Gravity-driven sliding and associated deformations along complex submarine slopes: a laboratory modeling approach based on constraints observed offshore Martinique Island (Lesser Antilles)

Since the 1960s, numerous field studies have described gravity-driven sliding processes occurring in sedimentary basins and continental margins (Wise, 1963; Cloos, 1968), as well as in mountain ranges, where sliding results in the formation of thrust faults (Hudlestone, 1976, 1977, 1980; Talbot, 1979, 1981; Kligfield, 1979; Graham, 1981; Merle, 1982). Siddans (1984) has suggested that the moving direction of these thrusts is determined by the surface and basal slopes of the affected sedimentary piles. Burollet (1975), Brun and Choukroune (1983) and Vendeville (1987) were among the first authors to propose that the deformation within passive margins results from gravity-driven sliding that propagates above décollement layers.

Considerable attention has been placed on deformation induced by gravity in sedimentary basins over the past several decades. Based on experimental patterns, Rettger (1935) demonstrated that local sedimentary input into a system may trigger a gravity-driven deformation within the underlying layers. Since this work, other studies have shown that a sedimentary pile may slide and spread, even down low slopes, under the effect of its own weight. These displacements can be accommodated either through translational motions (i.e., pure slide) when the basal décollement layer is submitted to an elevated pore pressure (Hubbert and Rubey, 1959; Hsü and Siegenthaler, 1969; Merle, 1982), or through rotational motions (i.e., slump) when the basal layer is made up of low-strength lithologies (i.e., shales, clays, evaporites; Kehle, 1970; Fletcher and Gay, 1971; Price, 1977; Ramberg, 1977; Déramond, 1979).

Today, it is well-recognized that the deformation induced by gravity in sedimentary basins is characterized by sub horizontal basal décollement layers as well as both extensional and compressional deformation patterns that develop upslope to downslope respectively (Crans et al., 1980; Letouzey et al., 1995). Typical normal faults, horsts and grabens form upslope, regardless of the lithologies within the layers, the density, viscosity and thickness ratios, or extensional velocity values (Vendeville and Jackson, 1992). Meanwhile, fold and thrust systems develop downslope. The intensity of the extensional and compressional deformations also appears to be conditional upon the sediment inputs within the whole system. When the sedimentary inputs stop, the sliding layer and the internal deformation progressively cease.

Many studies have investigated the mechanisms triggering the gravity-driven sliding of an unstable sediment layer along a slope. Nowadays, it is acknowledged that the driving load for sliding increases as sedimentary inputs preserved upslope cause the layer to progressively thicken. Sliding starts once the resistance downslope (i.e., the thickness) cannot support the driving charge upslope any longer. These processes have been largely analyzed through a large number of field examples (e.g., in the Niger Delta: Doust and Omatsola, 1990; Cohen and McClay, 1996; Haack et al., 2000; Corredor et al., 2005; Cobbold et al., 2009; Rouby et al., 2011), analogue modeling experiments (Cobbold and Szatmari, 1991; Koyi, 1996; Szatmari et al., 1996; Ge et al., 1997; McClay et al., 1998, 2003; Gaullier and Vendeville, 2005; Vendeville, 2005; Mourgues et al., 2009) and numerical simulations (e.g., Cohen and Hardy, 1996; Gemmer et al., 2004, 2005; Albertz et al., 2010; Ings and Beaumont, 2010). The role of fluid overpressures at the base of sedimentary layers during compression has also been highlighted (Cobbold et al., 2009).

The main characteristics of the observed submarine landslide occurring offshore Martinique Island (Lesser Antilles arc, Fig. 1) (Le Friant et al., 2015; Brunet et al., 2016) have been used to constrain the experimental device used in the present study and some of the experiments. As usual, for analogue modeling studies, the approach required simplifying the complexity of the natural system by respecting, in the laboratory, the scaling laws for the geometry, kinematics and dynamics of the processes (Hubbert, 1937; Ramberg, 1981). Despite these simplifications, and over a period of time lasting up to several hours and at a manageable size, the experiments conveniently reproduce long and slow natural processes (so slow that they may appear to be frozen at the scale of a human’s lifetime) involving objects of huge sizes. These experiments can be used to (i) test different settings and physical parameters and (ii) better understand the influence of each individual parameter on the behavior of the whole system. Therefore, the results of the present study apply beyond the geological context of the Montagne Pelée and its offshore submarine flank. Other volcanoes from the Lesser Antilles, other insular arcs, as well as intraplate volcanic islands, are concerned by flank collapses and submarine landslides. It is thought that these results apply in any geological context that involves slides of poorly consolidated sediments just above complex submarine, lacustrine, or aerial slopes.

Several studies have used analogue modeling to investigate the evolution of volcanic edifices, and the associated mass-movement processes such as the spreading of volcanoes and debris avalanches resulting from flank-collapse events (Merle and Borgia, 1996; Walter and Troll, 2003; Walter and Amelung, 2006; Oehler et al., 2005; Delcamp et al., 2008, 2012; Byrne et al., 2013; Kervyn et al., 2014). However, to the best of our knowledge, no studies have investigated the offshore consequences of these onshore instability processes, such as the development of large-scale submarine landslides (Watt et al., 2012a, b, 2021; Brunet et al., 2016). Based on our present level of knowledge so far, no analogue modeling studies have investigated submarine landslides triggered by volcano flank-collapse events and their consequences on the deformation of marine sediments. In this framework, this study investigates the mechanisms that govern the submarine landslide propagation induced by massive and sudden sedimentary input (i.e., debris avalanche) over simple and complex submarine slopes.

The Lesser Antilles volcanic arc results from the subduction of the Atlantic plate beneath the Caribbean oceanic plate since approximately 40 Ma ago (Martin-Kaye, 1969; Wadge, 1984; Bouysse et al., 1990; DeMets et al., 2000). Along the active arc, several volcanoes experienced collapsed flanks during their evolution, resulting in mass wasting deposits mostly within the Grenada Basin located to the west of the arc (e.g., Boudon et al., 2007; Le Friant et al., 2009; Lebas et al., 2011; Watt et al., 2012a, b; Crutchley et al., 2013; Trofimovs et al., 2013; Cassidy et al., 2014).

The Montagne Pelée volcano on Martinique Island (Fig. 1a) is one of these. It has been active since roughly 130 ka and underwent three major flank-collapse events affecting the western flank of the volcano (Deplus et al., 2001; Le Friant et al., 2003a, b; Boudon et al., 2005, 2007; Germa et al., 2011). In 2012, during the IODP Expedition 340, the submarine landslide deposits were drilled and as a result, the mass wasting processes that were involved were reinterpreted (Le Friant et al., 2015; Brunet et al., 2016). The submarine mass wasting deposit is 50 km long and 40 km wide and covers an area of 2100 km² (Fig. 1a). First interpreted as debris avalanche deposits (Le Friant et al., 2003a), it was determined that it mostly corresponded to submarine landslide deposits, except upslope where debris avalanche deposits were clearly identified (Fig. 1a). As a result, Brunet et al. (2016) proposed a new depositional model linking onshore processes (i.e., Montagne Pelée flank-collapse events and associated debris avalanches) with submarine instability processes in Martinique. The authors suggest that the first flank-collapse event produced a large debris avalanche flowing down to the Grenada Basin, which weakened and then initiated seafloor-sediment failure, triggering a major submarine landslide downslope. Then, the second and third debris avalanche deposits remobilized the unconsolidated surficial sediments within the proximal part of the submarine landslide deposit.

However, several questions remain and motivate this study: is it possible that a sudden sedimentary input could trigger a submarine landslide? Could a regular input trigger one? If so, in which conditions and how does it propagate? If not, is there some other possible initiating factor? Does the slope’s geometry – and especially slope break occurrences – influence landsliding? To what extent?

Using numerical simulations, Brunet et al. (2017) demonstrated that debris avalanches are slowed down when flowing over a slope break, suggesting that debris avalanche deposits have a limited lateral extent, as shown off Martinique Island (Fig. 1a). Based on bathymetric data, the slope geometry offshore the island shows two slope breaks (SB1 & SB2, Fig. 1b). From the volcano summit to the landslide’s toe, the main slope is: (i) 15° to 10° in the continuity of the aerial volcano flank, (ii) 5° to 2° beyond a first slope break (SB1, located approximately 7.5 km away from the coastline), and (iii) < 2° beyond a second slope break (SB2, approximately 30 km away from the coastline). Seismic reflection and drilling data were used to constrain the submarine landslide morphology, which is characterized by a chaotic unit showing a morphological front at the surface and a seismic reflector interpreted as a décollement layer at the base (Fig. 1c). The landslide thickness reaches a maximum of 450 m in the axial part but can be 110 m in both the distal and lateral parts (Fig. 1c). The estimated volume is roughly 300 km³. Based on the flank collapse structures on land, the estimations of the debris avalanche material uprooted from the volcano during the flank-collapse events vary from 16.7 km³ up to 40 km³ (Le Friant et al., 2003a; Germa et al., 2015). Therefore, this cumulated debris avalanche volume represents only approximately 13% of the total slide volume.

Cores from the IODP Expedition 340 were drilled within both the central and distal parts of the submarine landslide deposit (see Figs. 1a and 1c for the location). Hemipelagic sediments dominate in the slide unit in both cores. They alternate with volcanoclastic turbidites, which may form layers up to 10 m thick, with interspersed debrites (i.e., debris flow deposits) and tephra layers of variable thicknesses. These lithologic alternations result in a multi-layered structure with many contacts that may have acted as décollement layers in the system. In addition, the cores revealed that the relative proportion of clay minerals, likely resulting from the alteration of former ash tephra, is larger in the vicinity of the island than in the Grenada Basin (80% vs. 40%, respectively) and may have contributed to the landslide process (Brunet et al., 2016).

Our goal is to model the sliding of a layer of known thickness along slopes, for which the angle values may vary downslope. The geometrical scaling of the experiments was defined from the example illustrated in Figure 1. An experimental length scale ratio (L0) of 1 cm for 1 km in nature has been chosen (Tab. 1). The slope values and their layouts in the experiments were also calibrated to be representative of the natural example (Tab. 2). In nature, the heterogeneous lithological composition of the submarine landslide and its basal décollement layer are documented by the deep drilling cores from the IODP Expedition 340. First, the landslide is mainly composed of volcanic sandy turbidites (Le Friant et al., 2013, 2015; Brunet et al., 2016), which gives a fragile rheological behavior to the landslide layer, which is commonly modelled by using sand material (Faugère and Brun, 1984; Nalpas et al., 1999; Barrier et al., 2013). The basal décollement layer is characterized by intercalated tephra layers in hemipelagic sediments sequences. The strength contrast between both lithologies induces low shear strength levels acting as preferential weak layers for sliding under significant constraints. This low strength behavior of the basal surfaces is widely modeled in the literature using a ductile material such as silicon putty (Vendeville et al., 1987; Cobbold et al., 1989; Mauduit et al., 1997; Mourgues et al., 2009). Therefore, the heterogeneous nature of both the landslide and basal décollement layer had to be simplified using a two-layer system made of silicone putty and sand.

The modelling techniques used are similar to those usually used for experiments dealing with brittle-ductile systems in the Laboratory of Experimental Tectonics of Geosciences Rennes (Rennes University, France) and which have been described in numerous studies (e.g., Faugère and Brun, 1984; Vendeville et al., 1987; Davy and Cobbold, 1991; Cobbold and Castro, 1999, 2009). Dry Fontainebleau quartz sand (SIBELCO NE34) with Mohr–Coulomb properties was used to model the brittle (i.e., frictional) behavior of the sedimentary rocks. It exhibits a mean grain size of approximately 250 µm, an internal friction angle within the range of 30°–35°, negligible cohesion, and an average density close to 1.5 (Tab. 1; Krantz, 1991; Klinkmüller et al., 2016). The silicone putty (uncolored SGM 36 manufactured by Rhône-Poulenc, France) is used to model the ductile (i.e., viscous) behavior of weak sedimentary rocks. It was chosen because of its density of 0.96 and an almost Newtonian viscosity of about 10⁴ Pa · s under low stresses, at room temperature. As the natural detachment layer is not viscous, the basal silicone layer only represents a technical expedient to enhance sliding that cannot be correctly scaled to nature. Regardless, models provide valuable information regarding the deformation style. In most of the experiments, the sliding brittle layer mostly consists of a 5.10⁻³ m-thick sand layer corresponding to poorly consolidated sediments in nature that deform in a brittle manner under sufficient stress. It overlies a 2.10⁻³ m-thick silicone putty layer simulating a basal décollement layer that allows sliding to occur.

The sedimentary input (i.e., debris avalanche) was represented by dry Fontainebleau sand and therefore, during some of the experiments, the thickness of the sliding layer varied from 5.10⁻³ m up to about 10⁻² m in order to simulate punctual and periodical debris avalanches.

The geometric characteristics (thickness and length of the deposit, slope, sediment inputs, etc.) of the experimental device have been set-up using the interpreted marine geophysical data (Brunet et al., 2016) acquired offshore Martinique (Deplus et al., 2001, 2002; Le Friant et al., 2003a; Fig. 1). Although the area and volume of the slide models are proportional to the real ones, it is important to note that the slide area – and indirectly the volume – are limited by the dimensions of the underlying silicone layer (Figs. 2 and 3). The modeled slide propagation distance is proportionally shorter than the natural one (50 cm in the model, which is equal to 25 km in nature instead of 50 km), but still is within the same order of magnitude. Only the thickness “deposit” may be one order of magnitude higher compared to the real deposit.

In addition, the study aims to investigate the gravity-driven sliding dynamics and deformation propagation over submarine slopes. Each experiment was systematically carried out in dry and underwater conditions, in order to get first-order observations on the potential influence of hydrostatic pressure on sliding.

In total, 20 experiments were carried out (Brunet, 2015); six of them were set-up to work out the experimental device and protocols (not presented in this study). Among the 14 other experiments presented herein, constant parameters were applied (silicone and sand two-layer sizes; the upslope slope-angle value was fixed at 12°, Tab. 2). Some other parameters varied from one experiment to another (number, lengths and angle values of the downslope plate or plates; planed, additional sand input or inputs; dry or aqueous conditions; Tab. 2).

Wall edges measuring 1 cm-high laterally limit the experimental device, which is 75 cm-long and 42 cm-wide, and an additional fixed wall is located downslope (Fig. 2a). The experimental device is composed of rigid basal plates set up at different angle values from the horizontal (Figs. 2b, 2b’ and 2c). For all the experiments, from upslope to downslope, the angle values decrease. Two types of experiments were performed: two plate experiments exhibit a single slope break (labelled SB1 between plates P1 and P3), whereas three plate experiments exhibit two-slopes breaks (labelled SB1 and SB2 between plates P1 and P2, and P2 and P3, respectively; see Tab. 2 for the angle values for each experiment).

Before each experiment, a 0.2 cm-thick and 20 cm-wide silicone putty basal layer was set in the middle of the plates, all along the slopes (Fig. 2). Then the whole surface of the plates, including the silicone putty layer, was covered with sand to reach a total thickness of approximately 0.7 ± 0.2 cm. Thus, only the axial part of the experimental device was covered by the silicone plus sand two-layer. This configuration was chosen in order to minimize potential rigid sidewall effects. Above the silicone, the sand layer was comprised of white sand with an intercalated black sand marker (Figs. 2b, b’ and c) so as to be able to analyze the internal deformation at the end of the experiment via the cross-section views. Black sand was also sprinkled on top of the sand layer, and white sand was used to draw parallel markers on the sand surface to visualize the displacement (e.g., Figs. 3a and 3c). The colored sand had the same physical properties as the white sand.

The three slope models involving additional sand inputs, upslope (over the P1 zone), during the course of the experiments, were constrained to redesign the surficial white makers after each new sand supply. The inputs were set either only once at the start of the experiment (a single sand input of 150 or 450 g) or periodically by adding 75 g of sand every 12 h so as to accumulate six sand inputs (450 g in total) for the whole duration of the experiment (Tab. 2). The quantity of added sand was scaled from estimations of the cumulated debris avalanche volumes generated by the Montagne Pelée volcano (i.e., 40 km³), compared to the submarine slide mass volume (300 km³), representing 13%, with 450 g of sand corresponding to the total volume of debris avalanche deposits. We ran three experimental scenarios: the first scenario, with an initial input of 150 g of sand (G18), and a second scenario, with an initial input of 450 g of sand, were tested to represent the deposition of the three cumulated debris avalanche deposits. The third scenario tested multiple deposition events. The preliminary tests showed that the sliding was not moving anymore after 84 h of the experiment. Next, we decided to distribute the sand input (450 g) periodically throughout this time frame (84 h), which was equal to 75 g of sand every 12 h, representing six inputs during the G15 experiment.

All other conditions being equal, each model was carried out once in a dry condition and once under water (aqueous condition). For the dry conditions, the experimental device was left in the open ambient air. For the aqueous conditions, the device was installed in a tank (110 × 70 × 40 cm) filled with water to cap the sand layer upslope. Therefore, the experiments are labelled GAX or GDX: X refers to the experiment number, and A or D indicates either Aqueous or Dry conditions, respectively. From upslope to downslope, the slope breaks (where appropriate) and the surficial markers were labelled SB1, SB2 and M1, M2, M3, …, M14, respectively (Fig. 3).

A NIKON D90 camera with an AF-S DK NIKKOR 18–55 mm f/3.5–5.6G VR lens and light spots suspended above the model were used to take pictures of the model surface at regular time intervals (15 min). This was done to follow the evolution of the sliding during each experiment.

In practice, each run lasted 84 h, i.e., three and a half days. The length of the runs was determined according to the preliminary experiments showing that beyond that amount of time, for some experiments, the displacements became almost indiscernible and were < 1 mm during the last 12 h. Then, the water was removed from the tank (aqueous conditions only), and the models were entirely covered with new sand that was sprinkled on top along with water (dry conditions) in order to preserve the structures by making serial cross-sections parallel to the slope (direction of the sliding) of the models.

Photographs were used to measure the displacements of each surficial marker with time for each experiment, with an uncertainty of ± 1 mm (Fig. 3). These data were used to estimate the displacement velocities. The final displacement of the first marker (M1) was used to estimate the total shortening across each model (Tab. 3). The last photographs taken (surface and cross-section) were used to determine the distribution of the observed successive, surficial deformation zones and their lengths from upslope to downslope (Tab. 3). They also localize the final position of the observed morphological front given by the axial distance from the top of the upslope plate to the farthest downslope topographic positive anomalies encountered in the compressional zone: C. This distance is referred to as the morphological front distance (Fig. 3 and Tab. 3). The final spatial organization and the characteristics of the faults that developed close to the morphological front for each experiment were analyzed using each axial cross-section (located at the center of the model).

Tables 2 and 3 provide the experimental conditions and the main results for the experiments, respectively. All the experiments underwent spontaneous sliding and deformation of the silicone-sand layer. Given the experimental set-up, sliding is located in-between two strike-slip faults (Fig. 3a) that initiated as soon as the experiments were set up and which developed along the margins of the axial silicone layer. Simultaneously, normal faults formed upslope and fold and thrust systems developed downslope. The most downslope of these systems is referred to as the morphological front. The cross-sections reveal that the morphological fronts were made of several reverse faults, which are said to be synthetic when verging downslope and antithetic when verging upslope (Fig. 4a). During an experiment, the reverse faults may have successively formed as a sequence from upslope to downslope (Fig. 4b) or may be out of sequence when a newly formed fault developed upslope relative to an earlier one (Fig. 4c).

Irrespective of the experimental conditions, a typical distribution of the deformation zones developed. It shows, successively, from upslope to downslope: (1) an extensional zone with normal faults, (2) a translational zone with no fault, (3) a compressional zone corresponding to the morphological front with thrusts and (4) a damping zone with no fault but where tiny displacements of the superficial markers fade away, downslope (Fig. 3). Alternatively, but similarly, an additional translational-and-then-compressional zone may have formed downslope, usually upslope from the damping zone (e.g., Figs. 3e–3h). Overall, most of the obtained slides propagated relatively rapidly from the very beginning of the experiments, with most of the displacements occurring during the first 12 h, and then progressively slowing down. Although they became particularly slow as time went on, it was rare for the sliding systems to completely stop over the course of the entire experiment (84 h). Nevertheless, the local and temporary slowing down or stopping of the displacements of some surficial markers were observed during all the experiments. For the sake of convenience, the results of each experiment are given as series providing direct comparisons for the dry and aqueous conditions.

For the G10 and G11 series of experiments, two slopes were imposed from either side of the single slope breaks. The angles of the plates were the same but the total lengths of the plates, upslope, differed (Figs. 2b and 2b’ and Tab. 2). During sliding, both series provided the typical successive zones that are distributed from upslope to downslope (Figs. 5a–5c) or which displayed two translational and compressional zones (Fig. 5d). The final lengths of both the extensional and compressional zones are up to five times larger for the G11 experiments than for the G10 models (Figs. 5a–5c and Tab. 3). Accordingly, the distance of the morphological fronts, surfaces and volumes of the G11 sliding deposits are significantly larger than the G10 ones (Tab. 3). The G11 experiments also exhibit at least two times as many, or even more, thrust faults than the G10 experiments (Fig. 5). Folds, faulted folds and ramps occurred within the compressional zones and were responsible for the thickening of the sliding layer up to 2.5-cm in the GA11 experiment (Fig. 5d and Tab. 3). Overall, the G11 experiments underwent significantly more deformation than the G10 ones. For both series, the thrusts first appeared close to the slope breaks and then they developed either out-of-sequence during the G11 experiments or in sequence for the G10 experiments ones. In the GA11 experiment, a thrust also formed close to the very end of the device downslope.

Typically, the displacement record for the surficial markers shows that the sliding slowed down as time proceeded, usually decelerating at a higher rate in the dry experiments than in the corresponding aqueous models (Fig. 6). Correlatively, the maximum velocities were slightly higher for the dry experiments than for the aqueous ones (Tab. 3). In detail, the displacements of the markers located upslope underwent acceleration and deceleration (Fig. 6) that were observed to correlate in the compressional zones with the formation of new faults and local arrests of deformation along ramps, respectively. In addition, the final numbers of thrusts and folds in dry conditions are larger than for aqueous models, despite total shortening values that were, about the same or smaller for the dry experiments (Tab. 3).

On either side of both slope breaks and in-between, the plates for the G7 and G12 series were of the same length but exhibited different slopes (steeper slopes for the G07 models; Tab. 2). Slides also occurred in both series resulting in the typical distribution of the deformation zones from upslope to downslope (Figs. 5e–5h). Overall, the lengths, surfaces and volumes of the compressional zone(s) were larger for the G07 models than for the G12 ones (Tab. 3). Correlatively, the number of faults and folds and the shortening values were also larger in the G07 compressional zones. In dry conditions, the distance of the morphological front is significantly larger for the GD07 experiment compared to the GD12 experiment, but the morphological front reached approximately the same locations in the GA07 and GA12 models under aqueous conditions (Tab. 3). For both series, the thrusts first appeared close to one of the slope breaks (more often SB2 than SB1) and then randomly developed either out-of-sequence or in sequence, independently of the controlled experimental conditions.

Under water, the final numbers of thrusts, the shortening values, and the distances of the fronts, surfaces, volumes and maximum thicknesses of the slides obtained are larger than – or, as large as – those obtained for the dry conditions (Fig. 5 and Tab. 3). The final distance of the morphological front for the GA12 experiment is roughly twice that than for the GD12 one, but it is only about the same as GD7 for GA7 (Fig. 5).

The analysis of the displacements of the superficial markers versus time (Fig. 6) reveals the same main results as for the G10-G11 series. Again, the highest velocities were reached within the first 12 h of the experiments, most often showing a velocity peak after 6 h (Fig. 6). The GD12 curve in Figure 6 illustrates that sliding was almost completely stopped starting from 36 h after the beginning of the experiment. Under water, the first faults were synthetic for both GA12 and GA07 and in dry conditions for GD12 (Tab. 3). For the GD07 experiment, the first fault was antithetic, indicating that these features hardly correlate with the controlled experimental conditions.

The three series of experiments all involved two slope breaks in-between three plates of the same length and with the same slope values (Tab. 2). The G18 and G14 series underwent a single additional sand input of 150 g and 450 g, respectively, poured upslope (approximatively over plate P1) as soon as the experiments started. For the G15 series, a cumulative mass of 450 g of sand was also poured upslope; this was done as six successive inputs of 75 g of sand each periodically every 12 h.

Sliding was observed during all the experiments, again resulting in the typical distribution of the deformation zones from upslope to downslope (Fig. 7). Overall, the observed deformations have roughly the same intensities in the G14, G15 and G18 models (Fig. 7 and Tab. 3). The sizes, surfaces and volumes of the extensional and compressional zones, the morphological front distance and the number of structures in the compressional zone(s) are roughly the same for all of the experiments (Tab. 3). On main difference concerns the overall shortenings (displacements of the M1 marker, Tab. 3), which are larger in the G15 successive input models than the G18 and G14 single input experiments. Three to six reverse faults formed in sequence during the G14 and G18 experiments and out-of-sequence during the G15 ones. The first thrust was systematically synthetic and formed either in the vicinity of the SB1 slope break in dry conditions or in the vicinity of SB2 under water (Tab. 3).

In aqueous conditions, the GA14, GA15 and GA18 models also displayed larger lengths and surfaces for the compressional zones, larger volumes for the sliding deposits and, to a lesser extent, larger overall shortening values, compared to their dry counterparts (Fig. 7 and Tab. 3).

The G14 models exhibited the highest maximal displacement velocities for the superficial markers (Tab. 3 and Fig. 7). As for the experiments without additional sand input, the velocities for the G14 and G18 models progressively decreased with time after the maximum velocity was reached within the first six hours of the experiments. Conversely, the G15 models show much smaller velocity variations with almost linear displacement curves (Fig. 7). However, temporary decelerations and accelerations were recorded in all experiments of the three series. Some were observed to correlate with the stops of motions along the ramps and the creation of new faults, respectively, in the compressional zones. In addition, during the G15 models, some of the accelerations also correlate with some of the periodical sand inputs (Figs. 7b and 7d).

As sliding was observed in all the experiments, the results demonstrate that the gravitational driving forces systematically overcame the resisting forces (Locat and Lee, 2002), due to the presence of an underlying décollement layer along the slopes and a relatively thin initial sand layer. However, in the GD12 experiment, sliding stopped quite rapidly (36 h) showing that the strength of the thickened layer became dominant over the driving force during the course of the dry experiment. Together with the decelerations and then the accelerations of the sliding observed in all the experiments, this shows that the driving and resisting forces were about the same orders of magnitude. The starting experimental conditions mostly explored the influences of the various slope geometries (lengths and angle values), additional sand inputs and dry versus aqueous conditions on the dynamics of sliding.

Upslope, the slope lengths exert a significant influence on the slide dynamics. Longer upslope slopes result in larger displacements and, overall, more deformations in the systems than shorter upslope slopes (Figs. 5a–5d). An additional observation was that the first faults that formed often developed in the vicinity of a slope break. Given the experimental set-up, the distance separating the first fault from the upslope extremity of the device also places a limit on the upslope volume fraction of the slide, and therefore its mass fraction, which initially drove the sliding process. For instance, in the GD12 experiment where the first fault was roughly located above SB1, there was consistently less displacement and deformation in the end than for the GD07 model in which the first fault was near SB2, at a larger distance from the upslope box extremity (Figs. 5e and 5g). The G07 models also exhibited steeper slopes above plate P2 than the G12 models, which also result in larger final displacements and shortening values for the G7 experiments (Tab. 3 and Figs. 5e–5h). Indeed, steeper slopes favor sliding.

In addition, slope breaks localize the primary thrust system in most of the experiments. In models with two slope breaks, the first thrust faults essentially form at SB1 in the dry settings and at SB2 in the aqueous models. This localization of deformation at SB2 correlates with a further deformation propagation distance in aqueous conditions most of the time (Figs. 5 and 7). As a comparison, in the one-slope break models, the first thrust fault forms at the SB (i.e., the G10 and G11 models) before propagating further. Therefore, a successive two-slope break system tends to localize compressional deformation zones at the furthest downslope slope break, and then it contributes to sliding over a longer runout distance. The geometry, and especially the number and distribution of the slope breaks, is a key parameter of propagation distance for sliding processes.

The G7 models are the only ones to show different results that might be due to two different causes. First, local variations in the sand layer thickness may induce a local strength variability into the system. A locally thinner layer is less resistant and favors rupture. Second, the distribution of the grains within a sand layer may also control the location of the emerging rupture zone. The development of thrusts either in sequence or out-of-sequence during sliding illustrates the effects of slope break location, sedimentary input and initial thickness variability. When the sand thickness is constant, the rupture zone only occurs where constraints are building up (in front of the sliding). However, when faults develop in an out-of-sequence way, this suggests that the thickness is not constant, and may favor the rupture and formation of faults behind the sliding, and not necessarily at the exact place where the constraints are accumulating.

Our observations with regards to the influence of the slope geometry on the sliding masses dynamics correlate with a previous study investigating the runout distances of debris avalanches offshore Martinique Island (Brunet et al., 2017). Based on numerical simulations using two complementary depth-averaged thin-layer continuum models (Shaltop versus Hysea), our study shows that an accurate description of the complex topography is a crucial component in the dynamics and deposition of debris avalanches, perhaps even more so than its interaction with the water column.

It is challenging to run experiments on gravity-driven sliding processes under water. As far as we are aware, no such experiments have been performed before now (the present work), especially for the research topic of submerged slides. This is less due to the technical difficulties encountered when setting up an experimental device under water, and related more to quantifying and understanding the interactions between water and the other material used in the experiments.

At a given depth, the hydrostatic pressure corresponds to the water column between this depth and the surface. It is 0.4 times the lithostatic pressure and leads to rock densities up to 2500 kg/m³. Interstitial pressures higher than the hydrostatic pressure may be related to various mechanisms such as rapid sediment compaction, deformation or dehydration because of the pressure and temperature variations. In order to take interstitial pressure into account in the strength calculation, it is necessary to modify the Mohr–Coulomb criteria (1) as follows (von Terzaghi, 1923):

where τ is the shear stress, c is the cohesion, σ is the normal stress, p is the interstitial pressure and σ − p is defined as the effective normal stress σ’ and δ is the friction angle. In addition, the main effective stresses are σ1’ = σ1 − p and σ3’ = σ3 − p. Thus, with fluid pressure, the effective strength σ1–σ3 decreases to give σ1’–σ3’, which facilitates the deformation and sliding (Fig. 8).

Our experiments have demonstrated significant differences between the aqueous and dry models (Figs. 5 and 6). Most of our results clearly show that deformation propagates further in aqueous models compared to the dry counterparts (Tab. 3 and Figs. 6 and 9), which is consistent with the principle stated above. Therefore, we assume that the sand cohesion in our models is very low and the hydrostatic pressure induced by the water column above the sand level is both a facilitator mechanism and a catalyst for deformation propagation.

Lastly, contrary to sediment inputs that increase locally as well as temporary gravity constraints on sliding – and thus driving forces –, the hydrostatic pressure permanently induces a decrease in the effective strength σ1’–σ3’ thereby facilitating the movement on faults that favors sliding and deformation. This is correlated with the continuously progressing sliding in aqueous models compared to dry ones. Only the G11 models show an opposite trend, but they were characterized by higher gravity forces (longer and steeper slopes) (Figs. 6 and 9).

The increase of deformation facilitated by fluid overpressure in sliding processes has been extensively studied and demonstrated, through either field observations or experimental modeling with compressed air as a pore fluid in sandbox models (Cobbold and Castro, 1999; 2001, 2009; Mourgues and Cobbold, 2003; Mourgues et al., 2009). Among the mechanisms producing an increase in fluid volume are thermal dilation, mineralogical transformation and hydrocarbon generation (Osborne and Swarbrick, 1997).

In the previous paragraph, we notice that for steeper and longer slopes upslope (i.e., the G12, G10, G7 and G11 models, respectively), the propagation distance increases correlatively to the cumulated amount of shortening (Tab. 3 and Fig. 5), illustrating a growing gravity power in these models. Similarly, it appears that a steeper slope over a long distance (i.e., G11 models) has more impact in terms of shortening and propagation distance rather than the decrease in the effective strength σ1’–σ3’ (Tab. 3 and Fig. 5). This demonstrates that increasing slope angles favorably affect the sliding propagation distance, however, it also shows that the length of the slopes may modify the slide features so that when the slopes are longer, the driving gravity power for sliding becomes greater. The hydrostatic pressure and the induced decrease in the effective strength in the sand layer facilitate the movement on faults; this was observed experiment GA11 where there are fewer reverse faults in the compressional zone than in GD11, but the fault throw is larger.

Our experiments demonstrate the large control exerted by the system’s geometry over the sliding dynamics and the hydrostatic pressure with respect to the organization of the faults.

Regardless of the type of sedimentary input, our experiments have demonstrated that this input represents a driving mechanism rather than a triggering mechanism as per the sliding dynamics. With or without inputs, sliding initiates at the very beginning of the experiments. However, sedimentary inputs allow deformations to propagate over a longer distance. During the 12 h separating each input, the displacements may significantly slow down to the point where it appears that the sliding is about to stop.

The shortening/displacement rate – and therefore the induced deformation – increases in the following order: the G18, G14 and G15 models. This observation suggests that regular sediment inputs deposited upslope have more impact in terms of deformation of the upper slope (i.e., displacement rate) than those deposited all at once (Tab. 3 and Figs. 7 and 9). In addition, shortening is most often higher in aqueous models (Tab. 3 and Figs. 7 and 9), and is related to hydrostatic pressure which permanently causes a decrease in the effective strength σ1’–σ3’ and favors sliding and deformation.

With regards to the propagation distance, it is important to take the formation of locking fronts during sliding into account, as they tend to decrease the runout distances. These fronts appear during most of the experiments and decrease the propagation of the faults both locally and temporarily, whereas the upslope translation is still progressing. Therefore, the stresses accumulate at the fronts and once they exceed the resistivity stresses, the locking front gives way under stress and the sliding propagates again. This “unlocking” process leads to a slight increase in velocity.

In addition, the initial upslope sediment inputs in the G18 and G14 models (three times larger in G18 than in G14), result in more shortening in the G14 model than in the G18 model (approximately 15% higher, in both the dry and aqueous models). Similarly, the propagation distance is longer in both aqueous models compared to the dry ones (almost doubled between GA14 and GS14) (Tab. 3 and Figs. 7 and 9).

As previously mentioned, either with or without sediment inputs, it is common for sliding to not stop in aqueous conditions. However, in dry models, only some of the models without sediment inputs stop sliding (i.e., GG10 and G12 models). Periodic inputs make it possible to maintain sliding over time (Fig. 10). As expected, this shows that a regular input of sediments in the sliding process is a driving mechanism, but it is even more significant when its effects are combined with those of hydrostatic pressure (Fig. 10). In addition, the sliding process stops in the following order: GS12, GS10, GS18 and GS14. These examples clearly illustrate that the initial sediment inputs result in additional driving stresses compared to models without sediment inputs, but they do not prolong sliding for as long as the regular inputs do. Indeed, during sliding, two processes occur: (1) the downslope series become thicker because of shortening, hence they become stronger; (2) the length of the mid-slope sliding block becomes shorter. Therefore, the driving force decreases while the resisting force increases, leading to slow down or even a stop in deformation. By adding new sediments in the upslope area, the sliding block retrieves its original length, “restarting” the driving force (see Fig. 11).

This positive correlation between sedimentary supply and deformation intensity is also observed at different scales and settings. For instance, the gravity-driven deformation observed at the Niger Delta has been widely investigated and related to overpressured shales (e.g., Damuth, 1994; Wu et al., 2000; Bilotti and Shaw, 2005; Corredor et al., 2005; Briggs et al., 2006; Cobbold et al., 2009; Mourgues et al., 2009; Maloney et al., 2010), and has also been associated with the spatial variations in the sedimentary supply that cause spatial variations in the deformation intensity and rate: when the progradation length is longer, the intensity of the gravity-driven deformation is higher (Rouby et al., 2011).

Based on previous observations and interpretations, our experiments have highlighted three main factors contributing to the sliding deformation: slope geometry, hydrostatic pressure and sediment supply. A large number of studies have investigated the influence of the slope geometry on the sliding propagation, i.e., focusing on physical parameters such as the slope steepness and the type of décollement layers (Cobbold and Castro, 1999; Cobbold et al., 2009; Mourgues and Cobbold, 2003; Mourgues et al., 2009), showing the sliding potential of slopes (inclined only a few degrees) when associated with a specific décollement layer (i.e., salt tectonics, Fort et al., 2004; Brun and Fort, 2011). In the present study, we go further by demonstrating how slope breaks between the slope portions can localize the formation of the first thrusts and their propagation depending on the driving forces at play. Our experimental observations fully support another study demonstrating how submarine slope breaks constrain the runout of the debris avalanches offshore Martinique Island using a numerical simulation (Brunet et al., 2017). Therefore, the slope geometry could be considered as a preconditional and triggering factor for sliding when combined with other factors such as sediment loading, seismicity or specific décollement layers. With regards to the hydrostatic pressure, our models have clearly demonstrated its catalyst effect on sliding mobility over time, followed by the capacity to increase the sliding and deformation runout distance (Figs. 6 and 9). Numerical simulations on debris avalanches when they entered sea were also used to test the involvement of hydrostatic pressure on sliding dynamics, but the results showed that it is complex topographies rather than hydrostatic pressure that seem to have an impact on gravity-sliding processes (Brunet et al., 2017). Lastly, the experiments with sediment inputs demonstrated that irrespective of the type of sedimentary input and the experimental conditions (dry or aqueous), the inputs represent a driving mechanism and not a triggering mechanism with regards to the sliding dynamics. With or without inputs, sliding is initiated from the very beginning of the experiments (Figs. 6 and 9). However, inputs allow deformations to propagate over longer distances (Figs. 9 and 10). Sediment inputs may also trigger temporary sliding accelerations just after deposition resulting in increased deformation within the sliding mass. This positive correlation between high sediment input and gravity-driven deformation has been widely recognized at different geological scales and settings, i.e., deltaic fronts or glacial continental margins (Adams and Roberts, 1993; Imbo et al., 2003; Bryn et al., 2005; Leynaud et al., 2007; Rouby et al., 2011), and involves the generation of excess pore pressure. In the present study, the experiments have also shown that a regular supply of sediment in the sliding process is the major driving mechanism, but it is even more significant when combined with another catalytic/driving mechanism such as hydrostatic pressure (Fig. 10).

The geometric characteristics (thickness and length of the deposit, slope, sediment inputs, etc.) of the experimental device have been set up based on the interpreted marine geophysical data (Brunet et al., 2016) acquired offshore Martinique (Deplus et al., 2001, 2002; Le Friant et al., 2003a; Fig. 1).

With regards to the structures, the morphological front identified using bathymetric and seismic reflection data (Fig. 1) may be correlated to the formation of multiple thrusts as observed in our experimental models (Fig. 12). This surficial structure is related to the deep frontal ramp highlighted in Brunet et al. (2016), and as is typical of frontally emergent slides (Frey-Martínez et al., 2006), it is similar to the frontal ramp observed in some of our models (Fig. 4c). In addition, the thrusts and folds system described in the distal part of the slide are also systematically observed in our models and can be compared to those observed on seismic profiles (Fig. 1). The resolution of the seismic data is limited compared to the experimental cross-sections, and therefore caution should be taken when comparing the structures, e.g., fold structures may be confused with diffraction hyperbola for instance. Thus, analogue modelling can also potentially provide structural information when the current available data cannot. Lastly, microstructures identified on the drilled cores, such as micro-faults, inclined bedding or reverse grading (Brunet et al., 2016), indicate major compressional movements that may be associated with either large-scale folds (straight, overturned or bedded), or thrusts with variable strikes and dips (Fig. 12).

As previously mentioned, the second slope break locates the thrust formation (especially in aqueous conditions) and the subsequent propagation of the deformation downslope. This is also observed in the real context, with the main front located approximately 25 km downslope from the second slope break. In addition, the maximum thickness may reach 500 m, which lies perfectly within the measurement range for the modeled slides. Frontally emergent slides are also characterized by a surficial morphological front, a deep frontal ramp and a distal translational sliding mass (Frey-Martínez et al., 2006). These additional features were not reproduced in the laboratory, only a few frontal ramps appear in the GA11 and GA15 models. In nature, these ramps do not exceed a height of 150 m, whereas they can reach 0.5 to 1 cm in the models (equivalent to 500–1000 m high). It is significant that a difference of one order of magnitude is observed; this shows that the models do not exactly reproduce what is observed in nature, although it is similar in other respects (Fig. 12). Nevertheless, this difference can be explained by the unique oversized décollement layer in the models compared to the multiple ones that probably occur in nature (Lafuerza et al., 2014; Hornbach et al., 2015; Mencaroni et al., 2020; Knappe et al., 2020; Llopart et al., 2021). Therefore, in the laboratory, the formation of frontal ramps will involve the whole overlying sand stratigraphic interval (1 cm-thick) whereas in nature, the layout of the potential multi-décollement layers will involve the displacement of thinner stratigraphic intervals, resulting in smaller frontal ramps. Although observed in the seismic data (Fig. 1c), a distal deformed unit beyond the frontal ramp is missing in our models; this may be due to a lack of sufficient inertial energy allowing the slide mass to emerge beyond a stratigraphic level and to be translated over a long distance (Frey-Martínez et al., 2006).

The submarine landslide occurring in Martinique is not constrained yet in terms of timing. It might have been emplaced in few hours or in few days, or could even be still active today, either continuously or intermittently. The scenario proposed by Brunet et al. (2016) suggests a submarine landslide triggered by the emplacement of debris avalanches next to a major volcano flank-collapse event dated at 127 ka and reactivated at least once at 36 ka (Solaro et al., 2020; Boudon and Balcone-Boissard, 2021) by successive debris avalanche deposits loading the submarine volcano slope. As demonstrated by some experiments, a sudden sediment input can temporarily accelerate or reactivate a slide, which is compatible with the proposed scenario. Moreover, the experiments showed that all types of sediment inputs contribute to sliding process in one way or another. Applied to natural conditions, these inputs may not only correspond to debris avalanches but also to turbidity currents, debris flows or regular large-scale Grenada basin sedimentation.

So far, there are no temporal constraints with regards to the dynamics of the submarine landslide from Martinique Island, but the experiments do provide some kinematic proxies. Slides are still slightly progressing in most parts of the models after 84 h (i.e., the end of the experiment), but we can infer that the speed of these displacements is close to zero. Applied to natural conditions, our experiments suggest that the last and smaller volcano flank-collapse event, recently dated at 36 ka (Solaro et al., 2020) and the related debris avalanche volume estimated at 2 km³ (Le Friant et al., 2003a), would not have allowed the slide to propagate up to the point where it is observed today. We show that the first inputs, i.e., debris avalanches from the first flank-collapse event, dated at 127 ka (Germa et al., 2011) and with an estimated volume between 14.7 km³ and 25 km³ (Le Friant et al., 2003a; Germa et al., 2015), tend to have higher impacts on the sliding propagation, especially when they involve large volumes (G14 model vs. G18 model). When applied to the natural system, these results suggest that the last and smaller flank-collapse event might have had a smaller impact, if any, on the submarine landslide deposit. This observation supports studies that consider the first volcano flank-collapse event and the related debris avalanche as not only the largest in Montagne Pelée’s geological history (Le Friant et al., 2003a, b; Boudon et al., 2007; Boudon and Balcone-Boissard, 2021; Solaro et al., 2020) but also the one responsible for the large-scale submarine landslide identified offshore Martinique Island (Brunet et al., 2016).

The complex submarine slope to the west of Martinique Island, in the Lesser Antilles arc, exhibits various types of gravity-driven submarine landslides and debris avalanches. Laboratory modelling experiments designed to test the emplacement dynamics in either dry or aqueous conditions, as well as to assess the respective contributions of slope geometry, hydrostatic pressure and sediment input indicate that:

• the slope geometry controls the distribution and propagation of the deformation patterns. In particular, the slope breaks control the location of the first thrusts. Further propagation is observed in the case of multiple slope breaks;

• with increasing slope steepness and length, the influence of gravity is higher and the sliding goes further;

• hydrostatic pressure decreases the effective strength, thereby facilitating the movement on faults and favoring sliding and deformation in dry conditions;

• based on the sliding dynamics, it is more likely that sediment inputs are a driving mechanism and not a triggering mechanism. Sediment inputs increase the gravity forces and sliding deformation. However, they allow deformations to propagate for longer distances;

• when occurring only periodically, sediment inputs can temporally and locally accelerate sliding.

Funding for this research was provided by the INSU, ANR CARIB and the Bureau IODP-France.

The authors wish to thank S. Mullin for proofreading the English content. They are also grateful to Bruno Vendeville for his valuable comments allowing to significantly improve the manuscript, being one of his last contributions for our community.