Submarine channels share morphological similarities with rivers, but observations from modern and ancient systems indicate they are formed under processes and controls unique to submarine settings. Morphologic characteristics of channels—e.g., width, depth, slope, and the relationships among them—can constrain interpretations of channel-forming processes. This work uses morphometric scaling relationships extracted from high-resolution seafloor bathymetry to infer connections between morphology and process in submarine channels. Analysis of 36 modern channels in five geographic regions shows that channel widths vary regionally (from <100 m to >10 km wide) but occupy the same range of aspect ratios (∼10:1–100:1). This suggests an autogenic control on aspect ratio, perhaps resulting from feedback processes in levee growth and/or bank erosion, and allogenic (e.g., sediment supply, grain size) controls on channel width. Submarine channel aspect ratios tend to decrease with increasing dimensions, while the opposite relationship has been observed for fluvial channels, likely due to opposing relationships between flow discharge and channel distance. Additionally, observation of an apparent lag between channel thalweg and levee responses to gradient changes suggests that thalweg and levee deposition and erosion may be partially decoupled due to the vertical structure of turbidity currents, with thalweg evolution driven by the basal, higher-shear-stress portion of the flow and levee evolution by the dilute upper portion. The data presented here provide a basis for predicting channel metrics in exploration scenarios, in which data coverage may be sparse. This documentation of a diverse suite of channels also captures the range of scales and variability exhibited globally by submarine channel systems, providing context for local studies.


Through decades of study employing ever-improving technologies, researchers have demonstrated the important role submarine canyons and channels play in shaping the seafloor, building the sedimentary rock record, and transporting sediment and nutrients from coastal to deep-marine environments. However, fundamental questions remain regarding how these channels form, and what controls the diversity of scales and morphologies of submarine channels around the world. Despite advances in direct observation of active flow events in submarine canyons and channels (Paull et al., 2002; Cooper et al., 2013; Sumner and Paull, 2014; Xu et al., 2014; Azpiroz-Zabala et al., 2017), these efforts are hindered by difficulty of access, high expense, and infrequency of events (e.g., Clare et al., 2016).

Due to the difficulties of studying submarine channels directly, numerous authors have attempted to understand submarine processes and dynamics by drawing on fluvial channels and networks as analogs (e.g., Straub et al., 2007; Brothers et al., 2013). Although submarine channels can be an order of magnitude larger than rivers (Konsoer et al., 2013), their cross-sectional and planform morphologies bear striking similarities (Fig. 1) with many submarine channels exhibiting levees (Hansen et al., 2015), erosional terraces (Babonneau et al., 2004), and/or migrating meander bends and cutoffs (Kolla et al., 2012; Maier et al., 2012). Given the morphological similarities between submarine channels and rivers, can techniques developed in fluvial geomorphology shed light on the processes that control submarine channel morphologies?

A century of study in fluvial geomorphology has led to the discovery of numerous scaling relationships among morphological parameters of rivers. Channel properties such as width, depth, and catchment area scale with river length, reflecting the importance of precipitation and cumulative discharge along tributary networks in fluvial systems (Leopold and Maddock, 1953). Fluvial channel width and depth follow a power-law relationship, such that an increase in width is associated with a predictable increase in depth (Wilkerson and Parker, 2011). Leopold and Wolman (1960) show that meander dimensions scale systematically with channel width not only for rivers, but also for glacial meltwater streams and ocean currents. Flood and Damuth (1987) and Clark et al. (1992) demonstrate that similar width-wavelength scaling relationships exist in submarine channels, supporting the possibility of a universal behavior or property of channelized flows that dictates their geometries.

Although the subaqueous turbidity currents that shape submarine channels share some physical similarities with fluvial flows, in many respects they are fundamentally different, and direct comparisons of fluvial and submarine channels may not be appropriate (e.g., Keevil et al., 2006). Both fluvial flows and turbidity currents are vertically stratified in terms of their sediment concentration and flow velocity, but the shapes of their respective velocity and concentration profiles differ (Fig. 1; Choux et al., 2005; Le Roux, 2005; Xu, 2010). Turbidity currents typically contain a thin, high-velocity, coarse-grained basal portion of the flow where the maximum bed shear stress occurs (Azpiroz-Zabala et al., 2017; Symons et al., 2017), with a fine-grained upper portion that may not be fully confined by the channel along which the flow travels (e.g., Piper and Normark, 1983; Abd El-Gawad et al., 2012). Turbidity currents may also reach extraordinary thicknesses of tens of meters to over 100 m (Talling et al., 2007; Xu, 2010) due to the small density contrast between the sediment-laden flow and the ambient seawater. Turbidity currents are driven by gravity acting on this excess density, and contrary to fluvial flows, cannot travel downslope without entrained sediment. Turbidity currents are thought to typically last hours or days, although week-long flows have been documented (Cooper et al., 2013; Azpiroz-Zabala et al., 2017), and occur relatively infrequently in deep-water settings (a few times per year to every few hundred years; Paull et al., 2014; Stevens et al., 2014; Jobe et al., 2018). This contrasts the perennial flows of many fluvial channels, although highly variable discharge in some rivers may result in relatively few flows exhibiting disproportionate control on channel morphology (Wolman and Miller, 1960; Plink-Björklund, 2015).

In this study, we analyze morphometric scaling relationships in submarine channels to investigate the extent to which morphological similarities between fluvial and submarine channels reflect similarities in formative processes. We hypothesize that fluvial scaling relationships that arise from terrestrial processes, such as downstream increases in discharge driven by precipitation and overland flow, will be different in submarine systems where such processes do not occur. This study presents a rich data set of submarine channel morphometrics that can be further employed in numerical modeling of turbidity currents (e.g., Sequeiros, 2012; Traer et al., 2015, 2018a, 2018b), offshore energy resource exploration, seafloor infrastructure hazards assessment (e.g., Piper et al., 1999), and marine habitat and ecosystem research (e.g., Vetter and Dayton, 1998). By directly testing whether certain fluvial scaling relationships are also present in submarine channels, we address whether morphological similarities reflect analogous processes, and assess the relevance of fluvial analogs in deep-water systems (Mitchell, 2004; Keevil et al., 2006; Hughes Clarke, 2016).


Data Sources and Descriptions

This study utilizes publicly available and industry-provided high-resolution bathymetric data sets to map 36 channels from five regions around the world (Fig. 2A). Lateral resolution of the digital elevation models (DEMs) ranges from 12.5 m per pixel (from the seafloor reflection of 3-D seismic data) to 380 m per pixel (from multibeam bathymetry). In all cases, the resolution of the DEM imparts a bias toward detection and mapping of channels larger than ∼3 pixel widths across, but the Amazon fan bathymetry is the only data set used here that approached this limitation. Some of the channels mapped in this study have previously been analyzed; references and correlations to previously published channel names are noted where appropriate.

We highlight four well-imaged channels ranging in streamwise length from 56 km to over 700 km for detailed analysis (Figs. 2 and 3): (1) channel 1 of the Amazon fan, offshore northeast Brazil in the Atlantic Ocean (Pirmez and Imran, 2003), (2) channel 1 of the Bengal fan, in the Bay of Bengal in the northeastern Indian Ocean (Schwenk et al., 2003), (3) channel 12 of the eastern Gulf of Mexico, off the southern coast of the USA (Posamentier, 2003), and (4) channel 9 of the Niger Delta continental slope, offshore Nigeria in the Atlantic Ocean (hereafter called the “Niger slope”; Pirmez et al., 2000; Jobe et al., 2015). We also briefly discuss glacially influenced channels in the Gulf of Alaska, offshore the southern coasts of Alaska, USA and Canada in the Pacific Ocean, which are exceptionally large (up to 18 km wide and 300 m deep; Swartz et al., 2015). We also document dozens of smaller, isolated channel segments captured in the multibeam bathymetry from a single ship track (Amazon and Bengal channels) and relatively short, slope-confined channels lacking any visible connection to a submarine canyon or shallow marine sediment source (western Gulf of Mexico, northwestern Niger slope channels).

Measured and Calculated Values

We mapped 36 channel thalwegs and their associated margins in the geographic information system ArcGIS and extracted morphometric measurements in Matlab. Following the convention of “bankfull” width and depth measurements (e.g., Konsoer et al., 2013), channel margins are here defined as the highest point of the external levee crest (Kane and Hodgson, 2011; Hansen et al., 2015) for leveed channels, and the rollover point of the eroded edge for non-leveed or poorly leveed channels (Figs. 4A and 4C). Although alternative methods for measuring channel width and depth hold promise (e.g., “GLORIA width” of Pirmez and Imran, 2003), the bankfull width method is beneficial for areas with relatively poor-quality imagery, where levee crests may be resolvable while internal channel characteristics are not. Channel widths were measured perpendicular to the thalweg as the distance between the two channel margins (Fig. 4). Channel depths were measured as the maximum vertical distance between the channel margins and the thalweg (i.e., for asymmetrical channel cross-sections, the larger of the two depths was recorded; Fig. 4). Width and depth measurements were extracted from thalweg-perpendicular elevation profiles at two times the DEM resolution to avoid duplicating data from a single pixel. To filter out erroneous measurements (e.g., anomalously large width measurements arising from bend geometries; Fig. 4E), any cross-section whose width was more than 150% or 170% greater than the minimum width of the three previous measured cross-sections (the precise threshold value was determined manually for each channel) was excluded.

Channel measurements are referenced to the downstream distance along the thalweg, calculated from the first mappable point on the channel. We began mapping channel thalwegs at either the farthest upstream point at which the channel is detectable in the data set or the farthest upstream point at which the channel margins are easily defined (e.g., without significant tributary networks). While sinuous submarine channels are the focus of this work, the transition from erosional submarine canyon to leveed channel is typically gradual and difficult to pinpoint (e.g., Pirmez and Imran, 2003). Therefore, we do not attempt to robustly distinguish between canyons and channels. Some data sets presented here, particularly those from the Gulf of Alaska, may include measurements from features that could be classified as canyons. We calculate cross-sectional area as the area between bankfull height and the channel base for each channel profile (gray shading in Fig. 4A). We calculate an average whole-channel sinuosity for each channel, where sinuosity is the ratio between the along-channel (streamwise) distance and straight-line distance between the first and last measured points. Aspect ratio is defined as the ratio of width to depth (i.e., large aspect ratios are wide and shallow).


We extracted 28,921 width and depth measurements from 36 submarine channels in five geographic regions (Fig. 2A). We summarize the distribution of observations by reporting the tenth, fiftieth, and ninetieth percentile of measurements for each channel (P10, P50, and P90, respectively; Table 1) and plotting distributions of width, depth, and aspect ratio for each geographic region (Fig. 5). The P50 (median) channel width values range from 195 m to 6.8 km; median depth values range from 4 m to 132 m; and median aspect ratios range from 8:1 to 146:1 (Table 1). Average whole-channel sinuosities range from 1.0 to 4.2. Although there is significant overlap among aspect ratios throughout the data set, the channels from each geographic region appear to cluster together by width (Fig. 6E). The largest aspect ratios (>100:1) were primarily recorded from the Amazon fan, the Bengal fan, and the Gulf of Alaska (Fig. 5C) and the smallest aspect ratios (<10:1) were found in the Niger slope and the Gulf of Mexico, but each region spans a similar range of aspect ratio variability (Fig. 6E). Notably, the median aspect ratio for each geographic region falls between 10:1 and 100:1; aspect ratios greater than 100:1 and less than 10:1 are uncommon (Figs. 5C and 6).

Submarine channel aspect ratios, both within individual regions and across the data set as a whole, decrease as a function of increasing channel dimensions (Fig. 6). This trend is expressed by the exponent b of the power-law regressions of width as a function of depth, w = adb, where w is width, d is depth, and a is a coefficient. Noting that, in log space, the variability in depth at a given width is relatively constant (Fig. 6), we estimate b and its uncertainty in linear fits to log-transformed data to provide an even weighting to measurements that span multiple orders of magnitude. We find that b<1 for the submarine channel data, while b>1 for leveed fluvial channel data pulled from existing literature (Wohl and David, 2008; Wilkerson and Parker, 2011; Trampush et al., 2014). Uncertainties on the regression exponents are ±0.01 to ±0.03; none of the submarine exponents overlap the fluvial exponent within uncertainty.

Amazon Fan, Channel 1

The Amazon fan, offshore northeast Brazil in the Atlantic Ocean, initiated around late Miocene time (Damuth and Kumar, 1975; Figueiredo et al., 2009) and is primarily sourced by the Amazon River, the largest river in the world by water discharge and drainage basin area (Milliman and Meade, 1983). The Amazon fan is situated on a passive tectonic margin, though glacial-interglacial sea level changes are balanced by hinterland tectonic influences on sediment supply (Figueiredo et al., 2009). We measured the modern channel of the Amazon fan (Amazon 1; “Amazon channel” of Pirmez and Imran, 2003; and Jegou et al., 2008) for 584 km of streamwise distance at 380 m resolution beginning at around 1800 m water depth, the shallowest point at which the channel is resolved. As mapped, the channel spans at least two thirds of the total fan length (Jegou et al., 2008). Large bounding levees are clearly distinguishable for the entire length of the mapped channel, and channel migration behavior is evidenced by the presence of meander loop cutoffs (Fig. 2D). The channel is ∼4.5 km wide and 140 m deep at its upstream end, and systematically decreases in both width and depth downstream to a minimum of ∼1 km wide and 10 m deep (Fig. 7). The channel aspect ratio increases downstream from ∼25:1 to ∼100:1 (Fig. 7E). The rate of decrease in channel width and depth increases dramatically from ∼90 km to 170 km downstream distance. The reach of rapidly decreasing width and depth initiates at the site of a subtle slope change in the channel thalweg profile, where the channel steepens slightly before assuming a shallower, and decreasing, gradient for the remainder of its measured length (Fig. 7).

Bengal Fan, Channel 1

The Bengal fan, in the Bay of Bengal in the northeastern Indian Ocean, is the largest submarine fan in the world. It serves as the sediment repository for the India-Asia tectonic collision and is sourced largely by the Ganges-Brahmaputra river system, which drains a large portion of the Himalayas (Curray et al., 2003). Channel 1 of the Bengal fan (Bengal 1) is the longest mapped channel in our data set, with 706 km of streamwise distance measured at 200 m resolution. We began measuring at 2886 m water depth, the farthest upstream point at which the channel is clearly imaged. Upstream of this point, the channel is progressively buried. As mapped, Bengal 1 spans about one third of the total fan length, from the middle fan to the upper part of the lower fan (Curray et al., 2003; Schwenk et al., 2005). The channel is levee-confined and exhibits evidence for migration in the form of meander loop cutoffs (Schwenk et al., 2005). The channel mouth appears to be imaged in the bathymetry (within the limits of resolution of the DEM), and is characterized by subtle widening and possible bifurcation of the thalweg (Fig. 2B). The median channel dimensions are 2261 m wide and 80 m deep, with a median aspect ratio of 35:1 and an average sinuosity of 1.5 (Table 1). The channel width fluctuates between ∼1 and 3 km for its entire length, exhibiting less variability (∼1.5–2.5 km) downstream of ∼400 km streamwise distance (Fig. 8). By contrast, the channel depth increases initially from 50 m to just under 100 m, stabilizes at that depth for nearly 300 km, then decreases steadily back to 50 m over ∼200 km of streamwise distance. In the final 80 km of channel length, the depth decreases rapidly to ∼10 m, then stabilizes around that depth for more than 50 km before a final rapid decrease toward zero (Fig. 8C). Localized extreme lows in channel width at ∼300, ∼450, ∼500, and ∼575 km (Fig. 8C) are due to edge effects near gaps in the bathymetric DEM.

Eastern Gulf of Mexico, Channel 12

Channel 12 in the eastern Gulf of Mexico, off the southern coast of the USA (GoM 12; “Joshua channel” of Posamentier, 2003; “Gulf channel” of Kramer et al., 2016), is poorly studied relative to the other channels highlighted in this work, but is thought to be late Pleistocene in age and likely became abandoned ca. 29 ka (Posamentier, 2003; Kramer et al., 2016). We measured this channel for 214 km at 12.5 m resolution, beginning at 2864 m water depth. The upstream extension of the channel is detectable in the DEM for well over 100 km upstream of the first mapped point, but was excluded from mapping because it is partially buried (Posamentier, 2003). Levees are clearly visible in cross-section and map view along the entire mapped channel length, superelevating the channel by tens of meters to over 100 m above the surrounding seafloor. The outer slopes of the levees create a ridge-like topography surrounding the channel that may be due in part to differential compaction effects (Fig. 2C; Posamentier, 2003). The GoM 12 channel has a median width of 787 m, median depth of 22 m, median aspect ratio of 36:1, and average sinuosity of 2.1 (Table 1) and exhibits numerous recurved and compound meanders (Figs. 2C and 9A). It maintains a remarkably steady width for its entire mapped length, deviating by only 15%–25% from its median width of 787 m (Fig. 9B). An initial, rapid depth increase from 10 m to 30 m over 25 km of streamwise distance (likely related to the partial burial mentioned above) is followed by a subtle and gradual increase to 36 m over the remaining 200 km of the channel (Fig. 9C). The steady width coupled with increasing depth produces a gradually decreasing aspect ratio, from ∼100:1 at the upstream end to ∼25:1 downstream (Fig. 9E).

Niger Slope, Channel 9

The Niger Delta, offshore Nigeria in the Atlantic Ocean, is a large and gravitationally unstable feature on the West Africa continental margin with up to gravel-sized sediment supplied by the Niger River (Heiniö and Davies, 2007). Channel 9 of the Niger slope (Niger 9; “X channel” of Pirmez et al., 2000; and Jobe et al., 2015, 2017) lies within the “translational” structural regime of the western slope of the Niger Delta, where shale diapirism produces complex seabed topography (Damuth, 1994). We mapped the Niger 9 channel for 57 km at 12.5 m resolution, beginning at 370 m water depth. It has poorly developed external levees, and some cross-sections show inset terraces, recording a history of downcutting, lateral erosion, and inner levee deposition (Jobe et al., 2017). Width and depth vary significantly along the channel, with widths ranging from 406 m to 1131 m and depths ranging from 32 m to 73 m (P10 to P90 values; Table 1), reaching a minimum of ∼8 m deep where the channel rapidly loses confinement at its mouth (Figs. 3A and 10C). The width and depth changes are not gradual or systematic along the length of the channel, but rather occur as discrete, localized adjustments. In some channel reaches, width and depth increase or decrease together, while in other reaches, they appear to change in opposition to each other (e.g., the 15–25 km reach and near the channel mouth; Figs. 10B, 10C). The channel cross-sectional area and aspect ratio vary along with changes to width and depth (Figs. 10D, 10E). The aspect ratio ranges from ∼6:1 to ∼30:1 (Table 1), with the highest (>P90) aspect ratios up to ∼80:1 occurring along exceptionally wide reaches of the channel, including the channel mouth (Fig. 10E).

Gulf of Alaska Channels

Although the Surveyor fan in the Gulf of Alaska, offshore the southern coasts of Alaska, USA and Canada in the Pacific Ocean, is not connected to any major fluvial inputs, it has received large volumes of sediment associated with glacial erosion in the St. Elias Mountains since late Miocene time (Reece et al., 2011). The Surveyor channel is unique among deep-water channel systems in that it terminates into the Aleutian subduction trench (Reece et al., 2011). We mapped the Surveyor and Chirikov channels and their associated tributary legs (Fig. 3B), with 131–271 km mapped for each channel (Table 1). With median widths of 3.3–6.8 km and depths of 56–132 m, the Gulf of Alaska channels are exceptionally large relative to other channels mapped in this study (Table 1; Figs. 5A, 5B, and 6E). These glacially influenced channels likely represent upper end-members of the range of submarine channel morphologies. Due in part to their large scale, determining the transition point between canyon and channel is difficult for the Gulf of Alaska channels. Additionally, modification of the levees by sediment waves obscures the location of the true levee crest, so channel margins were mapped using the rollover point (as for incised channels; Fig. 4C). Given these uncertainties, we consider the Gulf of Alaska channels as a group rather than performing a detailed analysis on any single channel from the region. The channels exhibit broad bends rather than classic meander loops, and their average whole-channel sinuosities (1.1–1.3) are among the lowest recorded in this study (Table 1). The P10–P90 aspect ratios occupy a similar range of values as those of the Amazon and Bengal fans, from a low of 22:1 to a high of 145:1 (Table 1; Figs. 5C and 6E).

Western Gulf of Mexico Channels

Channels mapped in the western Gulf of Mexico, off the southern coast of the USA, range from 8.2 to 51.8 km in length, mapped at 12.5 m lateral resolution (Table 1). Channels 1–8 and 10 are associated with the Rio Grande delta, offshore southernmost Texas, USA (Banfield and Anderson, 2004), and channels 9 and 11 are part of the Brazos-Trinity slope minibasin system (Mallarino et al., 2006). Many of these channels appear to initiate and terminate on the slope, but they exhibit weakly to well-developed levees and sinuous planform geometries, distinguishing them from submarine slope gullies (e.g., Field et al., 1999; Shumaker et al., 2017). Some of these channels, most notably GoM 5, GoM 6, and GoM 9, traverse variable slope gradients caused by salt tectonics, and as a result exhibit dramatic changes in width and depth with downstream distance, similar to the Niger 9 channel (Fig. 10). Some channels crossing steep seafloor gradients become deeply incised, resulting in low aspect ratios. Given that the western Gulf of Mexico channels (GoM 1–11) are more significantly impacted by salt tectonics than the GoM 12 channel in the eastern Gulf of Mexico, we consider them separately in data compilations (Figs. 5 and 6).


Hydraulic Geometry of Submarine Channels

Numerous properties of fluvial channels (e.g., width, depth) exhibit power-law scaling with channel discharge (and by analogy, length; Leopold and Maddock, 1953; Wohl and David, 2008). A fundamental process that links fluvial channel width, depth, and drainage area to channel length is the accumulation of discharge through tributary input along a trunk stream (Leopold and Maddock, 1953). The processes of precipitation and overland flow, which are important factors in the formation of tributary geometries and cumulative downstream discharge, do not have obvious analogs in deep-water environments. Indeed, tributary geometries are common at submarine canyon heads but are rarely observed for submarine channels on continental slopes and basin plains. Instead, our understanding of turbidity current dynamics predicts that discharge will decrease with downstream distance, as the current gradually loses sediment load to overspill and levee construction. This process should result in a progressively smaller channel with increasing downstream distance, as has been observed in some instances (Fig. 7; Babonneau et al., 2002; Pirmez and Imran, 2003). Thus, submarine channels should exhibit scaling relationships between width and length, and depth and length, opposite those observed in fluvial channels. Konsoer et al. (2013) documented submarine channel width-depth and depth-discharge relationships that follow a power law similar to that calculated for fluvial channels. However, the width-depth correlation is poorly constrained due to limited data and the derived depth-discharge relationship is dependent on flow sediment concentration, a value that has never been measured directly in natural systems (Talling et al., 2015). Additionally, Konsoer et al. (2013) note a bias toward larger channels in their data set (widths ranging from 300 m to 30 km) due to limitations of data resolution. Below, the width-depth relationships exhibited by channels in this study are examined from the perspective of regional and global trends in aspect ratios, and changes in width and depth in individual channels downstream.

Channel Aspect Ratios

Each geographical region documented in this study exhibits a similar range of channel aspect ratios, despite variability in channel dimensions (Figs. 5 and 6). The median aspect ratios for each region range from 17:1 (Niger) to 63:1 (Amazon); excluding the two regions with strongly incisional channels (Niger slope and western GoM), the range is 30:1 (Bengal) to 63:1 (Amazon; Fig. 5C). The similarity of channel aspect ratios despite variations in width among systems suggests that channels of all scales may have a preferred geometry, but this geometry may vary slightly depending on channel dimensions. The upper and lower bounds on the range of observed aspect ratios could be caused by self-correcting feedbacks in channel-forming processes (Fig. 11). For example, very low aspect ratio channels (approaching 1:1) are not likely to be maintained or preserved because they require steep channel walls that would be prone to mass failure (e.g., Hansen et al., 2015), which promotes channel widening. Very high aspect ratio channels (approaching 1000:1) have a diminished ability to confine turbidity currents, driving either overspill and levee aggradation (which would serve to decrease channel aspect ratio), or full loss of confinement (i.e., channel-lobe transition; Wynn et al., 2002). The autocyclic process of levee aggradation driven by low channel relief is consistent with interpretations of basinward channel propagation (Babonneau et al., 2010; Hodgson et al., 2016; Bengal 1 channel in this study). Kane et al. (2007) noted possible autocyclic feedbacks between thalweg aggradation and upward increases in levee event bed thickness. Similar methodologies could be used to test the ideas discussed here, particularly given paired core and shallow seismic reflection data sets through submarine levees.

The observed trend of decreasing aspect ratios with increasing channel scale (Fig. 6) aligns with expectations for submarine channel behavior—for instance, we anticipate shallowing and widening at channel mouths as confinement is lost and flows can spread laterally (e.g., Bengal 1 and Niger 9 channels; Figs. 8 and 10). Additionally, channels are typically largest at their upstream end, where due to steeper slopes they may be partially confined by erosion, and therefore have lower aspect ratios (e.g., Amazon 1 channel; Fig. 7). The scale-dependency of aspect ratios among fluvial channels is opposite that observed for submarine channels in this study—that is, fluvial channels exhibit increasing aspect ratios with increasing scale (b>1), while the submarine channels studied exhibit decreasing aspect ratios with increasing scale (b<1; Fig. 6E). Although aspect ratios in fluvial systems are observed to vary with other controlling factors such as slope (Palucis and Lamb, 2017), we assume that the broad range of both fluvial and submarine channel dimensions examined here accounts for such variables. Additionally, the relationship of decreasing aspect ratio with steeper slopes observed for rivers (Palucis and Lamb, 2017) is opposite that observed for the Amazon 1 channel, which exhibits increasing aspect ratio with decreasing slope downstream (Fig. 7). We suggest the differing relationships between channel scale and aspect ratio in fluvial and submarine systems may be explained by the greater ability for submarine channels to build high levees relative to fluvial channels, due to the great thickness of turbidity currents driven by the low density contrast between the currents and ambient seawater (Imran et al., 1999). Fluvial channels can only aggrade levees during bankfull flow, at which time there is also increased likelihood for lateral bank erosion. Fluvial levee growth is also necessarily limited, as overtopping flows can only be incrementally deeper than the levees that contain them. Therefore, fluvial channels accommodate increasing discharge predominantly through widening (Leopold and Maddock, 1953; Savenije, 2003). In contrast, submarine channel levees can aggrade tens to hundreds of meters above the channel thalweg (and even higher above the surrounding seafloor; Hansen et al., 2015; Jobe et al., 2016) and are composed of compacted layers of sand and mud, potentially limiting their ability to shift laterally via erosion. This may drive submarine channels to accommodate changes in discharge primarily through changes in depth, either by levee growth/erosion or thalweg deposition/erosion, rather than through changes in width. The relatively stable width observed in the Bengal 1 and GoM 12 channels, even with changing depth (Figs. 8 and 9), supports this idea.

Downstream Variations in Width and Depth: Linkages to Submarine Channel Evolution

Although the data suggest that the range of channel aspect ratios may be independent of geographic region, and therefore likely driven by autogenic channel processes, the clustering of channel width measurements by geographic region (Figs. 5 and 6) implies some component of allogenic, system-specific forcings on width. The Amazon, Bengal, and Gulf of Alaska channels exhibit the largest dimensions, possibly related to the large water and sediment discharge supplied to these systems (Milliman and Meade, 1983; Ludwig and Probst, 1998; Swartz et al., 2015), although the relationship between channel dimensions and catchment parameters is not yet clear (Sømme et al., 2009; Pettinga et al., 2018). The four highlighted channels in this study each display distinct width and/or depth trends with downstream distance, although all except the GoM 12 channel show overall increasing aspect ratio downstream. These trends are likely partially influenced by incomplete bathymetric coverage of many of the mapped channels (i.e., most cannot be mapped entirely, from canyon to channel mouth). Some channels (e.g., western GoM, Niger 9) are influenced by shale or salt tectonism and thus are not expected to have an equilibrium downstream trend. Since the Amazon 1 channel is nearly completely imaged, from canyon-channel transition to near the channel mouth (Jegou et al., 2008), we can use this channel as a benchmark against which to compare the observations from the Bengal 1 (Fig. 8) and GoM 12 (Fig. 9) channels. Given the large scale of the Bengal submarine fan, we have likely documented roughly the lower half of the Bengal 1 channel (Schwenk et al., 2005), including the channel mouth, while the mapped portion of the GoM 12 channel spans an arbitrary portion near the middle of its length. The lower half of the Amazon 1 channel (from ∼200 km streamwise distance) exhibits much less variation in width and depth than its upper half (Fig. 7), indicating that the stable widths and aspect ratios observed in the upper Bengal 1 channel and the GoM 12 channel may be typical of medial to distal portions of submarine channels. This is consistent with the modeling results of Traer et al. (2018a, 2018b) that suggest turbidity currents can achieve flow equilibrium within the upstream reaches of a channel and maintain it over long distances, producing a flow filtering effect and consistency of flows that traverse the full length of the channel. The dramatic depth decrease at the end of the Bengal 1 channel can be attributed to the channel-lobe transition (see below), which is not mapped in the Amazon 1 and GoM 12 channels. The more gradual depth decrease over more than 200 km of streamwise distance in the Bengal 1 channel initiates around the same point at which the channel transitions from relatively high sinuosity to relatively low sinuosity (Fig. 8). The lower sinuosity reach also has fewer instances of meander migration, and no visible meander loop cutoffs, suggesting that it may be a younger, less mature portion of a progradational channel (e.g., Babonneau et al., 2010). If so, the data would suggest that channel width is established earlier than channel depth—i.e., once levees begin to form, their width remains relatively fixed, while their relief is able to increase through levee aggradation and/or thalweg erosion (Fig. 11).

Assessing whether channel depth changes are accommodated by deposition/erosion of levees, thalweg, or both may be important for understanding partitioning of sand and mud in submarine channels and their resulting deposits. Decrease of levee growth downslope indicates a progressive loss of fine-grained sediment load, while shallowing or steepening of the thalweg profile implies a changing propensity for deposition or erosion, respectively. While depth measurements alone cannot differentiate between levee and thalweg adjustments, analysis of thalweg and levee crest longitudinal profiles may provide some insights. For example, the Amazon 1 channel exhibits a slow and roughly steady decrease of levee relief (i.e., vertical distance of the levee crest above the channel thalweg) downslope, as evidenced by the steeper longitudinal profile of the levee crest compared to the thalweg (Fig. 7A). This is consistent with the interpretation of decreasing discharge with downstream distance in an equilibrium system, which would cause diminishing levee aggradation downstream (Pirmez and Imran, 2003). The gradual decrease of levee relief between ∼400 and 600 km distance of the Bengal 1 channel (Fig. 8A) could reflect the possible immaturity of the downstream portion of that channel, and suggests that levee aggradation occurs more slowly than downstream propagation of the thalweg. The rapid depth decrease at the channel mouth in both the Bengal 1 and Niger 9 channels is accommodated primarily by loss of levee relief, as indicated by the relative stability of the thalweg profiles through those zones (Figs. 8A and 10A). The Bengal 1 channel also shows rapid adjustments to the thalweg profile over short distances in the final ∼50 km, perhaps suggesting localized deposition and erosion near the channel mouth, but these irregularities in the thalweg profile are not reflected in the levee crest profile (Fig. 8A). If calibrated with core or seafloor backscatter data from a few modern channels, these morphological details could aid in better understanding and predicting how channel thalwegs and levees behave at the channel-lobe transition zone.

The Niger 9 channel provides a counterpoint example to the three leveed channels on relatively smooth slopes analyzed above. The Niger 9 data highlight the behavior of thalweg and levees as the channel traverses an uneven slope. Assuming that channel cross-sectional area reflects flow discharge, we expect to see a relatively stable or decreasing cross-sectional area along a given channel (e.g., Figs. 79). Instead, the Niger 9 channel exhibits fluctuating cross-sectional area due to the inconsistent relationship between width and depth along the channel—i.e., width increases are not always associated with depth decreases (Fig. 10). The thalweg and levee crest profiles show a possibly related pattern: there is an apparent lag in the response of the levee crest gradient to thalweg gradient changes, leading to higher levees at the upstream end of steep reaches, and smaller levees at the upstream end of low-gradient reaches (Fig. 10A). Catterall et al. (2010) interpreted unusual downstream thickening of levees in the Noor channel-levee system, offshore the Nile Delta of Egypt in the Mediterranean Sea, as evidence of structurally controlled changes in sediment entrainment and deposition, highlighting the complexity of levee relief changes in structurally complex areas. However, a similar lag between levee crest and thalweg slope changes can be seen in the middle of the Bengal 1 channel (Fig. 8A), where local structural perturbations of the slope are not observed. The fluctuating cross-sectional area and possible disconnect between levee and thalweg adjustments to slope could indicate that levee-forming processes operate over longer timescales than thalweg-modifying processes, causing channel geometry to depart from flow discharge in non-equilibrium settings. A slower rate of levee modification relative to thalweg changes is consistent with the vertical stratification of turbidity currents, such that the most energetic and concentrated part of the flow directly impacts the thalweg and can erode or deposit thick packages of sediment, while the levees are affected primarily (or exclusively) by the dilute upper part of the flow, which cannot deposit or erode as rapidly (e.g., Fig. 11; Conway et al., 2012). Turbidity current modeling efforts by Traer et al. (2018a, 2018b) show that currents may take tens to almost 100 km to achieve flow equilibrium (between entrainment and overspill or flow stripping) following a perturbation to channel slope, depth, or other parameters. These findings are consistent with our observations from the Niger 9 channel and suggest that the long adjustment period of flows relative to topographic irregularities and channel length could be reflected in the morphologies documented here.

Morphometrics of the Channel-Lobe Transition Zone

Of the four main channels studied, only the Niger 9 and Bengal 1 channels are imaged across the channel mouths. Niger 9 loses confinement around 57 km streamwise distance (Fig. 10) and deposits a lobe where it intersects a large, low-gradient section of seafloor caused by mobile shale tectonism (Fig. 3A; Jobe et al., 2017). Bengal 1 dissipates for no detectable reason around 706 km streamwise distance while a separate but adjacent (inactive, partially buried) channel of similar scale continues downslope (Fig. 2B). Both channels exhibit a rapid (1–5 km streamwise distance) width increase of over 800 m (from 350 to 1165 m in the Niger 9 channel, and from 1370 to 2260 m in the Bengal 1 channel), although in neither case does the channel width near the mouth exceed the maximum recorded width from cross-sections upstream. As noted above, both channels also undergo rapid loss of levee relief over a short distance, and the Bengal 1 channel shows evidence for localized thalweg deposition and erosion (Fig. 8). The Bengal 1 channel maintains its smaller geometry over ∼100 km of streamwise distance, while the Niger 9 channel dissipates more abruptly. The differences in depth-distance behavior at the mouths of these two channels could reflect the differences between slope gradient disturbances and flow properties as key factors dictating the end of a channel. Perhaps the Bengal 1 channel diminished gradually at the natural runout length of the majority of its flows, while the Niger 9 channel was forced to terminate due to an abrupt (fault-controlled) shallowing of the seafloor that drove flows to collapse (Fig. 3A). Sandy submarine lobe deposits associated with the Niger 9 channel (Jobe et al., 2017) corroborate the interpretation that flows abruptly exited the channel at disequilibrium. Core samples and high-resolution, shallow subsurface seismic profiles would help ascertain whether the channel-mouth deposits of the Bengal 1 channel differ significantly from those of the Niger 9 channel, as might be expected from the observed differences in the channel termination morphologies. The interpretation of a gradual flow dissipation for the Bengal 1 channel does not explain the similarly abrupt decrease in depth ∼100 km upstream from the channel mouth (Fig. 8C), but agrees with the previous interpretation that the channel is young, immature, and progradational as indicated by its low sinuosity and the limited evidence for migration at its downstream end (see above).

Although both the Bengal 1 and Niger 9 channels visibly appear to widen near their termini (Figs. 2B and 3A), the change in channel depth relative to the initial depth is far higher than that in channel width; depths approach zero while widths fluctuate no more than observed elsewhere along the channel (Figs. 8 and 10). This further supports the interpretation that channels more readily adjust their depth than their width to changes in flow properties and discharge (Fig. 11). This behavior at the channel-lobe transition may additionally reflect the expected transition from erosive or non-depositional behavior to fully depositional behavior in the basal portion of the flow as it approaches the channel mouth, such that the thalweg undergoes increasing aggradation while levee aggradation diminishes.


Morphometric analysis of modern submarine channels using existing seafloor bathymetry is a low-cost, efficient methodology for investigating potential autogenic and allogenic controls on submarine channel morphologies. Through a combination of manual mapping and automated data analysis, we have generated a database of nearly 29000 measurements of submarine channel width and depth from 36 channels in five geographic regions. Despite regional variations in channel width and depth, channel aspect ratios consistently occupy the same range of roughly 10:1–100:1, suggesting the potential for allogenic controls on channel dimensions and autogenic controls on aspect ratio. We propose that autogenic feedbacks related to levee aggradation can explain the narrow range of aspect ratios (Fig. 11). The data also suggest that bankfull width is established early in the evolution of a channel, and that channels adjust their depth more readily than their width, perhaps in part due to a disconnect between levee- and thalweg-modifying turbidity current processes. This interpretation is consistent with the observations from two channel mouths that show larger relative changes in depth than width, apparently driven by dramatic decrease of levee growth downstream. This is further supported by the apparent lag in levee relief adjustment to thalweg gradient changes observed in some channels.

Although submarine and fluvial channels share morphologic similarities, observations from this study caution the use of fluvial process analogs to understand or predict submarine channel behavior. We document an opposing relationship between aspect ratio and channel scale for fluvial and submarine channels, where fluvial channel aspect ratios increase with channel dimensions, and submarine channel aspect ratios decrease with larger channel sizes. This highlights the fundamentally different relationship between discharge and channel length in fluvial and submarine systems, as well as differences in levee-building capacity and lateral mobility for subaerial and subaqueous channels.

The database created in this work can be applied and built upon in future studies. Documenting the range of channel dimensions globally, as well as the variability within individual channels, provides useful constraints for numerical and analog models that attempt to predict sediment gravity flow responses to topographic changes, such as slope or base level. The trends observed here encourage future research into the local, allogenic controls on channel width, which may be best approached through a comprehensive study of individual submarine channel systems in the context of the data presented here.


The authors thank the Chevron Center of Research Excellence (CoRE) at the Colorado School of Mines, Golden, Colorado, USA, for funding this research. D. Cai was supported by the Undergraduate Research Fellowship at the Colorado School of Mines. We thank Ashley Harris, Fabien Laugier, Morgan Sullivan, and the CoRE research group for valuable insights and discussions. This work would not have been possible without the publicly available seafloor data provided by the Bureau of Ocean Energy Management, University of New Hampshire’s Center for Coastal and Ocean Mapping, Durham, New Hampshire, USA, and the National Oceanic and Atmospheric Administration’s National Centers for Environmental Information, Asheville, North Carolina, USA. Bathymetric data from the Bengal Fan were collected during RV SONNE expeditions SO125 (BMBF Grant 03G0125A) and SO188 (BMBF Grant 03G0188A) in cooperation between the University of Bremen, Germany, and the BGR Hannover, Germany. The data were processed with the open-source software MB-System and provided by Tilmann Schwenk and Volkhard Spiess (both University of Bremen). Comments and suggestions from reviewers Dr. Ian Kane and Dr. Esther Sumner, associate editor Dr. Andrea Fildani, and U.S. Geological Survey internal reviewer Dr. Jon Perkins greatly improved this manuscript.

Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.

Science Editor: Shanaka de Silva
Associate Editor: Andrea Fildani
Gold Open Access: This paper is published under the terms of the CC-BY-NC license.