Modeling Lithospheric Thickness Along the Conjugate South Atlantic Passive Margins Implies Asymmetric Rift Initiation

The lithospheric architecture of passive margins is crucial for understanding the tectonic processes that caused the breakup of Gondwana. We highlight the evolution of the South Atlantic passive margins by a simple thermal lithosphere‐asthenosphere boundary (LAB) model based on onset and cessation of rifting, crustal thickness, and stretching factors. We simulate lithospheric thinning and select the LAB as the T = 1,330°C isotherm, which is calculated by 1D advection and diffusion. Stretching factors and margin geometry are adjusted to state‐of‐the‐art data sets, giving a thermal LAB model that is especially designed for the continental margins of the South Atlantic. Our LAB model shows distinct variations along the passive margins that are not imaged by global LAB models, indicating different rifting mechanisms. For example, we model up to 200 km deep lithosphere in the South American Santos Basin and shallow lithosphere less than 60 km in the Namibe Basin offshore Africa. These two conjugate basins reflect a strong asymmetry in LAB depth that resembles variations in margin width. In a Gondwana reconstruction, we discuss these patterns together with seismic velocity perturbations for the Central and Austral Segments of the margins. The shallow lithosphere in the Namibe Basin correlates with signatures of the Angola Dome, attributed to epeirogenic uplift in the Neogene, suggesting an additional component of post‐breakup lithospheric thinning.

(LAB) model, which is specially designed for the South Atlantic passive margins and images the thermal structure of the lithosphere.
Passive margins are the end-member of rifting, where parts of the crust and lithosphere diverge from each other, accompanied by crustal and lithospheric stretching. The McKenzie model of rifting is a widely accepted model that explains uniform thinning of the continental crust and lithosphere by instantaneous stretching in pure shear mode, followed by thermal subsidence (McKenzie, 1978). Jarvis and McKenzie (1980) introduced a time-dependent analytical model that relates variations in heat flow and subsidence history to the rate of extension. In a traditional view, the formation of nonvolcanic passive margins were attributed to passive upwelling of buoyant sublithospheric mantle material, driven by far field extension forces, for which the McKenzie rifting model is chosen (Geoffroy, 2005;Sengör & Burke, 1978). Contrary to that, volcanic passive margins have been interpreted as the result of active upwelling of a mantle plume, associated with a thick crust due to magmatic underplating and the formation of Seaward Dipping Reflectors (SDRs; e.g., Geoffroy, 2005;Mutter et al., 1982). The occurrence of volcanic rifted margins does not necessarily require a pronounced thermal anomaly in the mantle related to a plume (e.g., Bown & White, 1995), but can be explained with transient small-scale mantle convection underneath the lithosphere (Nielsen, 2002;Simon et al., 2009) or plume-rift interaction (Morgan et al., 2020).
Numerous studies over the last years have shown that architecture of passive margins is more complex than previously thought (Peron-Pinvidic & Manatschal, 2019). The strict division between volcanic and nonvolcanic passive margins can no longer be maintained and passive margins should rather be characterized by their magmatic content as magma-rich versus magma-poor (Franke, 2013;Peron-Pinvidic & Manatschal, 2019). Tugend et al. (2018) propose that timing of decompression melting may be more important than estimates of the magmatic budget of passive margins to understand their evolution and variability. Another important observation is the asymmetry in the width of conjugate margin pairs Svartman Dias et al., 2015). Early models explaining margin asymmetry are based on simple shear (Wernicke, 1981), which generates detachment faulting along low-angle normal faults, cutting through the entire lithosphere (Lister et al., 1986). In contrast, Brune et al. (2014) showed that margin asymmetry can instead be generated by rift migration, causing the development and lateral migration of a low viscosity pocket between upper and lower crust into which the faults are detaching. There is consensus that crustal and mantle processes interact during passive margin formation. However, the contribution of the mantle lithosphere in these processes is to a large amount unconstrained (Peron-Pinvidic & Manatschal, 2019).
In this study, we focus on the lithospheric structure of the South Atlantic passive margins, which are characterized by a wide range of different margin types. Triassic dyke swarms and rift basins may indicate earliest tectonic extensional forces (Borsa et al., 2017;Clemson et al., 2007;Peyve, 2010). However, it is unclear whether these structures are directly related to the subsequent extension and opening of the South Atlantic. In the Late Jurassic, rifting started and caused the disintegration of Western Gondwana, leading to the opening of the Falkland and Austral Segments of the South Atlantic (e.g., Heine et al., 2013;Moulin et al., 2010;Rabinowitz & LaBrecque, 1979). Rifting propagated northward, forming the Central Segment in the Aptian, which is positioned between the Rio Grande Fracture Zone in the south and the Chain Fracture Zone (CFZ) in the north (Moulin et al., 2010). In the Late Aptian/Early Albian, the equatorial part of the South Atlantic opened Moulin et al., 2010), characterized by a higher degree of oblique rifting (Brune et al., 2018).
In recent years, several global and regional lithospheric thickness models have been published that cover the South Atlantic passive margins. Global models are, for example, derived from surface-wave dispersion maps (Pasyanos et al., 2014), conversion of seismic tomography to thermal LAB (Steinberger & Becker, 2018) or multi-probabilistic joint inversion (Afonso et al., 2019). The global LAB models have a wide depth range, partly depending on the data sets and regularizations used in establishing the models. Due to the narrow and elongated margin geometry, many of these global models are not capable of mapping the LAB in this region. Finger et al. (2021) presented a regional model for the South American continent based on combined density, thermal, and compositional modeling. However, their modeling approach focusses on the continental platform and the passive margins are not precisely represented.
We model the thermal evolution of lithosphere during rifting in the South Atlantic and define a LAB depth after rifting ceased, representing the present-day LAB depth. Our model is specially designed for the margin area of the South Atlantic. The thermal LAB depth is derived from three input parameters: stretching factors, duration of rifting, and crustal thickness. Stretching factors are calculated by dividing unthinned crust by thinned crust, using published crustal models. Together with duration of rifting and crustal thickness, we then calculate a thermal model of the 10.1029/2021TC006828 3 of 18 extended lithosphere. The present-day LAB depth is defined by extrapolating the linear geotherm from the crust throughout the lithosphere after rifting and lithospheric cooling. Next, we discuss the evolution of the thermal LAB for the conjugate South Atlantic passive margins in a Gondwana reconstruction and evaluate differences between conjugate margin basins by correlating the predicted LAB depth with margin width and seismic tomography.

Methods
We calculate lithospheric thickness as a function of duration of rifting, crustal thickness, and stretching factors. For that, we use the python code RiftSubsidence based on a software that was originally designed to calculate theoretical subsidence curves for rifting scenarios in 1D (White et al.,pers. com.). In RiftSubsidence, the subsidence is calculated based on the amount and timing of pure shear lithospheric extension, as well as on the thermal and density structure of the lithosphere, following the model of McKenzie (1978).
In this approach, the lithospheric thickness is derived from the thermal structure after the lithosphere has been stretched. The temperature of the model is calculated by 1D advection and diffusion using finite differences. The top of the model is defined at sea level, while the base of the model is defined as the LAB. At these boundaries, the temperature is fixed throughout the entire rifting period.
We adopt the duration of rifting as implemented in the deforming plate model of Müller et al. (2019). As simplification, early oceanic crust of the Austral and Central Segments of the South Atlantic are considered to be of the same age, ranging between 141 and 120 Ma (Müller et al., 2019). These two values represent onset and cessation of rifting for the Austral and Central Segments. At the Equatorial Segment, rifting initiated at 121 Ma and stopped at 107 Ma (Müller et al., 2019). Even though the exact timing of rifting may vary internally in each segment, the simplification is sufficient for application in thermal lithosphere modeling. In a synthetic example, we show that onset and cessation of rifting have the lowest impact on the modeled lithospheric thickness ( Figure S1 in Supporting Information S1).
Prior to lithospheric stretching, LAB depth z LAB and LAB temperature T LAB must be defined. z LAB is balanced isostatically against a reference Mid-Ocean-Ridge (MOR). Figure 1 shows the isostatic model of the reference MOR on the left and passive margin on the right. The vertical column of the passive margin is defined by five layers: water, sediments, crust, mantle lithosphere, and asthenosphere. Assuming that the thickness of the mantle lithosphere h m is the only unknown parameter, an isostatic equation can be defined as follows:  Table 1.
The assumed values of density and thickness of each layer are listed in Table 1.
The densities of the sediments are calculated using depth-dependent equations provided by Sykes (1996), accounting for isostatic corrections of the sediment load. The sediment layers are vertically subdivided every 0.1 km. For each vertical column a density value is calculated, which contributes to the isostatic equation Equation 1.
The crustal thickness varies spatially for the passive margin. Accordingly, the LAB depth of the isostatic balance, from now on referred to as z LAB,iso, is individual for each point and is defined as LAB, = ℎ + ℎsed + ℎ + ℎ . For crustal density, we calculate the average value over the margin area (see Table 1). The value of = 2.81 g/cm³ is obtained by isostatically balancing thicknesses and densities of crystalline crust and sediments of the crustal model of Finger et al. (2021). Thus, it represents a mean value of the entire crust.
The isostatic balance in Equation 1 assumes a constant mantle density. As the South Atlantic passive margins are bounded by both continental and oceanic lithosphere, the mantle density is expected to be heterogeneous. Estimates from global and regional inversions indicate varying mantle densities along the South American passive margin (e.g., Afonso et al., 2019;Finger et al., 2021). Therefore, the mantle density is the second unknown parameter of the isostatic column.
Prior to rifting the thermal state of the lithosphere can be regarded as purely conductive with a linear geotherm. Assuming a constant linear geotherm for the entire lithosphere (blue lines in Figure 2), we calculate the mantle density and thickness of the mantle lithosphere in an iterative scheme: 1. Select a starting value of h m,0 . We select h m,0 = 60 km. 2. As the geotherm is linear throughout the lithosphere, the average mantle temperature can be directly derived from T LAB and the temperature at the Moho T Moho is = LAB − Moho 2 (2) Figure 2. Geotherms and the lithosphere-asthenosphere boundaries. Dashed line: initial geotherm prior to onset of rifting, representing initial isostatic LAB depth z LAB,iso and temperature at the LAB, T LAB . Circles, uplifted geotherm after rifting and cooling; solid blue line, present-day extrapolated linear geotherm based on crustal thickness, z Moho ; and temperature at the Moho, thermal structure of crust (filled circles). The depth at which the extrapolated geotherm reaches T LAB provides an approximate estimate of present-day lithospheric thickness z LAB,pres .
With h m,i = thickness of mantle lithosphere at iteration step i.
3. Assume mantle density as a function of temperature, based on the volumetric coefficient of thermal expansion α (e.g., Turcotte & Schubert, 2018): Solve the differential as follows: Solve Equation 5 for mantle density: With ρ m,i = mantle density at the iteration step i.

Update h m based on Equation 1 and mantle density of Equation 6. 5. Repeat
Step 2-4.
The process is iterated until the density change reaches the threshold ‖ +1 − ‖ < 0.001 g/cm³. For T LAB and α, we choose standard values of T LAB = 1,333°C and α = 3.28*10 −5 1/K (e.g., Jarvis & McKenzie, 1980;Parsons & Sclater, 1977). z LAB,iso represents the depth of the LAB prior to rifting. If the lithosphere is not stretched, a linear geotherm can be assumed for the entire lithosphere. However, rifting causes lithospheric extension with subsequent nonlinear displacement of the geotherm. After extension ceases, the lithosphere cools and the geotherm relaxes back to the linear state at infinite time. Depending on initiation and end of rifting and the amount of stretching, the geotherm will deviate from its initial linear state ( Figure 2).
In this approach, we simplify the geothermal gradients inside the lithosphere after rifting ceased. Assuming that conductive heat transport in the crust is the dominant heat source and that thermal expansion is constant both for the crust and lithosphere, we can extrapolate the geotherm of the crust throughout the mantle lithosphere. As a result, z LAB,iso is shifted upward (Figure 2). The difference between the initial z LAB,iso and present-day lithospheric thickness z LAB,pres, , based on the extrapolated crustal geotherm, is defined as Δz LAB . z LAB,pres defines the LAB depth after rifting ceased and the lithosphere has been cooling down.
The amount of nonlinearity of the distorted geotherm in Figure 2 depends on the elapsed time and on the amount of stretching. Based on the concepts described by McKenzie (1978), we define the stretching factor ß as the thickness of unthinned crust divided by the thinned crust: h i defines the initial crustal thickness, whereas h s is the crustal thickness after stretching. Müller et al. (2019) used this approach to derive stretching factors globally for all deforming regions. However, their values are calculated based on uniform stretching and do not consider crustal thickness gradients.
Synthetic tests show that stretching factors and crustal thickness have the largest impact on the modeled LAB depth ( Figure S1 in Supporting Information S1). Therefore, we derive new stretching factors for the South Atlantic passive margins. The proximal extension of the passive margins is defined by the landward limit of stretched continental crust, referred to as unstretched continental crust limit (UCCL; Williams et al., 2011). Here, we use the extension as it is captured in the deforming plate model of Müller et al. (2019). The distal extension of stretched continental crust is defined as landward limit of oceanic crust (LaLOC; Heine et al., 2013). We use the LaLOC geometry of the plate kinematic model of the South Atlantic (Heine et al., 2013), as it better integrates regional crustal-scale seismic data. Where SDRs sequences go beyond the margin area, hyperextended crust can be expected. Therefore, we adjust the geometry of the LaLOC by acknowledging the recently imaged SDR sequences for the Austral Segment of the passive margins (Chauvet et al., 2021;McDermott et al., 2018). This gives us the final geometry of the passive margin area, which is from now on referred to as continent-ocean-transition (COT) zone.
For the calculation of the stretching factors, we consider the thickness of crystalline crust by removing water and sediment layers ( Figure S2 in Supporting Information S1). Two different crustal models are used: one for the inner continent, representing unthinned crust h i , and one for the COT, representing thinned crust h s . To obtain unthinned crust h i , the inner margin of the COB is extended 500 km toward the continent. For South America, we select the continental crustal thickness model of Finger et al. (2021), which is based on a geostatistical kriging approach using available seismic determinations. For Africa, we use the CRUST1.0 model (Laske et al., 2013). Based on crustal thickness patterns and the distribution of tectonic domains, we define different segments and calculate mean values for each segment. This is sufficient for a first-order approximation of crystalline crustal thickness of unthinned crust, which is required for the estimation of stretching factors.
For thinned crust in the COT, we select the CRUST1.0 model in a resolution of 0.5° for both margins. To ensure a smooth transition to oceanic crust, we seed synthetic points in ∼50 km distance along the LaLOC with a thickness of h s = 8 km, representing average thickness of oceanic crust ( Figure S3 in Supporting Information S1). Figures 3a and 3c show the crystalline crust for both margins and the inner extension area. The resulting stretching factors (Figures 3b and 3d) are discussed in the next section. Figure 4 shows the modeled z LAB,pres for the South Atlantic passive margins. The Rio Grande Fracture Zone (RGFZ) separates the Austral and Central Segments of the South Atlantic, while the CFZ separates the Central and Equatorial Segments (Moulin et al., 2010). As a general trend for both margins, z LAB,pres is getting deeper toward the UCCL and shallower toward the LaLOC. However, the along-margin and across-margin patterns of z LAB,pres vary for the different segments and basins of the South Atlantic passive margin.

LAB Structure Along the South Atlantic Passive Margins
Along the Austral Segment of the South American passive margin, the LAB depth reflects intermediate depths from 90 to 120 km ( Figure 4a). In this area, only minor across-margin gradients in lithospheric thickness are modeled. In some parts of the Colorado and Punta del Este Basins, the modeled z LAB,pres is almost flat, reflecting values at around 100 km depth. Toward the Santos Basin, the variation of z LAB,pres is getting stronger. Along a wide COT area, the LAB deepens up to 180 km depth toward the UCCL. For the Central Segment, z LAB,pres is uplifted up to 40 km toward the LaLOC, resulting in pronounced across-margin gradients. This is evident for the narrow Bahia Basin and the northward continuing Sergipe Alagoas Basin, where the range of z LAB,pres varies from 40 to 200 km. In this area, the UCCL is extensively shifted toward the inland Borborema Province, which hosts isolated Early Cretaceous rift basins, causing lithospheric deformation during the opening of the South Atlantic (Heine et al., 2013). Compared to the Central and Austral Segments most parts of the Equatorial Segment are characterized by a deeper z LAB,pres toward the UCCL. Only the narrow Potiguar and Barreirinhas Basins reflect a very shallow z LAB,pres of 40-80 km. For the northward continuing Foz do Amazonas-Marajo Basin the COT is widely characterized by deep lithosphere between 120 and 160 km, but with a sudden uplift to 40 km toward the LaLOC.
On the African side, large parts of the Austral Segment of the COT show characteristics in lithospheric thickness patterns similar to the South American side. For the Southwest African Basin z LAB,pres reaches values from 120 to 140 km ( Figure 4b). However, an important difference to the South American conjugate is the distinct thinning of z LAB,pres toward the LaLOC, similar to parts of the Equatorial Segment of South America. This lithospheric pattern is maintained along the Central Segment, independent of the margin widths. In some parts, the UCCL cuts through the adjacent Congo Craton, resulting in values of z LAB,pres deeper than 150 km. Similar to the Borborema Province, the COT in this area represents extended lithosphere along reactivated basement structures (Heine et al., 2013). At the northern boundary of the Central Segment, both UCCL and LaLOC are located onshore. In this area, the determination of the COT is inaccurate due to ongoing deposition of sediments in the Niger Delta, as well as overlapping signatures of the Central and West African Rift Systems, expressed as the Benue Trough. This causes large uncertainties in the modeled z LAB,pres . These uncertainties are propagated throughout the Equatorial Segment, which is characterized by a very thin margin, challenging the identification of lithospheric structures both along and across the margin. For a comparison of conjugate margin segments in a Gondwana framework, we therefore focus on the Central and Austral Segments.
In Figure 5a, we show z LAB,pres rotated back to 115 Ma, together with depth profiles for selected conjugate margin pairs that are stitched together at the LaLOC of both margins (Figures 5b-5d). Thus, the value of maximum margin width corresponds to the location of the UCCL. For absolute comparison, we pick the LAB depth in the center of three selected margin profiles to show the lithospheric architecture across the margin: 1. Bahia/Congo margin. 2. Santos/Kwanza margin.

Salado/Southwest African Margin
We calculate the width of each passive margin profile by measuring the distance between the coordinates of UCCL and LaLOC, estimated by flowlines, which are implemented in the GPlates software . The conjugate Bahia/Congo Basins (Figure 5b) are characterized by a narrow margin of ca. 100 km width on the South American side and an intermediate margin of 200 km width at the African counterpart. z LAB,pres varies between 80 km at the LaLOC and 120 km at the UCCL for the Congo Basin. For the Bahia Basin, the structure of z LAB,pres is similar, but 10 km shallower. Since the Bahia Basin is narrower than the Congo Basin, the across-margin gradient in z LAB,pres is more expressed. This gradient is already apparent in the initial stretching factors ( Figure 3) and represents orthogonal rifting from the Base Aptian with a relatively constant thinning of z LAB,pres toward the LaLOC.
The Santos/Kwanza margin ( Figure 5c) reflects a strong asymmetry with an ultra-wide margin on the South American side (more than 700 km) and a narrow margin on the African side (ca. 120 km). The asymmetry in this area has been attributed to steady state rift migration with larger extension velocity toward the Santos Basin , accompanied by an oblique orientation of rifting that results in very wide margin profiles across the Santos Basin. The asymmetry of the conjugate margin pairs is also reflected in the lithospheric structure across the margins. z LAB,pres varies between 70 and 200 km for the Santos Basin (Figure 5c). Given the enormous width of the Santos Basin, the across-margin gradient is rather smooth, especially for the first 550 km proceeding from the LaLOC, where z LAB,pres increases from 80 to 100 km. Just in the last 150 km toward the UCCL, z LAB,pres drops from 110 to 200 km depth. This deep lithosphere is already present in the isostatic LAB depth z LAB,iso (Figure 7a), due to thick crust ( Figure 3a) and shallow water depth ( Figure S2 in Supporting Information S1), which is only partially compensated by the thermal modeling. For the conjugate Kwanza Basin, z LAB,pres is in stark contrast to the South American counterpart. In this area, z LAB,pres is constant at around 55 km depth across the entire margin. However, since the Congo Craton directly borders to the Kwanza Basin, deeper cratonic lithosphere can be expected. One reason for the shallow lithosphere in the Kwanza Basin is higher stretching factors across the entire COT area (Figure 3d). Another reason might be Neogene epeirogenic uplift of the onshore Angola Dome (white star in Figure 5), which has been attributed to thermomechanical thinning of the lithospheric mantle (Klöcking et al., 2020). The distribution of the +20 mGal free air anomaly contour (white dashed line in Figure 5) indicates that the dynamic support of the lithospheric mantle is located more southward of the Angola Dome. Together with our modeled z LAB,pres , which represents the thermal state of the lithosphere, these observations provide evidence for post-breakup lithospheric thinning in this area.
As part of the Austral Segment, the conjugate Salado/Southwest African Basins represent an earlier rifting stage in the Early Cretaceous. Therefore, the margins already have diverged from each other at 115 Ma (Figure 5a). Rifting occurred presumably in orthogonal direction, and both margins are characterized by intermediate margin widths of moderate asymmetry, where the Salado Basin is 100 km narrower than the Southwest African Basin. A moderate asymmetry is also visible in the LAB structure with values from 90 to 120 km for the Salado Basin and 60-130 km for the Southwest African Basin. z LAB,pres in the center of the margin profiles is, however, similar for both margins (100-110 km), representing typical values for cooled oceanic lithosphere (e.g., Richards et al., 2020). The observed asymmetry is in agreement with a study of Chauvet et al. (2021), who relate the asymmetry of conjugate SDR sequences in the Austral Segment to variations in the thermal structure.
In comparison to the other input variables, the lateral variations of lithospheric thickness are caused by varying stretching factors. In our synthetic example, we show that the amount of stretching causes large variations in the modeled lithospheric thickness ( Figure S1 in Supporting Information S1). As an approximation on global scale, Müller et al. (2019) showed stretching factors as part of their deforming plate model that represent "wide rifts that lack margin-orthogonal strain rate and crustal thickness gradients." Our approach is based on laterally varying crustal thickness. This is captured as along-and across-margin variations in the stretching factors (Figures 3b and 3d), making it an important enhancement of the stretching factors that are delivered with the Müller et al. (2019) deformation plate model. In most areas, stretching increases toward the LaLOC. Highest stretching factors are obtained close to or at the LaLOC. The maximum value of stretching is ß = 4.6, implying that continental crust is almost five times thicker than the crust close the LaLOC. This is evident for large parts of the LaLOC of both margin sides (Figures 3b and 3d).
Thermal lithospheric thickness models are closely related to seismic tomography models. Assuming that variations in seismic velocities are only related to temperature variations, seismic tomography models can be converted to thermal LAB models (e.g., Cammarano et al., 2003;Steinberger & Becker, 2018). Therefore, we qualitatively compare seismic velocities from S-wave tomography (Celli et al., 2020), LAB depths, and conjugate margin widths for the Central and Austral Segments (Figure 6b-6d). In general, margin width is proportional to LAB depth, that is, a thin margin corresponds to shallow LAB depth. Only the LAB depth in the center of the South American Santos Basin deviates from this trend. The patterns in seismic tomography, however, are different.
For the African part of the Austral Segment, LAB depths from 90 to 110 km correspond to slight perturbations in S-wave velocity dV s around −0.5% to 0%. Along the South American margin, the imaged z LAB,pres is similar to the African part, but the across margin distribution and velocity pattern changes (light gray areas in Figures 6c and 6d). In addition, a high velocity anomaly between 2% and 3.5% is observed, which is unusual for passive margin lithosphere. Celli et al. (2020) interpret the high velocities in this area by a deep lithospheric root, caused by isostatic negative buoyancy. The deeper across margin distribution of z LAB,pres indicates that there might be a contribution of lithospheric buoyancy. However, this area also coincides with compositional changes in the lithospheric mantle, which are not imaged by seismic tomography (e.g., Afonso et al., 2019). Our results indicate a combination of both thermal and compositional anomalies along the Austral segment of the South American passive margin.
Toward the Rio Grande Fracture Zone and the transition to the Central Segment, both z LAB,pres and dV s deviate stronger. On the South American side, high velocity anomalies are maintained, coincident with a wide range of lithospheric thickness in the Santos Basin. Further north in the Campos and Bahia Basins dV s drops to −1.5%, accompanied by shallower LAB depths (80-100 km). These structures might represent anomalously hot oceanic lithosphere due to activity of the Trindade Hotspot (Celli et al., 2020). At latitudes 5° S and further north, LAB depths between 60 and 160 km coincide with distinct velocity anomalies higher than 2%. Because the UCCL is shifted landward, these anomalies represent to a large extent continental lithosphere of the Borborema Province.
Strikingly, between latitudes 20° S and 15° S of the African passive margin, z LAB,pres and dV s are anti-correlated. Very shallow lithosphere less than 60 km in the Namibe Basin is associated with a high velocity anomaly of 2%. At the transition to the northern Kwanza Basin, z LAB,pres increases, while dV s drops to 0.5%. Only for latitudes north of 10° S, throughout the Congo and Gabon Basins, high velocity perturbations correlate with deeper lithosphere. Similar to our interpretation of the shallow lithosphere, the drop in S-wave velocity is related to a thermal anomaly in the lithospheric mantle, expressed as the Angola Dome (Klöcking et al., 2020). The anti-correlation between z LAB,pres and dV s in this area less indicates a dichotomy of the two models, but rather a relative of a few hundred kilometers in north-south direction. In the South Atlantic Basins, the tomographic model has the lowest lateral resolution in the lithosphere (Celli et al., 2020). Therefore, horizontal smearing of seismic velocities can be expected. However, further research is required to explain the differences between our lithospheric model and seismic tomography in more detail.

Isostatic Balance and Ground Truthing of Crustal Thickness
To quantify the effect of linear extrapolation of the geotherm, z LAB,iso and the difference of z LAB,pres and z LAB,iso , Δz LAB , are calculated for the South American passive margin. We also investigate the estimated mantle density and number of iterations. Here, we show the results for the South American passive margin. Results for the African passive margin show similar features ( Figure S4 in Supporting Information S1).
Many features of z LAB,pres are already present in z LAB,iso (Figure 7a). Across margin gradients of lithospheric thickness are established with shallower lithosphere toward the LaLOC. A prominent feature is the deep lithosphere along the Santos Basin and the continental Borborema Province. However, Δz LAB reflects important differences, changing the structure of lithospheric thickness significantly. Highest values of Δz LAB are 30 km and are found in the Santos Basin, the proximal part of the Bahia Basin and the distal parts of the Punta del Este, Salado, and Colorado Basins (Figure 7b). In the Central and Equatorial Segments, Δz LAB decreases toward the LaLOC. Highest differences can be expected in regions with a low ratio between crustal thickness and isostatic thickness of the mantle lithosphere, because the difference between the linear geotherm approximation and the initial geotherm is  strongly expressed (see Figure 2). The Santos Basin is characterized by deep z LAB,iso up to 200 km and intermediate crustal thickness around 30 km and the distal part of the southern basins is characterized by very shallow crust less than 10 km ( Figure 3a). Therefore, Δz LAB is high in these regions. In the other regions, the difference of crustal thickness and isostatic thickness of the mantle lithosphere is too small to generate a significant variation in Δz LAB . This is evident, for example, in large parts of the Pelotas and Bahia Basins.
The second unknown parameter of the isostatic balance is the mantle density ρ m . Figure 7c displays the mantle density of the isostatic column prior to rifting. Notably, ρ m is modeled 3.35-3.36 g/cm³ for most areas, which gives a difference of 0.05 g/cm³ compared to the density of the asthenosphere ρ a (3.3 g/cm³). This reflects the temperature-dependent decrease of density with depth from lithosphere to asthenosphere. Most of the points reach the density threshold after i ≤ 4 iterations ( Figure 7d). Only where lithospheric thickness is very low, like in the distal part of the Equatorial Segment, the algorithm needs more iterations to converge. The results show an inherent stability between mantle density and thickness of the mantle lithosphere, which are the variable parameters of the isostatic balance.
Even though many studies of the crustal structure of the South Atlantic passive margins have been carried out, a coherent and consistent crustal model for the individual passive margins is not publicly available. Since our approach is designed on a continental scale, we select crustal thickness values from the global CRUST1.0 model (Laske et al., 2013). However, to acknowledge the continuous growth of available seismic refraction data that has been acquired in the last decades, we compare the modeled LAB depth z LAB,pres, based on CRUST1.0 crustal thickness data, to a modeled LAB depth, which is based on crustal thickness values for a given seismic refraction profile in the Santos Basin. We select the profile SB01, which is a wide-angle seismic profile that was acquired during the Santos Basin (SanBa) experiment in 2011 (Evain et al., 2015). By comparing both LAB models to the seismic profile, we can ground truth our modeled LAB depth.
For the isostatic calculation, we keep the other input parameters and data sets as listed in Table 1 and interpolate sediment thickness and water depth on the seismic profile. Stretching factors are recalculated by updating thinned crust with the crustal thickness values of Evain et al. (2015). The values for unthinned crust, as well as onset and cessation of rifting remain the same. The modeled LAB depth z LAB,pres at the SB01 profile reflects a very smooth structure with values between 90 and 110 km (Figure 8a), which is in accordance with the plate cooling model . Even toward the LaLOC this structure is maintained. Interpolating the previ- ously estimated z LAB,pres onto the seismic profile gives the differences to the obtained values on the SB01 profile ( Figure 8b). Toward the distal part of the Santos Basin, the gridded z LAB,pres is up to 40 km shallower than z LAB,pres on the seismic profile. In this area, synthetic crustal thickness of h c = 8 km at the LaLOC may cause a bias toward shallower LAB depth. Nevertheless, the LAB structure of both modeling approaches is fairly similar. Differences are in a range between ±10 km throughout the vast part of the profile, illustrating that crustal thickness of the CRUST1.0 model is well-suited to model the LAB structure along the Santos Basin.
The Santos Basin is characterized by large SDR sequences, which vary in type and thickness across the margin, indicating a variable budget of magmatic material that was erupted during continental breakup (McDermott et al., 2018). The heterogeneity in SDR sequences is not reflected in the modeled LAB depths, neither in the gridded z LAB,pres , nor in z LAB,pres along the SB01 profile. Therefore, a characterization of different SDR types and the subsequent determination of the magmatic budget remains a task that is beyond the resolution of the presented LAB depth models.

Comparison With Global LAB Models
While our lithospheric model is designed for passive margins only, other lithospheric models capture passive margins as part of a larger, in many cases global, model. In this section, the lithospheric thickness of our approach is compared to two global lithospheric models ( Figure 9). The first model, LithoRef18, is derived by joint inversion of multiple data sets (Afonso et al., 2019). The second model from Steinberger and Becker (2018) is derived by conversion of seismic tomography to a thermal LAB. We used the mean LAB depth of several seismic tomography models as published in Steinberger and Becker (2018).
For LithoRef18, a two-part structure of deeper lithosphere in the Equatorial Segment and shallower lithosphere in the Central and Austral Segments can be observed (Figure 9a). Compared to our model, the across margin gradients of LAB depth are less pronounced. In general, the LAB of LithoRef18 is deeper than our model. Thicknesses up to 180 km at the Equatorial Segment are reached, which is 80 km deeper than our model (Figure 9b). Also, the Central Segment of LithoRef18 is 40-80 km deeper. As an exception, a shallow anomaly in the southern Santos Basin images lithospheric thickness of around 50 km. Controversially, this structure coincides with the deep anomaly of our lithospheric model. In the LithoRef18 model, this area is characterized by anomalously low densities in the lithospheric mantle (∼3.27 g/cm³), forcing the LAB to be lifted up. In the Austral Segment, the difference in LAB depth between our model and LithoRef18 are in a range of ±40 km, indicating that both models image the same lithospheric structures. For the African margin, the differences are of the same magnitude as for the South American margin ( Figure S5 in Supporting Information S1). With depths between 110 and 180 km the thermal LAB of Steinberger and Becker (2018) shows a smoother distribution of lithospheric thickness than the other two models (Figure 9c), representing the laterally coarse resolution and attenuation of global seismic tomography models. Values smaller than 100 km are not observed. In general, the distribution of lithospheric thickness is similar to LithoRef18. The most notable difference is the missing shallow lithosphere in the southern Santos Basin. In this area, the thermal LAB of Steinberger and Becker (2018) (Schaeffer & Lebedev, 2013). The shallow lithosphere in the southern Santos Basin is a consequence of the preferred selection of Litho1.0, where shallow lithosphere is very pronounced.
Our model does not imply any preference which seismic tomography model should be used to convert seismic velocity to lithospheric thickness. As our model is purely thermal, pressure and compositional effects are not accounted for in the modeling process, whereas the presented global LAB models (partly) account for that. Our regional analysis rather shows a more variable structure of lithospheric thickness that global seismic tomographies cannot resolve due to their relatively coarse resolution. This makes our model relevant for studying the entire crustal and lithospheric architecture of passive margins, where tectonic signatures that are linked to lithospheric deformation vary on small scale.

Conclusions
We have derived a new present-day thermal lithospheric thickness model for the South Atlantic passive margins. Our model is calculated as a function of onset and cessation of rifting, crustal thickness, and stretching factors. The stretching factors are obtained by dividing unthinned crust by thinned crust, using published crustal models for the South American and African continent. The new stretching factors account for across rift crustal gradients at the passive margin and are a refinement compared to the Müller et al. (2019) model.
From our model, the LAB structure along the South Atlantic is more precisely determined than from global models. We model distinct variations in LAB depth in the range of 40-200 km along and across the passive margins, indicating different rifting mechanisms that lead to the opening of the South Atlantic. As a general trend, the LAB deepens toward the proximal part and shallows toward the distal part of the margin area. The amplitude, however, varies for individual basins. In the Austral Segment of the South American basins, the LAB is rather constant between 90 and 110 km, whereas for the conjugate Southwest African Basin, the across-margin gradient is expressed by an abrupt shallowing toward the LaLOC.
Comparing the LAB depths along the margins with the width of margin profiles show a structural correlation: thin lithosphere coincides with narrow margins. Analyzing the LAB depth in a Gondwana reconstruction reveals an asymmetry for conjugate margin profiles. Significant differences in margin width correlate with significant differences in LAB depth for conjugate margin pairs. This is especially evident for the conjugate Santos/Namibe Basins, implying a substantial component of asymmetric rifting. As an additional factor, thin lithosphere in the Namibe Basin matches with signatures of the Angola Dome, indicating post-breakup lithospheric thinning. To which extent the Angola Dome contributes to lithosphere thinning remains unclear. The same holds for quantification of magmatic underplating, which is beyond the resolution of our LAB model.
Future efforts should address these open questions, once a comprehensive data set outlining underplated crustal thickness on both conjugate margins is available. This could be easily included in the governing isostatic equation. The modeling of the thermal structure can be extended to 2D rifting scenarios instead of the 1D approach that we are using. Given the potential of improvements, we are confident that our approach opens a new pathway for more extensive analysis of the lithospheric structure of passive margins. Our procedure can be easily adapted to other passive margins on the globe. Ultimately, this would fill the gaps of reconstructed lithospheric models for the Gondwana Supercontinent.