Flood and tides trigger longest measured sediment ow that accelerates for thousand kilometers into deep-sea

Here we document for the rst time how major rivers connect directly to the deep-sea, by analysing the longest runout sediment ows (of any type) yet measured in action. These seaoor turbidity currents originated from the Congo River-mouth, with one ow travelling >1,130 km whilst accelerating from 5.2 to 8.0 m/s. In one year, these turbidity currents eroded 1-2 km 3 of sediment from just one submarine canyon, equivalent to 14-28% of the annual global-ux from rivers. It was known earthquakes trigger canyon-ushing ows. We show major river-oods also generate canyon-ushing ows, primed by rapid sediment-accumulation at the river-mouth, but triggered by spring tides weeks to months after the ood. This is also the rst eld-conrmation that turbidity currents which erode can self-accelerate, thereby travelling much further. These observations explain highly-ecient organic carbon transfer, and have important implications for hazards to seabed cables, or how terrestrial climate change impacts the deep-sea.


Introduction
Flows of sediment that move along the seabed (called turbidity currents) form the largest sediment accumulations, deepest canyons and longest channel systems on Earth 1,2,3 . The scale of individual turbidity currents can also be exceptionally large (Table 1). For example, an earthquake-triggered turbidity current that occurred in 1929 in the NW Atlantic carried over 200 km 3 of sediment, and ran out for >800 km, at speeds of up to 19 m/s 4,5 . This single turbidity current carried over 20 times the annual sediment ux from all of the world's rivers 6 , and its volume exceeded the largest documented subaerial landslide in the last ~350,000 years 7 . It was previously thought that directly measuring powerful turbidity currents that reached the deep-sea was impractical 8 . However, here we describe direct monitoring of deep-sea turbidity currents in the Congo Canyon offshore West Africa 9 , whose timing was captured by an array of seabed moorings and seabed telecommunication cable breaks (Figs. 1 and 2). On January 14-16th 2020, one of these ows travelled for over 1,130 km from the mouth of the Congo River, measured along the sinuous axis of the submarine Congo Canyon and Channel (Fig. 1). This is the longest runout sedimentdriven ow yet measured in action on our planet, with a runout distance exceeding that of the 1929 NE Atlantic turbidity current 4 , and longest known terrestrial debris ow 10 , snow avalanche 11 or volcanic pyroclastic ow 12 (Table 1). Two further long runout turbidity currents broke seabed telecommunication cables again on March 9th 2020 and 28-29th April 2021 (Supplementary Table 1), whilst the mooring array recorded twelve slower and shorter runout turbidity currents within the upper canyon during a ~4 month period in 2019-20 (Fig. 2). The scale of turbidity currents ensures that the sediment-mass carried by these ows rivals that of any other process on Earth 1,13 , including rivers 6 or glaciers, or settling from the surface ocean 14 (Table 1).
Turbidity currents are thus important for a wide variety of reasons. For example, turbidity currents play a globally signi cant role in terrestrial organic carbon burial 15 that affects atmospheric CO 2 levels on long time scales, and other global geochemical cycles. It was once thought that terrestrial organic carbon was incinerated primarily on continental shelves 16 . More recent studies 15,17 proposed that transfer and burial of terrestrial organic carbon in the deep-sea by turbidity currents could be highly e cient, based on similarities in organic carbon abundances, compositions and ages within sediment samples collected from river-mouths and deep-sea oor. However, these studies 15,17 did not document how such e cient sediment and organic carbon transfer actually occurred. Here, we use novel direct observations to explain why transfer of sediment and associated organic carbon from rivers to the deep-sea can be so e cient.
Organic carbon forms the basis for most sea oor food webs, and rapid and sustained deposition of organic-rich sediment by turbidity currents can creates distinct ecosystems, such as at the end of the Congo system 18, 19 . These sometimes very powerful ows can also scour life from oors of submarine canyons 20 , and this study therefore also illustrates how turbidity currents affect deep-sea life in disparate ways.
Turbidity currents are also important geohazards 21 . In particular, they break seabed telecommunications cable networks that now carry over 99% of intercontinental data tra c 22 , which underpin the global internet and many other aspects of our daily lives worldwide 23 . The January 2020 ow described here broke both telecommunication cables (Figs. 1 and 2) connecting to West Africa, causing the internet to slow signi cantly from Nigeria to South Africa 9 , and these cables were broken again by turbidity currents in March 2020, and April 2021 (Supplementary Table 1), including during coronavirus (CoV-19) related lockdown when internet bandwidth was particularly critical. These cables had not been damaged by turbidity currents in the previous 18 years. Understanding why these cables broke in 2020 and 2021 is crucial for assessing the hazard to submarine cables and other seabed infrastructure. This paper will also address why turbidity currents broke some cables, but left intervening cables intact 9 (Figs. 1 and 2), as seen in other locations worldwide 24,25 . It has been proposed that turbidity current deposits (called turbidites) can provide long term records of other major hazards, including earthquakes, typhoons or river oods 26-28 . Such records are potentially valuable, as they extend further back in time than most records on land. This study provides the most detailed information yet on how long runout turbidity currents are related to river oods, and how oods are recorded in the deep-sea.
Despite their importance, there are remarkably few direct measurements from turbidity currents, ensuring they are poorly understood 1 . This is a stark contrast to far more numerous and widespread direct measurements from of other major sediment transport processes 6,14,29 . Recent pioneering work has shown how short runout (< ~50 km) turbidity currents can be monitored in shallow water, typically using moorings with sensors, such as acoustic Doppler current pro lers (ADCPs) that measure pro les of ow velocity and sediment backscatter [30][31][32][33][34][35] . However, detailed monitoring is still only available for turbidity currents at <10 sites worldwide, all in water depths of <2 km [30][31][32][33][34][35] . Previously there were no detailed direct measurements from turbidity currents that ran out for >50 km, which ush submarine canyons and dominate longer-term sediment transfer. This situation ensured that fundamental questions remain. For example, previous studies showed that major earthquakes can sometimes trigger canyon-ushing turbidity currents that carry very large amounts of sediment 4,20 . However, it was not clear if river oods could also generate such large canyon-ushing events, and therefore how major rivers are connected by turbidity currents to the deep-sea. It was also theorised that turbidity currents might potentially behave in a very different way to rivers; as turbidity currents that erode the seabed could become denser and faster, and erode yet more sediment and become even denser, causing turbidity currents to self-accelerate or 'ignite' 36 . However, sustained ignition was yet to be documented clearly in submarine ows 35

Aims
Here we present the most detailed measurements yet from turbidity current in the deep (2-5 km) sea, which combines information from cable breaks with that from 11 moorings along a 900 km length of the Congo Canyon and Channel 9 (Fig. 1). The overall aim of this contribution is to understand how turbidity currents can connect major river mouths to the deep-sea. To do this, we address two speci c questions. First, how are unusually powerful and long runout turbidity currents initiated that ush submarine canyons, and what controls their timing and frequency? In particular, how are ows linked to major river oods and tidal cycles. Second, how do turbidity currents then behave, and why do some ows accelerate and runout much further than others? Answers to these questions underpin a generalised model for how turbidity currents transfer globally signi cant sediment volumes from major rivers to the deep-sea. Finally we outline the wider implications of this study, for e ciency of organic carbon transfer to the deepsea 15,16 , predicting future hazards to seabed telecommunication cable networks 9,22,23 , and how future climate or land-use changes may impact the deep-sea.

Study Area
The head of the Congo Submarine Canyon lies within the estuary of the Congo River (Fig. 1d), which has the second largest water discharge and fth largest particulate organic carbon export of any river 6 . The canyon incises deeply into the continental shelf and slope, before transitioning in a less-deeply incised conduit termed a submarine channel 37,38 (Fig. 1a-c). The channel terminates at a water depth of ~4,800 m, beyond which there is an area of sediment deposition termed a lobe 37,39 . Previous deposit-based studies suggest that long-term sediment transfer through the canyon and channel is e cient, with ~30% of the total sediment mass located in lobe deposits 40 . Exceptionally rapid deposition of organic carbonrich (3-4% TOC) sediment of mainly terrestrial origin (70-90%) leads to e cient organic carbon burial on the lobe 41 , with methane-rich uids due to diagenesis of this organic matter leading to unusual chemosynthesis-based ecosystems 18, 19 .
Past work along the Congo Canyon produced some of the rst measurements from turbidity currents, albeit with less-detailed sensors 38,42 , or at just one site in the upper canyon [31][32][33] . Initially, current meters recorded velocities at a single height above the seabed, at three sites along the canyon-channel 38, 42 . (Figs. 1 and 2). Eleven ADCP-moorings were deployed at depths of 1,650 to 5,000 m ( Fig. 1), with each mooring containing one or more downward-looking ADCPs, located 30 to 150 m above the seabed 9 ( Supplementary Fig. 1).

Initial causes of powerful and very long run-out turbidity currents
A series of 12 ows restricted to the upper canyon were recorded by ADCP-moorings between September 2019 and early January 2020 (Fig. 2), causing three moorings to break. A much longer and more powerful ow then occurred on 14-16th January 2020, breaking the eight remaining moorings and two seabed telecommunication cables in a sequence from shallower to deeper water ( Fig. 2; Supplementary Tables 1   and 2). Data from 9 of 11 ADCP-moorings were recovered successfully, despite considerable challenges as surfaced moorings drifted across the sea-surface, amid CoV-19 related lockdowns. Further cable breaks due to turbidity currents occurred on March 9th 2020, and April 28-29th 2021 (Supplementary  Table 1). There had been no cable breaks in the preceding ~18 years, despite one or more cables being present in the canyon during this period (Supplementary Table 1), suggesting that the three cablebreaking turbidity currents in 2020-2021 were unusually powerful.
None of the turbidity currents recorded by the ADCPs or cable breaks coincided with earthquakes, and there is no clear relation to offshore wave heights (Supplementary Material). However, these cable breaking ows are associated with the largest oods of the Congo River since the early 1960's, and they occurred after an 18 year period without cable-breaks or comparable oods. A 1-in-50 year ood 43 occurred with a peak discharge of ~70,883 m 3 at Kinshasa on December 21st 2019 (Fig. 4), with the ood peak 43 most likely arriving ~2-4 days later at the river-mouth estuary. The rst cable-breaking ow occurred on January 14-16th, two weeks after the ood peak when river discharge was high, and the arrival times of this turbidity current were captured by eight ADCP-moorings just before they broke. The second cable-breaking ow on March 9th 2020, occurred 10 weeks after the ood peak whilst river discharge was much lower (Fig. 3b). A second major (1-in-20 year) ood 43 occurred the following year, with a peak discharge of 67,210 m 3 in Kinshasa on December 13th 2020. This was followed by a third cable-breaking ow on April 28-29th 2021, some 4.5 months after the December 2020 ood. However, there were signi cant time delays between the ood peaks and the cable-breaking ows, with all three cable-breaking ows coinciding with subsequent spring tides (Fig. 4). It appears that oods supplied large amounts of sediment that primed the river mouth to produce powerful and long runout ows (Fig. 3), which were triggered nally at spring tides (Fig. 4).

Subsequent ow behaviour
Changes in turbidity current transit (front) speeds, and ow behaviour, are documented by arrival times at ADCP-moorings and cable breaks. These data show that the front of the January 14-16th turbidity current progressively accelerated as it ran out for over 1,130 km (Fig. 5a). The ow front initially moved at 5.0-5.2 m/s for its rst 500 km, before reaching a velocity of 8.2 m/s over 1,000 km from source, albeit with a small decrease in front speed between ~880 and 1,000 km (Fig. 5a).
A further 13 ows were recorded by ADCP-moorings between September 2019 and January 2020 (Figs. 2 and 5). Twelve of these ows terminated in the upper canyon, and these events had front velocities of <4 m/s (Fig. 5a). One ow on January 5-15th travelled for >800 km, initially with a front speed of 4.4 m/s, but this ow decelerated to speeds of <1 m/s in deep-water, and terminated before the nal mooring (Fig. 5a). Cable breaks on 28-29th April 2021 recorded a long runout ow travelling at 4.0 m/s, although no ADCP-moorings remained to capture this event in detail (Fig. 2). Thus a broad pattern emerges; ows with initial front speed exceeding 4 m/s ran out for long distances (>1,000 km), and accelerated if their show that ~1.04 km 3 was eroded from resurveyed reaches of the upper canyon and deep-water channel (Fig. 7). The resurveyed reaches comprise only ~50% of the total canyon-channel length (Fig. 1a), so the total amount of eroded sea oor sediment may be ~2 km 3 . This is an exceptionally large amount of sediment. For comparison, the total annual sediment ux from all of the world's rivers is ~7 km 3 (Table   1) 6 . The unusually powerful and long runout turbidity currents in January and March 2020 presumably caused this erosion. The amount of sediment eroded along the ow pathway most likely greatly exceeds that initially within the ow, as the eroded volume is ~35-70 times the average annual sediment supply from the Congo River (0.028 km 3 ) 6 .

Discussion
Here we discuss the rst detailed direct measurements from turbidity currents in the deep (> 2 km) sea. These unique measurements show how sediment can be transferred e ciently from a major river mouth to water depths of ~5 km, by the longest runout sediment ows (of any type) yet measured in action on Earth.
It was previously known that major earthquakes could generate long runout turbidity currents that transfer very large volumes of sediment to the deep-sea 4,20 . However, here we document directly for the rst time that major river oods also generate extremely large turbidity currents that ush submarine canyons. Indeed, the turbidity currents that ushed the Congo Canyon-Channel in January and March 2020 eroded ~1-2 km 3 of sediment from the seabed. This volume is equivalent to 14-28% of the global annual sediment ux from all rivers 6 , and it was carried down a single submarine canyon-channel, probably by just two turbidity currents (Fig. 2). The 1929 event in the NW Atlantic 4 involved a much larger sediment volume (>200 km 3 ), but the amount of sediment carried by the ood-related events in Congo Canyon rivals or exceeds other canyon-ushing ows due to earthquakes, such as those offshore New Zealand in 2016 (~1 km 3 ; M w 6.8 Kaikōura earthquake 20 ) or Japan in 2011 (0.2 km 3 ; M w 9.1 Tōhoku earthquake 44 ). The turbidity currents in 2020 and 2021 that ushed the Congo Canyon-Channel were linked to two river oods with recurrence intervals of 20 and 50 years 43 . This ood recurrence interval is signi cantly shorter than recurrence intervals of major earthquakes (100-300 years) that were previously proposed to trigger canyon-ushing events elsewhere 20,44−48 .
Turbidity currents that ushed the Congo Canyon were initiated by a combination of oods and spring tides. Past work has shown how elevated river discharge and tides can combine to generate much shorter runout (1-50 km) turbidity currents offshore from smaller rivers 30,49−51 . However, this is the rst study to show that oods and tides generate far larger turbidity currents offshore from one of the world's largest rivers, and in an estuarine setting. This suggests that oods and tides may trigger turbidity currents in a wider range of settings than previously thought, which then transfer globally signi cant sediment volumes.
A notable observation is that delays of several weeks to months can occur between an initial river ood and turbidity currents that ush the Congo Canyon (Fig. 3). Previous work has documented signi cant delays between river oods and associated turbidity currents, but only for hours 50 to days 24,25 , not weeks to months. Moreover, older cable breaks (1883 to 1937) in the Congo Canyon indicate that clusters of cable breaks may occur after one or more years of elevated river discharge 9 (Supplementary Fig. 2). This suggests that the Congo Estuary can store ood sediment for up to several months, and maybe years, and thus act as an e cient 'capacitor', before eventually releasing sediment in one or more long-runout turbidity currents.
Past work on how turbidity currents are generated by oods often focussed on a model in which the oodwater has enough sediment to become denser than seawater, so that the river-plume plunges to move directly along the seabed as a 'hyperpycnal ow' 27,28,52 . However, this model can be ruled out for the turbidity currents that ushed the Congo Canyon, because of the signi cant delay between peak ood discharge and these submarine ows (Fig. 3). The Congo River also has relatively low suspended sediment concentrations, making it unlikely to trigger hyperpycnal ows 53 .
However, two other models could explain how oods and spring tides combine to generate these canyonushing ows. In the rst model, major oods drive large amounts of sand-dominated bedload across the submarine canyon head ('x' in Figure 1d). This causes the canyon-lip to prograde rapidly, and then collapse, thereby forming a powerful turbidity current 30,50 . However, a signi cant time delay occurs between ood peaks and all three canyon-ushing ows (Figs. 3 and 4). Thus, although rapidly deposited ood-sediment may prime the canyon-head for failure, it must remain close to failure for weeks to months after the ood, until a minor perturbation associated with spring tides triggers nal failure 30,50 .
Those perturbations might be due to expansion of gas bubbles in sediment 54 , or increased bedload transport at a spring ebb tide 55 .
A second model is that major river oods supply large amounts of ne-grained mud, which is then stored within the river-mouth estuary for weeks to months, before being released at spring tides ( Supplementary   Figure 4c and 5). This mud is initially dispersed via surface plumes 56 (Fig. 1d), but settles onto the seabed across the entire estuary (Fig. 1d). Field observations (R. Nunny, pers. comm., 2021) from an extensive shallow water plateau upstream of Soyo ( Supplementary Fig. 4a) show that a mud layer accumulates throughout the year (Fig. 5). During periods of elevated river discharge, and especially when spring ebb tides also occur, the freshwater plume touches-down across this shallow water plateau. This causes mud to be resuspended, forming highly-mobile uid-mud layers 57 , which can be several meters thick (Supplementary Figs. 4 and 5). These uid-mud layers then drain into tributary canyon-heads, where they may either directly generate turbidity currents, or produce unstable deposits that fail to produce even larger turbidity currents ( Supplementary Figs. 4 and 5). Near-bed estuarine circulation may also help to trap ne sediment in this second model 58 . It is unclear which of these two models generated the canyon ushing turbidity currents in 2020 and 2021, as we lack suitably detailed observations of what occurred at the river-mouth, and the two models are not mutually exclusive.
To understand how turbidity currents transfer sediment from river-mouths to the deep-sea, we also need to understand why some turbidity currents increase in power and runout for exceptional distances into the deep-sea, whilst other ows terminate in shallow water. It has been theorised that turbidity currents which erode sediment become denser, and thus accelerate, causing increased erosion, and further acceleration (a process termed 'ignition' 36,59−60 ). Alternatively, turbidity currents that deposit sediment decelerate, leading to further deposition ('dissipation'). These positive feedbacks could produce thresholds in behaviour that depend on small differences in initial ow state 36, 59-60 . It has also been proposed that ows could achieve a near-uniform state in which erosion is balanced by sediment deposition, termed 'autosuspension' 36, 59-60 . However, it was previously contentious whether ignition or autosuspension were reproduced in relatively slow laboratory-scale turbidity currents 61-62 , and ignition had not been documented clearly in the eld. This is rst study to document unambiguously that full-scale turbidity currents can ignite in the oceans, and that ignition can occur over exceptionally long (1,000 km) distances (Fig. 5a). This acceleration cannot be explained by changes in seabed gradient that decreases with distance, or narrowing of the canyon-channel (Fig. 5b,c), and it is associated with very large (1-2 km 3 ) volumes of seabed erosion. However, once ows erode and accelerate, it is important to understand how rapidly ignition occurs, and whether ows then tend towards a new equilibrium state. The rate at which the front of the January 2020 ow accelerated is relatively slow, as it took ~1,000 km to accelerate from 5.0 to 8.2 m/s, despite a very large amount (1-2 km 3 ) of seabed erosion. Front speeds were sometimes relatively constant for long distances, suggesting a near-uniform frontal state. This near-equilibrium may arise if the mass of sediment eroded and incorporated into the dense frontal part of the ow ('frontal cell' 32 ), is balanced by loss of comparable sediment-mass from the frontal cell into the slower-moving body (Fig. 5a). Local decreases and increases in front speed may be due to patchy seabed erosion.
Spatial changes in submarine turbidity current speeds have only been measured in detail at ~5 locations 25,34,35,63,64 (Fig. 6) However, a comparison between these data suggests that a common pattern emerges for ows con ned in canyons and channels 34 , which may be a fundamental property of turbidity currents (Fig. 6). Flows with initial front speeds in excess of ~4 to 5 m/s tend to runout for longer distances.
These ows either sustain speeds of 5-8 m/s (autosuspend), or accelerate from ~5 to 8 m/s (ignite), albeit it for variable distances of 30 to 1,100 km (Fig. 6). It is these ows that carry sediment and organic carbon much further and pose the greatest hazard. Conversely, ows in canyons and channels whose fronts travel at <4 m/s tend to decelerate and dissipate, as do initially faster ows (e.g. NW Atlantic turbidity current) which exit canyons and channels, and then decelerate markedly as they become uncon ned and spread laterally 4 .
Previous theory predicts that sediment grain size, and thus settling velocity, plays a key role in determining whether a turbidity current ignites or dissipates 36,60−62 . Thus, a notable new result is that similar threshold initial front speeds (4.5 to 5 m/s) for ignition are observed in locations with very different grain sizes (Fig. 6). For example, the Congo Canyon is fed by a ne-grained and muddy river 37 , and the canyon oor is often mud-dominated 32 , whilst at the other end of the spectrum, Monterey Canyon is fed beach-sand via long-shore drift 34,35 and has a sandy oor 34 . It thus appears grain size is a weak control on ow speed needed for ignition. Previous theories for ignition are based on energy balances 59,60 or series of equations 36 that often assume ows are dilute (<< 10% sediment volume), such that sediment grains settle individually, although settling may become hindered as sediment concentrations increase.
An alternative model is proposed here (also see 35 ) in which faster turbidity current fronts comprise a dense (>20-40% volume) near-bed layer, in which grains do not settle individually, and which is weakly turbulent. Field evidence from Congo Canyon and elsewhere suggests faster (>4-5 m/s) turbidity currents contain such a dense near-bed layer at their front, whilst slower moving (<2-3 m/s) ows lack a dense layer 33,34,63 . Behaviour of this dense layer may depend more on variations in excess pore pressures, or rates of sediment erosion from the bed 65,66 , rather than the settling velocity of individual grains. Indeed, experiments show how substrate character and erosion processes can determine whether a dense sediment ow grows and accelerates, or dies out 65,66 .
We now present a new generalised model for how turbidity currents transfer globally signi cant volumes of sediment from a major river to the deep-sea, through a submarine canyon (Fig. 8). Previous studies suggested that frequent and smaller turbidity currents deposit sediment within canyons, which are then ushed by much more infrequent and powerful ows, which occur every few thousand years and are most likely triggered by earthquakes 4,20,44−48 . Here we show that numerous smaller ows in ll the Congo Canyon; indeed these ows are active for 30% of the time in the upper canyon 32,33 (Fig. 2a). Far more powerful and infrequent ushing events then excavate very large volumes (1-2 km 3 ) of sediment from the canyon-channel oor (Fig. 7). However, contrary to previous models [37][38][39][40][41][42][43][44][45][46][47][48] , this study shows that canyonushing events can be triggered by major river oods as well as earthquakes, with clusters of multiple canyon-ushing events occurring after a single major ood over a period of weeks to months. Recurrence intervals for these major oods is 20-50 years 43 , whilst previous work documented ushing events every few hundred to thousand years [45][46][47][48] . It also appears that the amount of sediment carried into the deepsea by a ushing event is comparable to that supplied by the Congo River between ushing events. The Congo River supplies ~43 Mt of sediment each year 6 , so the volume of sediment (1-2 km 3 ) excavated by the 2019-20 canyon ushing ows is comparable to sediment supply from the river over the last ~35-70 years, assuming a density of ~1,500 kg/m 3 for eroded seabed material. Thus, although sediment is mainly stored for up to several decades in the upper canyon, it is then e ciently ushed beyond the canyon-channel into the deep-sea (Fig. 8).
This new understanding of how river mouths are connected to the deep-sea by turbidity currents (Fig. 8) explains why organic carbon transfer and burial can be highly e cient 15,17 . Fresh organic carbon from major oods can reside in the river-mouth for only a few weeks or months before being ushed into the deep-sea, together with a far larger volume of organic carbon from canyon-lling deposits that accumulated over several decades. The supply of organic carbon by turbidity currents can also have profound impacts on seabed life. For example, distinctive chemosynthesis-based ecosystems occur on the lobe fed by the canyon-channel, where sediments rich in (mainly terrestrial) organic-carbon are rapidly buried 18, 19 . This new study illustrates how large amounts of organic-matter-rich sediment are delivered episodically to this lobe. It also emphasises how turbidity currents physically disturb benthic fauna, as tens of meters of sediment may be removed locally along the canyon-channel oor (Fig. 7).
Seabed telecommunications cables now carry >99% of global data, underpinning daily lives 23 . Cable routes are generally chosen to avoid submarine canyons, but this is not always possible, such that ~2.8% of global cables involve canyon crossings. Cable-breaking ows in this study are associated with exceptional oods along the Congo River, and these oods may thus provide an early warning of elevated risks to cables. This elevated risk may persist for a signi cant period after the ood peak, and a single major ood can generate multiple cable-breaking ows (Fig. 8). A key decision for cable routing is how far offshore the cable should be located from the river-mouth. The overall frequency of turbidity currents decreases strongly with distance as initially slower events dissipate within the upper canyon. However, the January 2020 event also shows how infrequent and longer runout turbidity currents may selfaccelerate with distance (Fig. 5). Thus, only for larger and more infrequent ows, there may be an increased hazard to cables located further offshore, as they will experience the fastest ow-front speeds in such events.
This study indicates turbidity currents with frontal speeds in excess of ~5.5-6 m/s (Fig. 5a) are needed to damage cables, and this is broadly consistent with information from cable breaks elsewhere 25 (Fig. 6).
However, a notable observation is that although some cables broke in the January and March 2020 and April 2021 ows, other cables survived despite being impacted by turbidity currents with similar front speeds (Figs. 3 and 5; Supplementary Table 1). Thus, there are local conditions that can prevent a cable from breaking, whilst neighbouring cables break. This in turn suggests there may be ways to route cables in more advantageous positions to reduce cable breaks. Time-lapse surveys may provide an explanation for why some cables break, whilst others do not. These surveys show that seabed erosion during turbidity currents is very patchy, over distances of just a few kilometres (Fig. 7). In particular, deep (20-40 m) erosion may be associated with knickpoints 67,68 , de ned as zones of locally steeper gradients along the canyon or channel oor (Fig. 7), and such localised deep erosion will tend to undermine cables and cause breaks or faults 9 .
It has previously been suggested that turbidity current deposits (turbidites) may provide a record of major oods 27 , which could be valuable if it goes further back in time than records on land. This study con rms that major river oods can indeed be recorded by deposits of unusually powerful and long runout turbidity currents, although a single major ood can generate multiple turbidity current deposits. The best submarine record of major oods is located near the end of the canyon-channel system, as smaller-scale turbidity currents complicate the ood-record closer to the river-mouth (Fig. 8).
Finally, this study provides the clearest evidence yet that river oods can directly and rapidly impact the deep-sea 28 , weeks to months after a ood. Climate change is predicted to produce a more active hydrological cycle, with global changes to ood frequencies 43,64 . Future changes in Congo River discharge are uncertain but potentially signi cant 65 . Here we show how such changes in terrestrial climate and river-ood frequency may affect how organic carbon is ushed into the deep-sea, associated functioning of deep-sea ood webs, and hazards faced by sea oor cables. Dam construction, deforestation and changes in land-use can also substantially affect sediment ux to river-mouths 66,67 , and this too may change the frequency of turbidity currents. This study of the longest runout sediment ow yet measured in action thus illustrates why changes affecting terrestrial continents may also have signi cant impacts for the deep-sea oor.

Field deployment of moorings
Eleven moorings with ADCPs were deployed ( Supplementary Fig. 1) at points along the oor of the Congo Canyon-Channel 9 (Fig. 1), with locations con rmed to within +/-~15 m by ultra-short baseline acoustic positioning. Three moorings were damaged by smaller ows in the upper canyon, and surfaced before a much larger turbidity current occurred on January 14-16th 2020 (Fig. 2). The remaining eight moorings surfaced on January 14-16th due to this exceptionally powerful cable-breaking ow (Fig. 2). Nine of the 11 moorings were then eventually recovered via emergency vessel charters.

Arrival times of turbidity currents at moorings and cables
The arrival times of turbidity currents at ADCP-moorings were de ned using the time series of velocity pro les recorded by 75, 300 and 600 kHz ADCPs every 11-to-45 seconds (Supplementary Table 2). The arrival times of turbidity currents were marked by an abrupt increase in near bed velocities above ambient values of ~0.3 m/s. The timing of faults on submarine telecommunication cables were also used to de ne turbidity current arrival times (Supplementary Table 1), and this assumes the cables were damaged by the arrival of the ow front. Cable breaks were recorded to the nearest minute.

Flow front (transit) speeds between moorings or cables
The speed of the ow front between moorings or cables was calculated by dividing the distance between sites and the difference in arrival times. Distances were measured along the oor of the canyon-channel using bathymetric survey data.

Time at which turbidity currents are triggered
The rst mooring is located ~80 km from the river mouth (Fig. 1). It was thus assumed that turbidity currents originated at the mouth of the Congo River, and that the ow speed from the river mouth to the rst mooring was the same as that between the rst and second moorings. For faster moving turbidity currents with speeds over 2-3 m/s between the rst two moorings, the uncertainty of when the ow originated is likely to be less than a few hours (i.e. the time taken for the ow to travel 80 km at speeds of >4 m/s). Thus, although the original times of these turbidity currents cannot be reliably compared to individual low and high tides, those times can be compared to longer term cycles of spring and neap tides. Uncertainties in the time taken by ows to travel from the river mouth to the rst mooring site are much larger for slow moving ows, and may be several days for ows travelling at <1 m/s (and see Supplementary Information). Thus it is more challenging to determine if these slower moving ows are also triggered by spring-neap tidal cycles, and they too cannot be linked to individual low or high tides.

River discharge
The timing of turbidity currents was compared to uctuations in water discharge from the Congo River at the Kinshasa gauging station 43 (Fig. 3), located ~400 km from the river mouth, as measured by the Règie des Voies Fluviales (RVF) at Kinshasa, Democratic Republic of Congo.
Tidal elevations at the river mouth Daily tidal data (Fig. 4) were obtained for Santo Antonio do Zaire near the port of Soyo, at the Congo River mouth (Fig. 1a).
Time-lapse sea oor surveys and eroded volumes Swath multibeam surveys of sea oor bathymetry were collected in September 2019 and October 2020 using a Kongsberg EM122 (1° x 1°) system operating at 12 kHz for two areas (Fig. 1a). Highest resolution data was generated by setting the swath width to the narrowest setting (45° from the nadir), and having large overlaps between adjacent swaths. Sound velocity pro les (SVPs) were taken through the water column at the start of most surveys, and a second SVP was performed halfway through some longer survey. The rst area of repeat surveys was along the upper canyon in Angolan waters, whilst the second area was the deeper-water channel in international waters (Fig. 1a). The accuracy of these sounding data were <0.3-0.5% of the water depth.
The multibeam bathymetric data were processed in CARIS HIPS and SIPS and corrected for the ship's motion and for differences in sound velocity in the water column (using SVP data), before being gridded with a horizontal grid cell resolution of 5-15 m that depended on water depth. A bathymetric difference map was then produced by subtracting October 2020 bathymetric data from September 2019 bathymetric data. The total volume of eroded sediment was determined from the sum of the (negative) vertical difference in elevation for each grid cell. This value for the difference in seabed elevation at each grid cell was multiplied by the grid cell area, and then summed, to derive the volume of eroded sediment.
The eroded volume of 1.0 km 3 was derived only from the canyon and channel oors in the two areas that were repeat surveyed in September 2019 and October 2020. This ensured that areas with the largest likely errors due to steep topography (i.e. the canyon walls) were not included. The rst area comprises the upper Congo Canyon, whilst the second area is located along the distal deep-sea channel (see Fig. 1a).  Timing and runout distance of turbidity currents measured from September 2019 to January 2020 along the Congo Canyon and Channel system. (a) ADCP time series of velocities measured at mooring M9, showing occurrence of turbidity currents. (b) Plot of event timing against distance from Congo River mouth, as measured along the sinuous canyon-channel axis. Red vertical lines denote ow events, and indicate their runout distances, with the most powerful January 14-16 th event in bold. Dotted horizontal lines denote a mooring site or submarine cable. The times of mooring deployment are shown, together with when moorings or cables broke due to turbidity currents. Two moorings (M4 and A3) were not recovered ('NR'); ow timings at these two sites are derived from when mooring reached the ocean surface, and an assumed rise rate of 150 m/min (as seen during earlier work). Cable breaking turbidity currents coincide with spring tides. (a) Time series of velocity pro les recorded by an ADCP at mooring site A2 in 2019-20 (Fig. 1), with warmer colours indicating turbidity currents.
Superimpose are the arrival timings of long-runout, cable-breaking turbidity currents on January 14 th and March 9 th 2020 (red lines), and slower moving ows restricted to the upper canyon (thin black lines). (b) Time series of daily maximum tidal range, and daily lowest low tide, at Soyo in the estuary at the mouth of the Congo River. (c) Box and whisker plots showing median, rst and second quartiles of daily tidal range values for (i) all days in which ADCP-moorings were in the Congo Canyon in 2019-2020, (ii) days on which turbidity currents occurred at the ADCP moorings, (iii) days on which no turbidity currents occurred at ADCP moorings, and (iv) days on which the ve fastest non-cable breaking ows occurred at ADCP moorings. Each box and whisker plot shows the median tidal range (x), tidal ranges on given days (o), and the 95% percentile of the distribution of tidal ranges for speci ed days (-). Stars indicate the maximum daily tidal range for the days on which the 3 cable-breaking ows occurred on January 14-16 th and March 9 th 2020, and April 28-29 th 2021. (d) Time series of daily tidal coe cients at river mouth (Soyo) showing times of three cable-breaking turbidity currents (red stars), and peak of major river oods (blue arrows). The larger the tidal coe cient, the greater the tidal range. Period in which ADCP-moorings deployed shown by yellow box.  Table 1), and the speed of a turbidity current between moorings in 2004 42 . Flows with front speeds >4-5 m/s (grey box) tend to self-accelerate or sustain those front speeds over long distance, whilst ows with front speeds < 4 m/s tend to decelerate and dissipate.
(b) Changes in water depth and (c) sea oor gradient with distance along the oor of the canyon-channel.
(d) Changes in canyon-channel width with distance measured at crests of con ning levees or rst terrace.

Figure 6
Changes in turbidity current front (transit) velocity with distance for ows that are con ned in canyon and channels, from the four locations worldwide where such data are currently available. These eld sites are   Generalised model for how the turbidity current pump operates from river mouths to the deep sea, showing ow timing and frequency, and spatial behaviour and evolution. (A) Schematic pro le along a generalised submarine canyon-channel from the river mouth to deep-sea. Numerous smaller-scale turbidity currents that in ll the canyon (in blue). Much more infrequent, powerful and longer runout turbidity currents then erode the sediment in ll from the smaller ows, and ush the canyon (in red). (B and C) Time series (vertical axis) showing a sequence of smaller canyon lling ows (in blue) and larger canyon ushing ows (in red), based on this study of the Congo system. Part B shows canyon lling and ushing ows over a ~2 year period, together with river oods and tidal cycles. Canyon ushing ows occur 2 weeks to 5 months after major oods (Fig. 3), and coincide with spring tides (Fig. 4). Part C shows a longer 100 year period in with canyon ushing ows are associated with major oods occurring every 20-50 years.