Remarkably little is reliably known about the sedimentology of river avulsions: we know they occur, but few have been described in detail. Most accounts in the literature are interpretations of the rock record based on hypothesis not empirical evidence. The little detailed evidence we have comes mostly from humid and temperate regions. Should we expect the avulsion behaviour of dryland rivers to be similar? If not, then what differences are likely?

Here, I summarise the position as I see it from the literature. Then I propose a series of questions I'd like to see discussed.

Please contact me with your own ideas. Do you agree with my summary of the current positon? Have I missed key literature? How should we proceed?

Our lack of understanding is surprising given the unanimous acceptance that avulsion is a primary process that determines channel location over the longer term, and therefore is a controlling influence on the large-scale distribution of river sediment. The significance of avulsion has been explicitly recognized in alluvial architecture simulations for a long time, both experimental (e.g. Schumm, 1977) and computer-based (e.g. Allen, 1978; Leeder, 1978; Mackey & Bridge, 1995).

One reason for the lack of data on avulsions is that, by human timescales, they occur infrequently and take a long time to reach completion. Another reason may be the gap between geomorphologists and sedimentologists: the former work at too short a time span, on too small a section of river system, and don't work enough on the issues of preservation potential; and the latter ignore geomorphological studies, and consider work on the Quaternary to be 'gardening'.

The three main causes of avulsion are (Jones & Schumm, 1999):

1. Decrease in gradient of existing channel, such as produced by increasing sinuosity or tectonic uplift;
   
   
2. Increase in gradient away from the channel, such as occurs when the river builds an alluvial ridge;
   
   
3. Reduction of the capacity of the channel to carry all the water and sediment, such as caused by log or ice jams, increased sediment input from tributaries, slope failure, or aeolian incursions.

In principle, all of these can operate in drylands, but, in practice, the likelihood of occurrence is different for some factors. Alluvial ridges are less common because clay is not produced in large amounts in drylands (chemical weathering is much reduced), so river banks are less stable and levees are much reduced in height. On the other hand, evaporation and infiltration of water mean dryland rivers commonly deposit large amounts of sediment prematurely in-channel, so locally reducing channel gradient and bringing the system nearer to conditions suitable for avulsion to occur.

There is no example of avulsion from drylands as extensively studied as that for humid regions of the Saskatchewan River (Smith et al., 1989; Pérez-Arlucea & Smith, 1999; Morozova & Smith, 1999; Morozova & Smith, 2000) - there are in total few documented examples of avulsion deposits. To understand avulsions in drylands, we have to piece together snippets of information from several places.

RIO GRANDE: 1

One of the most illuminating is the analysis by Mack & Leeder (1998) of the channel shifting of the Rio Grande in southern New Mexico, prior to intensive human interference in the flow and floodplain of that river (dams, irrigation canals, and so on). Using historic records, they have mapped out the avulsions that occurred in the period 1844-1912. Arguably, this is still too short a period to draw far-reaching conclusions, but it is the best we have.

The Rio Grande flows through a subtropical semi-arid region, but has seasonal discharge variations, in this case fed by snowmelt in the mountains of southern Colorado and northern New Mexico. In the period of analysis, mean annual temperature was about 16°C and mean annual precipitation about 220 mm (half of which fell in July-September). During most of the year, discharge was less than 10,000 cfs, and occasionally the river stopped flowing all together. But peak discharges, in May and June, exceeded 25,000 cfs, sometimes reaching 50,000 cfs, and it was in these periods that major shifts in channel pattern took place.

The historic Rio Grande was a pebbly sand bedload stream that displayed a wide range of channel widths (100 - 1,300 m) and sinuosities (1.2 - 1.9). This variation seems to be intrinsic, there is no evidence of tectonic or climatic control. So the rock record also may contain juxtaposed sand bodies of markedly different architecture without the influence of external factors.

The Rio Grande avulsions shifted the channel laterally a distance not exceeding 5km, a limit which is controlled by valley confinement (the Saskatchewan was free to migrate laterally a much greater distance). Downstream avulsion length (to reoccupation point) was 17 - 30km (the Saskatchewan 1873 avulsion was 55km). Sometimes the new course cut across the old course without rejoining it, which seems counter-intuitive. This also runs counter to the observations for other ancient rivers where it appears an avulsed river commonly rejoins the original course further downstream (Mohrig et al., 2000)

The key point about these historic Rio Grande avulsions is that they left negligible amounts of avulsion deposits because the river established a new course extremely rapidly.

RIO GRANDE: 2

Some 600km back upstream, the Rio Grande is much coarser grained (cobble to pebble) but in places is still a meandering river. Jones & Harper (1998) used historic survey records and aerial photos from 1875 to the present to ascertain the avulsion history of this river in southern Colorado, which also is a semi-arid region. They particularly looked to see what had happened as discharge decreased due to progressively increasing amounts of water being withdrawn for irrigation (analogue for climate change?). Rio Grande mean annual discharge decreased by 60%-70% from 1875 to 1925 because of irrigation withdrawals upstream from the study area.

This study has produced many thought-provoking observations, some of the key ones being:

• Along four reaches, relatively abrupt shifts of the channel to a new location (avulsions) have been common, but in five other reaches the channel has been laterally stable.
  
  
• In the unstable reaches, repeated avulsions have led to the development of subequally spaced, lozenge-shaped internodes composed of coarse-grained point-bar deposits and both coarse-and fine-grained abandoned-channel fill.
  

• In the stable nodes that separate the active internodes deposition has been minimal. In map view, the large-scale depositional geometry of these nodes and internodes is analogous to a string of lozenge-shaped, linked sausages, the links representing the nodes and the lozenges representing the internodes. See Fig. AFx1, which also speculates on the architectural implications.
  
  
• However, study of aerial photos reveals abandoned meander loops in the region of the nodes, suggesting that over time periods much longer than the 120 years of this study the position of stable reaches is transitory.
  
  
• Over this period, meander wavelength decreased from about 500 m to about 320 m, and sinuosity increased from about 1.2 to 1.7. Empirical data from other studies suggest that this was probably a result of decreased discharge: the river becomes less able to transport all the sediment, so the channel aggrades, in turn triggering an increase in sinuosity. The number of two-channel reaches appears to have decreased, probably as a direct result of the avulsion process.
  
  
• During an avulsion, the new channel evolves from low to high sinuosity due to rapid meander growth at the same time that discharge shifts from the old to the new channel. As a result, early in an avulsion two channels of different sinuosity generally are present, but later a single high-sinuosity channel develops.
  
  
• After an avulsion, coarse-grained deposits quickly fill the upstream end of the pre-avulsion channel, while mostly fine-grained sediment slowly fills the remainder of the pre-avulsion channel.
  
  
• On the new channel, point bars grow quickly.
  
  
• The number of avulsions that occurred decreased from about 19 (1875-1941) to 2 (1941-present). Data are inadequate to show what, if any, relationship exists between the decrease in discharge and the change in avulsion frequency.
  
  
• The avulsions here are NOT due to elevation of the channel in an alluvial ridge. It is surmised that coarse-grained sediment accumulates at and above highly sinuous, inefficient and low slope reaches, tending to block flood discharge, and tipping the system towards an avulsion. This is avulsion category 1 of Jones & Schumm (1999) as mentioned above.

Jones and Harper have produced an intriguing animation of the evolution of the Rio Grande over 120 years by combining channel and point bar positions from the numerous maps and aerial photos. It is available for download (various formats) from their website, at http://research.gg.uwyo.edu/joelh/rio/index.html

On the Rio Grande (a dryland river), the immediately post-avulsion channel is relatively straight, yet the pre-avulsion channel was highly sinuous. This is true also for the Thompson River in Australia (a humid region river) (Brizga & Finlayson, 1990). The change in channel morphology is intrinsic to the system, and is not directly due to changes in hydrologic regime. One cannot therefore rely on palaeochannel criteria (e.g. width, depth, sinuosity) as indicators of environmental conditions (climate, discharge, and so on), contrary to received wisdom.

The common feature with both these rivers, and the Rio Grande in southern New Mexico (Mack & Leeder, 1998), is that, despite the difference in climate regime, avulsions were completed extremely rapidly. Yet the Saskatchewan river avulsion is still not complete after 126 years.

ANCESTRAL RIO GRANDE

In southern New Mexico, it is possible to constrain very tightly in time the Plio-Pleistocene deposits of the ancestral Rio Grande by means of magnetostratigraphy and isotope dating of contained volcanic material reworked from surrounding areas of extrusives. These deposits can therefore be used as a proxy for modern day rivers.

In the Pliocene, the Rio Grande region was warm (palaeolatitude 32°N, presence of giant tortoise fossils), yet semi-arid (calcic palaeosols require greater than 100mm but less than about 700 mm precipitation (Royer, 1999)) with seasonal rainfall (wedge-shaped peds, slickensides, deep desiccation cracks).

In the Box Canyon region south of the Robledo Mountains, near Las Cruces, southern New Mexico, the ancestral Rio Grande deposits contain complexes of avulsion deposits preceding deposition of multistorey channel deposits (Pérez-Arlucea et al., 2000).

Avulsion breakout channels are clearly seen, invariably succeeded by and cut by laterally extensive scoured (primary) channel bases. The sand-filled breakout channels are 0.8-2 m deep, and 7-50 m wide, steep-sided, and cut into often pedogenically-modified mudrocks. Many of these channels contain near-vertical cyclindrical burrows, some with spreite, which are interpreted to be escape burrows, possibly of crustaceans.

Although in this region the ancestral Rio Grande contains laterally extensive avulsion deposits, there are notable differences from the Saskatchewan avulsion model. No coarsening-up grainsize trends are observed. Soil types indicate generally well-drained conditions (calcic palaeosols) though gleying in some units suggests the water table was occasionally elevated.

The time taken to complete avulsion, and re-establish 'normal' fluvial processes, is highly variable. The shorter the duration, the less the record left behind (so the deposits can be ignored in modelling). The Rio Grande in southern Colorado typically completed avulsion in less than 11 years; the Thompson River in Australia in four years; the Rio Grande in southern New Mexico in one flood season. Yet the Saskatchewan at Cumberland Marshes is still not complete after 126 years.

The recurrence interval is much harder to estimate. The Rio Grande in southern New Mexico historically has a high frequency of avulsion, with three major avulsions in 68 years in an 80 km reach (recurrence interval 22 years). The Rio Grande in southern Colorado experienced 21 avulsions in 125 years in a 20 km reach (recurrence interval 6 years). Both of these are very much more frequent than the mean avulsion interval (400-1000 years) typically used in alluvial stratigraphy simulations which assume the river builds an alluvial ridge prior to avulsion (e.g. Mackey & Bridge, 1995).

In contrast, the humid-region Saskatchewan River at Cumberland Marshes has experienced at least nine major avulsions during the last 5400 years (recurrence interval 600 years) (Morozova & Smith, 1999). This is about the same frequency as has been used in simulations.

Following the damming of the Missouri River in the 1950s, the lower Niobrara River, a tributary of the Missouri, has been subject to a base-level rise of about 2.9 m. Prior to this, the Niobrara was a wide and shallow, sand-bed braided river. The modest base-level rise has caused the lower reaches to aggrade, increased sinuosity, and caused a rapid increase in the frequency of avulsions (Ethridge et al., 1999). Although this base-level rise was induced by human- activity, such small changes can easily be induced naturally, such as by minor tectonic activity or capture of tributaries. Yet such apparently small changes in river regime can have pronounced effect on fluvial processes, and trigger an entirely new sedimentological style.
 

It is generally assumed, and implicit in all the standard facies models, even recent ones (e.g. Miall, 1992; Miall, 1996; Galloway & Hobday, 1996), that most of the sedimentation on the floodplain occurs through crevasse-splay events and true overbank flooding (discharge exceeding bankfull, so producing out-of-channel and unconfined flow of water and sediment).

The new work on avulsions, aided by better understanding of soils, suggests this view is very misguided, and that in practice the deposits of avulsions dominate the floodplain and the stratigraphic record (Aslan & Autin, 1999). The majority of the floodplain sediment may be accreted immediately after an avulsion, and before the development of a major single channel. In humid regions, avulsion deposits occupy a large portion of the valley (e.g. Aslan and Blum 1999) and stratigraphic interval (e.g. Kraus & Wells 1999 for the Bighorn Basin). This style is not included in any published fluvial simulations.