Channel systems: meandering, braided, distributary, submarine

Why map view is diagnostic

A channel 100 to 500 m wide and a few metres to a few tens of metres thick, set in a much thicker succession of mudstone, is a small anomaly on a vertical section. It is easy to miss, and a section that cuts it obliquely makes it look wider than it is: crossing a channel of width $w$ at an angle $\alpha$ from square, the section spans $w/\cos\alpha$.

In map view the same channel is a sinuous trail across the landscape, and its plan form is what identifies it. A horizon slice, the amplitude extracted along a mapped horizon or in a window that follows it, stays at the stratigraphic level of the channel. A time slice cuts the volume at one two-way time; where the strata dip, it crosses the channel's level only in a band and shows the channel in pieces. That is why channels are mapped on horizon slices (Brown 2011).

Two things make the trail visible. The channel sand has a different acoustic impedance from the mud around it, commonly lower in young, shallow sections, so its top reflects as a trough. And below the tuning thickness, where the reflections from the top and the base of the sand overlap into a single wavelet, a section can no longer resolve the sand's thickness, yet the amplitude of that wavelet still grows with thickness (Widess 1973). A horizon slice therefore maps channels that no section resolves, brightest where the sand is close to the tuning thickness.

Exercise, the four channel types

• The figure opens on the meandering river. Its sinuosity, in the table, is the length of the channel over the 12 km of valley it crosses; above 1.5 a channel counts as meandering. The banded sand inside each bend is the point-bar belt the migrating bends left behind, and the dark loop is a clay plug: a bend whose neck closed, was cut off and filled with mud.
• Section A–A′ starts across the clay plug. In (c) the mud-filled loop cuts the point-bar sand into separate bodies, and the inclined lines inside the point bars are lateral-accretion surfaces, each a former inner bank. Drag A–A′ along the valley, or press Walk A–A′ downstream, and watch the count of separate sand bodies.
• Switch to Braided. Each thread is nearly straight, its sinuosity under 1.5, yet several cross every line across the valley: the braiding index counts them. In (c) the threads are shallow cuts in one connected sand sheet. Braided rivers carry mostly bedload and are wide and shallow, meandering ones narrower and deeper for their width (Schumm 1985): compare the width-to-depth ratios in the table.
• Switch to Distributary and walk A–A′ from the apex toward the sea. The channels split and never rejoin, narrowing at every split, and the section breaks into more and smaller sand bodies seaward; past the shoreline it cuts mouth bars. The abandoned distributary, filled with mud, shows in (a) and is missing from the horizon slice in (b).
• Switch to Submarine. Its sinuosity is in the range of meandering rivers, so sinuosity alone cannot tell the two apart; the levees can. They are dim wings on the horizon slice and wedges thinning away from the channel belt in (c), the gull-wing section of deep-water channels. Then lower the frequency, or switch (b) to a time slice, and compare what seismic shows with the deposits in (a).

Sinuosity as a diagnostic

Sinuosity is the length of a channel divided by the length of the valley it runs down, $P = L_{\mathrm{channel}}/L_{\mathrm{valley}}$. Leopold and Wolman (1957) took 1.5 as the threshold for calling a channel meandering, and the convention has stuck. Together with the number of threads it gives the standard plan-form classification (Miall 1996): single-thread channels are straight below 1.5 and meandering above it, and channels with several threads are braided below 1.5 and anastomosing above it. Typical ranges:

• Meandering river: above 1.5, commonly up to about 3; one thread, with point bars on the inside of its bends.
• Braided river: below 1.5 for each thread; several threads that split and rejoin around bars, so the braiding index, the mean number of channels a line across the valley crosses (Egozi and Ashmore 2008), is above 1.
• Distributary: commonly 1.0 to 1.3 along each path; the diagnostic is that the channels split downstream and never rejoin.
• Submarine slope channel: from nearly straight to about 3, overlapping the meandering range; the diagnostic is the levees, wedges that thin away from the channel on both banks, and the deep-water setting.

Measuring sinuosity on a horizon slice is quick: trace the channel's centre line, measure its length and divide by the length of the valley over the same reach. Because the meandering and submarine ranges overlap, sinuosity alone does not separate a river from a slope channel; the levees, the section and the regional setting do.

Watch a meander belt build itself

Everything above describes the finished planform. Figure 4.4.2 grows one from a nearly straight channel, with the kinematic rule of Howard and Knutson (1984), the rule Sylvester's open-source meanderpy implements. Every point of the channel moves sideways at a rate set by the local curvature plus a weighted sum of the curvature upstream, because the flow needs distance to respond to a bend. That one rule produces what an interpreter maps on a timeslice: bends grow and travel downstream, the inner bank builds a point bar while the outer cut bank erodes, loops pinch off at the neck and become oxbow lakes, and the abandoned channel positions pile up into scroll bars and clay plugs.

Plate (a) is the deposit record seen from above, the middle 12 km of a 20 km valley, the river flowing from left to right. The river is drawn at its true width, 200 m at the start, and every deposit keeps the width it was laid down with. The point-bar sand it has swept is striped every 16 model years, so each stripe is a time line and a wide stripe means fast migration. A cut-off loop is an open oxbow lake until mud fills it, and then a clay plug. On a real amplitude extraction the clay plugs are often the most visible elements of a belt, because a mud-filled loop sits acoustically apart from the sand around it. The map follows the river across the valley, so the channel stays on it as the belt widens; when the channel spans more than the map is tall, a note on the map says so.

The figure opens on a young, nearly straight channel, because the growth of the first bends is the part worth watching. For about two centuries little seems to happen; then a slight irregularity steers the flow against one bank, that bank erodes, the bend sharpens and the growth feeds itself, so the bends appear within about a hundred years. The migration rate $k_l$ sets the pace. It is the bank-erosion rate per unit of the dimensionless curvature $W/R$, channel width over bend radius, so at the default 25 m/yr the fastest point of the sharpest bend moves about 15 m a year. The slider's 10 to 40 m/yr is the range in which this model runs cleanly, not the range of real rivers: faster, and the model's step limit rather than the rule begins to set the pace of the sharpest bends. Rivers span far more. The Ucayali in Peru, 600 to 700 m wide, migrates tens of metres a year on average and more than 750 m a year at some bends (Schwenk et al. 2017); small streams move far less.

Flow friction is the one control whose meaning is not obvious from its name, and it is the physical heart of the model. It is the drag coefficient $C_f$ that the bed exerts on the flow. In the Howard and Knutson rule the curvature a distance $s$ upstream counts with the weight $e^{-2C_f s/D}$, so the flow carries the memory of a bend for about $D/2C_f$, where $D$ is the channel depth: 278 to 397 m across the slider for a channel 200 m wide. That memory sets how long the bends are, at the default friction about 10 channel widths from a bend to the next on the same bank, close to the 11 widths Leopold and Wolman (1960) found on natural rivers, and how soon their necks close. More friction means a shorter memory, shorter and tighter bends and more cutoffs; less friction means longer, lazier bends.

The channel width $W$ scales the whole rule. The depth is $W/18$, so the memory is $W/36C_f$, and a neck is cut off when its limbs come within 2.2 widths. A wider river therefore builds longer bends in proportion, still about 10 widths from one to the next across the slider's 140 to 260 m, and because $k_l$ fixes how fast its banks move in metres a year, it takes longer to build them in the same proportion. Beyond those widths the model's grid and step limit, not the rule, would start to shape the bends. A width changed while the river runs acts at once, and what is already laid down keeps the width it had.

Plate (c) shows what the rule computes at the sharpest bend in the reach. The tinted ribbon is where the rule alone, without the noise, resampling and cutoffs of the full run, carries the channel in fifteen years, and every arrow is a migration vector from a point of the channel to its place on that ribbon. The point of maximum curvature and the point of maximum migration do not coincide: the migration peak sits downstream of the bend apex, by about one and a half memory lengths, and the dashed line along the channel measures the lag. It is not a numerical quirk. Sylvester, Durkin and Covault (2019) found the same downstream offset on rivers of the Amazon basin: once it is allowed for, migration rises steadily with curvature, and the sharper the bend, the faster it moves.

Plate (b) keeps the score. The sinuosity of the reach climbs from 1.0 past 1.5, the threshold for a meandering channel, then saws up and down: each neck cutoff shortens the river at a stroke, and the bends grow back. Once cutoffs begin, the reach mostly runs between about 2 and 4, higher than most natural rivers, because the model has no valley walls, chute cutoffs or avulsions to straighten it.

Exercise, run the river

1. Press the first setup under Figure 4.4.2 and watch plate (b): the sinuosity sits near 1.0 for two centuries, passes 1.5 at about 345 model years, and the first neck in the reach closes at about 470. From then on it saws up and down, the history every mature meandering river lives on.
2. Run the river with the fastest migration and the shortest memory, then with the slowest migration and the longest (the second and third setups). The first cuts off a loop about every ten model years; the second needs about 2,200 years to cut off its first, and only 8 in the 4,800 years the record holds. Migration rate sets the clock; friction sets how long the bends are and how soon their necks close.
3. Set the channel width to 260 m and press New river, then do the same at 140 m. The wider river's bends grow about twice as long, still about 10 widths from one bend to the next, and its first neck in the reach closes about twice as late, typically near 600 model years against 300: every length in the rule scales with the width, while $k_l$ fixes how fast the banks move.
4. Turn the mud supply down and oxbow lakes stay open water for 275 model years; turn it up and they plug into clay-filled scars within 25. In the subsurface that difference decides whether abandoned loops act as barriers inside a sand belt.
5. With the river paused by the fifth setup, raise the flow friction from 0.014 to 0.020 and watch the lag in plate (c) shrink, from about 630 m to 450 m on that bend: the offset exists because the flow carries an upstream memory of the bend, and friction sets how long that memory is.

Net-to-gross by channel type

• Meandering channel belt: NTG 15-45% typically. Point bars and channel fills are sand; overbank (floodplain) is shale. Amalgamation of multiple generations can raise NTG.
• Braided channel belt: NTG 70-90%. Braided rivers carry mostly bedload, so the belt is typically amalgamated sand or gravel with minor mud interbeds.
• Distributary network: NTG 25-60%. Mix of distributary channel fills + distributary mouth bars (sandy) separated by interdistributary mud.
• Slope channel complex: NTG 30-70%. Channel-axis sands are high-quality reservoir; levee wings are muddier. Multi-stage channel complexes can amalgamate to high NTG at the axis.

Interpretation workflow

When you have a 3D volume and want to identify channel systems:

• Pick a regional horizon that likely captures a depositional system of interest (e.g., a sequence boundary with lowstand-fan fill above, or an MFS with overlying HST deltas).
• Generate a horizon amplitude extraction or RMS amplitude extraction in a window above/below the horizon. Channels typically show as bright anomalies against a dimmer background: a strong trough where the sand is softer than the mud around it, or a high value on an RMS extraction.
• Examine the plan-view pattern. Is it a single sinuous trail (meandering or slope)? Multiple sub-parallel threads (braided)? A bifurcating radial network (distributary)?
• Cross-check with section view. Confirm the channel on an inline or crossline that cuts across it. Look for the characteristic U-shape or V-shape cut with fill.
• Map the channel belt in 3D. Use the amplitude extraction to trace the channel laterally and identify tributary / distributary branches.
• Estimate reservoir potential. Channel type + width + length + likely NTG → estimate the reservoir volume. Combine with rock-physics-derived fluid prediction (Part 5) for full characterization.

Pitfalls

• Mistaking mass-transport deposits for channels. Submarine mass-transport complexes (MTCs) can also produce bright amplitude anomalies with curvilinear patterns. Distinguish by shape: channels are confined ribbons; MTCs are bulk-deposited masses with irregular shapes.
• Confusing modern river geomorphology with ancient systems. Modern rivers on satellite imagery are sharp and crisp. Ancient channels on seismic are blurred by seismic resolution and fluid effects. Calibrate your expectations.
• Ignoring compaction effects. A mud-filled channel 50 m thick when deposited may be 30 m thick after burial; a sand-filled one compacts much less than the mudstone around it. This differential compaction can warp the channel geometry on seismic, making it look more undulating than it actually is.
• Assuming channel fill = reservoir. Not all channels are reservoirs. Some are filled with MUD (abandoned channel fill) not sand. Look for amplitude brightness + check impedance via AVO or rock-physics analysis.
• Over-interpreting isolated channel segments. A horizon extraction may show one channel segment; without confirming it extends regionally, you may be looking at a single isolated channel rather than a trunk system. Map carefully before committing.