Landscape responses to dynamic topography and climate change on the South African source-to-sink system since the Oligocene

The South African landscape displays important lithological and topographical heterogeneities between the eastern, western margins and the plateau. Yet the underlying mechanisms and timings responsible for this peculiar layout remain unclear. While studies have proposed a post-Gondwana uplift driver, others have related these heterogeneities to a more recent evolution induced by deep mantle flow dynamics during the last 30 million years. This theory seems supported 10 by the rapid increase of sediment flux in the Orange basin since the Oligocene. However, the triggers and responses of the South African landscape to dynamic topography are still debated. Here we use a series of numerical simulations forced with Earth data to evaluate the contribution of dynamic topography and precipitation on the Orange river source-to-sink system since the Oligocene. We show that, if the tested uplift histories influence deposits distribution and thicknesses in the Orange sedimentary basin, they poorly affect the large-scale drainage system organisation and only strongly impact the erosion across 15 the catchment for two of the four tested dynamic topography cases. Conversely, it appears that paleo-rainfall regimes are the major forcing mechanism that drives the recent increase of sediment flux in the Orange basin. From our simulations, we find that climate strongly smoothed the dynamic topography signal in the South African landscape and that none of the currently proposed dynamic topography scenarios produce an uplift high enough to drive the pulse of erosion and associated sedimentation observed during the Palaeocene. These findings support the hypothesis of a pre-Oligocene uplift. Our results 20 are crucial to improve our understanding of the recent evolution of the South African landscape.


Introduction
Conflicting timings and drivers have been proposed to explain the singular South African tilted topography as well as the subsequent increase of the sediment flux in the Orange basin over the last 30 Ma. Previous studies proposed dynamic topography uplifts associated with the African superplume (Lithgow-Bertelloni & Silver, 1998, Gurnis et al., 2000Nyblade 25 andSleep, 2003, Braun et al., 2014), with the African plate motion (Burke, 1996;Burke & Gunnell, 2008) or induced by lithospheric heating (Tinker et al., 2008b;Stanley et al., 2013) to explain both the landscape and the sedimentation history of the region. If the underlying mechanisms are still under discussion, the timings and number of uplifts remain also highly debated. Thermochronological data (Brown et al., 1990;Gallagher & Brown, 1999a;Brown et al., 2002, Tinker et al., 2008bFlowers and Schoene, 2010, Wildman et al., 2015, 2016 and proposed rift models (Gilchrist et al., 1994), are in favour of an 30 early uplift phase (around 130 Ma) induced by thermal uplift and rift inherited topography. Other thermochronological studies combined with increasing sedimentary flux estimates of the Orange and southern sedimentary basins (Tinker et al., 2008a, Rouby et al., 2009, Guillocheau et al., 2012, Baby et al., 2018;2020) support a significant and rapid phase of erosion affecting a large area of the South African plateau in the mid-Late Cretaceous (around 100-60 Ma). The two main proposed mechanisms behind this major pulse of erosion are attributed to 1) a renewal of dynamic topography uplift induced by the LLSVP (Braun Here, we focus on new studies that show a pulse of sedimentation in the Orange basin over the last 25 Ma (Baby et al., 2018(Baby et al., , 2020 combined with drainage studies (Roberts & White, 2010, Paul et al., 2014, geomorphological features (Dauteuil et al., 2015), and thermochronology data (Green et al., 2017) which uphold a late Cenozoic uplift. These studies link the last increase of sedimentary flux with either dynamic topography associated with LLSVP (Gurnis et al., 2000) or small-scale convection 40 (Burke, 2008). If some authors suggest that the Cenozoic uplift is directly caused by mantle plume upwelling, others argue that the associated dynamic topography amplitude would be too low (less than 10 m/Myr; Gurnis et al., 2000) and happened too early (South Africa was over the superplume about 70 Ma ago) to justify the rapid erosion of the plateau (Braun et al., 2014).
Various mantle flow models and associated dynamic topography scenarios have been published recently. In this study, we 45 quantify the impact of four dynamic topography scenarios on the Orange River source-to-sink system in order to evaluate its most recent phase of denudation (Partridge and Maud, 1987). These uplifts histories account for different mechanisms: 1-the drift of South Africa over a mantle upwelling inducing a small and 2-long-wavelength uplift (Gurnis et al., 2000, Cao et al., 2019, 3-or inducing a higher and narrower uplift (Hassan 2020) and 4-a tilted east-west uplift induced by the passive northeastward motion of Africa over a buoyant mantle (Moucha et al., 2011, Braun et al., 2014. 50 We decided to integrate but not to test various flexural uplifts as we consider that during the last 30 Myrs, the denudation and large volume of sediment shed from the continent (up to 1-2 km for Braun et al., 2014 and up to 1 km for this study) is more likely to be driven by the increase of precipitation than associated with flexural uplift.
Besides dynamic topography, we also explore the role of precipitation as a driver of increasing sediment flux supported by Baby et al., 2018 and2020. Despite the lack of climate synthesis in the region, we used a compilation of paleo precipitation 55 indicators (Appendix A) to define three paleoclimate maps accounting for spatial and temporal rainfall changes. We did not use climate models; their spatial resolution is too low compared to the 5 km resolution of the landscape used and does not have the same reconstruction as the ones used in the mantle flow simulations. We defined a first episode (Fig. A1a), from 30 to 20 Ma of semi-arid to arid conditions based on Macgregor 2010Macgregor , 2013. The second episode (Fig. A1b), from 20 to 15 Ma, integrates the observations from Pickford and Senut, 1999, Braun 2014, Salman & Abdula 1995 in increased 60 pluvial stage in the actual Namib desert and 2) a sub humid area in the southern part of South Africa during the Burdigalian (Appendix A). The last episode (Fig. A4), from 15 to 0 Ma, is based on the present-day rainfall record (Burke and Gunnell 2008

75
The tested dynamic topography scenarios capture different wavelengths of mantle flow as they are generated with four different methods (Table A1). The dynamic topography estimations of the model M1 results from a global seismic tomography-based backward-forward adjoined model under Boussinesq approximation and is calculated using CitcomS mantle convection model Gurnis 2008, Muller et al., 2018). The scenarios AY18 and M2 are also derived from CitcomS mantle flow calculations under the extended Boussinesq approximation (Zhong, 2006) and are driven by plate reconstruction, subduction 80 history and basal heating (Cao et al., 2019, Hassan et al., 2020. The dynamic topography values of the models M1, M2 and AY18 are calculated by removing the first 250 km (Muller et al., 2018;Cao et al., 2019;Hassan et al., 2020) of the upper mantle. The last scenario, model TX08, is extracted from a global joint inversion of seismic and geodynamic data model, which better simulates horizontal velocities and captures higher resolution of mantle flow structures (Forte et al., 2010, Moucha & Forte 2011. These four models have been chosen as they contain dense material similar to present-day Large low-shear-85 velocity provinces which are impacting the African plate.

Dynamic topography changes
Model AY18 has an increasing and homogeneous dynamic topography variation for both the Orange river mouth and the plateau (

Building the initial paleo-elevations
Considering that the tectonic history of South Africa was quiet during the Oligocene, we choose to generate the initial paleoelevation of the landscape evolution model by inverting the dynamic topography histories presented in Fig. 1. We extract this signal and subtract it from the actual topography model ETOPO1 (NOAA). We also remove the estimated load of sediments for the last 30 Ma as well as the associated isostasy component (Appendix A; Fig. A3, Table A1). We did not add the volume 110 of eroded sediments (i.e., sediments accumulated on the margins) in the Orange catchment basin as we consider that 1) the volume of eroded material vs. the area of the catchment basin is small (2000 km wide with an extent of more than 5.e5 km 2 ), 2) the source system is poorly constrained and 3) our study focuses on rates of sediment fluxes changes more than on their specific values. We recognise that this bias will lower the present-day topography of the primary source areas as well as lower the resulting marine deposit thicknesses. Despite this limitation, this approach produces for each mantle dynamic histories a 115 corresponding high-resolution paleo-elevation maps (Fig.2a).

Forward simulations
After inverting the dynamic topography impact, we use the four-resulting paleo-elevations to simulate the landscape evolution associated with each mantle driven uplift-subsidence scenarios using Badlands landscape evolution model (Salles et al., 2016; 2018; Appendix A). To insulate the dynamic signal from the climate forcing in the sedimentation record, we first test a first 125 series of experiments with a uniform reference precipitation of 0.6 m/yr., approximating the present-day rate of precipitation in the Orange catchment basin. From produced initial paleo-elevations and based on the different forcing conditions presented above, we then run a series of landscape elevation models accounting for the flexural isostatic responses associated with plate loading and unloading induced by erosion and sedimentation (Wickert et al., 2016, Appendix A, Table A1). Our simulations do not include the effect of chemical weathering as well as rollover, mass loss which implies a bias on sediment thickness and 130 sediment fluxes compared to data driven models (Rouby et al., 2009, Guillocheau et al., 2012, Dauteuil et al., 2013, Baby et al., 2020.

Topography
We first observe the contrasting initial topographies induced by the four tested scenarios. The models with the higher uplift 135 rates (AY18 and M1; Figs. 1a and 2a) have the lowest initial elevation. The initial topography of M2 shows a homogeneous east-west topography induced by the heterogeneity of the east-west dynamic topography variations. M2 and TX08 show a regression of several kilometres on the western South African margin and TX08 shows a Lesotho massif culminating at 600 meters higher than its current elevation.
We run a first series of landscape evolution models with a fixed rainfall of 0.6 m/yr. The results show that after 30 Myr of 140 evolution, the four simulated topographies are close to the present day (less than 100 m of variation in elevation within the Orange river catchment; Figs. 2b & c.), while the margins exhibit more differences induced by both sediment loading, isostatic adjustment (initial and temporal) and erodibility approximations bringing less material on the source area. The simulated location of the Orange River (and catchment; Fig. A4) are in accordance with its actual location but incisions are deeper than present day (up to 200 m of erosion). We attribute this mismatch to the initial surface elevation reconstruction approach that 145 we used where sediments removed from the Orange sedimentary basin have not been redistributed across the catchment (Appendix A). Nevertheless, this underestimation does not bias our results as we focus on the variation of sediment fluxes through the last 30 Ma. To be specific and better quantify this bias, we compare the actual erosion rate of our models with cosmogenic radionuclide-based bedrock erosion rates. The final erosion rates predicted by our models range from 0.5 m/Ma for M1 (Fig. A5) to a maximum of 10 m/Ma for TX08. These values fit with published erosion rates inferior to 10 m/Ma 150 (Cockburn et al., 2000, Kounov et al., 2007, Erlanger et al., 2012, Decker et al., 2013, Scharf et al., 2013, Dirks et al., 2016. It is worth mentioning that this comparison between models values vs. cosmogenic analyses can only be used as a first order estimation as it was performed in various erodible bedrocks. When considering landscape evolution models with a fixed rainfall, the four sediment fluxes extracted at the Orange river 160 mouth do not vary significantly through time (Fig. 3a, coloured lines). These values have the same magnitudes as the fluxes from Baby et al., 2020 but they do not fit with the increasing predicted fluxes over the Miocene, Eocene and Neogene. Our models, instead, capture finer but stable variations of sedimentary flux (every 100 000 years) around 5000 km 3 /Ma. The simulated values are comparable with the stable post rift accumulation rate of Guillocheau et al., 2012 andBraun et al., 2014. The mismatch between both values, simulated and previously published from earth-modelled fluxes, can be attributed to the 165 correction of carbonate proportion, volcanic production, porosity removal and chemical weathering (Guillocheau 2012, Rouby et al., 2012. For example, the estimated volume of sediments for the Orange basin in Rouby et al., 2012 is doubled assuming that half of the volume of sediment removed has been lost to chemical weathering (Larsen et al., 2014). Simulated fluxes for the different dynamic topography histories have similar trends. As this first series of tests considers fixed rainfall, this suggests that none of the tested dynamic topography models are able to drive the large pulse of erosion observed across the South 170 African plateau over the last 25 Ma (Baby et al., 2020).  (Fig. 3b). Nevertheless, this increase is contained between 20 and 15 Ma in our case (based on the climate maps that we derived from available paleo-dataste) and does not have a constant increasing pattern. This can be explained as the climate is 185 more humid on the eastern margin over the last 15 million years but overall drier on the western margin where the imposed erodibility is higher.

Correlations between erosion, climate, and dynamic topography
To quantitatively analyse the importance of each forcing mechanism, we extract different Spearman correlation indices (Appendix A) to measure the degree of association between 1) the instantaneous erosion, 2) the dynamic topography impact 190 and 3) climate in the Orange river catchment (Fig. 4).
We test correlation between erosion and the average of dynamic topography as well as dynamic topography at specific altitudes (850m, between 850 and 1350m and above 1350m; Fig. A5) under uniform and spatially variable precipitation rates. These specific altitudes have been chosen as the uplift associated with dynamic topography can vary between the western and the eastern part of South Africa ( Fig. 1) but also because they correspond to different erodibility provinces; for example, Karoo 195 basin contains sandstones intruded by dolerites which are more subject to erosion than the high overlapping basalts of Lesotho.
Even if we look at the correlation between instantaneous erosion and dynamic topography in the catchment, the simulated erosion in our simulations is always the product of three components: uplift, erodibility and rainfall.  For the model AY18, when precipitation is uniform, there is a strong positive correlation between the uplifted areas above 850m ( Fig. 4: 0.57 and 0.60) and the instantaneous erosion of the catchment. This model contains the most important dynamic 210 topography uplift, especially around the Lesotho (Fig. 1). In contrast, when the climate variation is turned on, the correlation between erosion and dynamic topography (Fig. 4: -0.28) is anticorrelated and non-significant compared to the moderate positive relationship between climate and erosion.
The model M1 results in a small uplift, around 50 meters over the last 30 Ma, of the eastern part of the catchment (Fig. 1) which did not act on the erosion of the catchment whilst an anticorrelation between uplift and erosion is strongly positive on 215 the subsiding western part of the catchment ( Fig. 1; Fig. 4). When the precipitation maps are turned on, the eastern and most elevated parts of the plateau, which experienced the most humid conditions between 0 and 15 Ma, highlight a stronger relationship between uplift and erosion (0.32 and 0.40: strong and moderate positive correlations).
The model TX08 experienced the highest subsidence rate of all tested dynamic topography scenarios. We see that for this model, the anticorrelation rates between uplift and erosion are highest, strongly positive anticorrelations for the area below 1350m, and moderate positive for the highest areas (Fig. 4). When we add temporal and spatial precipitation to this model, we can see that average climate and erosion are strongly correlated (0.64, Fig. 4).
In the last M2 scenario, we observe a very strong correlation between the erosion in the lowest area of the catchment and the imposed subsidence. Nevertheless, when the spatial and temporal precipitation variations are added into the model, this positive correlation between uplift and erosion is replaced by a strong positive correlation between climate and instantaneous 225 erosion in the Orange river catchment.
Here we observed that dynamic topography impact in the models AY18, TX08 and M2 is smoothed by the impact of spatial and temporal precipitation variations. The dynamic topography influenced on the South African erosion also appears to be non-significant for model M1, especially for the areas above 850 m. Here, not only the imposed uplift/subsidence associated 230 with dynamic topography vary between the eastern and western part of the Orange catchment but also does the climate condition, the erodibility as well as the slopes of the initial and evolving landscape. All these parameters have an impact on the instantaneous erosion which differs depending on their prevalence through time. For example, if the erodibility variation is fixed for the model TX08, then the correlation between erosion and uplift decreases in the highest area and increases in the area below 1350m (Fig. A7).

Depocenters locations
We saw that climate is a major driver of sedimentary fluxes on the Orange Basin for the model M2, TX08 and AY18, especially when the eastern and western part of the Orange catchment have the same uplift history (Model AY18). In Fig. 5, we see that depending on the uplift forcing, the location and the thickness of deposits change. Increased marine accumulations are observed when considering models M2 and TX08. Models AY18 and TX08 contain less deposited sediments in the southern basins than 240 models M1 and M2. Only models AY18 and TX08 show a preferential accumulation of sediments in accordance with actual observations, i.e., close to the Orange river mouth (Rouby et al., 2012, Braun, 2014.

Conclusions
Here we explored different scenarios of South African mantle uplift differing by their amplitudes and timings. We inverted mantle-driven forces to obtain the paleo-elevation of South Africa 30 Myrs ago. None of the tested dynamic topography 250 scenarios allow to generate the same variations of sediment flux as data-based estimations. We recognise that this might be due to an underestimation of the uplift associated with dynamic topography calculations (removal of the first 250 km of upper mantle for example). We actually found that after doubling the dynamic topography estimation (Fig. A6), we could observe a strongest correlation between dynamic topography and erosion of the Orange river catchment.
Here we demonstrate that if rainfall variations through space and time can generate the same pulse amplitude as sediment flux 255 observed in the Miocene (Baby et al., 2018;2020), our actual knowledge of the South African climate cannot predict the entire increasing pattern. Incorporating climate variations in landscape evolution models showed how climate smoothed the dynamic topography impact, especially if in scenarios where differential uplift also affected the eastern and western margin of South Africa (model TX08 for example). We show the importance of taking climate, dynamic topography as well as erodibility together as forcing parameters and not only focusing on one main driver as they appear to be completely interdependent.
We agree that mantle dynamics can cause rapid uplift associated with drainage reorganisation, and denudation over a short period of time (Ding et al., 2019). Braun et al. 2013, also showed that low slopes associated with the motion of a continent over a fixed source of dynamic topography may lead to substantial, i.e., kilometre scale, and rapid surface erosion, through continental-scale drainage reorganization. Erosion is linked with dynamic topography scenarios in this study, however, none of them can generate an increase of sedimentary fluxes compatible with Baby et al., 2018. Our results suggest that sediment 265 accumulation in the shelf might have happened as a consequence of an earlier plume impact (Gurnis et al., 2000) that might have induced a more significant uplift.
If the remaining mantle flow is not able to generate major changes in the source-to-sink system of the Orange River catchment over the last 30 Myrs, it strongly influences the location of depocenters in the Orange basin. This paper includes new modelled sedimentary flux data which allows us to understand the uplift history of South Africa as well as the climate impact over the 270 last 30 Ma. The method used here is a first step to reconcile Earth data and landscape evolution models and proposes a new framework to fill the data gap for the South African surface evolution. This work opens new perspectives to understand the interplay between tectonic, mantle, and surface processes in the context of climate, erodibility and sea level change, not only in South Africa but all around the margin basins. (1) where 30 is the initial paleo-topography (m), 0 is the actual topography (m), dt is the uplift value associated with the geodynamic models, is the timesteps between dynamic topography timesteps (yr.), is the sediment thickness for the last 30 Ma (m) and is the deflexion (m) induced by the load of sediments removal. We compiled existing sediment thickness maps (Seranne and Anka 2006, Rouby et al., 2009, Kuhlmann et al., 2010, Maystrenko et al., 2013, Baby et al., 2018 as well as 285 cross-sections along the Orange basin (De Vera et al., 2010, Guillocheau et al., 2012, Dauteuil et al., 2013, Baby et al., 2020 to produce a sediment thickness map for the last 30 Ma (Fig. A3). The flexural parameters used to calculate the deflexion induced by the removal of these sediments are synthesized in the Fig. A3 and Table A1. To take the effect of the load or removal of sediments into account on the initial deflection and flexural isostasy through time, we used the python module gFlex with finite difference approximation as we used non-uniform lithospheric elastic thickness values (Wickert, 2016).
Landscape evolution models. The surface evolution models have been generated using the open-source code Badlands (Salles, 2016, Salles and Hardiman 2016, Salles et al., 2018 which allows to simulate surface processes including fluvial, hillslope, and wave-induced sediment transport, as well as the growth of coral reefs (Salles et al., 2018).  Table A2) which is based on the South African lithologies.
Flexure. As detailed above, we account for flexural isostasy responses induced by the erosion and deposition via the python 310 package gFlex (Wickert, 2016) included in Badlands. The detailed parameters are available in Table A1 and and Senut, 1999;Braun 2014, Salman & Abdula 1995, with pluvial stage in the actual Namid desert (Bamford and Dewitt 2013). To integrate these data, we extended a sub humid area on the south of South Africa (Fig. A1b).
As authors consider more arid events (with a relatively humid plateau) around 15 Ma (Pickford and Senut, 1999;Ségalen et al., 2006, Braun et al 2014, Salman and Abdula 1995, we used the actual rainfall range from 15 to 0 Ma (Burke and Gunnell 2008). This incorporates the conclusion of Senut and Pickford that the climate was more arid in the Neogene (23-3Ma). Even if some authors argue that there is a short wetter period during the Plio-Pleistocene (after 5.3Ma, Ward and Corbett 1990, Bamford and 2013), it is difficult to constrain this event and we consider that the arid climate is still active (Fig. A1c).
Correlation methods. The heat maps presented in this study compiled robust statistical cross-correlation coefficients between 330 instantaneous erosion and the rainfall or dynamic topography variations. We used the Spearman rank correlation to measure the degree of association between these variables as it does not require any assumptions on data distribution. It is usually a more robust and stable estimate than the Pearson analysis.
Model limitations. Some of these limitations are intrinsic to all numerical landscape evolution models. For example, the nature 335 of simulated sediments is temporally and spatially uniform (Salles and Hardiman, 2016). Another limitation is the spatial and temporal resolution of the models (5 km and 200,000 yrs. time steps), which prevent the integration of local and short-term features as small catchments and river reorganisations for example. The time steps also imply non-realistic and drastic changes in the sedimentary fluxes variations (Fig. 3). The diffusion and slopes coefficient are another approximation impacting the sedimentation and the landscape elevation. These types of imposed parameters are not directly supported by data. For example, 340 the spatially variable erodibility values as well as the fixed ones (Fig. A7) are not constrained by data but by an erodibility index (Table A2) and a series of models to match both present-day elevation and sediment distribution of South Africa. Another reason these models must be tempered is that the material removed from the margin in the backward step is not added back in the Orange catchment basin before the forward step. Therefore, the analysis presented here focused on sediment fluxes, relative differences in sediment volumes between models and not absolute sediment volumes. Other imposed data come from 345 interpretations or models where the uncertainties are not quantified such as the dynamic topography estimations. The uplift rates or sedimentary fluxes used as "earth data" themselves are inferred from apatite fission track, thermochronometric data exhumation rates, seismic reflections, introducing another level of uncertainty.
Another limitation is that the sea-level curve used (Hacq 1987) has a temporal resolution finer than the timestep resolution of the models, yet the coastal sedimentation is also biased. The impact of longshore drift or bottom current like the Aguilas one 350 is not considered while it is essential to obtain a more accurate sedimentation prediction on the eastern and southern margin of South Africa. These limitations do not affect the results of our study as we focused on the evolution of the western margin.
The paleo precipitation maps were constrained by the limited availability of data resulting in a patchwork of data (Appendix A; Climate) and do not account for altitude variation. Their level of accuracy cannot be quantified whilst precipitation is one of the major controls on erosion rates. Even if the uncertainties involved in the creation of such the paleo precipitation maps 355 may have contributed to differences between data and model results, here we do not focus on values but on the impact of rainfall variations on erosion and sedimentation.

415
The authors declare that they have no conflict of interest.