Madagascar’s landscape A tale of two rifts

Authors: Romano Clementucci, Egor Uchusov, Yanyan Wang, Sean D. Willett

Abstract

The topography of Madagascar reflects a dynamic history of water divide migration, initially driven by rift events on both coasts separated by ~80 million years and later modified by Late Cenozoic volcanic and tectonic activity. This study examines how rifting-induced tilting of an escarpment-plateau landscape led to changes in drainage area and water divide positions. Combining topographic analysis and erosion rates inferred from cosmogenic isotope concentrations, we document a westward-tilted low-relief plateau with sinuous remnant escarpments along the western margin and a linear, high-relief escarpment corresponding to the modern drainage divide on the eastern margin. Our numerical modeling demonstrates that landscape asymmetry and contrasting escarpment morphology can result from a shift in the main water divide across the island during the second rifting event. These mechanistic processes allow us to reconstruct the landscape evolution of Madagascar over more than 100 million years with profound implications for the topography, hydrology, and biodiversity development of passive continental margins.

INTRODUCTION

The evolution of Earth’s landscapes in response to tectonic and surface erosional processes gives rise to the major characteristics that make life possible, from the supply of mineral nutrients, through the formation of soils, and by the creation of the diverse environmental habitats that make possible the diversity of biological species (1–3). Most complex topography is the result of tectonic activity, with convergent mountain belts being the prime example of high topography (2, 4) associated with high geo- and biodiversity (1, 2). Passive continental rift margins are often regarded as less topographically interesting, but in many cases, the opening of a new ocean basin leads to a complex landscape response as new drainage basins form a landscape escarpment that propagates inland with time [old escarpment (5–7)]. Madagascar is a prime example of a continental rift margin with high topography, high-relief escarpments, heterogeneous habitat, and high biodiversity. In contrast to convergent mountain belts, the relatively simple tectonic history of Madagascar provides an opportunity to reconstruct and quantitatively model the landscape evolution, thereby providing a set of predictions useful for studies of environmental conditions, paleoclimate, or biogeography.

Although Madagascar is regarded as a passive continental margin, it contains two, independent margins formed by rifting, first from Africa [170 to 140 million years ago (Ma) (8–10)] and, subsequently, from India and the Seychelles [~90 Ma (10–13)]. We argue here that it is the interaction between these two events, in particular the tilting associated with rifting, which has produced the characteristic landscape with multiple escarpments, and highly asymmetric topography. Adding a final phase of extension and volcanism in the Late Cenozoic to this history, we can explain all the major physiographic features of the island in a quantitative and chronologic sequence.

The formation of topographic escarpments is a consequence of short, steep rivers heading against a water divide and driving that divide inland (7, 14–16). This horizontal retreat of the water divide triggers a series of landscape reorganizations with implications for the spatial patterns of erosion rate and landscape evolution (17, 18). The typical “divide-type” river originates at the main water divide at the top of the escarpment, resulting in a “shoulder-type margin” (7, 14, 19, 20), although there are also rivers that initiate in the continental interior, flowing across the escarpment, exhibiting a stepped or “knickpoint-type” river profile, and where these rivers define an “arch-type margin” (Fig. 1) (6, 7, 21). The river profile form determines its erosional efficiency and whether the escarpment retreats uniformly, becomes more sinuous with time, or is degraded (22, 23).

Previous studies of landscape evolution in Madagascar have focused on the elevations of the surfaces and plateaus (23, 24) or on the shapes of the river profiles (25–27). However, most of them did not account for changes in water divides and catchment size through time, a fundamental driver of landscape evolution, which calls for a reevaluation of previous interpretations incorporating this observation.

In Madagascar, we observe examples of multiple styles of rift escarpment and evidence that these morphological styles have changed with time. Both rifting events produced rift escarpments, and the crustal thinning resulted in tilting of the preexisting landscape (28). Madagascar’s central and northern regions are characterized by a westward-tilted plateau, part of the “African erosion surface” (19, 29). Although the modern water divide is located near the east coast, there is evidence of a highly sinuous, remnant escarpment along the western edge of the plateau, and most west-flowing rivers show a stepped morphology (Fig. 1). In contrast, most of the eastern drainages comprise a low-sinuosity, shoulder-type escarpment along the eastern rifted margin. This pattern is not uniform. Many regions, particularly the northeast and central regions, have experienced surface uplift, rift activity, and volcanism during the Late Cenozoic, leaving an imprint on the river pattern, divide positions, and biodiversity distribution (Fig. 1) (27, 30–32). Here, we explore the hypothesis that the modern landscape reflects these two main phases of rifting, each with an escarpment-forming event, but also regional tilting, resulting in the interference of the uplift patterns and changes in the drainage divide position. We show that the western Madagascar escarpment has been degraded and is characterized by isolated topographic remnants and highlands but that the original escarpment is still recognizable (Fig. 1B). The eastern side, in contrast, presents a steep and linear escarpment edge resulting from escarpment retreat driven by divide-type rivers, excepting those regions affected by Late Cenozoic tectonics (Fig. 1B) or recent river capture (Fig. 1) (25). We posit that the western escarpment originally served as the main drainage divide associated with the western rift. However, following the second phase of rifting in the east, Madagascar experienced down-to-the-west tilting, and the water divide jumped to near the east coast. Drainage of the upland plateau shifted to westward, and the rivers crossing the western escarpment rapidly changed from divide-type to knickpoint-type, leading to its rapid demise as a coherent geomorphic feature. In a final phase of drainage adjustment, segments of the main drainage divide have shifted inland in response to Late Cenozoic tectonic and volcanic processes in the central and northern regions, leading to the present-day geomorphic configuration. To investigate these processes, we characterize the eastern and western margins using ^10^Be-derived erosion rates and regional geomorphic analysis and construct a numerical landscape evolution model showing the effects of a large divide jump. These large-scale divide jumps have left a distinct imprint on Madagascar’s landscape, leading to substantial geomorphic disequilibrium, as shown by differential erosion rates, divide migration rates, and relief patterns from west to east. Geomorphic disequilibrium has implications for patterns of erosion, weathering processes, and ecological habitat distributions. Our results suggest that Madagascar’s landscape evolution contributes to its exceptional species richness and local endemism (33, 34), supporting earlier work (17).

Geological evolution of Madagascar

Madagascar was originally situated in the eastern half of the Gondwana supercontinent and experienced two main phases of continental breakup (Fig. 1A) (8, 35, 36). The separation of Madagascar from Africa occurred in the Middle Jurassic (170 to 140 Ma) and was interpreted as (i) a distinct episode of rifting followed by rapid continental drift and subsequent translation along a north-south transform zone (8–10, 37) or (ii) a progressive, southwest-propagating phase of oblique rifting and seafloor spreading (38, 39).

The second major tectonic event is the breakup between Madagascar and Seychelles-India in the Early Cretaceous, accompanied by extensive volcanism on the western and eastern sides of the island [~90 Ma (10–13)]. This phase is marked by voluminous magmatism along both rifted margins, with basalt flows covering most of the coastal areas and Precambrian basement, likely related to the Marion plume (40).

Following the major rifting events, Madagascar has remained largely tectonically inactive. However, in the past 30 to 60 Myr, there have been vertical motions associated with mantle dynamics, and since 30 Ma, there have been renewed volcanism and local extension associated with the southern extension of rifting in East Africa (30, 41). Dynamic uplift related to subplate mantle convection can account for hundreds of meters of uplift in the central and northern sector, respectively (fig. S1A) (27, 42–44). Observed uplift rates are ~10 m/Myr since the Eocene and Miocene and are locally up to 20 to 70 m/Myr since Marine Isotope Stage 5e, as estimated by uplifted marine deposits and terraces, respectively (Fig. 1A) (27, 45, 46). Renewed extension is evident in the Neogene to Pleistocene volcanism and crustal extension in the Alaotra-Ankay Graben system (AAG), the Ankaratra volcanic complex (AV), and the Antongil Basin (AB) (Fig. 1B) (30, 31, 41, 47). Instrumental seismicity indicates tectonic activity in central and northern Madagascar, where three clusters of earthquakes are observed, corresponding to the active extensional basins and volcanic complexes (fig. S1B) (47, 48).

Geomorphological characteristics of Madagascar

Madagascar is characterized by three main morphological a low-gradient coastal plain, a low-relief high-elevated erosional plateau, and steep topographic escarpments, almost ringing the island. The eastern topographic escarpment can reach more than 2000 m of relief, with a rapid elevation gain in a short distance from the coastline (tens of kilometers). The western side Madagascar is characterized by lower relief (Fig. 1B) and a more gradual transition from the coastal plain to the plateau, typically over hundreds of kilometers (Fig. 1B).

The topographic escarpments and tilted plateau are mainly composed of Precambrian paragneiss and orthogneiss. Locally more resistant granitic and volcanic rocks occur sporadically in the plateau and western margin of the island (fig. S1). The coastal plain of the western margin is dominated by Paleo-Mesozoic volcanic and sedimentary rift deposits that are uplifted and tilted during the Cenozoic vertical movement (fig. S1C) (23, 49).

The eastern escarpment of southern and central Madagascar exhibits primarily divide-type rivers and highly asymmetric topography across the divide, with a tilted plateau on the western side (Fig. 1B, swath C). There are two exceptions to this morphology. At about 23°S, the divide shifts to the west, deviating from the escarpment. This is associated with the Mananara River, which is hypothesized to have been recently captured and reversed from west-flowing to east-flowing (25). The second exception is at about 20°S, where the main divide again jumps to the west, to a position atop the AV (Fig. 1B, swath D). In northern Madagascar (north of 19°S), the topography is less asymmetric, showing a highly dissected plateau with high-relief topography near the western margin of the plateau, which we interpret as escarpment remnants. The eastern slope shows the same steep escarpment as in the south, but in the north, the plateau-escarpment morphology has been cut by the active faults of the AAG (Fig. 1B, swath E) such that the divide has jumped westward to the west flank of the graben system. The grabens contain a longitudinal river system with a secondary plateau to the east and the original escarpment on the eastern margin of this plateau. North of 17°S, the topography near the divide reverses in symmetry, showing a steeper slope to the west, suggesting that the divide follows the older, western escarpment. On the west, there are many isolated topographic highlands, with remnant fragments of plateau topography fully contained in the westward drainage basins. Conversely, the eastern side shows a coherent linear escarpment, removed from the water divide and a well-preserved plateau (Fig. 1B, swath F).

Madagascar has experienced a relatively stable climate over the Cenozoic, with only a small northward drift since the Early Paleogene (10, 36, 50). Latitudinal shift has resulted in a decrease in seasonality and an increase in tropical monsoonal rainfall on the eastern side of the island (50). Regionally, precipitation and vegetation patterns are dominated by orographic precipitation on the eastern escarpment and main drainage divide, which acts as a topographic barrier to the easterly Indian monsoon (fig. S1D). Surface temperatures generally decrease from the lowland coastal plains to the central highlands, with the lowest values occurring near the drainage divide location (50, 51).

Analysis of escarpment and water divide migration

To unravel the complex topographic characteristics of the rifted margin landscape, we conducted a detailed analysis of the escarpment morphology, stream network, and water divides of the island. Our approach (see Materials and Methods) focused on (i) identification of active (i.e., the main drainage divide is located at or near the escarpment edge) and relict escarpment segments; (ii) characterization of divide asymmetry and stability [e.g., Gilbert metrics (52)]; (iii) interpretation of river profiles (e.g., identification of knickpoints), combined with fluvial metrics (χ and ksn) (53, 54); (iv) estimation of rates of landscape change, plateau erosion, and horizontal escarpment retreat based on published and new cosmogenic isotope concentrations; and (v) testing of process scenarios through comparison with landscape evolution models (55).

RESULTS

Topography of northern Madagascar

Geomorphic mapping, particularly of normalized channel steepness (ksn) and normalized channel length (χ), shows the existence of isolated remnant escarpments along the western edge of the central plateau (Fig. 2A and fig. S2). These features are discontinuous, characterized by high ksn along their flanks and high χ at their top, reflecting a high degree of sinuosity. They typically coincide with water divides between western catchments and are mostly found far from the main water divide. These remnant escarpments are characterized by short and small river basins, with slope-break knickpoints rimming the flanks. At these locations, normalized distance (χ) values at the channel heads of the small basins draining the local highs are typically larger than those of the surrounding rivers that drain from the main divide (Fig. 2B). In contrast, the eastern margin of Madagascar features a well-preserved plateau and slope-break knickpoints that are found between 500 and 1000 m in elevation and exhibits lower escarpment sinuosity (Fig. 2, A and B, and fig. S2). The plateau-escarpment transition is marked by the location of the slope-break knickpoints, with abrupt variation in ksn values along the stream network (Fig. 2A).

The plateau surface is locally disrupted by the faults of the AAG between 19°S and 17°S and again in the AB north of 15°S, where active faults have led to a series of normal fault–bounded basins. In these regions, the main drainage divide lies on the western uplifted flank of the graben shoulders with fluvial metrics, indicating contrasting patterns of divide migration. The cross-divide channel steepness values suggest a short-term eastward migration of the main drainage divide, in contrast to the long-term westward migration direction inferred by the χ metric [red and black arrows in Fig. 2 (A to D)]. In this sector, the regional divide lies near to the more landward western remnant escarpments, where rivers drain a small portion of the preserved central plateau. In some cases, the main divide even coincides with a steep, westward-facing escarpment (Fig. 2D). In contrast, the eastern margin exhibits knickpoint-type rivers, with the main divide located 25 to 60 km from the eastern escarpment-plateau edge.

Drainage divide stability and remnant escarpment edges

On a regional scale, Madagascar exhibits strongly east-west asymmetric topography, with a gently sloping western flank and a steep eastern escarpment, as well as a high contrast in χ values across the divide (Fig. 3A). Southern Madagascar (south of 20°S) features divide-type rivers on the eastern great escarpment (Fig. 1B, swath C) with an inferred westward migration of the drainage divide, as indicated by normalized distance (χ) and Gilbert metrics, although the latter suggest increasing stability over shorter timescales (Fig. 3, A to C). In the area of the Mananara River capture (MC), a large offset between the drainage divide position and the eastern escarpment-plateau edge is observed (25). Here, the cross-divide contrast in χ, relief, elevation, and slope suggests a more stable yet still westward migration of the main divide, although the river profiles and divide contrast in χ are affected by the reversal of the Mananara trunk and transient incision of its tributaries (Fig. 3C and table S3). The western side features remnant fragments of plateau topography (Fig. 3A).

In central Madagascar (20°S), in the region of Ankaratra Volcano (AV), the main drainage divide is located at the top of the volcanic edifice, which is up to 80 km west of the eastern escarpment edge (Figs. 1B, swath D, and 3B).

Northernmost Madagascar (north of 20°S), topography is more symmetric, with a decrease in across-divide contrast in χ values. In this region, the main drainage divide is located far from the eastern escarpment edge, with minimum and maximum distances of 25 and 100 km, respectively, and is closer to the western remnant escarpment edge, reversing the pattern observed in southern Madagascar (Fig. 3, A and B). Here, the cross-divide contrast in topographic metrics indicates a stable to eastward migration of the divide, with a small cross-divide χ contrast, indicating westward migration (Fig. 3C and table S3).

The ^10^Be-derived rates of escarpment retreat increase from south to north along the eastern escarpment of Madagascar, with minimum values of 170 m/Myr in the southernmost sector and maximum values up to 3800 m/Myr in the AV and AGG regions and ranging between 1000 and 2500 m/Myr in the northern region (Fig. 4 and fig. S3).

A similar distribution is observed for species richness along the eastern escarpment, with minimum values of up to 1250 in the southern sector and higher values of up to 2300 in the central and northern areas (Fig. 4). Basin-averaged annual precipitation rates are evenly distributed along the eastern escarpment, ranging from 1200 to 1600 mm/year (Fig. 4 and table S1).

Landscape evolution model

We investigated the fluvial network evolution of the Madagascar landscape using the landscape evolution model Divide and Capture (DAC) (55), simulating a scenario with two rifting phases of a continental block with two escarpments (see Materials and Methods). The effects of a rifting event in our model are kinematically represented by dropping the base level to sea level of the rifted boundary and applying a flexural deflection within the model domain. The flexural isostatic response to the synthetic topography is to kinematically represent the unloading as a result of crustal thinning. Crustal thinning is regionally compensated using an elastic-plate model, resulting in flexural uplift of the rift shoulder (56, 57). The drop in local base level, combined with flexural uplift in response to crustal thinning, localizes the water divide at the edge of the rift zone, forming an asymmetric erosional escarpment that retreats away from the rifted margin (Fig. 5 and movie S1) (58–60). We initiate a second rift at our model domain’s “eastern” boundary with opposing polarity from the first rift, which is at the “western” boundary to simulate a continental fragment like Madagascar (Fig. 5 and fig. S4). The initial topography in our model is characterized by a 1000-m-high plateau with little internal relief, and the main water divide located near the first rifting boundary. Other model parameters and boundary conditions are given in table S4 and section S1.

Following the first rifting at the “western” margin of our model domain, the main drainage divide has an initial elevation of 1400 m at the escarpment edge (Fig. 5A), presenting relief of 400 m with respect to the adjacent continental interior. The escarpment on the western margin displays low sinuosity and an average rate of retreat of 0.93 km/Myr (fig. S5). The main drainage divide is retreating with the escarpment-plateau edge (Fig. 5 and fig. S5B).

In the second phase, flexural compensation and rift flank uplift tilt the topography and the main divide shifts from the western to eastern escarpment (Fig. 5B). At this stage, the plateau is tilted to the west and the older margin shifts from divide-type to knickpoint-type rivers. The eastern margin displays a linear escarpment drained by divide-type rivers. The sudden shift of the main divide and the formation of knickpoint-type rivers on the western margin lead to transient river entrenchment and progressive dissection of the plateau (Fig. 5, B and C). The river knickpoints retreat rapidly but at different rates depending on the varying distribution of drainage area among the rivers at the time of the divide jump. In contrast, the eastern margin of the model displays divide-type rivers, maintaining low sinuosity with evenly distributed rates of erosion and escarpment retreat over the remaining duration of the model. The slope-break knickpoints extracted from the model synthetic topography and from western Madagascar exhibit similar shapes in their longitudinal profiles and show a wide range of χ values from 5 to 25 m (Fig. 6).

DISCUSSION

Landscape evolution and divide jump

Treated in isolation, the inland retreat of topographic escarpments dominates the landscape evolution of passive margins (60) with rates reaching hundreds of meters per million years (25, 61). In Madagascar, two escarpments can be identified, exhibiting pronounced morphological differences. These differences can be attributed primarily to the sequence and timing of rift formation but within one flexural wavelength of each other. The flexural tilting of Madagascar, first to the east and then to the west, resulted in a displacement of the main water divide from the older escarpment to the younger one. The more recent westward tilt reflects crustal thinning from continental rifting, combined with an uplift gradient possibly enhanced by the deposition of sediments along the western passive margin (22, 49, 62).

The resulting shift in the drainage divide has left a clear imprint on the landscape, expressed in the dissected and discontinuous nature of the western escarpment. The divide jump and landscape response are illustrated well by our numerical model (Fig. 5). In the simulation, tilting and rapid shift in the main water divide reorganize the plateau rivers, redirecting them from an initial configuration flowing away from the western escarpment to the one crossing it (Fig. 5B and fig. S4). The west-flowing rivers develop a knickpoint morphology, steepening as they incise across the former escarpment (Fig. 6, A and B). However, not all rivers have the same drainage area; the rapid reorganization of the plateau rivers results in a wide variance in basin size (Fig. 5B), including many small basins that still have their headwaters in the western escarpment. Large basins incise the plateau rapidly, with a fast upstream knickpoint propagation, whereas smaller basins, particularly those still headed against the former escarpment, incise more slowly, continuing its gradual degradation (Figs. 5C and 6). The contrast in incision rate and knickpoint propagation rate leads to plateau dissection, an increase in the sinuosity of the western escarpment, and the formation of isolated remnants behind the erosional front defined by the knickpoints of the large basins (21, 22). Alternatively, the knickpoint-type river profiles can result in response to an increase in uplift rate relative to the base level but with a constant drainage basin area distribution. The alternative models can be differentiated on the basis of the χ normalized river profiles. In both models, the celerity of knickpoints depends on the drainage area and water discharge, both of which vary spatially (across the catchments) (63, 64). However, in the case of changing the uplift rate, all knickpoints initiate at the base level and travel at the same rate in the χ space so that knickpoints should align on χ plots (fig. S7). If, instead, the knickpoints are a consequence of a divide jump, they initiate at the plateau edge but different values of χ (Fig. 6B). Observations from western Madagascar show that knickpoints are highly dispersed across values of χ (Fig. 6), which is consistent with the model of a divide jump but not with a model of time-dependent uplift rate. Their distribution, along with topography and χ patterns, mimics what is observed in the northwestern Madagascar landscape (Fig. 6; compare Figs. 2 and 5D; compare Fig. 2 and fig. S6), suggesting that their formation is connected to the process of plateau degradation following the rift along eastern Madagascar.

The landscape and morphological features of the model explain other geomorphic observations (Fig. 7). Specifically, Madagascar’s western margin is characterized by a discontinuous, dissected plateau edge (Figs. 1B and 2), while the eastern escarpment is straight and mostly corresponds to the main water divide; some important exceptions are discussed later. The inferred increase in erosion rates in response to divide migration is supported by changes in the sediment supply along Madagascar’s western margin, where a thick Albian-Turonian prograding sequence was deposited in the Majunga Basin, following the second rifting event (49).

In northern Madagascar, between about 17°S and 14°S, the western escarpment corresponds to the main divide. Although it is possible that the divide did not jump in this region and has always been associated with the western escarpment, it is more likely that in the largest basins, the knickpoints have reached the divide. Although the escarpment is at the main divide, there are many topographic remnants to the west, supporting the idea that the escarpment has not propagated as a uniform feature. In addition, had there been a larger drainage area to the east-flowing rivers, we would expect a more sinuous eastern escarpment.

Impact of Late Cenozoic geodynamic activity

Given that the last rifting event was between Madagascar and India, the simple model prediction is that the eastern escarpment should form a shoulder-type margin, with the main water divide located near its edge. In southern Madagascar, this is largely true if one accepts some complexity because of the existence of a thick weathering layer on the high plateau (25). The one major exception is the hypothesized capture of the Mananara River from west-flowing to east-flowing, and the formation of a major knickpoint has not yet propagated to the main divide (25).

In central Madagascar, there is a large excursion of the divide from the escarpment to the top of the Ankaratra volcanic field (Fig. 3). The AV was emplaced through multiple events starting around 28 Myr ago (12, 65), with extensive extrusive volcanism. We interpret the resulting volcanic edifice as having formed in formerly west-flowing river basins, creating a topographic barrier that blocked the west-directed rivers, redirecting them to the east, as suggested by large river deviation of plateau rivers (Figs. 1B, swath D, and 7C).

In northern Madagascar, the eastern regions of the high plateau are affected by crustal extension, with surface faulting initiated around 30 Myr ago with the formation of the AAG (62). Extension continues to the north into the AB, although it is less well documented in either the AB or the region between the AAG and the AB (47). The formation of these morphological depressions redirected plateau rivers from west-flowing to longitudinal, flowing into the grabens and parallel to the main divide before exiting to the east. This constitutes a shift in the water divide from the top of the eastern escarpment to the westmost fault footwall block (Fig. 7C). In the northernmost sector, the influence of Cenozoic volcanism and localized uplift (27, 45), possibly linked to magmatic intrusion and dynamic topography, may have affected the drainage divide position by anchoring it near the region’s highest and most pronounced peaks (Fig. 1B). While the mantle dynamic support is a key factor in sustaining high topography in the northern tip of the island (26, 27, 45, 66), stratigraphic evidence from the syn- and postrift sequences and Eocene sediments suggests a prolonged, steady Cenozoic uplift across much of Madagascar without a substantial spatial gradient (27, 49). Previous studies (26, 27) interpreted many of the pronounced river knickpoints as geomorphic markers of this dynamic uplift but did not consider potential shifts in drainage divides, which can also create systematic knickpoints (Figs. 6 and 7). Although dynamic topography likely increased the average elevation by a few hundred meters during Late Cenozoic, its long-wavelength uplift is unlikely to have triggered changes in drainage basin configuration or be the cause of all the disequilibrium river profiles, as discussed in the previous section (Fig. 6).

Overall, these Cenozoic changes in flow direction resulted in a high degree of geomorphic disequilibrium, as indicated by knickpoint-type rivers (Figs. 1G and 7C) and the contrast in short- and long-term divide stability metrics across the impacted divides (Fig. 3C). Further evidence is provided by the perturbed χ profiles along the constant erodibility geomorphic feature of the eastern escarpment, which show a systematic shift in channel profiles toward lower χ values northward, consistent with a gain in upstream drainage area (fig. S8) (54, 67). In northern Madagascar, the Cenozoic water divide shift led to (i) a reduction in the across-divide χ contrast by decreasing the difference in basin areas between the eastern and western margin catchments and (ii) an increase in the across-divide contrast in Gilbert metrics, as the main divide shifted closer to the western remnant escarpments, which feature a sharp and steep landscape compared to the low-relief plateau located to its east (Figs. 2D and 3C).

Our work underscores the critical role of drainage divide migration in shaping Madagascar’s topography through escarpment migration but also through discrete jumps in position in response to continental tilting, volcanism, and surface faulting. Shifts in the water divide control the relative erosion rates, the changes in planform river networks, and patterns of topographic degradation.

Divide migration controls on surface processes and biodiversity

An important consequence of landscape transience in Madagascar is the control that transient geomorphic processes appear to have on the spatial patterns of biodiversity. The migration rate of regional drainage divides and the downstream cascade of landscape change are arguably primary drivers of species richness at a mountain scale (2, 18), and our landscape model for Madagascar shows marked changes in drainage basin configuration. The transient changes in habitat connectivity lead to population isolation and increased speciation rates (17). Escarpments form in response to isostatic rebound, but once established, their retreat becomes an episodic and transient process through plateau area capture. This mechanism may enhance biodiversity, as observed by the relationship between plant diversity and the geomorphic process along the eastern escarpment (Fig. 4). We can use our new data and analysis to extend the study of ref. (17) along the escarpment with quantitative estimates of the escarpment retreat rate. Our analysis indicates pronunced along-strike variations in morphology and inferred retreat rates, with a positive correlation between plateau area and escarpment retreat rates (fig. S3B), χ-perturbed profiles (fig. S8), and up to a fourfold increase in average retreat rates from south to north (Fig. 4). Plant species richness was shown to be high along the eastern escarpment (17), but with our new data, we find that even the along-escarpment variations in surface process rates correlate with biodiversity (Fig. 4 and fig. S1B; Pearson coefficient of 0.67). This relationship could be explained by (i) faster rates of escarpment retreat toward the north, which triggered a higher degree of habitat disruption and contributed to the observed spatial pattern of diversity, and (ii) jumps in the main water divide in response to volcanic and tectonic processes in the north, which may have enhanced habitat fragmentation and reconnection, increasing landscape transience and acting as a speciation pump. The timing of diversification is difficult to estimate, but there is some evidence for an increase in diversification rate starting at ~40 to 30 Ma (17, 68), coinciding with the time of Cenozoic geodynamic processes.

Other taxa in Madagascar [lemurs, amphibians and reptiles (33, 34, 69)] also show a latitudinal segregation along the eastern escarpment, suggesting that enhanced diversification is not limited to plants. The positive relationship between geomorphic disequilibrium and biodiversity has been documented in other rifted margins where major escarpments are found, such as the Western Ghats of India, Brazil, Australia, and South Africa (70, 71), highlighting the importance of the process of escarpment retreat in controlling rates of biological diversification in passive continental margins. Tectonic processes in other settings, for example, in the western United States, have also been related to diversification rates, presumably through changes in landscape characteristics and habitat heterogeneity (4).

MATERIALS AND METHODS

Stream profiles and topographic and escarpment analysis

To investigate the topography, extract the river network, and map the main topographic features, we used a 90-m HydroSHEDS digital elevation model (DEM) (72) and 30-m Shuttle Radar Topography Mission DEM (http://srtm.csi.cgiar.org). Topographic and morphometric maps and escarpment-plateau edge mapping were obtained in ArcGIS (e.g., Spatial Analyst Toolbox). The river network, swath profiles, and drainage divide stability were analyzed using MATLAB software packages in TopoToolbox (73) and additional MATLAB functions (52, 74). Swath profiles sample the topography, presenting the mean, minimum, and maximum elevations along a 20-km-wide transect.

To evaluate divide stability, we used a combination of different topographic metrics [e.g., Gilbert metrics (52)] and fluvial metrics [χ and ksn (53, 54)]. Gilbert metrics and ksn provide information about the short-term horizontal motion of drainage divides (52), whereas χ provides a measure of the long-term stability of the drainage basins. In particular, ksn is the channel steepness normalized for the downstream increase in drainage area (53); χ is an integral function along the channel network, derived by the stream-power model (75, 76), assuming steady-state condition and spatially invariant uplift and erodibility (54, 77)z(x)=z(xb)+(UKA0m)1n∫xbxA0A(x)mndx=z(xb)+ksA0−(mn)χ(1)χ=∫xbxA0A(x)mndx(2)where z(xb) is the elevation at the river network’s base level at x = x0; A is the upstream drainage area; U is the rock uplift rate; K is the erodibility coefficient related to bedrock lithology, climate, and sediment load; A0 is an arbitrary scaling area; and m and n are empirical constants that depend on basin hydrology, channel geometry, and erosion physics (75). ks is the channel steepness index, which can be normalized (ksn) assuming a reference value of concavity index (m/n = 0.45 in this study), allowing a comparison of catchments with varying area and slope. Normalized distance (χ) provides a metric for (dis)equilibrium in river basins and serves as a transformation of river longitudinal profiles (χ plot). The χ plot serves as an effective tool for visualizing ksn variation along stream segments, revealing possible knickpoints, which can be used as geomorphic markers. The χ map and χ values across drainage divides allow to identify patterns of drainage reorganization at multiple scales. As the catchment gaining area experiences an increase in average erosion rate and χ value at the divide, the area loss will show a decrease in erosional power and χ (54). The ksn and χ were extracted from the DEM, with a critical drainage area of 1 km^2^ considered for channel head initiation.

Knickpoints were manually identified from a χ-elevation plot using MATLAB software packages in TopoToolbox (fig. S2) (78–80). Minor and lithological knickpoints associated with heterogeneities usually do not show a sharp break in the χ space, in contrast to slope-break knickpoints, which separate stream segments with different ksn values (78). Profiles with a spatially coherent slope-break knickpoints were used to map the transition between steep escarpment flanks and adjacent the low-relief plateau (fig. S2).

Escarpment mapping is performed by integrating the distribution of slope-break knickpoints with slope, ksn, and available geological maps. Escarpment edges were manually delineated in ArcGIS using these combined datasets. Specifically, hillslope gradient values were used to trace escarpment edges along slopes, while ksn values and knickpoint locations helped delineate edges along valleys. This approach allowed us to map the locations of active and relict escarpments and to identify isolated, detached remnants. These mapped escarpment edges are presented in Fig. 2.

^10^Be-derived erosion rate

We compiled a database of basin-averaged erosion rates derived from ^10^Be cosmogenic nuclides collected from the eastern escarpment and plateau of Madagascar. We selected 23 catchments draining the great eastern escarpment from published data (16, 81, 82). In addition, we provide four new samples for the northernmost sector of the eastern escarpment to complement the existing data along the eastern margin (table S1). The sampling locations are located inland to avoid anthropogenic activities, and the upstream catchments primarily drain the steep escarpment and parts of the plateau upstream of the slope-break knickpoints, consistent with the sampling strategies in (16, 81). The ^10^Be concentrations were extracted following the protocol in (83) including (i) sieving and extraction of 250- to 500-μm fraction; (ii) magnetic separation and leaching with a HCl and H2SiF6 solution to remove all remaining minerals except quartz; (iii) decontamination from atmospheric ^10^Be in the quartz and total dissolution of the grains in HF; (iv) extraction of the final amount of Be using cation and anion exchange chromatography, which was used to eliminate iron, aluminum, manganese, and other elements; and (v) BeOH that was precipitated and transformed into BeO at 1000°C. The clean quartz was extracted at the Cosmogenic Isotope lab at the Geological Institute at ETH following standard procedures (84). ^10^Be/^9^Be ratios were measured at the ETH AMS facility on the MILEA system (85) and normalized to the ETH in-house AMS standards S2010N (^10^Be/^9^Be = 3.3 × 10^12^), which were calibrated to the ICN 01-5-1 standard (86). All the ^10^Be/^9^Be ratios were corrected for processing blanks.

Escarpment retreat rate calculation

Basin-averaged cosmogenic nuclide concentrations can be converted into horizontal retreat rates following the methodology in (61). The volume of eroded sediments can be generalized as the erosional flux of rock relative to the basin surface (16, 61). This approach is effective when studying highly heterogeneous catchments characterized by low-relief and very steep topography, such as a topographic escarpment. In these settings, erosion rates are highly variable within the catchment, creating lateral rock flux relative to the surface from erosion over long-term timescales. Concentrations of ^10^Be are converted to a conventional basin-averaged erosion rate, e, using the vertically projected area, Av, or to horizontal retreat rates (v) using the horizontal projection of the basin area (Ah) (61)M=∬SΛP0(x,y)dxdyρC=eAv=vAh(3)where P0(x,y) (atoms g^−1^ year^−1^) is the production rate of the cosmogenic nuclide at the surface at any given geographic location (x,y), S is the basin surface, ρ (g/cm^3^) is the density of the rock, and Λ (g/cm^2^) is the free path absorption length. The spatially integrated volume flux (M), expressed as the volume of rock per unit time, is the sum of horizontal and vertical components of erosion, calculated over their respective projected areas. We can partition the total flux into its two components to isolate the horizontal retreat rate of the steep escarpment, given a known vertical erosion rate of the plateau. The corrected horizontal retreat rate (vc) is then calculated using the average plateau erosion (ep) along with the vertically projected area (Ap) and a horizontally projected area (Ar) (table S1) (16)M=epAp+vcAr(4)

The weighted average plateau erosion of 7.3 ± 0.5 m/Myr was calculated by using the catchments draining the plateau areas upstream of slope-break knickpoints (table S2 and fig. S3A). In northern Madagascar, four horizontal rates have been modeled using the best fit between the drainage area above escarpment and ^10^Be-derived retreat rates (fig. S3B).

Last, we extracted basin-averaged species richness (17) for the selected catchments draining the east escarpment to compare them with ^10^Be-derived retreat rates (table S1).

Landscape evolution model

Madagascar’s topography results from two main phases of rifting, followed by Late Cenozoic tectonic and volcanic activities. In (i) the first rifting event occurred along the western margin with Africa during Middle Jurassic and (ii) the second rifting took place along the eastern margin in the Early Cretaceous, separating Madagascar from India and the Seychelles. To test the topographic and surface responses to these phases of rifting, we simulate an erosional low-relief plateau surface and two phases of rifting in a landscape evolution model. We used the landscape evolution model DAC (55) modified to incorporate additional rifting-induced uplift on a preexisting low-relief plateau (25). In DAC, the incision and erosion follow the stream-power law, and upstream from a threshold drainage area, processes are defined by a hillslope diffusion process (55). We set uniform substrate erodibility, n, and m of 5 × 10^6^ m/year, 1, and 0.4, respectively (see table S4 and section S1). The rivers are free to drain to all boundaries. The background isostatic uplift rate is set to 10 m/Myr, consistent with the background erosion and uplift rates of plateau and coastal areas in Madagascar (16, 45). A flexural deflection of the continental lithosphere associated with crustal thinning and mechanical unloading of the lithosphere is applied, promoting the formation of a flexurally uplifted rift flank (56, 57, 87, 88). The major load at a rifted margin is the formation of a basin (or displacement of the crust to form a basin), which leads to the permanent uplift of the rift margin and initiates the formation of a water divide at the top of the newly formed escarpment (59, 89). We modeled the crustal thinning as a semi-infinite, negative load, which was calculated by integrating the solution for a line load over the half-space. The deflection (wj) to a line load (i) of hi is (90, 91)wj=hiλ2Ke−λa(cosλa+sinλa)(5)andλ=(ρm−ρa)g4D14(6)where λ is the inverse flexural parameter, D is the flexural rigidity and depends on Young’s modulus and the effective elastic thickness (57, 87, 92). ρm and ρa are the densities of the mantle and air, respectively; g is the average gravity; and K is the hydrostatic restoring force, given by (ρm − ρa)g (see table S5 for details). Rather than calculate a load magnitude, which varies across a margin, we used a uniform load that produced a maximum rift shoulder uplift of a specified magnitude. For these models, we prescribed a shoulder uplift of ~400 m to impose several hundreds of meters of shoulder elevation with respect to the adjacent plateau (60, 93), as observed in passive margin topography worldwide (5, 60, 87, 94). To complete the initial prerift topography, we prescribed an initial elevated surface of 1000 m following estimates from the pre-Gondwana breakup surface (fig. S4) (29, 93). The flexural deflection of the lithosphere as a result of simulated crustal thinning was calculated using a two-dimensional model in Matlab.