolution of the WAM exist. In this analogue modelling study we test various factors that may affect the development of a WAM-style passive margin: brittle crustal thickness, (en echelon) rheological contrasts, sedimentation and oblique extension. Our experimental results illustrate how marginal flexure due to a weak lower crust below Afar elegantly reproduces the structural features of the WAM. Brittle crustal thickness controls what structures occur: a thinner brittle crust accommodates flexure internally, whereas increasing brittle thicknesses lead to faulting. Large escarpment faults develop early on, followed by late-stage antithetic faulting and marginal grabens. A thicker brittle crust also causes more subsidence, and higher strength contrasts between lower crustal domains leads to more localized deformation. Basin-wide sedimentation causes enhanced subsidence, as well as longer activity along large (escarpment) faults. Finally, oblique extension clearly prevents the development of marginal grabens, which only form in near-orthogonal extension. These results support a tectonic scenario involving initial oblique extension due to Arabian plate motion, creating echelon synthetic escarpment faults along the WAM. After the Danakil Block started its independent rotation, near-orthogonal extension conditions occurred, allowing (enhanced) marginal flexure, antithetic faulting and marginal graben formation along the older en echelon escarpment. Differences in extension obliquity may also explain the differences in structural architectures between the WAM and SAM. The characteristics of the WAM are typical of magma-rich passive margins, and the margin has great potential for studying continental break-up and (magma-rich) passive margin formation. of the Arabian plate, leading to the development of an en echelon style escarpment, as seen in model K. The second phase of rifting involves near-orthogonal extension due to the independent rotation of the Danakil block, leading to flexure and the development of marginal grabens as observed in model L. Note that magmatism has remained active during the development of Afar. Modified after Bosworth et al. (2005), Bosworth (2015), Zwaan et al. (2020b).

The development of Afar started with the emplacement of one or multiple mantle plumes below East Africa from ca. 45 Ma (e.g. Ebinger & Sleep 1998;Rogers et al. 2000;Pik et al. 2006;Rooney et al. 2017), resulting in a massive outpouring of Trap series flood basalts over a 1 My time period around 30 Ma (Hoffmann et al. 1997) that covered large parts of the region (Mohr 1983a). Moreover, continental extension driven by the anticlockwise rotation of the Arabian plate initiated in the Gulf of Aden around 35 Ma and propagated towards Afar and subsequently the Red Sea basin at ca. 29 Ma and 23 Ma, respectively (Ukstins et al. 2002;Bosworth et al. 2005;Wolfenden et al. 2005, ArRejehi et al. 2010Leroy et al. 2010;Szymanski et al. 2016;Purcell 2017, and references therein Fig. 1c). This rift configuration caused oblique extension along the WAM and other parts of the Red Sea-Gulf of Aden system (Smith 1993;Zwaan et al. 2020b, Fig. 1c). Around 11 Ma the Danakil Block, initially a part of the rift floor, began to independently rotate anticlockwise (McClusky et al. 2010), resulting in a local stress field change in Afar. Along the Western Afar Margin (WAM), this is thought to have resulted in NE directed extension changing to near orthogonal E-W extension (Zwaan et al. 2020b, Fig. 1d). The (northern) MER only formed around 11 Ma, long after the other two branches of the Afar triple junction were well established (Wolfenden et al. 2004). formation remain poorly understood. Analogue and numerical modelling techniques are a reliable means to explore these questions. For instance Corti et al. (2015b) suggest that magmatic loading can generate large-scale crustal flexure, but these numerical models not reproduce the more detailed structural features of the margin. Another potential mechanism for WAM-style margin formation is a (synthetic) basement fault, inducing flexure and marginal graben formation by deforming a brittle-viscous overburden as proposed in previous models (e.g. Naylor et al. 1994, Dooley et al. 2003, Hardy et al. 2018 and Gabrielsen et al. 2019). However these models simulate structures in the uppermost crust (i.e. deformation above an evaporite layer) and are difficult to scale up, especially since there is little evidence for a similar (lithospheric-scale) fault below the WAM (e.g. Hammond et al. 2011;Zwaan et al. 2020a). Furthermore, Livio et al. (2019) provide a method to test flexure of brittle layers, but only focus on the uppermost crust. Furthermore, flexure is applied symmetrically (i.e. doming), which is different from the situation along the WAM (Fig. Fig. 1b, e). Finally, Zwaan et al. (2019) apply rapid extension and a velocity discontinuity, inducing lower crustal stretching and upper crustal flexure. Yet, the resulting rift is symmetric, with flexure on both sides, whereas Afar is strictly asymmetrical since flexure does not occur along the Danakil block or Yemen margin. Previous works are thus not directly applicable to the WAM because they focus on much smaller-scale systems, do not properly reproduce the tectonic setting, or do not provide sufficient insights into the detailed structural development of the margin. In this study we therefore test factors that may lead to the development of a WAM-style passive margin by means of centrifuge experiments. We run a first series of models exploring the effects of brittle crust thickness, rheological contrasts between the rift and its shoulder as well as syn-tectonic sedimentary loading in orthogonal extension. A subsequent set of experiments aims to reproduce the tectonic setting during the first and second phases of WAM development, including en echelon margin structures and oblique extension. We subsequently compare our results with previously published modelling work, the WAM itself and established magma-rich passive margins.

Centrifuge modelling
Applying a centrifuge for analogue modelling allows the manipulation of model gravity and the use of different (viscous) materials from those in normal-gravity analogue experiments of lithospheric deformation (e.g. Corti et al. 2003Corti et al. , 2004Agostini et al. 2011, Corti, 2012Philippon et al. 2015;Fig. 2a, b). The relatively stiff viscous materials applied in enhancedgravity models make model handling and preparation easier than in the case of normal-gravity experiments (Ramberg 1981;Schellart, 2002;Corti et al. 2003). When using a centrifuge, both the model and an equal counterweight are placed within the apparatus. During a model run, both the model and the counterweight start rotating with high velocity around the central axis, while their containers adapt to the increasing centrifugal forces. These forces translate to an enhanced gravity (18 g) in the model, thus deforming the model materials (see section 2.2.4 and Appendix A2), by means of gravitational collapse, rather than externally applied extension. Interior of the centrifuge during a model run. The machine contains an inner construction with two pods that can rotate about a vertical axis. The model is placed in one pod, and a counterweight in the other. The model is deformed by rapidly rotating the inner part, generating centrifugal forces that simulate enhanced gravity. During rotation, the pods, which are also suspended by horizontal hinges, tilt into a vertical position so that the centrifugal forces act vertically on the model surface. (c) 3D depiction of the general model set-up. The model materials are contained between two sets of removable spacers and two L-shaped blocks. The weak viscous ASC7020 bottom layer allows free deformation of the overlying materials. The relatively strong viscous PP45 and weak viscous SCA705 represent the stronger and weaker lower crust (LC) below the Ethiopian Plateau and Afar, respectively. An optional thin layer of PP45 on top serves to reduce the rheological contrast between the two domains. Finally, a layer of very fine feldspar sand (FS900SF) represents the brittle upper crust. In order to impose asymmetric extension on the model, one of the spacers is removed and the model put into the centrifuge (b), where the enhanced gravity induces gravitational collapse to fill in the space left by the spacer. Tape is used at various locations to prevent boundary effect. (d) Map view geometries of the rheological contrast. Left: model set-up for orthogonal extension and general parameter testing (series 1 models). Middle and right: specific setups to mimic the tectonic situation during the first and second phases of WAM development (series 2 models).

Model design
J o u r n a l P r e -p r o o f Models are contained within a plexiglass box with 25 cm x 15.8 cm inner dimensions (Fig. 2c). At both far ends plexiglass spacers and L-shaped blocks are placed, the latter with a 2 cm high base, leaving a 14 cm x 15.8 cm space at the model base (Fig. 2c). Here, a 1 cm thick bottom layer consisting of a very weak viscous mixture (SCA7020) is placed, flanked by 1 cm bars of sturdy plasticine (Giotto Pongo). The next 1 cm thick stratum contains two domains, both 7 cm long, the first consisting of a relatively strong viscous material (PP45) and the second of a relatively weak viscous mixture (SCA705). These two domains represent the stable and weak lower crust below the Ethiopian Plateau and Afar, respectively (Fig. 1a), and are covered by a ca. 2 mm sheet of the same PP45 mixture that extends onto the L-shaped blocks. This thin intra-crustal level serves to reduce the rheological contrast between both domains. Finally, a cover of feldspar sand (FS900SF) represents the brittle upper crust in the Afar region (Fig. 2c). A cm in our models translates to ca. 15 km in nature. For detailed material properties, see Appendix A1. During a model run, we step-by-step remove the 1.5 mm thick spacers on the right-hand side to allow space for the model to deform asymmetrically, as we only simulate one margin of the Afar rift system (Fig 2c). The centrifuge applies 18 g to the model, thus forcing the model material into the space left open by the spacer. The lower layer, which does not directly correspond to a lithospheric layer, allows deformation of the crustal analogues: the weak lower crust is expected to stretch and collapse, whereas its resistant counterpart remains stable. As a result, the overlying layers warp downward into "Afar", creating a crustal flexure (Fig. 1e). By applying scaling laws (e.g. Hubbert 1937;Ramberg 1981;Weijermars & Schmeling 1986) we demonstrate that our models are well suited for simulating the Afar crust (detailed scaling calculations are presented in Appendix A2). Specifically, the models have a geometric scaling ratio of 6.7•10 -7 , such that 1 cm correspond to 15 km in nature. The extension velocities in our models (ca. 2.5•10 -5 m/s) translate to ca. 10 mm/y in nature, close to velocities found in the Afar region (i.e. between 5 and 20 mm/y, e.g. McClusky et al. 2010;Saria et al. 2014). Also dimensionless scaling ratios are very similar between model and nature, so that scaling requirements are fulfilled (Appendix A2).

Model parameters
A total of 14 experiments testing the influence of various parameters on crustal flexure and margin development in orthogonal extension are completed (series 1), of which an overview is presented in table 1. In these, the brittle cover thickness is varied (from 0.6 to 2.0 cm). Furthermore, the effect of an enhanced rheological contrast is studied by removing the thin intra-crustal level in selected experiments. We furthermore include sedimentation to test the effects of loading (somewhat similar to the magmatic loading mechanism proposed by Wolfenden et al. 2005). This is simulated by filling in the model topography to the predeformed level with the same FS900SF feldspar sand used for the brittle cover after every second time-step (i.e. t 2 or 3.0 mm, t 4 or 6.0 mm and t 6 or 9.0 mm extension, respectively). Every sedimentation step is marked by a thin (<1 mm) layer of dyed sand for analysis of vertical displacements in final cross-sections. Similar thin layers are also added to the sand layer representing the upper crust. The total extension in our series 1 models amounts to 10.5 mm (ca. 15 km) over 7 intervals. An additional set of oblique extension models serves to directly simulate the different phases of WAM development (series 2, Fig. 2d, table 1). Two models involve a -35˚ oblique extension (i.e. angle α, measured between the normal to the rheological contrast and the extension direction, Fig. 2d, middle), which reflects the tectonic setting during the initial phase of WAM development (Fig. 1c, Smith 1993;Zwaan et al. 2020b). A further experiment involves 5˚ oblique extension with respect to a right-stepping rheological contrast, which represents the en echelon structural arrangement of the WAM resulting from the first tectonic phase (Fig. 2e, right, Zwaan et al. 2020b). These en echelon structures were reactivated under nearorthogonal extension due to more recent Danakil Block rotation (Fig. 1d, Zwaan et al. 2020b). Our series 2 models contain a thin intra-crustal layer, but no sedimentation is applied. Total extension deviates from the series 1 models: 15 mm and 12 mm, respectively (Table 1).

Model analysis
We apply various methods to analyze internal and external model evolution. Top-and side view photographs taken after completion of each deformation interval provide a first-order record of model evolution. Furthermore, pictures taken from various angles serve to reconstruct model topography over time through photogrammetry software (Agisoft Photoscan). The resulting digital elevation models (DEMs) are further processed using Global Mapper and QGIS software for a detailed and quantified assessment of model topography development. Finally, after model completion, the sand layers are soaked, frozen and cut to obtain cross-sections. Thin levels of dyed sand in both the cover layer and the sedimentary infill subsequently reveal (syn-sedimentary) internal deformation.

Results of series 1 models 3.1. Final structures
An overview of final model surface structures from series 1 is presented in Fig. 3. There is a clear distinction between models with and without syn-rift sedimentation, as the former have a flatter topography due to the repeated infilling of the rift basins. A further overview of final cross-sections is provided in Fig. 4, showing that the weak lower crust collapses in all models, leading to deformation and faulting in the brittle cover above the rheological contrast between the strong and weak lower crustal analogues.. Distinct structural differences are associated with the increasing brittle layer thickness in models A-D (with intra-crustal layer, without sedimentation, Figs. 3a-d, 4a-d). Whereas model A (0.6 cm thick brittle cover) develops very limited normal faulting as the brittle layer is warped, model B and C (1.0 and 1.5 cm brittle cover, respectively) produce marginal graben structures (Figs 3a-c, 4a-c). The associated antithetic normal fault in model B is not readily visible on the cross-section (Fig. 4b), possibly due to a very small throw or the soaking of the sand before freezing and cutting, but is clearly present on the top view photograph (Fig. 3b) and in the case of model C also visible in section (Fig. 4c). Model C furthermore contains a larger antithetic fault more basinward (Fig. 4c). Increasing the brittle layer thickness to 2.0 cm leads to the formation of a second graben structure downslope. Note that the flexural domain in models A-D is dominated by large synthetic faults, whereas (visible) antithetic faulting is constrained to the marginal graben boundary faults (Figs. 3b-c, 4b-c).

J o u r n a l P r e -p r o o f
Models G and H (without intra-crustal layer or sedimentation, and a 0.6 and 1.5 cm sand cover, respectively, Figs. 3g, h, 4g, h) fit the trend described above; the former develops a minor marginal graben similar to that of model B, while the latter contains a wider fault zone reminiscent of the structures in model D, although no clear antithetic faulting is involved (Figs. 3b,d,g,h,4b,d,g,h). The structures in Models G and H are generally more developed in comparison to their direct equivalents with the same brittle cover thickness, but with an intracrustal layer (Models A and C,Figs. 3a,c,g,h,4a,c,g,h). The models with sedimentation (models E, F, I and J, Figs. 3e, f, i, j, 4 e, f, i, j), form the same general features as their counterparts without sedimentation, except that the accommodation space created in the latter is almost fully filled with sediments. The only important difference is that the models with sedimentation develop a slightly higher total subsidence and synthetic fault offset is increased (see section 3.3).

J o u r n a l P r e -p r o o f
The final top views and cross sections illustrate that models without intra-crustal layer form more developed structures than their counterparts containing such a layer (Figs. 3, 4). These differences in structural maturity are highlighted when analyzing the surface evolution of selected experiments without sedimentation (models A, C, G and H, the former two with, and the latter two without intra-crustal layer, respectively, Fig. 5). In all these models, some surface deformation is apparent after 1.5 mm of extension. Faulting appears after 3.0 mm of extension (t 2 ) in model C, whereas model A does not develop any clear faults at all (Fig. 5a,  b). By contrast, models G and H, without intra-crustal layer, develop distinct faulting early on (between 1.5 and 3.0 mm of extension [t1-2], Fig. 5c, d). Furthermore, initial faulting is synthetic, forming an escarpment fault system between rift shoulder and rift floor, eventually followed by (in most cases) antithetic faulting and marginal graben formation (Figs. 3-5). Only models A and H differ, as the former develops pure flexure, whereas the latter is dominated by the main escarpment fault system and synthetic faulting (Fig. 4a, h).

Topographic analysis
Photogrammetry-derived DEMs allow a detailed analysis of model topography and subsidence, best shown in cumulative subsidence graphs along the central axis of the experiments (Fig. 6). For models without sedimentation, these are equivalent to their topography. In the case of models with sedimentation, topography remains rather flat over time (Fig. 3) and the cumulative subsidence values are obtained by summing up the topographic difference between sedimentation intervals instead. Most subsidence plots are very similar, in that they are dominated by an (synthetic) escarpment fault, established during the initial phases of deformation (i.e. t 1-3 , Fig. 6), accommodates a large amount of the model's vertical displacement due to the downwarping of the margin. Only Model A does not develop such a dominant fault, instead it shows a gradual flexure towards the basin (Fig. 6a). In contrast, the marginal grabens, if present, only appear during the final stages of model development (t 5-7 , Fig. 6). Furthermore, the maximum subsidence in all series 1 experiments shows a clear positive correlation between the thickness of the brittle cover and the subsidence (Fig. 7a). We also observe that models without an intra-crustal layer of PP45 (i.e. with a more pronounced contrast between both lower crustal domains) as well as models with sedimentation tend to have higher degrees of maximum subsidence (Fig. 7b-e).

Results from series 2 models
Next to the orthogonal extension series 1 models, we run a set of models to more specifically represent the two phases of WAM evolution (Series 2,Figs. 1c,d,8). For these models, the layering of model C is adopted (i.e. an intra-crustal layer and a 1.5 cm thick the sand cover). Total applied extension is slightly higher than in the series 1 experiments (15 and 12 mm for models K and L, respectively, Fig. 8b, g). Model K reproduces the first phase of WAM development involves 35˚ sinistral oblique extension with respect to the inferred N-S striking lower crustal rheological contrast along the margin (Figs. 1c, 2d, 8a). This model develops a series of right-stepping en echelon normal faults (Fig. 8a-e), which are dominantly synthetic and develop largely sub-perpendicular to the general extension direction, Their en echelon arrangement allows the general escarpment to follow the rheological contrast. The lack of antithetic faulting leads to the absence of marginal grabens along the escarpment (Fig. 8a-e), although some faulting deviates from this trend where the rheological contrast is farthest away from the moving sidewall (around Section K1). Model L simulates the second phase of WAM evolution and involves a right-stepping rheological contrast mimicking the right-stepping character of the margin, inherited from the first phase of oblique extension simulated in Model K (Figs. 1a, 2d, 8). Extension is slightly oblique (5˚ with respect to the direction of the right-stepping segments and 10˚ to the general N-S orientation of the WAM (Figs. 1d, 8). The rheological contrast focuses deformation and J o u r n a l P r e -p r o o f flexure, developing a distinct right-stepping en echelon series of escarpment faults and marginal grabens (Fig. 8f-j). Fig. 8. Surface evolution and cross-sections of models K and L, aiming to simulate the tectonic setting along the Western Afar Margin during its first and second development phase, respectively (Fig. 1c, d).
In section view, the strong and weak lower crustal analogues are light grey and black, respectively. Dotted lines indicate the transition form a strong LC (lower crust) to a weak lower crust analogue. MG: marginal graben.
J o u r n a l P r e -p r o o f 4. Discussion

Synopsis of model results
Our models, of which an overview is presented in Fig. 9, illustrate how the stretching of a locally weaker ductile lower crust leads to rift margin development. Generally we observe flexure of the overlying brittle layer that deforms as a factor of its thickness and the rheological contrast between the lower crust below the plateau and the rift floor (Fig. 9a). Pure flexure occurs with a thin brittle layer and a moderate rheological contrast; due to the relatively low brittle strength flexure can be accommodated by small-scale deformation, rather than by large faults (Figs. 3a, 4a, e, 5, 6a, e, 9a). In contrast, a large change in the lower crustal strength causes strong localization of deformation above the rheological transition and the development of large synthetic faults to accommodate subsidence (Figs. 3, 4g, h, i, j, 5, 9a). The larger structures observed with increasing brittle cover thickness are due to the increased strength of the brittle layer preventing marginal flexure from being accommodated via smallscale internal deformation only (Fig. 9a). Therefore, flexure leads to escarpment development, antithetic faulting and even (double) marginal graben formation, when the lower crustal rheological contrast is moderate (Figs. 3d, 4d, j, 5, 9a). For higher strength contrasts, the enhanced localization of deformation forces the development of synthetic faults (Figs. 4g, i, 5, 9a). The models containing marginal grabens follow a distinct evolution (Fig. 9b). Initially, differential subsidence leads to marginal flexure without clear faulting at the surface: minor deformation can still be accommodated internally (Figs. 5, 6, 8b, t 1 ). The first clear faults are synthetic and occur along the developing escarpment (Figs. 5, 6, 8b, t 2 ). Continued flexure forces the development of antithetic faulting and marginal grabens that represent the "keystone" in the flexural arch, as explained by the Abbate & Sagri (1969) structural model (Figs. 1e, 9b, t 3 ). Based on this concept and supported by field data (Figs. 1b, e), we should also expect the development of pervasive antithetic faulting in our models. Such structures likely have too small an offset to discern on our model scale. This is highlighted by the smaller (antithetic) normal faults observed in model map views, which are often not readily visible in section (Figs. 3, 4, 8).
The addition of syn-rift sedimentation does not significantly alter large-scale model features.
The associated enhanced subsidence must be due to increased loading above the weak lower crustal analogue (Figs. 7, 9c, Zwaan et al. 2018). This loading effect also explains why maximum subsidence intensifies with increasing sand cover thicknesses (Fig. 7b). This effect is enhanced by a higher rheological contrast, which allows for easier fault development and rapid localized subsidence, rather than more widespread flexure (Fig. 7c-e). Furthermore, the increased throw along the main faults when sedimentation occurs is described in previous modelling studies (e.g., Corti et al., 2010;Poliakov et al. 2014;Zwaan et al. 2018). Finally, our experiments suggest that extension direction has an important impact on margin development. Marginal grabens are typical of (near) orthogonal extension systems (Figs. 3-7, 8e-I, 9d), whereas oblique extension leads to the creation of en echelon synthetic faults (Figs. 8a-e, 9d). These results are in agreement with analogue models by Agostini et al. (2011) and Corti et al. (2013), which also develop marginal graben-like structures, but only when (local) extension conditions are near-orthogonal. We conclude that orthogonal extension provides the best conditions for margin "collapse" as it would create maximum space. By contrast, in oblique extension systems the margin is stabilized by the hanging wall block, which moves partially along the margin, rather than 90˚ away from it ( Fig. 9d) (Chorowicz et al. 1999).  (Fig. 1b). Note that the gradient from mode 1 (pure flexure) to mode 3 (double graben) is in a sense also a gradient in deformation intensity. (b) Sequence of developments during marginal flexure: initial flexure (t1) is followed by synthetic escarpment development (t2) and finally clear antithetic faulting and marginal graben development. (c) Contrast between models with and without syntectonic sedimentation; syn-rift sedimentation leads to continued deformation along large faults, and the loading due to the extra material causes increased subsidence. (d) Effects of extension direction on margin style. Oblique extension causes en echelon escarpment faulting, whereas orthogonal extension allows marginal grabens to develop.

Implications for the evolution of the Western Afar Margin
When comparing our models with the WAM, the orthogonal extension models with a 1.5 cm brittle layer and moderate rheological contrast present a good fit (Figs. 1a, b, 3c, 4c). The brittle cover thickness scales up to ca. 30 km, which is consistent with the maximum depth of earthquakes along the WAM (Illsley-Kemp et al. 2018, Zwaan et al., 2020b. Furthermore the early escarpment fault development, followed by late antithetic faulting and marginal graben initiation (Fig. 8b) is in agreement with the eroded nature of the WAM escarpment, which must have accommodated large amounts of deformation at an early stage, but seems currently relatively inactive (Tesfaye & Ghebreab 2013;Illsley-Kemp et al. 2018;Zwaan et al. 2020b, Fig. 1a, b). In contrast, the relatively fresh and active fault scarps of the antithetic faults, and the thin sedimentary infill of the marginal grabens (< 500 m, often much less) along the WAM, suggests these features are relatively young (Abbate et al. 2015;Illsley-Kemp et al. 2018;Zwaan et al. 2020b). This interpretation is also supported by geochronological analysis on volcanic rocks found along the WAM (Wolfenden et al. 2005;Rooney et al. 2013). However, a comprehensive interpretation must include the multiphase evolution of Afar. We therefore present a tectonic reconstruction, somewhat similar to a scenario proposed by J o u r n a l P r e -p r o o f Chorowicz et al. (1999), involving the different rift phases and their impact on the WAM (Fig.  10). Initially, the rotation of the Arabian plate causes sinistral oblique extension along the WAM (Smith 1993;Zwaan et al. 2020b), leading to the development of the large-scale, rightstepping en echelon escarpment faults reproduced in model K (Figs. 8a-e, 9d, 10). Subsequently, the shift in extension direction from sinistral oblique to near-orthogonal due to the individualization of the Danakil Block around 11 Ma (McClusky et al. 2010;Zwaan et al. 2020b) allows enhanced flexure. This causes antithetic faulting, as well as marginal graben development along the previously established escarpment arrangement as simulated in model L (Figs. 8f-j, 9b, d, 10). Our model results fit well with the features observed along the WAM, yet there is some potential discrepancy. Afar has been highly magmatically active for most of its geological history (e.g. Rooney et al. 2013;Stab et al. 2016 and references therein). These processes may not only have weakened the Afar lithosphere as simulated in our models, but may have locally added mass through intrusion of mafic material, causing (additional) subsidence and flexure (Wolfenden et al. 2005, Corti et al. 2015b, Fig 11a). Instead, we induce flexure by differential stretching of the lower crust rather than by a localized (magmatic) loading. Nevertheless, our results remain valid; even if magma loading would be the driving force of flexure along the WAM, the effects on the deforming brittle crust and the associated surface expression should be similar to those in our models. Fig. 10. Evolution of the Western Afar Margin within its regional context as derived from our model results. The first syn-rift phase involves sinistral oblique extension due to the counterclockwise rotation J o u r n a l P r e -p r o o f of the Arabian plate, leading to the development of an en echelon style escarpment, as seen in model

Conclusions
Our analogue modelling efforts exploring the development of WAM-type rift margins through crustal flexure lead us to the following conclusions:  We find that marginal flexure can occur due to the differential extension of a weak lower crustal domain, potentially enhanced by magmatic loading, and the concept of marginal flexure can elegantly explain the structural features observed along the WAM.  The thickness of the brittle crust controls the type of structures marginal flexure causes. A thinner brittle crust can easily accommodate flexure, leading to a gentle basinward flexure. In contrast, increasing layer thicknesses lead to faulting, mainly of large synthetic escarpment faults. Flexural deformation of thicker layers also create antithetic faulting and marginal grabens. A thicker brittle crust also causes more subsidence due to increased loading.  A stronger contrast between the competent and weak lower crust due to the absence of an intra-crustal layer causes more localized deformation and enhances subsidence along the resulting synthetic faults.
J o u r n a l P r e -p r o o f  Most deformation is accommodated by large synthetic escarpment faults that develop early on. Antithetic faulting and marginal grabens are late and relatively minor features within a rift margin that undergoes flexure.  Basin-wide sedimentation leads to more loading and enhanced subsidence, as well as longer activity along large (escarpment) faults. These effects are however different from those predicted by the magmatic loading model (Wolfenden et al. 2005), probably because the latter type of loading is much more localized along the rift axis.  Oblique extension leads to the development of en echelon synthetic escarpment faults, but prevents the occurrence antithetic faulting and marginal grabens, the latter can only form in near-orthogonal extension conditions.  Our results support a scenario in which the early evolution of the WAM was characterized by oblique extension and en echelon synthetic escarpment fault development, due to the orientation of the WAM and the direction of Arabian plate motion. Only after the Danakil Block started rotating independently, near-orthogonal extension conditions were established, allowing enhanced flexure, antithetic faulting and marginal graben formation along the previously created en echelon escarpment.  The influence of oblique extension explains why the WAM and SAM have such different structural characteristics, and why marginal grabens are not often observed in nature. The other characteristics of the WAM (flexure, antithetic faulting, potential SDRs) are typical of magma-rich passive margins. The WAM has thus great potential for improving our understanding of the processes involved in continental break-up and (magma-rich) passive margin formation.

Data availability
Rheological measurements of the brittle and viscous materials applied in our models are available in the form of two open access data publications via GFZ Data Services (Zwaan et al. 2020c and2020d, respectively). A third open access data publication available via GFZ Data Services (Zwaan et al. 2020e) contains additional images, digital elevation data and videos of our experiments. Links to these datasets are provided in the references.
The lowermost model layer, which allows deformation of the overlying layers consists of the same components as the SCA705 mixture, but in a different ratio. Instead of a 100:70:05 ratio between Dow Corning putty, corundum sand and oleic acid, we apply a 100:70:20 ratio. The resulting SCA7020 mixture has a density of 1610 kg/m 3 , and a power law exponent of 9.1 (details in Zwaan et al. 2020d). These values are valid for modelling temperature (20˚C). When temperatures increase, the viscous materials generally become less viscous (Zwaan et al. 2020d). Therefore, the centrifuge apparatus has an internal temperature control system allowing constant modelling conditions.

A2. Model scaling
The crust is represented by a brittle and viscous layers. The brittle layer thickness ranges between 0.6 and 2 cm, whereas the viscous layer is always 1 cm thick. Following the discussion by Zwaan et al. (2019), we consider the models with a brittle-to-thickness ratio of 1.5 (i.e. 1.5 cm sand cover) as the most realistic and use this as a base for our scaling calculation.

J o u r n a l P r e -p r o o f
Geometric scaling in the models is 6.7•10 -7 , such that 1 cm in the models correspond to 15 km in nature (ratio convention: model/nature). The stress ratio between model and nature (σ*) is obtained via the following equation (Hubbert 1937;Ramberg 1981): σ* = ρ*⋅h*⋅ g*. Here ρ*, h* and g* represent the density, length and gravity ratios respectively, and the equation yields a σ* of ca. 7•10 -6 . Dynamic similarity in the brittle layers can be validated using the the Rs ratio (Ramberg 1981;Mulugeta 1998;Bonini et al. 2001): Rs = gravitational stress/cohesive strenght = (ρ•g•h) ⁄ C, where ρ is the density, g the gravitational acceleration, h the height and C cohesion. The calculations yield Rs values of 22 and 15 for the model and nature, respectively, which is very similar ensuring dynamic scaling of the brittle behavior. Moreover, the internal friction angle of the sand, is very similar that of upper crustal rocks (e.g. Byerlee 1978, Table A2). For the scaling of viscous materials, the stress ratio σ* and viscosity ratio η* produce the strain rate ratio * with the subsequent formula: * = σ*⁄η* (Weijermars & Schmeling 1986) and subsequently, the velocity ratio v* and time ratios t* through the following equations: * = v*⁄h* = 1⁄t*. Given that every time step represents 1.5 mm of extension over a steady state 18 g interval of ca. 1.5 min (i.e. an extension velocity of 2.5•10 -5 m/s*) and an initial model width of 14 cm (Fig.  2c), strain rates in our models are in the order of 1•10 -4 /s*. For the strong lower crust analogue (PP45), this value translates to a viscosity of ca.