For the past few decades, plate–boundary dynamics have been, to a first order, well understood. This is not the case for intraplate regions, where short-term (103–105 years) regional strain rates are low and the responsible dynamical processes are still in debate (e.g., Calais et al., 2010, 2016; Vernant et al., 2013; Tarayoun et al., 2017). Intraplate deformations evidenced by seismic activity are sometimes explained by a transient phenomenon (e.g., glacial isostatic rebound or hydrological loading). However, to explain the persistence through time of intraplate deformation and explain the high finite deformation that we can observe in the topography in many parts of the world as for instance the Ural Mountains in Russia, the Blue Mountains in Australia, or the French Massif Central, one needs to invoke continuous processes at the geological timescale.

Located in the southwestern Eurasian plate (Fig. 1), the French Massif Central is an ideal case to study these processes because a high-resolution digital elevation model (DEM) encompasses the whole region, and widespread karstic areas are present along its southern and western edges, allowing the possibility to quantification of the landscape evolution rates thanks to terrestrial cosmogenic nuclides (TCN) burial ages. The region is characterized by a mean elevation of 1000 m with summits higher than 1500 m. Such a topography is likely to be the result of recent active uplift, and as the Cévennes mountains experiences an exceptionally high mean annual rainfall (the highest peak, Mont Aigoual, records the highest mean annual rainfall in France of 4015 mm) it raises the question of a possible link between erosion and uplift as previously proposed for the Alps (Champagnac et al., 2007; Vernant et al., 2013; Nocquet et al., 2016). This region currently undergoes a small but discernible deformation, but no significant quantification can be deduced due to the scarcity in seismicity (Manchuel et al., 2018). In addition, GPS velocities are below the uncertainty threshold of GPS analyses (Nocquet and Calais, 2003; Nguyen et al., 2016).

In this study we focus on the Cévennes mountains and the Grands Causses (Fig. 1) area, where cave systems with trapped sediments are known over a widespread altitude range. South and west of the crystalline Cévennes mountains, prominent limestone plateaus, named Grands Causses, rise to 1000 m and are dissected by few canyons that are several hundreds of meter deep. The initiation of incision, its duration, and the geomorphic processes leading to the present-day landscape remain poorly constrained. A better understanding of the processes responsible for this singular landscape would bring valuable information on intraplate dynamics, especially where large relief exists.

The oldest rock units in the study area were formed during the Variscan orogeny (late Paleozoic, ∼300 Ma; Brichau et al., 2007) and constitute the crystalline basement of the Cévennes. Between 200 and 40 Ma (Mesozoic and middle Cenozoic), the region was mainly covered by the sea, ensuring the development of an important detrital and carbonate sedimentary cover, which can reach several kilometers of thickness in some locations (Sanchis and Séranne, 2000; Barbarand et al., 2001). During the Mesozoic Era, an episode of regional uplift and subsequent erosion and alteration (called the Durancian event) is proposed for the origin of the flat, highly elevated surface that persists today across the landscape (Bruxelles, 2001; Husson, 2014).

The area is affected by the major NE–SW trending Cévennes Fault system, a lithospheric-scale fault, inherited from the Variscan orogen. This fault system was reactivated several times (e.g., as a strike-slip fault during the Pyrenean orogen or as a normal fault during the Oligocene extension). During the Pyrenean orogeny, between 85 to 25 Ma (Tricart, 1984; Sibuet et al., 2004), several fault and fold systems affected the geological formations south of the Cévennes Fault, while very few deformations occurred farther north within the Cévennes and Grands Causses areas (Arthaud and Laurent, 1995). Finally, the Oligocene extension (∼30 Ma) led to the counterclockwise rotation of the Corso-Sardinian block and the opening of the Gulf of Lion, reactivating some of the older compressive structures as normal faults. The main drainage divide between the Atlantic Ocean and the Mediterranean Sea is located in our study area and is inherited from this extensional episode (Séranne et al., 1995; Sanchis and Séranne, 2000).

Afterwards, during the Pliocene–Quaternary period, intense volcanic activity affected the region, from the Massif Central to the Mediterranean shoreline. This activity is characterized by several volcanic events that are well constrained in age (Dautria et al., 2010). The last eruption occurred in the Chaîne des Puys during the Holocene (i.e., the past 10 kyr; Nehlig et al., 2003; Miallier et al., 2004). Some authors proposed that this activity was related to a hotspot underneath the Massif Central that led to an observed positive heat-flow anomaly and a possible regional Pliocene–Quaternary uplift (Granet et al., 1995; Barruol and Granet, 2002). Geological mapping at a different scale can be found at: http://infoterre.brgm.fr/ (last access: 21 February 2020).

Despite this well-described overall geological evolution, the onset of active incision that has shaped the deep valleys and canyons (e.g., Tarn or Vis river, Fig. 1) across the plateaus and the mechanisms that controlled this incision are still debated. One hypothesis proposes that canyon formation was driven by the Messinian salinity crisis with a drop of more than 1000 m in Mediterranean Sea level (Moccochain, 2007). This, however, would not explain the fact that the Atlantic watersheds show similar incision. Other studies have suggested that the incision is controlled by the collapse of cave galleries that lead to fast canyon formation mostly during the late Quaternary, thus placing the onset of canyon formation at only a few hundreds of thousands of years ago (Corbel, 1954). More recently, it has been proposed (based on relative dating techniques and sedimentary evidence) that incision during the Quaternary was negligible (i.e., less than a few tens of meters) and that the regional morphological structures seen today occurred around 10 Ma (Séranne et al., 2002; Camus, 2003).

In this paper, we provide new quantitative constraints on both the timing of incision and the rate of river downcutting in the central part of the Cévennes and the Grands Causses that has resulted in the large relief between plateau and channel bed.

We employ two methods to infer allochthonous karstic infilling age and associated river downcutting. First, we use quartz cobbles to measure the concentration of cosmogenic 10Be and 26Al isotopes. The 10Be/26Al ratios provide burial ages of these karstic infillings. Second, paleomagnetic analyses of clay deposits are used to obtain paleo-polarities. In both cases, vertical profiles among tiered caves systems and horizontal galleries could provide local incision rate information. By analyzing a high-resolution DEM (5 m), we show that the region is affected by a southeastward regional tilting. Our results allow quantification of the role of the Pliocene–Quaternary incision on the Cévennes landscape evolution and constraint of the numerical modeling from which we derive the regional uplift rates and tilt of geomorphological markers.

If incision is initiated by uplift centered on the north of the area where elevations are at a maximum, it will lead to tilting of fossilized topographic markers as strath terraces. Our research approach provides an opportunity to discriminate between three possible explanations for the current terrain morphology. The first is based on old uplift and old incision (Fig. 2a). In this case, apparent incision rates would be very low. For instance, if incision commenced 10 Ma (Sérrane et al., 2002), we would find surface tilting, but cosmogenic burial dating with 10Be/26Al, which cannot discern ages older than ∼5 Ma due to excessive decay of 26Al, would not be possible. The second possibility (Fig. 2b) is that the uplift is old, and incision consequently follows but with a time lag. Here the incision rate would be rather fast, but no tilting is expected for the river-related markers because no differential uplift occurs after their formation. Finally, the third possibility (Fig. 2c) is that uplift and incision are concurrent and recent (i.e., within the timescale of cosmogenic burial dating), and thus we would expect burial ages <5 Myr, relatively high incision rates, and tilting of morphological markers. These different proposals for the temporal evolution of the region will be compared using numerical modeling.

Both geomorphological and geochronological evidence suggest a Pliocene–Quaternary uplift of the Cévennes area. The origin of such an uplift could be associated with several processes: erosion-induced isostatic rebound, dynamic topography due to mantle convection, thermal isostasy, residual flexural response due to the Gulf of Lion formation, etc. For the Alps and Pyrenees mountains, isostatic adjustment due to erosion and glacial unloading has been recently quantified (Champagnac et al., 2007; Vernant et al., 2013; Genti, 2015; Chéry et al., 2016). Because the erosion rates measured in the Cévennes are similar to those of the eastern Pyrenees (Calvet et al., 2015; Sartégou et al., 2018a), we investigate by numerical modeling how an erosion-induced isostatic rebound could impact the southern Massif Central morphology and deformation.

We define a representative cross section parallel to the main topographic gradient (i.e., NNW–SSE) and close to the field investigation areas (Fig. 11). We study the lithospheric elastic response to erosion with the 2-D finite element model ADELI (Hassani and Chery, 1996; Chéry et al., 2016). The model is composed of a plate accounting for the elasticity of both the crust and uppermost mantle. Although the lithosphere rigidity of the European plate in southern Massif Central is not precisely known, vertical gradient temperatures provided by borehole measurements are consistent with heat flow values ranging from 60 to 70 mWm2 (Lucazeau and Vasseur, 1989). Therefore, we investigate plate thickness ranging from 10 to 50 km as done by Stewart and Watts (1997) for studying the vertical motion of the alpine forelands.

We choose values for Young's and Poisson parameters of 1011 Pa and 0.25, respectively, both commonly used values for lithospheric modeling (e.g., Kooi and Cloething, 1992; Champagnac et al., 2007; Chéry et al., 2001). This leads to long-term rigidity of the lithosphere model ranging from 1021 to 1025 Nm. Since the effect of mantle viscosity on elastic rebound is assumed to be negligible at the timescale of our models (1 to 2 Myr), we neglect the viscoelastic behavior of the mantle. Therefore, the base of the model is supported by an hydrostatic pressure boundary condition balancing the weight of the lithosphere (Fig. 11). Horizontal displacements on vertical sides are set to zero since geodetic measurements show no significant displacements (Nocquet and Calais, 2003; Nguyen et al., 2016). The main parameters controlling our model are the erosion (or sedimentation) triggering isostatic rebound and the elastic thickness.

The erosion profile (Fig. 11) is based on topography, our newly proposed incision rate, and other studies (Olivetti et al., 2016, for onshore denudation and Lofi et al., 2003; Leroux et al., 2014, for offshore sedimentation). This profile is a simplification of the one that can be expected, from Olivetti et al. (2016), and does not aim to precisely match the published data because of the following: (i) the explored time span (∼1 Myr) is not covered by thermochronological data (>10 Myr) or cosmogenic denudation rate (10s–100s kyr); (ii) we base our erosion rate on being linked with local (10s km2) slopes that are higher near the drainage divide. We, by this aim, can invoke any kind of erosion processes (e.g., landslides); and (iii) the model assumes a cylindrical structure, and consequently, high-frequency lateral variations in term or actual denudation rate or proxy (slope, elevation, etc.) must be averaged. Concerning this erosion profile, parametric study (highest erosion rate ranging from 1 to 1000 mMa-1) gives no difference in the interpretation, and for a few percent change in the used erosion rate, the model result (uplift rate) changes by a few percent.

The flexural rigidity controls the intensity and wavelength of the flexural response and ranges from 1021 to 1025 Nm. It can be expressed as a variation in elastic thickness (Te) ranging from 4.4 to 96 km (Fig. 12). We also test a possible Te variation between inland and offshore areas. For the following discussion, we use an elastic thickness of 15 km corresponding to a value of D (flexural rigidity) of 3.75×1023 Nm. In this case, the inland and offshore parts are largely decoupled and the large sedimentation rate in the Gulf of Lion does not induce a flexural response on the Cévennes and Grands Causses areas.

With a maximum erosion rate of 80 mMa-1 (Fig. 11), the models display uplift rates of 50 mMa-1 over more than 100 km. As previously explained, the finite incision is permitted by an equal amount of uplift considering that the incision is not due to regressive erosion.

Every model shows a general uplift. However, the uplift amplitudes are smaller than the expected ones. To obtain the same uplift rate as the incision rates, the applied erosion rate over the model must be increased. However, we assume that the landscape is at equilibrium, so, if the erosion rate is increased, it will be higher than the incision rate leading to the decay of relief over the area. No evidence of such evolution is found over the region, and if further studies need to be done to quantify the actual erosion rate, we mostly think that a second process is acting, inducing the rest of the uplift that cannot be obtained by the erosion-induced isostatic adjustment. Finally, models predict a seaward tilt of the surface at the regional scale (Fig. 13), in agreement with the observed tilting of morphological markers.

We assume that the sediments collected in the karst were deposited per descensum; i.e., we do not know if the galleries existed a long time before or were formed just before the emplacement of the sediments, but the more elevated the sediments are, the older their deposit is. If there is no evidence of an important aggradation episode leading to more a complex evolution as proposed for the Ardèche canyon (Moccochain et al., 2007; Tassy et al., 2013), we point out that small aggradation or null erosion period could, however, be possible. Some processes could explain such a relative stability, e.g., variation in erosion (due to climatic fluctuation) or impact of eustatic variations (in river profile, flexural response, etc.). Such transient variations have been shown for the Alps (Saillard et al., 2014; Rolland et al., 2017) and are proposed as being related to climato-eustatic variations and therefore should last 10 to 100 kyr at most.

Based on our sampling resolution, we cannot evidence such transient periods, and we must use an average base level lowering rate in the karst, which we correlate to the incision of the main rivers. The TCN-based incision rate derived from the Rieutord samples (83±35 mMa-1) is consistent with the one derived from the Garrel samples (U–Th ages: 85.83 mMa-1, according to the sole U–Th exploitable result; Camus, 2003) and from the Garrel–Leicasse combination (paleomagnetic approach: 84 +21/-12 mMa-1).

This mean incision rate of ca. 85 mMa-1 lasting at least 4 Ma, highlights the importance of the Pliocene–Quaternary period in the Cévennes and Grands Causses morphogenesis. Furthermore, the 300 to 400 m of incision precludes a relative base level controlled by a sea-level drop. Indeed, documented sea level variations are less than 100 m (Haq, 1988; Miller et al., 2005). Furthermore, the Hérault River does not show any significant knickpoints or evidence of unsteadiness in its profile as would be expected if the incision was due to eustatic variations. Therefore, we propose that the incision rate of ∼85 mMa-1 is due to a Pliocene–Quaternary uplift of the Cévennes and Grands Causses region.

Other river–valley processes could lead to a local apparent high incision rate, for instance a major landslide or alluvial fan (Ouimet et al., 2008). This hypothesis of an epigenetic formation of the Rieutord is irrelevant because (i) none of the possible causes had been found in the Rieutord canyon and (ii) the consistency of the TCN-based incision rate and the paleomagnetic-based incision rate for two other cave systems. Indeed, the use of two independent approaches and three locations is a good argument in favor of the robustness of our proposed mean 85 mMa-1 incision rate. Yet, using more data, particularly burial dating co-localized with clay samples and adding sampling sites would give a stronger statistical validation. In the Lodève basin (Point 4, Fig. 1), inverted reliefs allow another independent way to quantify minimal incision rate. The K/Ar and paleomagnetic dated basaltic flows spanning from 1 to 2 Myr old that were deposited at the bottom of the former valley (Dautria et al., 2010) are now located at ca. 150 m above the current riverbed, leading to an average incision rate of 77±10 mMa-1, in agreement with karst-inferred incision rates.

Furthermore, preliminary results from canyons on the other side of the Grands Causses (Tarn and Jonte) based on in situ terrestrial cosmogenic dating suggest similar incision rates (Sartégou et al., 2018b) and confirm a regional base level lowering of the Cévennes and Grands Causses region during the Pliocene–Quaternary. This is consistent with the similarities of landscapes and lithologies observed both on the Atlantic and Mediterranean watersheds (e.g., Tarn river).

Once the regional pattern of the Pliocene–Quaternary incision established for the Cévennes–Grands Causses area, the next question is how this river downcutting is related to the regional uplift? First order equilibrium shape and absence of major knick points in the main river profiles preclude the hypothesis of regressive erosion. Hence, back to the three conceptual models presented in part 1 (Fig. 2); we can discard, at first order, the models A (old uplift–old incision) and B (old uplift–recent incision) because the obtained incision rate shows recent incision and surface tilting tends to prove a current uplift. Therefore, the incision rate has to be balanced to the first order by the uplift rate. We add that eustatic variations are too low in magnitude (100–120 m) and cannot explain such total incision (up to 400 m). Furthermore, no obvious evidence of active tectonics is reported for the area, raising the questions about the processes responsible for this regional uplift. Very few denudation rates are reported for our study area (Schaller et al., 2001; Molliex et al., 2016; Olivetti et al., 2016), and converting canyon incision rates into denudation and erosion rates is not straightforward, especially given the large karst developed in the area. Using a first order erosion–sedimentation profile, that follows the main topography gradient direction we have modeled the erosion-induced isostatic rebound. If this process could create between half and two-thirds of the Pliocene–Quaternary uplift, a previously existing topography is needed to trigger erosion, so it cannot explain the onset of the canyon-carving or the full uplift rates. Other, processes have to be explored such as dynamic topography or thermal anomaly beneath the Massif Central; the magmatism responsible for the important increase in volcanic activity since ∼6 Myr (Michon and Merle, 2001; Nehlig et al., 2003) could play a major role, notably in the initiation of Pliocene–Quaternary uplift. Further studies should aim to address the problem of uplift onset, giving more clues concerning the stable continental area, but owing to the data we presently have, discussing such an onset is out of the scope of this paper.

Main results of this study are the following three points:

Mean incision rate of the Cévennes area is 83 +17/-5 mMa-1 during the last 4 Ma.

To the contrary of previous studies that focused on one cave, we have shown that combining karst burial ages and paleomagnetic analysis of clay deposits in several caves over a large elevation range can bring good constraints on incision rates. This multi-cave system approach diminishes the intrinsic limits of the two single methods: low sampling density (and analysis cost) for the TCN ages and a difficulty setting the position of paleomagnetic results. Our estimated paleo-base-level ages are Pliocene–Quaternary (ca. last 4 Ma) and allow the derivation of a mean incision rate of 83 +17/-5 mMa-1 for the Cévennes area. The landscape and especially the river profiles suggest a first-order equilibrium, allowing the consideration of the incision rate as an uplift rate.

We have shown using a geomorphological analysis that at least south of the Cévennes, several surfaces are tilted toward the SSE. This kind of study had been performed before on large structures (Champagnac et al., 2007) or endokarstic markers (Granger and Stock, 2004), but it is the first time that it is performed at such a scale with small markers. Numerical modeling yields the same pattern of SSE dipping, allowing more confidence in the geomorphometric results.

Our multidisciplinary approach brings the first absolute dating of the Cévennes landscapes and suggests that the present-day morphology is partly inherited from the Plio-Quaternary erosion-induced isostatic rebound.

We propose that related erosional isostatic adjustment is of major importance for the understanding of the southern French Massif Central landscape evolution and explains a large part of the uplift.

At a larger scale, we assume that the main conclusion of our study can be extrapolated to the majority of the intraplate orogens. That is to say, once the forces responsible for the initial uplift (e.g., plate tectonics or dynamic topography) fade out, the uplift continues thanks to erosion-induced isostatic adjustment.

An analysis at the scale of the Massif Central is now needed before nailing down our interpretations of the Massif Central dynamics.