Two distinct types of turbidity currents observed in the Manila Trench, South China Sea

Abstract

Sediment gravity flows are the most direct and efficient transport mechanisms for moving terrestrial sediments into deep oceans. Scarcity of firsthand measurements, however, has hindered the quantitative, even qualitative characterization of such flows. Here we present a unique year-long data record from ~4000 m depth in the Manila Trench that captured two very different gravity flows in terms of their hydraulic and sedimentary properties. The first flow was of slow speed (~40 cm s⁻¹) and long duration (~150 h), thus nicknamed ‘Tortoises’, and carried very fine sediment with low concentration (~0.01%). The fast (~150 cm s⁻¹) but short-lived (~40 h) flow, nicknamed ‘Hares’, carried much coarser sediment with higher concentration (~1.2%). Clay mineral compositions suggest that the ‘Tortoises’ originated from upstream canyon wall slumping, whereas the ‘Hares’ was likely submarine canyons southwest of Taiwan Island due to typhoon. Grain size is a key factor in determining evolution of turbidity currents.

Introduction

Oceanic turbidity currents are sediment-laden, gravity-driven, turbulence-supporting underflows primarily active at river delta fronts and in submarine canyons^1,2,3. They are known to transport large quantities of terrigenous materials (e.g., sediments and organic carbon) into the deep^4,5. Turbidity currents’ power and unpredictability make them difficult to measure directly in the field. Their flow properties, e.g., speed, thickness, sediment concentration, grain size etc., are most commonly either evaluated from laboratory experiments^6,7 or inferred from turbidite deposits at outcrop or in the deep sea⁸. In general, turbidity currents can be placed broadly into two different categories: surge-like flows with peak velocity waxing and then waning rapidly, and a prolonged quasi-steady flow that, as its name suggests, continues for much longer time. Both types of flows were well investigated in laboratory flumes^9,10 or numerical modelling^11,12, but it is not until the Acoustic Doppler Current Profiler (ADCP) became available, detailed hydraulic and sedimentary properties of field scale turbidity currents were directly measured¹³. The Monterey Submarine Canyon and Congo Submarine Canyon, two locations with very different river discharge and sediment sources, produced probably the most complete field data for the surge-like and prolonged quasi-steady flows^14,15. Yet, it remains a difficult task to make a direct comparison study of the two flows because of the different context within which these two types of flows have been observed. This paper presents results from a field investigation where both types of flows were directly measured on the same subsurface mooring in the Manila Trench. Here, we describe the detailed measurements and characteristics of the two types of flows. We then discuss the influence of tidal currents and grain size of particles on the evolution of the turbidity currents, infer their provenances and triggers, and calculate the sediment flux into the deep part of the Manila Trench. Lastly we describe methods of flow measurement and sediment sampling in the field, and subsequent analysis.

The Manila Trench

Located in the northeastern part of the South China Sea, the Manila Trench, is tectonically part of the subduction system where the Sunda plate in Eurasia goes beneath the Luzon volcanic island arc^16,17. The trench begins at the confluence of three submarine canyons (Gaoping, Penghu, and South Taiwan Shoal) and continues to the south for more than 350 kilometers, with a maximum water depth of nearly 5400 m (Fig. 1). Until recently, direct measurements of turbidity current in the Manila Trench are still very limited, even though existing geologic records, such as sediment waves around the trench^18,19,20 and core logs in the trench²¹, have revealed the extensive and frequent occurrence of gravity-induced flows. Heavy rainfalls, brought by frequent typhoons²², and the high weathering and denudation rate generates huge flux of freshwater and sediments from Taiwan’s rivers into the ocean²³. Liu et al. ²⁴ discovered that hyperpycnal plumes in the Gaoping Canyon could ignite turbidity currents, which were later found to correlate well with typhoons passing through the region several times per year²⁵. Also, frequent earthquakes of various magnitude exacerbate the occurrence of landslides, terrestrial or submarine, that (1) markedly increase the sediment supply into the Manila Trench, and (2) directly triggers turbidity currents into the Manila Trench. A series of subsea cable breakage in the Gaoping Canyon and Manila Trench also documented rapid (5–16 m s-1, particularly 5–8 m s-1 in the Manila Trench), and long-runout (>300 km), sediment-laden flows associated with earthquakes and typhoons^26,27,28. None of the above studies, however, was designed to investigate the detailed hydraulic and sedimentary properties such as flow structures and grain size distribution inside the turbidity currents.

Fig. 1: Mooring locations in the Manila Trench.

a Bathymetric map of the northeast South China Sea showing the mooring locations. The white dashed lines indicate the Manila Trench and three submarine canyons upstream. The red stars denote the mooring locations (S1–S4). The area inside the yellow rectangle is zoomed in (b). The green dots are the tracks of Tropical Storm Lupit happened in August 2021, and the corresponding wind speeds are shown. The magenta dots show the earthquake source centers before occurrence of the turbidity currents. b Multi-beam bathymetric measurements showing the seafloor topography covering the mooring site S2 (indicated by the red star). The blue square indicates the box sampling site X02. And the black line indicates the location of cross section of (c). c Cross-trench section of the thalweg channel at the mooring sites S2 (the red star). d Schematic diagram of the mooring. The depth, height in meters above the seabed (MAB), and the corresponding instruments are indicated. The detailed information of instruments is introduced in materials and methods.

The Manila Trench observation program started in September 2019 when four subsurface moorings (S1–S4) were deployed along the trench (Fig. 1a). All moorings were recovered in August 2020. After all the data were downloaded and new batteries installed, two of those four moorings (S1 and S2) were redeployed at their original locations respectively. The redeployed moorings had the same configurations as before except that one turbidity sensor was added on each mooring at 72 MAB. These two moorings were scheduled for recovery in early August of 2021 but were delayed for a week till mid-August because of Tropical Storm Lupit. S2 was successfully recovered, but S1 encountered technical problems and remained on the seafloor. No turbidity current signal was found in any sensors of the four moorings during the 2020 deployment, probably because no typhoons made landfall in Taiwan in this period, and there were no earthquakes within the Gaoping drainage. Consequently, here we only present the data from the S2 mooring of the 2021 deployment. The S2 mooring, at 3808 m water depth, recorded data for nearly 12 months.

Results

Characteristics of the turbidity currents observed in the Manila Trench

Two turbidity current events, occurred in April and August respectively, and can be clearly identified in the field data (Supplementary Fig. 1). The turbidity current in April 2021 (hereafter E1, Fig. 2) began at 23:10 on March 31 (Beijing Time, UTC + 8) and lasted for about 6.3 days (150 h). Notably, all measured parameters clearly show that E1 pulsated several times in tidal frequencies: the ebb tidal phase (flowing down trench) correlated well with increasing current velocity, water temperature, and the thickness of turbidity current (Fig. 2d). The maximum speed of E1, 44.5 cm s⁻¹, occurred during the second pulse, i.e., 9 h after the flow’s arrival at the mooring (Fig. 2a). The acoustic backscatter intensity (a proxy of turbidity, Fig. 2b) and the measured turbidity from the Seaguard recording current meter system (RCM) at 12 MAB (Fig. 2c) correlated well with the pulsating velocities. However, the turbidity measurements from the turbidity sensor (RBR-TU) at 72 MAB only gradually increased as E1 phased out (Fig. 2c), suggesting a long-time lag before sediment particles from the turbidity current diffused upward to the RBR-TU sensor. The temperature slowly increased, also modulated by tide, by as much as 0.08 °C, against a background fluctuation of ~0.02 °C (Supplementary Fig. 1e), before returning to the pre-event value at the end of E1 (Fig. 2e).

Fig. 2: Plots of the time-series measurements for the turbidity current E1 at the mooring site S2.

a Along-trench current velocity (cm s⁻¹) and (b) net acoustic backscatter intensity (counts) (averaged over four beams) recorded by the downward-looking ADCP. The influence of water attenuation and spherical spreading had been corrected. The area under the black dashed line was affected by the side lobe interference. c Time-series of turbidity measured by the RBR-TU at 72 MAB (blue line) and by the turbidity sensor of RCM at 12 MAB (orange line). d Along-trench current velocity (cm s⁻¹) measured by the RCM at 12 MAB (orange line), along-trench current velocity at 12 MAB recorded by the downward-looking ADCP (dark blue line), and predicted along-trench tidal current velocity (black line). e Time-series of temperature measured by CTD at 17 MAB (blue line) and by the sensor of RCM at 12 MAB (orange line). The red dashed lines indicate the beginning and end of the turbidity current.

The turbidity current in August 2021 (hereafter E2, Fig. 3) began at 14:12 on August 10 and lasted for 1.67 days (40 h). Pressure data measured by the Conductivity–Temperature–Depth recorder (CTD) at 17 MAB jumped instantaneously when E2 arrived (Fig. 3a), indicating that the mooring, at least the section below the sediment trap (see Fig. 1d), was severely tilted by the flow. The tilt angle reached the maximum value of 43° more than 2 h later (at 16:51) when the pressure perturbation was about 4.6 dbar (Fig. 3a). After the peak of the flow passed, the mooring remained tilted for nearly 8 h. Even after the mooring returned upright, the stabilized pressure was 0.3 dbar greater than the pre-E2 value, suggesting that the mooring might have been dragged to a slightly deeper location by the E2 flow. The instantaneous increase of the along-trench velocity (Fig. 3b) and backscatter intensity (Fig. 3c) from the ADCP marked the arrival of E2 at 14:12 August 10. While the backscatter intensity reached its maximum value at the beginning of the flow, the along-trench velocity did not. Rather, it stayed at an elevated speed of 20–60 cm s⁻¹ within a thin layer near the canyon seafloor for more than 40 min before it rose dramatically to a maximum speed of 145.1 cm s⁻¹ at 15:06 (Fig. 3b). Surge-type turbidity currents typically have their maximum speed at the head or the beginning of a Eulerian time-series data^6,29, thus the nearly 40 min of low velocity stage is rather unusual (see Discussion section). It is also worth noting that the small zone of low acoustic backscatter intensity inside the body of E2 from 15:00 to 19:00 on August 10 (Figs. 3c and  4c). This is probably caused by the high suspended sediment concentration above this zone which hindered the downward-looking ADCP’s acoustic penetration^30,31. Only the RBR-TU at 72 MAB recorded turbidity during E2 because the RCM ran out of battery. Similar to what is seen during E1, there is a long lag for the E2 turbid plume to reach the sensor at 72 MAB (Fig. 3d). The measured maximum turbidity in E2 (1333.2 FTU), however, is one order of magnitude greater than in E1. The sudden change in temperature around 14:57 also clearly marked the E2’s arrival (Fig. 3f). The temperature fluctuated around 0.03 °C above pre-E2 normal for nearly 4 days, long after the along-trench velocity had diminished.

Fig. 3: Plots of the time-series measurements of the turbidity current E2 at the mooring site S2.

a Pressure perturbation (dbar) measured by CTD at 17 MAB. b Along-trench current velocity (cm s⁻¹) and (c) net acoustic backscatter intensity (counts) (averaged over four beams) recorded by the downward-looking ADCP. d Time-series of turbidity measured by the RBR-TU at 72 MAB. e Along-trench current velocity (cm s⁻¹) at 12 MAB (dark blues line) recorded by the downward-looking ADCP and predicted along-trench tidal current velocity (black line). f Time-series of temperature measured by CTD at 17 MAB. The red dashed lines indicate the beginning and end of the turbidity current.

Fig. 4: Sediment concentration profiles of the turbidity currents E1 and E2 at the mooring site S2.

a Profiles of the inverted SSC for E1. The area under the black dashed line was affected by the side lobe interference. The yellow line indicates the local tidal oscillation (positive value is ebb phase). b Plots of the inverted SSC at 7, 12, 22, 42 and 62 MAB for E1, and the calculated SSC from RBR-TU at 72 MAB and RCM-TU at 12 MAB. Different colors represent the SSC in different heights those are indicated by white dashed lines in (a). c Inverted SSC profiles for E1. d Plots of the inverted SSC at 7, 12, 22, 42 and 62 MAB for E2, and the calculated SSC from RBR-TU at 72 MAB. Different colors represent the SSC in different heights those are indicated by white dashed lines in (c). Only the SSC after E1 and E2 happening was shown.

Suspended sediment concentration of the turbidity currents

Suspended sediment concentrations (SSC) were estimated from optical (turbidity sensor) and acoustic (ADCP’s backscatter intensity) data (see Methods section). The good correlation coefficients (>0.7) between the optically and acoustically converted SSC (Supplementary Fig. 2) suggest that these converted SSC are generally reliable even though the acoustic conversions are more dependent on particle sizes than the optical ones. E1’s maximum SSC (~0.01% or 0.26 g l⁻¹) was recorded at the lowest possible ADCP bin at 7MAB when the flow first arrived (Fig. 4a, b). Sensors at different heights ranging from 7 to 72 MAB recorded the sudden jump of turbidity almost simultaneously (Fig. 4b), suggesting that a thick, albeit slow, turbid plume engulfed the whole mooring. Close to the bed (indicated by ADCP at 7 MAB and RCM at 12 MAB; Fig. 4b), the SSC was relatively stable over the initial several days from April 1 to April 4—varying around 0.10 g l⁻¹ roughly in a frequency of local tidal oscillation. This variation with tide is more apparent in the upper part of the flow (Figs. 2, 4b). In contrast, E2’s SSC is much greater: the recorded maximum concentration at 7 MAB is ~1.2% (31.0 g l⁻¹), nearly 130 times larger than that of E1. Secondly, the fast but thinner flow arrived at the lower part of the mooring (7 MAB) nearly 4 h before the whole mooring was submerged by the turbidity current (72 MAB, Figs. 3, 4). The front of the flow arrived at 15:00 on August 10, well corroborated by the SSC (Fig. 4c, d) and the measured maximum velocity (Fig. 3b, e). The near-bed maximum concentration decayed rapidly over 10 h and then became relatively stable hovering around 1 g l⁻¹ for the next 3 days. Because of the poor signal-to-noise ratios mainly due to the side-lobe effect of the downward-looking ADCP, is 7 MAB. Thus the true SSC in the bottom 7 meters of the flow could be higher than the maximum SSC used in the above statistics.

Properties of sediment particles collected inside the turbidity currents

Both sediment traps collected sediment particles but the volume in the trap at 15 MAB is much greater than the one at 73 MAB (Fig. 5): the dry weights are respectively 1704 g and 32 g. In comparisons, the two sediment traps on a similar mooring placed at the same site for 1 year during a previous deployment (2019–2020) collected 6.1 g (15 MAB) and 7.4 g (73 MAB) of sediment. Given the fact that no turbidity current occurred during the 2019–2020 deployment, the large amount of sediment in the two traps during the 2020–2021 deployment can be interpreted as the result of the two turbidity currents in that period. It is also reasonable to designate the discontinuity in the trap sediment as the boundary between the two flows (Fig. 5b): accumulation below the unconformity resulted from E1 and accumulation above the discontinuity resulted from E2. Knowing the duration of the two flows and the respective dry weights of sediment collected during the two flows, the deposition rates during the two flows could be estimated: 312 and 5351 g m⁻² d⁻¹. These were respectively 0.97 × 10³ and 1.7 × 10⁴ times greater than the average deposition rate of the previous deployment (2019–2020) when no turbidity current occurred. The sediment flux of E1 and E2 are 0.039 Mt d⁻¹ (millionmetric tons per day) and 5.1 Mt d⁻¹, respectively. Although the transport rate of E2 is nearly 130 times than E1, considering the long duration of E1, the gross amount transported by E2 (8.5 Mt) is only 35 times than E1. And the sediment load of E1 and E2 accounts for about 34.3% of the yearly average sediment transport by turbidity currents in Gaoping Canyon (25.5 Mt, though the actual turbidity flux should be higher due to a deficiency of basal observations)²⁵, and 17.8% of the yearly average sediment flux of Gaoping River (49 Mt)³².

Fig. 5: Properties of sediments collected from the 2020 deployment, the 2021 deployment and the substrate at the mooring site S2.

a The gray-scale values of 3–7 cm of the sediment at 15 MAB are indicated by the blue line. b The gray scale maps are computed-tomography (CT) images of sediment trapped from E1 to E2. The coarser the grain size, the lighter is the color of the CT images. The blue circled numbers denote the positions of cross-section photos. c Median and mean grain size and sorting of the collected sediment. d The content of clay, silt and sand, (e) proportion of the clay minerals, and (f) the content of δ¹³C relative to PDB. In the legends, sediment properties from layers #7-#8’ denotes the finer sediment samples of layers #7-#8 (hereafter ‘_F’). Layers #7-#8’ located in the oblique interface between the deposits from the turbidity currents E1 and E2, containing the sediment from E1 to E2. The ‘Substrate’ in legends indicates the substrate sediment sampled in 2020 at the site X02 (see the location in Fig. 1b). The ‘Trap-2020’ and ‘Trap-73’ indicate the sediment trapped during 2019–2020 and at 73 MAB during 2020–2021, respectively.

In addition to the volume difference between the sediment accumulations from the two flows, other properties of the sediment particles are also visually different. Firstly, the sediment from E1, about 6 cm thick, has a yellowish-brown color. This is in strong contrast to the greyish-black color for the sediment from E2 that is about 25 cm thick. When subsampled, at 1 cm interval (From the bottom up, these are Layer #1-#31; Fig. 5b), for grain size and mineral analyses, the size differences between the sediment grains from the two deposits are very obvious (Note that because the interface between the two deposits in the trap was tilted, layers #7 and #8 each contains sediment particles from both flows. Each sample was manually partitioned to E1 and E2 sediment particles for grain size analysis. Layers #1–#8 from E1 are composed of nearly 90% of silt and <5% of sand. The highest clay content in the middle of the E1 deposit reached nearly 10%. The median/mean grain size of the major portion of the E1 deposit (layers #1–#6) is about 25 μm (Fig. 5c), but the fine portion of layers #7 and #8 are much finer, with a median size of only 9 μm. Because this is very close to the grain size of (1) samples from the upper sediment trap at 73 MAB, (2) trap sample from previous year when no turbidity current occurred, and (3) bed surface sample from outside of the trench (Fig. 5c), it probably represents normal pelagic settling rather than turbidity current E1.

Layers #7–#28 are composed of much greater proportion of sands and thus considered to be from turbidity current E2. The first five E2 layers (#7-#12) contain nearly 50% of sand (Fig. 5d). Above these very coarse layers the content of sand gradually decreases (and the content of silt gradually increases). Overall, the median and mean grain size of layers #7–#28 decrease from approximately 65 to 30 μm, displaying the distinct normal grading of grain size distribution that is common in turbidite. Layers #29–#31 are similar to the finer portion of layers #7–#8, with clay content of >10%, representing typical pelagic particles. Other particle parameters such as sorting (Fig. 5c) reflect the two very different settling process between E1 and E2. The much higher sorting of E1 and the very top portion of the E2 deposit resulted from fine and quite uniform size of particles settling in a weaker hydrodynamic environment than that in the beginning of E2. In contrast, the much coarser grains with poorer sorting indicate rapid deposition indicative of more energetic gravity flows.

Clay minerals in sediments collected at 15 MAB were mainly composed of illite and chlorite, accounting for more than 90% (Fig. 5e). The kaolinite content in E1 sediment (5–9%) are consistently higher than in E2 sediment (less than 5%). Smectite was only present in the bottom three quarters of E1 sediment (layers #1–#4), comprising 5–9% of total clay minerals, strongly suggesting that E1 came from a different source than E2. The averages of δ¹³C values of particulate organic carbon were −24.6‰ in E1, slightly more depleted than the average δ¹³C values of Trap-73, Trap-2020 and Substrate (around −23.5‰) (Fig. 5f). The sudden drop of δ¹³C values at the beginning of E2 (from −24.0‰ to −25.5‰ indicate a more terrestrial source of the sediment particles.

Discussion

Contrasting flow structures

A striking difference between the two flows is their arrival at the mooring, as clearly demonstrated in Fig. 4, where the first 4 days of converted sediment concentration profiles, in both backscatter intensity maps and line plots of SSC at several heights above bed, are plotted on an equal time interval. Turbidity current E1 arrived at sensors of all heights ranging from 7 to 72 MAB, almost synchronously, with merely 15 min delay to reach the highest sensor. This flow resembled a thick plume of a powder snow avalanche of low density and slow speed, that engulfed the mooring instantaneously. Turbidity current E2 resembled a flash flood in a stream with rising water level; It took 50 min to increase in thickness by 15 meters (from 7 to 22 MAB), 44 min to grow additional 20 meters (from 22 to 42 MAB), and another 136 min to overtop the sensor at 72 MAB. In another word, the head of E2 had traveled more than 20 km further down trench by the time the top of the flow finally reached highest sensor (72 MAB) on the mooring. These above contrasting behaviors could have resulted from different degrees of mixing with the ambient water in the flow front of E1 and E2^{33,34,35,36,37}.

Some turbidity currents with high velocity is more likely to have a fast-moving head with a dense basal layer whose SSC is high (generally > 10%), but how high is still an open question^14,37,38,39. Given its weak velocity and low SSC, turbidity current E1 certainly does not fit the above characteristics. Instead, E1 is likely a fully dilute turbulence-supported flow without a powerful basal layer. As for turbidity current E2, whose highest concentration at 7 MAB was ~1.2% (lower than 10%), it is qualitatively similar to flows that contains a dense near-bed dense layer^38,39. The SSC of this layer, assuming a thickness of few meters¹⁴, probably be much >1.2%, a value obtained from 7 MAB. In addition, both the multi-beam bathymetric measurements (Fig. 6a) and the ADCP beam’s acoustic footprints (Fig. 6b, c) indicated that the mooring was on the east wall of the trench, ~500 m away from and 25 m above the trench thalweg (Fig. 1c). We argue that, for the first ~40 min, between the time when ADCP beam 1 and beam 4 first detected high acoustic intensity signals at 14:12 and the time when significant increase of flow speed occurred (Fig. 6, Supplementary Fig. 5), the fast but thin (<25 m) turbidity current E2 was restricted at the bottom of the trench thalweg. The SSC of this flow front may thus have been significantly greater.

Fig. 6: The arrival process of the turbidity current E2 at the mooring site S2.

a Multi-beam bathymetric measurements around the mooring site S2 from Fig. 1b. b The vertical distance from the transducer to the seafloor in different directions (−180°–180°) measured by an individual ADCP beam (Beam 3). The center of the ADCP beam’s acoustic footprint locates on the right side of the trench floor. The red lines are the positions of four beams according to the initial heading of ADCP during E2, which is indicated in (e). The black and dark blue lines are the major flow direction of Periods 1 and 2 indicated in (d). c Net acoustic backscatter intensity (counts) from bin 50 (12 MAB) to 70 of four beam recorded by the downward-looking ADCP. The maximum intensity values indicate the seabed position. The red dashed rectangle indicates the arrival time of the turbidity current E2. d Current velocity vectors at 12 MAB measured by the down-looking ADCP at the initial stage of E2. The light yellow and green rectangles indicate Period 1 and Period 2, respectively. e The postures include the heading, pitch, and roll measured by the down-looking ADCP.

Sediment sources and possible triggers

Clay mineralogy is often used as an indicator of sediment provenance or origin^40,41. At least two distinct patterns can be drawn from the clay mineral composition of the sediment trap samples (Fig. 5e): (1) Illite and chlorite are abundant throughout the trap samples, including deposit from both turbidity currents E1 and E2; (2) Smectite only exists in sediment deposit from E1 where content of kaolinite is also higher. While illite and chlorite are the two most abundant clay minerals in the South China Sea region⁴², smectite is believed to be uniquely sourced by the volcanic rocks in the Luzon arc⁴³. In general, chemically-weathered smectite is discharged from the Luzon rivers and transported northward by the NW Luzon Coastal Current, then westward mainly via mesoscale eddies produced by the westward South China Sea Branch of the Kuroshio and Taiwan Warm Current^32,44.

The presence of smectite in the mostly fine particles of E1 sediment, along with E1’s weak flow velocity, strongly suggest that this turbidity current is likely to have originated from trench wall slumping analogous to the dilute flow monitored in Monterey Canyon caused by a similar mechanism⁴⁵. Collapse of mud drapes on the trench walls upstream of the mooring site could create dilute sediment clouds that move down the trench as gravity flows. Lack of momentum, due low density and small gradients, would allow the currents to be modulated by tides (Fig. 7a, details in next section). However, what triggered such a slump or slumps is still unclear—there was no record of either storms or earthquakes in the region prior to the turbidity current E1. The exact location of the slump is also unknown.

Fig. 7: The schematic diagrams of the turbidity currents E1 and E2 (not to scale).

a The formation process of the turbidity current E1. The sediment from mass wasting moves down the trench under gravity and is promoted by tidal currents. b The flow thickness of turbidity current E2 is gradually thickening when the front of E2 gets through the mooring site S2.

In contrast, turbidity current E2 is characterized by its larger velocities and coarse sediments. In addition, the absence of smectite in the E2 sediment, and the more negative δ¹³C values (Fig. 5f) of particulate organic carbon in E2 (−24.6‰) than in E1 (−24.0‰), both suggesting that E2’s sediment was more likely from a fresh source with abundant terrestrial organic matter such as Taiwan’s rivers⁴⁶. Previous studies have shown that frequent passage of tropical storms often bring very high precipitation to Taiwan, and consequently dramatic increase of sediment discharge into the South China Sea^47,48. For instance, after the typhoon Hagibis in June 2014, the Gaoping River’s discharge increased, and a turbidity current signal was detected by a mooring deployed in Gaoping Canyon²⁵. When Typhoon Soudelor passed through southern Taiwan in 2015, a turbidity current was detected not only in Gaoping Canyon²⁵ but also in the Manila Trench²⁸, and finally its front velocity reached 5.6 m s⁻¹ near the mooring site S2. Not only was the intensity and track of tropical storm Lupit of August 2022 comparable to Typhoon Hagibis of 2014, it also rapidly increased the water and sediment discharge of the Gaoping River to the ocean, reaching maximum values of 6980 m³ s⁻¹ and 3.21 g l⁻¹ respectively (Supplementary Note 1 and Supplementary Fig. 6). Thus, it is reasonable to argue that turbidity current E2 resulted from the enhanced discharge of terrestrial sediment from Taiwan rivers into, first the several canyons SW of Taiwan, then the head of the Manila Trench. Except the hyperpycnal flow, it also might be helpful for triggering the turbidity current that the direct sediment resuspension in the shallow shelf and transport to the canyon head by typhoon-induced waves and currents⁴⁹.

In summary, sediment slumping from the trench wall or upstream canyon triggered by earthquakes or sediment instability is one major source of turbidity currents. A second source, likely a more important one, is typhoon-generated flows from Gaoping or other submarine canyons southwest of Taiwan Island, that are fed by frequent landslides, terrestrial floods and combinations of storm waves and currents during the annual typhoon season.

Tidal modulation of weak turbidity current E1

For a weak turbidity current such as E1 whose speed is in the same order of magnitude of the local tidal current, its flow structure as well as the resulting deposit are inevitably modulated by the internal tide. The velocity and flow thickness of the turbidity current E1 exhibited periodic pulsation resembling tidal oscillations that happened to be in a transition from semidiurnal to diurnal (Fig. 2a, d). Model predictions showed that the local tide was at its neap phase when E1 arrived at the mooring, but the predicted tidal current in the along canyon direction grew from ~5 cm s⁻¹ at the beginning of E1 to ~12 cm s⁻¹ when E1 diminished several days later. The flooding (ebbing) phase of tidal current acted like a headwind (tailwind) that reduced (enhanced) the speed and decreased (increased) the flow thickness of turbidity current E1. This explains why E1’s peak velocity was recorded 9 h after the initial arrival of the turbidity current. Despite the tidal modulation, turbidity current E1 still moved suspended sediment downcanyon. The footprint of E1, estimated by the cumulative product of velocity and time, is about 140 km long. In another words, the body of turbidity current E1, at its maximum, stretched along the trench for 140 km.

E1 was inferred to originate from an upstream trench wall slumping (i.e., mass wasting). These collapsed materials moved down the trench wall by a gravity flow in direction nearly orthogonal to the trench strike (Fig. 7a). After a head-on collision with the trench bottom and opposing trench wall, substantial sediment was suspended to form a turbid cloud. This cloud, initially with low or zero initial along-trench momentum, slowly moved down the trench, driven by gravity. The headwind (tailwind) effect of tidal currents on turbidity current inevitably changed the flow’s sediment carrying competence as well as capacity. Researchers have used a Rouse-based criterion (${{{{{\rm{B}}}}}}={W}_{s}{/u}^{* }$) to determine whether a sediment particle of settling velocity ${W}_{s}$, should stay in suspension or fall to the bed given a shear velocity of u^*50,51. A value of ${{{{{\rm{B}}}}}}=0.3$ was also found to provide a best fit to data for equilibrium open channel flows⁵².

Here, the grain size data of Layer #1-#6 sediment (Fig. 5c) were used to calculate the average D₅₀ (median grain size), D₇₅, and D₉₀. Their corresponding settling velocities were 0.028, 0.068, and 0.148 cm s⁻¹ respectively. These settling velocities and the computed shear velocity in the flow (see Material and Method section for details) were used to obtain the Rouse-based criterion B. In the first 5 days of the E1, the computed B values were always smaller than 0.3 if D₅₀ was used, independent of tidal phases (Fig. 8a). For D₇₅ and D₉₀ size particles, while the B values during ebb tide (tailwind) were still <0.3, it became significantly greater than the critical value (0.3) during peak flood tide (headwind). For instance, the B values for D₉₀ and D₇₅ during the peak of several flood tides reached 0.6 and 0.4 respectively. The 12 h settling distances for the D₅₀, D₇₅, and D₉₀ particles are ~12, 29 and 64 m, respectively. the same order of magnitude as the changes of flow thickness (10–60 m). In addition, the estimated bulk Richardson number (Ri) increased during the flooding and decreased during the ebb (Supplementary Fig. 7). It suggests that the weak-turbulence current during flood was unable to support the sediment particles in suspension⁵³, so the larger particles, though fewer and fewer in the upper part of the flow, would fall toward the bottom and effectively reduce the thickness of the turbidity current (Figs. 4a and 8b). During the next ebb tide, some of these particles that had fallen into the lower part of the flow or even onto the seafloor would be lifted/resuspended by the enhanced turbulent shear resulted from the stronger flow velocity, and the thickness of the turbidity current also increased (Figs. 4a and 8c, d).

Fig. 8: Tidal modulation of weak turbidity current E1.

a Variations of the Rouse-based criteria B computed from different grain sizes (D₅₀, D₇₅, and D₉₀). The orange line and blue line are the vertical maximum velocity (u_M) and tidal current (positive value is ebb phase). b The sketch map of tidal modulation of weak turbidity current. The shadow indicates the depth-averaged flow thickness. The brown and black dots denote the fine and coarse particles, respectively. c and d Conceptual vertical velocity and sediment concentration profiles of E1 during the flood (red) and ebb (blue) phases.

Evidence of tidal modulation was also found in the sediment trap collection during E1. The alternating gray scale highs and lows in the CT images of the sediment trap (layers #1–#8) are believed to have resulted from differential deposition between tide-enhanced and retarded phase of the turbidity current (Fig. 5a). There are only 6 gray-scale couplets (borrowing a term from classical tidal sedimentology), about half of the total number of tidal cycles during the lifespan of E1. This perhaps indicates that the differential sedimentation pattern is only discernable for the tidal cycles in the early days of E1 when sediment concentration is very likely greater.

‘Tortoises’ or ‘Hares’ determined by particle sizes

Table 1 lists the measured and computed parameters of the two turbidity currents, E1 and E2. Two proverbial characters, Tortoises and Hares, represent the two flow types well: slow, steady, and long-lasting for the former; fast, rapid-decaying, and short-lived for the latter.

Table 1 Key parameters of the two turbidity currents.

As seen in the previous section, E1 was tidally modulated for several days. The peak current velocity and SSC, both occurred at each ebb tide, decreasing only slightly over that time (Fig. 2d and Fig. 4b). The long duration of E1 was primarily owing to its slow speed, but the stop-and-go movement caused by tidal modulation probably also contributed. E1’s flow properties such as large initial thickness (rapidly thickening and well mixed) and low SSC (dilute) of fine sediment particles within each ebb tide episode are similar to the “entirely slow, dilute, and well mixed” turbidity currents observed in Bute Inlet³⁷, except E1 lasted much longer duration (~150 h vs 2 h), probably due to its much finer-grained sediment particles (9–25 μm vs ~200 μm)⁵⁴. Prolonged turbidity currents such as E1 have also been reported in Congo Canyon⁵⁵. The 170 h lifespan of the Congo Canyon flow was believed to have been assisted by a fast erosive zone at the front that apparently caused flow stretching and the fine grains in the flow. This flow stretching mechanism was not present in E1. There is one property that is shared by these two long duration flows: very fine particles in suspension inside the flows (9–25 μm vs 4.23 μm).

The fast, thin, and short-lived E2 is an entirely different animal (metaphorically). When it arrived at the mooring, its flow speed first accelerated fast but then decelerated exponentially. The same flow structure as E2 were also observed in Monterey Canyon, Bute Inlet and Whittard Canyon^14,37,56, which were mostly characterized by a fast, dense, thin, and stratified head. Compared with these turbidity currents, E2’s slightly finer sediment (10–65 μm vs ~200 μm) and lower SSC (1.2%) had probably contributed to its longer duration (~40 h vs ~2 h).

Sediment grain sizes clearly played a key role in defining the different characteristics of the two turbidity currents, conforming with findings from other field observations^29,55, laboratory experiments^57,58,59, and numerical simulations^11,60. The interplay between hydraulic shear and particle settling velocity governs both the density structure and evolution of turbidity currents⁶¹.

Conclusions

Frequent occurrence of sediment gravity flows out of the Gaoping River and into the Gaoping Canyon have been reported^24,25, but how often those flows reach the deeper water of the Manila Trench is unclear. This field study has shown that two different turbidity currents passed the mooring site at 3800 m water depth, both in the second year of a 2-year investigation. Grain size and clay mineral analyses indicated that turbidity current E2 in Aug. 2021, which was characterized by coarse sediment with high content of illite and chlorite but no smectite and of flowing fast and thin thus nicknamed ‘Hares’, was originated from the Gaoping river/canyon system. Turbidity current E1 in April 2021, containing fine sediment with smectite and flowing thick and slow thus nicknamed ‘Tortoises’, had a very different source which we propose to be the slumping of trench wall material. Notably, tidal modulation in the deep trench had likely prolonged the lifespan of this ‘Tortoises’ flow. Those turbidity currents increase our understanding of the transport of terrigenous material to the deep waters. The rivers from Taiwan Island deliver a large amount of sediment and organic carbon due to the unique geographical and climatic conditions⁴⁷. Turbidity currents transport those material (including carbon) into the deep sea. Applying the average organic carbon concentration of 0.44% for the sediment carried by the hyperpycnal turbidity currents from Taiwan Island²⁴, the estimated organic carbon transport into the deep sea by the turbidity currents in this study was roughly 3.85 × 10⁴ t, accounting for nearly 16% of the annual organic carbon load in the Gaoping Canyon^47,62. But carbon transport and burial into the deep sea is more than just the strait shots by those ‘Hares’ flows to move terrestrial carbon. Rather, it’s more like a cascading process in which ‘Tortoises’ also play an important role in transporting marine carbon, typically resulted from marine particle settling, given that these flows have much longer lifespan and finer material composition. The two modes of transport form an effective route for the organic carbons of continental margins, both autochthonous (marine) and allochthonous (terrestrial), into the deep ocean for eventual burial. Not only that, they have important implications for the transport of pollutants (i.e., microplastics) and the warning of geological hazards.

Methods

Field data collection

The S2 substrate mooring was at 3808 m water depth and recorded data for a whole year (from 2020 to 2021). High-resolution multi-beam bathymetric data that were collected during the cruise on August 2022 clearly show the seafloor topography around the mooring site S2, including the longitudinal seafloor slope of the canyon (0.4% or 0.23°), the width of the trench thalweg (about 2.5 km), the average slope of canyon walls (left: 8% or 4.6°, right: 3% or 1.7°), and the depth of the channel thalweg (80 m) (Fig. 1b, c). A down-looking Workhorse Marine 300 kHz acoustic Doppler current profile (ADCP) was mounted at 65 MAB, to record a vertical profile of flow speed and direction every 60 s from an ensemble of five 12 s pings, with a spatial resolution (ADCP bin size) of 1 m (Fig. 1d and Supplementary Table 1). A Seaguard recording current meter system (RCM) placed at 12 MAB was set to record flow speed, flow direction, water temperature, conductivity, and turbidity. An SBE 37-SM Conductivity–Temperature–Depth recorder (CTD) was deployed at 17 MAB, with a sampling interval of 1 min. Two Anderson-type sediment traps, each composed of a fiberglass cone (90 cm in length, 26.8 cm in diameter) and a clear acrylic linear tube inside a Polyvinyl Chloride (PVC) sediment protection cylinder (150 cm in length), were placed respectively at 15 and 73 MAB on the mooring. Each sediment trap was equipped with an intervalometer (Timer) that dispensed Teflon disks, as time marks, into the liner tube at a preset interval of 15 days. A turbidity sensor (RBR-TU) was deployed at 72 MAB, with a sampling interval of 1 min. The accuracy of flow measurements for ADCP was 0.5% ± 1% for RCM. The CTD’s precision was 0.0028 °C for temperature, 0.003 millisiemens (mS cm⁻¹) for conductivity, and 0.1% of the full-scale range for pressure (about 7 m for CTD used in the study). The seafloor sediment near the mooring site S2 was sampled by box corer in 2020 compared to the sediment in the sediment traps. All raw measurements were shown in Supplementary Fig. 1.

Typhoon and earthquake records (Supplementary Table 2) are from the Typhoon Network of the Central Meteorological Observatory (NMCS, http://typhoon.nmc.cn) and the China Earthquake Networks Center (CENC, https://news.ceic.ac.cn), respectively. Bathymetry data of the South China Sea is from the GEBCO Gridded Bathymetry Data (https://download.gebco.net/).

Vertical velocity structure of turbidity currents

The measurements of flow velocities were rotated 159° clockwise to the along- and cross-trench components, and the down-trench direction is positive (Supplementary Fig. 3). In a typical velocity profile of the turbidity currents, the lower part below the velocity maximum is the wall region, where the velocity distribution exhibits a logarithmic relationship known as the ‘law of the wall’, and the upper part above the velocity maximum is referred to as the jet region, where the vertical distribution of velocity is nearly Gaussian relation. The vertical velocity profiles of E1 and E2 were plotted in Supplementary Fig. 4.

In addition, the normalized velocity profiles, using the [u/U, z/h] scheme where the U is the depth-averaged velocity, h is the depth-averaged thickness of the flow, z is the corresponding height of each layer velocity (u), were compared with other field measurements⁶³. The depth-averaged thickness (h) and velocity (U) were calculated using the moment equations⁶⁴. The vertical velocity distribution of a turbidity current essentially represents the balance between the logarithmic boundary layer and the Gaussian outer layer^6,65.

Inversion of suspended sediment concentration

Applying the relationship between backscatter voltage (V_rms) and the SSC proposed by Thorne and Hurther⁶⁶, the structures of the SSC of the turbidity currents were inverted from the acoustic backscatter intensity (E) measured by the 300 kHz ADCP:

$${{{{{\rm{SSC}}}}}}\left(r\right)={\left(\frac{{V}_{{rms}}\varphi \left(r\right)r}{{K}_{t}{K}_{s}}\right)}^{2}{e}^{4\left({\alpha }_{\omega }r\right)}$$

(1)

V_rms is calculated by the acoustic backscatter intensity according to ${V}_{{rms}}={10}^{{K}_{c}E/20}$, K_c is a measured constant for each of the transducers; φ(r) is a correction for transducer’s near field, r is the distance from the measuring point to the transducer; K_t should be constant and describes the sensitivity of the individual transducers; K_s is the parameter of the scattering properties of sediment with mixed mineralogy, which depends on the particle size, particle distribution characteristic and backscattering of suspended particles. Here the properties of particles collected in sediment trap at 15 MAB were used. α_ω is the absorption of sound by the properties of seawater, which depends on the measured temperature, salinity, depth and the assumed pH of 8 during the turbidity currents E1 and E2. The SSC derived by inversion of bin 50 (at 12 MAB) and bin 1 (the top layer at 62 MAB approximately represented the SSC at 72 MAB) was compared to the correlated SSC at 12 and 72 MAB during E1 and E2 (Supplementary Fig. 4), which was calculated from the turbidity measured by RCM and RBR-TU with a correlation relationship between the turbidity measured by RCM and in situ SSC from Zhang et al:^25,67. SSC (mg l⁻¹) = 0.96 × turbidity (FTU).

Notably, both the grain size and concentration influence the correlation relationship between the measured turbidity and SSC^68,69. Because the turbidity recorded by RBR-TU and RCM are not exceeding 1500 FTU, the relationship between the measured turbidity and SSC is still linear. Under the same turbidity, the bigger particle size corresponds to a larger SSC. The particles at 72 MAB in E2 are finer than those trapped at 15 MAB in E2, but coarser that those at 72 MAB in E1. The particles at 12 MAB in E1 also are coarser than those at 72 MAB in E1. This linear relation was adopted in the transformation of the turbidity recorded by RCM, which was deployed nearly 500 m vertically from the thalweg, and SSC during the long-term monitoring of turbidity currents in the Gaoping Canyon. So, the true SSC in the bottom of turbidity currents E1 and E2 may be underestimated.

Sediment analysis

Sediment samples collected in two sediment traps were processed following the protocol for typical sediment cores^24,70. They were first scanned at 0.5-mm resolution using an x-ray computed-tomography (CT) device developed at the Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences. The material in the traps was pushed up and out with a 1-cm stepper and resampled into plastic bags and stored in a refrigerator. There are a total of 31 samples from the sediment trap at 15 MAB. These samples, along with other sediment samples either from the first deployment or seafloor samples at locations nearby the mooring site, were then analyzed for grain size using a laser particle size analyzer (Mastersizer 3000), and for clay minerals using a Rigaku SmartLab (9 kW) diffractometer. Stable isotopes of total organic carbon (δ¹³C) were measured using an isotope ratio mass spectrometer (Thermo Science Delta Plus, USA) connected online to an elemental analyzer (Carlo Erba Instruments Flash 1112, USA).

Estimation of suspended sediment flux (SSF)

The SSF of the turbidity currents was calculated with the following equation:

$${{{{{\rm{SSF}}}}}}={\int }_{0}^{{Z}_{t}}{SSC}\cdot u\cdot L\cdot {dz}$$

(2)

where Z_t is the top of the flow where the velocities vanish or are close to zero, L is the averaged width of the trench. Because the SSC and current velocity near the seabed at the mooring site S2 and from the thalweg to the mooring site S2 were unknown, they were replaced by the SSC and current velocity of the distinguished lowest layer. It should be noted that this replacement may result in the SSF being underestimated.

Estimations of bulk Richardson number (Ri)

Following Park et al. ⁷¹, the bulk Richardson number (Ri) was calculated from

$${Ri}=\frac{{RCgh}}{{U}^{2}}$$

(3)

Where R is the submerged specific gravity (1.65), C is the depth averaged sediment concentration, g acceleration of gravity (9.81 m s⁻²), h is the depth averaged thickness, U is the depth averaged velocity.

Estimations of settling and shear velocities

With known median sediment grain size (D₅₀), the settling velocities (W_s) are calculated using Stokes’s equation:

$${W}_{s}=R\frac{g{\left({D}_{50}\right)}^{2}}{18v}$$

(4)

Where ν is the kinematic viscosity of the seawater (here set to 1.6 × 10⁻⁶ m² s⁻¹). Here we did not take the viscous characteristics of sediment into consideration because the turbulence in flow might break the flocculation. The turbulent shear, as expressed by the shear velocities (u^*), are determined using:⁷²

$${u}^{* }={u}_{M}k{\left({{{{{\rm{ln}}}}}}\left(\frac{{Z}_{M}}{0.1{D}_{90}}\right)\right)}^{-1}$$

(5)

Where u_M is the maximum velocity, Z_M is the height where u_M was found, D₉₀ is the sediment grain size at the 90th percentile.