The largest canyons on our planet occur on the seabed1, and the submarine flows that flush these canyons form the largest sediment accumulations on Earth2,3. Only rivers carry comparable amounts of sediment and organic carbon (OC) across such large areas4,5. However, unlike rivers, there are exceptionally few time-lapse surveys of submarine canyons. Only ~4 of 9,477 submarine canyons documented globally1 have time-lapse surveys, and this is the first time-lapse survey of a submarine canyon feeding a large (>150,000 km2) submarine fan in the deep (2–5 km) sea. There are also few direct measurements of the flows that flush submarine canyons6,7,8,9. This ensures that rates of geomorphic evolution, and the associated sediment and geochemical fluxes, including OC, are far more poorly understood than those for rivers. Previous time-lapse surveys of the upper reaches of much smaller submarine canyons have identified channel migration10, bedform movement6,9 and landslides11. We use repeat mapping of the Congo Canyon to show how canyon-flank landslides can affect canyon geomorphology and fluxes of sediment and carbon to the deep sea.

Landslides are important geomorphic agents in terrestrial environments. Where landslides flow into river channels they can lead to full blockages (landslide-dams)12,13, channel constrictions, diversions or avulsions of the river channel itself12,14,15. Once emplaced, the landslide-dam itself can fail instantaneously, leading to erosion and transport of sediment upstream and downstream of the dam16. Over longer timescales, and depending on their longevity, landslide-dams can result in the accumulation of sediments and organic material upstream of the dam13,17,18. Upstream aggradation and downstream incision can lead to knickpoint generation in river longitudinal profiles19. Frequent landslide-dams or large-magnitude events can therefore notably influence river profiles over long timescales19. From an ecological perspective, landslide-dams can impact water quality and aquatic habitats in rivers12,13 and are a key phenomenon in fluvial landscape evolution.

In this Article, we show how landslide-dams can impact submarine canyons. Submarine canyons are carved by slope failures and sediment flows called turbidity currents that can run out for hundreds of kilometres11,20. Indeed, landslide scars are ubiquitous on submarine canyon walls in many locations worldwide11,21,22, suggesting that landslide occurrence is common (see Supplementary Table 1). However, a nearly complete lack of repeat bathymetric surveys ensures their geomorphic effects and exact timing are poorly constrained. As a consequence, an a priori assumption often persists that these landslides commonly evolve from low-mobility slumps to higher-mobility flows and eventually turbulent turbidity currents11,20. Using repeat bathymetric mapping, we show how submarine landslides can impact canyon morphology and interact with turbidity currents in submarine canyons.

We then show how landslide-dams have the potential to affect the tempo of sediment and carbon transfer to the deep sea. Submarine canyons act as the primary conduits for sediment, oxygen, nutrient, pollutants and organic-matter transport from rivers and coasts to the deep sea1,23,24,25. They are commonly envisaged as highly efficient sediment transport conduits as well as locations for rapid burial and sequestration of substances such as OC23,26,27. It is therefore important to understand how landslide-dams, and their interaction with other sediment flows, can impact sediment and OC transfer through submarine canyons28.

Congo Canyon and canyon-flank landslide-dams

The Congo Canyon, offshore West Africa (Fig. 1), is directly connected to the Congo River, the second-largest river by water discharge on Earth4, and therefore acts as a continuous conveyor for sediment delivery from the river to the deep sea22. Annually, the river delivers ~43 Mt sediment and ~2 Mt OC into the canyon head, which is directly linked to the river mouth29. The canyon itself is 6–12 km wide and >800 m deep, and its floor is 50–800 m wide. The canyon walls have a largely concave pattern, with numerous individual headscarps recording past landslides22,30 (Fig. 2a). There are also multiple terrace levels that are likely to be areas of high sediment accumulation and may be prone to collapse31. Monitoring studies have shown that the canyon is highly active, with multiple long-duration (up to a week) turbidity currents traversing the canyon annually8,32,33,34.

Fig. 1: Study area and bathymetric map.
figure 1

a, Map of the Congo turbidite system. b, Bathymetry map of the upper Congo Canyon surveyed in 2019 on JC187. A sinuous canyon thalweg is present in the >800-m-deep canyon. Multiple terrace levels are also visible along the canyon. An, Angola; DRC, Democratic Republic of the Congo. a, Basemap from GEBCO; b, Congo Canyon/Fan bathymetry from Ifremer.

Fig. 2: Repeat bathymetric mapping of the upper Congo Canyon and a difference map.
figure 2

a, Bathymetric map of part of the upper Congo Canyon derived from data collected in 2005. b, Bathymetric map of the same area from 2019. A canyon-flank landslide that has dammed the canyon is shown. The landslide-dam has resulted in the trapping of sediment leading to infilling up-canyon of the landslide-dam. c, Difference map of the 2005 data and the 2019 bathymetry resampled to 25 m showing the canyon-flank landslide, the landslide-dam and the sedimentary infill. This is overlain on a hillshade map of the 2019 bathymetric data. It allows the patterns of erosion and deposition to be characterized.

Bathymetric surveys of the upper canyon were conducted in 2005 and 2019 using Kongsberg EM120 and EM122 instruments. Survey design and water depths resulted in optimum gridding of these datasets at 25 m and 5 m, respectively (Methods). The contrasting dataset gridding prevents small-scale changes (<25 m) from being confidently resolved. However, large-scale changes to canyon morphology can be assessed confidently. Comparison of these datasets shows significant morphological change has occurred over ~14 years (Fig. 2). These data show that a large landslide occurred on the southern canyon flank at 1,200 m water depth (Fig. 2b). The landslide headscarp has a perimeter of 2.5 km, a maximum width of 1 km and height of 390 m. The length of the associated deposit is ~800 m. Overall, the landslide is 1.7 km long, with an area of ~0.64 km2. The minimum excavated volume was ~0.09 ± 0.01 km3. A smaller landslide (0.006 ± 0.001 km3) also occurred ~2 km up-canyon (Fig. 2b).

The canyon-flank landslide deposit caused a meander-bend cut-off, which shortened the thalweg length by ~3 km. Up-canyon of the flank collapse, canyon infilling occurred. The deposited infill is up to 150 m thick closest to the landslide deposit (Fig. 2c). At the eastern edge of the survey area, 26 km up-canyon, the newly deposited infill is up to 40 m thick. The total sediment infill volume, excluding the landslide deposit, is ~0.4 ± 0.1 km3. A new channel has been incised into the infill (Fig. 2b). This channel is up to 20 m deep and 100 m wide and runs 6.2 km up-canyon of the landslide deposit. A series of knickpoints are associated with the landslide deposit and sediment infill. A 30-m-high knickpoint has formed where the new channel descends over the northwest end of the landslide deposit (Figs. 2b and 3). Two smaller knickpoints are also present where channels have incised into the sediment infill (Fig. 3). Down-canyon of the landslide deposit, a wedge of sediment with a volume of 0.02 ± 0.01 km3 has been deposited.

Fig. 3: Landslide-dam impacts on the Congo Canyon thalweg.
figure 3

a,b, Plan view of the thalweg profiles in 2005 (a) and 2019 (b). Knickpoints (K1–K4) identified in c are shown in b. c, Thalweg profiles normalized for distance for 2005 and 2019. The profiles show the large amount of infill that has occurred post-canyon-flank collapse. The 2019 profile also shows the presence of four new knickpoints. The largest knickpoint (K4) is situated where the newly incised channel descends over the landslide-dam. The smallest knickpoints (K2 and K3) represent the up-canyon initiation of the small channel visible in Fig. 2b. The proximal knickpoint (K1) is the result of a further channel incising into the infill where the canyon is most constricted.

Geomorphic change driven by canyon-flank landslides

Our data provide an opportunity to (1) test whether submarine landslide-dams may have comparable (or different) impacts to their subaerial counterparts; (2) derive models of canyon-flank landslide impacts on submarine canyon evolution and geomorphology (Fig. 4) and (3) identify the sedimentary signature of such events.

Fig. 4: Schematic model of possible morphological changes caused by large canyon-flank landslides and their stratigraphic signature.
figure 4

a,b, Morphological evolution if canyon-flank landslide dams the canyon. a,c, Morphological evolution if canyon-flank landslide constricts the canyon. a, Canyon walls are eroded and undercut by turbidity currents. b, Undercutting results in canyon-flank landslide that dams the canyon (top). Up-canyon of the landslide-dam, the canyon floor is infilled by trapped sediment transported by turbidity currents (middle). A knickpoint incises up-canyon, forming a new thalweg channel as a consequence of turbidity-current activity (bottom). The position of this new channel can lead to meander-bend cut-off and the formation of terraces. The channel may eventually erode to the pre-landslide base level. c, Undercutting results in a canyon-flank landslide that constricts the canyon-floor width (top). The constriction results in enhanced turbidity-current erosion and the development of a newly incised canyon-floor channel (middle). Headward erosion of a new knickpoint and continued down-canyon incision results in the growth of this new channel (bottom).

On land, where landslides block river channels, meander cut-offs have been observed where the river is forced to incise a new channel through the deposited material14,18,35. Sediment aggradation behind the landslide-dam can protect the channel bed, leading to slower bedrock incision rates, thereby impacting long-term longitudinal river profile evolution17,19. This disequilibrium can also lead to knickpoint formation as the channel adjusts to the perturbation14,36.

Our data document a similar process in submarine canyons (Figs. 4 and 5). Where canyon-flank landslides are sufficiently large, landslide-dams can be emplaced across the canyon floor (Fig. 4b (top)). Once emplaced, this material can lead to rapid canyon infilling as sediment carried by perturbed and decelerating turbidity currents is trapped behind the dam, starving down-canyon areas of sediment (Fig. 4b (middle)). Our data show that deposition also occurs, immediately, down-canyon of the dam. This is probably a consequence of over-spilling turbidity currents that no longer have sufficient velocities to maintain sediment in suspension (Fig. 5). Once infilling has occurred, the observed new channel shows subsequent turbidity current activity can begin to erode through the landslide-dam and the infill (Figs. 2b and 4b (bottom)). We envisage that this channel was formed by erosion from the landslide-dam knickpoint once the accommodation space behind the landslide-dam was sufficiently full, such that changes in seafloor gradient did not impact the flows and allowed turbidity currents to maintain their driving force (sediment load). The rate of future channel incision and knickpoint migration will depend on turbidity current magnitude-frequency and infill-consolidation rate. The long-term existence of the observed knickpoints will depend on the incision rate through the landslide deposit and infill. Incision of a new canyon thalweg can lead to terrace formation on the canyon floor (Fig. 4b (bottom)). Where canyon-flank landslides occur on a meander bend, incision of a new thalweg channel through the infill can result in meander-bend cut-off (Fig. 3). Indeed, our observations show the first documented meander cut-off caused by a submarine canyon-flank landslide. Sediment trapped behind the dam may be gradually released by smaller flows or eroded en masse by a larger canyon-flushing event20,37,38. On active margins, large earthquakes may trigger a canyon-flushing event37,38. However, on passive margins, an alternative trigger will probably be required. In the Congo Canyon, this may be an exceptional river flood.

Fig. 5: Schematic model comparing landslide-dam impacts on rivers and turbidity currents in submarine canyons.
figure 5

a, Terrestrial landslide-dammed river. Fluvial landslide-dam traps water and sediment. Trapped water forms a lake behind the dam, and deltaic sediment is deposited when river water enters the lake or within the body of the lake. Once accommodation space has been filled, river water is able to overtop the dam and continue downstream. b, Submarine landslide-dam. Landslide-dam blocks the submarine canyon, interrupting the passage of turbidity currents. c, Partial flow reflection and ponding. Landslide-dam results in trapping, ponding and reflection of turbidity currents. d, Turbidity current deposits backfill accommodation space behind landslide-dam. As turbidity currents depend on their velocity to keep sediment entrained, this sediment is rapidly deposited. Parts of the turbidity current that are able to overtop the landslide-dam quickly dissipate as their sediment load is now greatly reduced. e, Gradual sediment release. Once the accommodation space is filled, more-erosive, faster-moving parts of turbidity currents are able to overtop the dam. These are able to erode the dam, leading to knickpoint formation as a new channel is incised into the infill. Turbidity current in panel b is a scaled representation of the flows described in ref. 8 under a CC BY license.

Where landslides are not sufficiently large or mobile enough to block terrestrial rivers or submarine canyons, we also see parallels14. Partial infilling of river channels is known to cause increased erosion and meander migration14,36. In submarine canyons (Extended Data Fig. 1b), three possible processes are envisaged (Fig. 4a,c). First, canyon-floor constriction can increase turbidity-current confinement (Fig. 4c (top)). This confinement can increase the flow’s erosive potential, allowing the generation of scours and eventually the formation of a new incised channel39. This channel can further concentrate flows, resulting in up-canyon and down-canyon channel growth (Fig. 4c (bottom)). Second, the thalweg may be deflected. In this case, a more sinuous course would persist until the landslide deposit was eroded. Third, the landslide deposit may be too small to impact flow characteristics in the canyon thalweg. In this case, the deposit provides sediment that can be eroded by turbidity currents21. These smaller failures may be particularly prevalent in certain settings, such as seismically active regions where earthquakes may trigger frequent smaller-magnitude slope failures, thus reducing the frequency of canyon-damming events40.

Landslide-dams and the resulting infill should be recognizable in submarine settings. The landslide deposit itself will be resolvable in seismic cross section as a zone of chaotic reflectors/transparency that may also have evidence of folding, normal faulting or compression41 (Fig. 4b). Where a new thalweg has been cut post-infill, these features will probably be preserved as part of a terrace and buried beneath layered sediments (Fig. 4b (bottom)). Landslide-dams and their infill may also be recognizable from the presence of stepped changes in terrace height along a canyon, representative of the up-canyon extent of infilling resulting from the landslide-dam (Fig. 5).

Deep-sea sediment transport and submarine landslide-dams

Fluvial landslide-dams can result in the storage of large volumes of water and sediment14. Similarly, the observed ≥0.09 km3 (~120 ± 10 Mt assuming a wet density of 1,800 kg m−3) canyon-flank landslide-dam in the Congo Canyon has resulted in the trapping and storage of a large sediment volume (≥0.4 km3, equivalent to ~170 ± 40 Mt assuming a wet density of 1,300 kg m−3). Indeed, the volume of stored sediment could be much larger as it is still 40 m thick at the edge of the 2019 survey (Fig. 3). The annual sediment flux from the Congo River is estimated to be ~43 Mt (ref. 4), while ref. 33 calculates ~40 Mt is transported annually by turbidity currents in the canyon. The landslide occurred at some point between 2005 and 2019. The landslide-dam is therefore efficiently trapping and storing large volumes of sediment that would otherwise have been transported down-canyon.

Quantifying global submarine landslide magnitude-frequency remains challenging because so few repeat multibeam bathymetry surveys have been collected. Subaerial landslide-dams can be studied and dated far more easily, especially with the advent of remote-sensing techniques14. The largest inventory of subaerial landslide-dams contains >1,000 examples since 190014. In this Article, we show the scale of this Congo Canyon landslide-dam and sediment infill is comparable to the largest terrestrial examples. Despite their frequent occurrence globally, only 24 fluvial landslide-dams since 1900 have volumes >0.09 km3, and only 40 have dammed lake volumes >0.4 km3. The Congo landslide-dam is equivalent to the 98th percentile for volume and 95th percentile for stored volumes in this 120 yr subaerial inventory.

The key differentiator between subaerial and submarine landslide-dams is how sediment is stored. Initially, fluvial landslide-dams will trap water and sediment. Once filled, sediment may continue to be deposited as the river enters the dammed lake, but water is still able to overtop the dam and continue to erode and transport sediment downstream13,14,15,16,17,18. By contrast, submarine landslide-dams will result in rapid infilling of sediment, and starve down-canyon reaches of sediment, due to the trapping of turbidity currents by the dam. Unlike rivers, turbidity currents in submarine channels are driven by their sediment load, and their ability to keep sediment entrained depends on their velocity8. Turbidity currents that deposit sediment lose driving force, potentially leading to further deceleration and sediment deposition; this positive feedback does not occur in rivers. Recent monitoring studies7,8,33 have suggested that turbidity currents are driven by dense, near-bed frontal cells (Fig. 5b). These dense frontal cells will probably be retained behind the dam, perturbing the flow and potentially resulting in flow ponding and reflection, leading to rapid deposition42. Only when the accommodation space is sufficiently filled will parts of the turbidity current pass over the landslide-dam42 (Fig. 5c,d). However, the loss of sediment still leads to further deposition immediately down-canyon, reducing turbidity-current run-out. Once filled, further turbidity-current activity will rework the infill (Figs. 2b and 5e).

Submarine landslide-dams and OC sequestration

Burial of young OC, produced by photosynthesis in the biosphere, in marine sediments removes CO2 from the atmosphere, regulating the climate over geological timescales26. Indeed, submarine canyons are considered efficient sequesters of OC in the ocean23,26,27. However, few measurements of OC sequestration exist in these environments, and we are therefore yet to constrain long-term fluxes or burial efficiencies despite common submarine canyon occurrence on all continental margins.

Measurements of sediment from the Congo Fan averaged over multiple sampling locations29 suggest a total OC content estimated at 2.75–3.00% by weight29. Assuming such values, the canyon-flank landslide and trapped infill contained ~3.2–3.5 and ~4.6–5.0 Mt of OC, respectively. The landslide will re-exhume already buried OC, but the infill is likely to contain younger OC43. The Congo River annually supplies ~2 Mt terrestrial OC (Supplementary Table 2). The buried OC therefore represents a minimum of 3–4 years of terrestrial OC supply from the river and ~4% of the global total of OC delivered annually by rivers to the oceans5. Thus, although worldwide frequency-magnitude of landslide-dams in submarine canyons is uncertain, this single example buried 4% of the terrestrial OC flux to the global ocean (Supplementary Table 2).

The rapid burial (<14 years) of this mass of OC should prevent oxidation and thus preserve the OC. This supports the previous assertions that submarine canyons are effective carbon sinks23,26,27. However, subsequent reworking by turbidity currents of OC stored in the dammed wedge of sediment will probably expose much of this carbon to future oxidation. The OC drawdown, therefore, depends on the preservation of terraces after a new channel has been incised into the infill. A similar process is envisaged for the re-exhumed OC contained in the landslide deposit. Erosion and re-exhumation are intrinsic to a meandering channel, but recurrent mass sliding, as suggested by the identified landslide, and concave canyon walls (Figs. 1 and 2) clearly add to the amount of re-exhumation. This suggests that landslides and landslide-dams probably reduce the efficiency of OC burial in canyons and their perceived role as efficient carbon sinks.

The mass of OC buried in the Congo Canyon infill is comparable to that remobilized by the Tohoku-oki and Kaikōura earthquakes11,44. These earthquakes resulted in ~1.7 and ~7.0 Mt OC being remobilized, respectively. Delivery of OC to the deep sea by these and other earthquake-triggered events is believed to exert a strong control over regional deep-sea productivity and carbon sequestration11,44. By contrast, landslide-dams will temporally lead to reduced OC fluxes that nourish down-system benthic ecosystems45,46,47. Submarine canyon-damming therefore has the potential to modulate marine carbon transport and cycling. These impacts will probably be greater where canyons are directly connected to terrestrial sediment sources. However, landslide-dams in canyons that are starved of terrestrial inputs may also exert impacts on OC cycling due to the impacts on ecosystems that rely on down-canyon sediment transport1.

We identified a submarine landslide-dam that is comparable in scale to the largest landslide-dams of the past 120 years. It shows how submarine landslide-dams may affect sediment and OC transfer to the deep sea and submarine canyon geomorphic evolution. This single example had a comparable impact on sediment and OC fluxes to the deep sea as major earthquakes and trapped a sediment mass four times the annual flux from the Congo River, equivalent to ~4% of the annual supply of terrestrial OC to the global oceans. Similar landslide scars are commonplace on the flanks of other submarine canyons (Supplementary Table 1), including those connecting to very large rivers48,49. This suggests submarine landslide-dams could potentially be common phenomena, even away from active margins. Indeed, at least three other landslide-dams have been identified in modern bathymetry (Supplementary Table 1). However, time-lapse bathymetric surveys are currently available for only ~4 of the ~9,477 submarine canyons known globally. This study shows the value of repeat high-resolution bathymetric mapping of large deep-sea systems. Further work should now determine the frequency of submarine landslide-damming to constrain their impact on deep-sea ecosystems, sediment transport and OC burial on different margin types.


The results presented in this paper are based on analysis of bathymetry data of the upper Congo Canyon collected during cruises JC187 of the RRS James Cook (2019) and OPTIC-CONGO2 in 2005 on board the FS Beautemps-Beaupré. JC187 used a Kongsberg EM122 system, and OPTIC-CONGO2 used a Kongsberg EM120. Both operated at 12 kHz with a vertical sounding accuracy of 0.2% of water depth. The beam aperture was 0.5° × 0.5° for JC187 and 1° × 1° for OPTIC-CONGO2. Data were processed to correct for differences in sound velocity in the water column (using data from sound-velocity profilers) and the ship’s motion. JC187 data were processed in CARIS HIPS and SIPS software and gridded with a resolution of 5 m. JC187 data were able to be gridded at 5 m due to the high number of soundings that were retrieved due to the survey design. The canyon was surveyed using along-canyon lines crossed by traverse lines perpendicular to the along-canyon lines to minimize potential shadowing or reflection due to steep canyon walls. The survey paths were also designed to ensure the entire survey area was covered by the central third of the swath width to maximize the number of soundings. The OPTIC-CONGO2 survey used along-canyon lines. OPTIC-CONGO2 data were processed and gridded at 25 m in CARAIBES. Inter-survey positioning was validated by matching features in common areas of relatively stable seafloor.

Erosion and deposition of sediment between 2005 and 2019 were calculated by first resampling the JC187 5 m data to 25 m so as to be comparable with the OPTIC-CONGO2 data. A bathymetric difference map produced in ArcGIS was subsequently used to determine volumes of eroded and deposited sediment. Where canyon-flank landslides were observed, their volumes were calculated using depth changes associated with their headwalls. This was necessary because the emplaced deposit was often buried, and no seismic data penetrating the landslide deposits were available, preventing direct assessment of the deposit volume. The calculated landslide volumes are therefore probably only minimum volumes.

The bathymetric surveys were used to construct along-canyon profiles. The position of submarine canyon and channel thalwegs shift as they evolve, so down-canyon profiles were measured along the position of the thalweg in each survey. These profiles were subsequently normalized for comparison in the section of the canyon impacted by the canyon-flank landslide and infill. Normalized distances along canyon were calculated by taking a starting point (Fig. 3; furthest point west) where the thalweg was in the same location for both surveys and dividing each measurement of distance by the distance travelled along the thalweg to the end point of the profiles (Fig. 3; furthest point east).

The morphology of the Congo Submarine Canyon observed in the JC187 and OPTIC-CONGO2 bathymetry data was also compared with high-resolution bathymetric data that has been previously published for a range of submarine canyons around the world6,11,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85 (Supplementary Table 1). The volumes of sediment and OC stored in-canyon and up-canyon of the observed landslide-dam were also compared with a range of known events in marine and fluvial environments in terms of displaced masses and OC transport5,11,29,44,86,87,88,89,90,91,92,93,94,95,96,97,98 (Supplementary Table 2).