Alluvial plains along the Andean foreland represent a large part of the South American wetlands and seasonally flooded landscapes and provide important ecological services (Melack and Hess, 2011; Junk, 2013). These landscapes are characterized by fragile hydrological systems, increasingly threatened by climate change and human activity (Junk, 2013). These alluvial plains are built with the sediments that rivers bring from the eastern flank of the Andes and deposit on the Andean foreland basin. River activity is continuously reshaping the landscape, with far reaching implications for rural populations and biodiversity. Through meandering, the formation of crevasse splays, avulsions and backswamp sedimentation, rivers fill sedimentary basins (Slingerland and Smith, 2004); they create an irregular topography, favouring the formation of diverse ecological niches (Lewin and Ashworth, 2014); they generate the flood pulses that maintain the biota in river-floodplain systems (Junk et al., 1989); and they cause disturbance in forest structure, which, in turn, is key in creating and maintaining biodiversity (Salo et al., 1986; Nelson et al., 1994). River activity can cause important economic losses (Latrubesse et al., 2009b; Marengo et al., 2013) and greatly affect the livelihoods of rural communities, particularly indigenous people who are often settled along these rivers and depend on their resources (Pärssinen et al., 1996). Understanding what controls fluvial processes in the Andean foreland basin and how these rivers react to external forcing is fundamental in order to foresee how floodplains and alluvial plains will respond to future pressures (Thompson et al., 2013).

In the last few decades, an increasing number of studies in the Andean-Amazonian foreland basin have furthered our knowledge of river dynamics and flood-plain erosion/sedimentation processes and forest disturbance (do Nascimento Jr. et al., 2015; Dunne et al., 1998; Salo et al., 1986; Peixoto et al., 2009; Constantine et al., 2014; Aalto et al., 2003; Latrubesse et al., 2009a; Wittmann et al., 2009). In the southern Amazonian foreland basin (SAFB) (Espurt et al., 2007) it has been shown that large river floodplain sediment accumulation observed in ∼ 100 dated floodplain cores is primarily controlled by the El Niño/Southern Oscillation (ENSO) cycle (Aalto et al., 2003), with warm (El Niño) phases causing smaller shorter floods and low sedimentation rates and cold (La Niña) phases causing larger longer floods and high sedimentation rates (Aalto et al., 2003; Schöngart and Junk, 2007).

However, most of these studies have focused on the Amazon River's main tributaries, Strahler stream order higher than 7, overlooking the contribution of lower order tributaries, which account for more than 90 % of the total length of the Amazonian river network (Mayorga et al., 2005). Conclusions drawn from these studies of large Amazon rivers cannot be extrapolated to small rivers, because they differ in important aspects (Ashworth and Lewin, 2012). In the SAFB, the patterns of paleo channels show that it is not the large Río Mamoré but rather its tributaries that have deposited most of the sediments that form the modern alluvial plains (Lombardo et al., 2012; Lombardo, 2014; Hanagarth, 1993). Hence, it is important to further our understanding of the behaviour of these tributaries and the mechanisms controlling alluvial plain sediment accumulation.

Thanks to the availability of Landsat imagery with sub-annual temporal resolution covering the last 3 decades, it is now possible to document river spatial and temporal changes and make inferences regarding large-scale changes in hydrology, sedimentation patterns and river sedimentary loads (Buehler et al., 2011; Peixoto et al., 2009; Constantine et al., 2014). Here, I use several time series of LANDSAT images from 1984 to 2014 to analyse the behaviour of the 12 tributaries of the Río Mamoré which have their headwaters in the Andes: the Maniqui, Sécure, Moleto, Isiboro, Chipiriri, Chapare, Chimoré, Sacta, Ichilo, Yapacaní, Piraí and Grande (Fig. 1). The geomorphology of these rivers has never been studied before and hydrological and geochemical data only exist for four of them: the Grande, the Piraí, the Yapacaní and the Ichilo rivers (Guyot et al., 1994, 2007). In this paper I analyse the occurrence of crevasses, a breach in the river levee, and river avulsions, the abrupt abandonment of a channel for a new course at a lower elevation (Slingerland and Smith, 1998, 2004) and the link between these processes and strong to extreme ENSO events. I investigate how these rivers contribute to the formation of the alluvial plain and affect the local forest-savannah ecotone and forest disturbance. Although the analysis of optical remote-sensing imagery here presented does not provide quantitative data on sedimentary processes, it does allow a qualitative assessment of these processes and a re-interpretation of existing quantitative data. The impact of river dynamics on indigenous communities and the rural economy is also explored, with particular emphasis on how these highly active rivers may affect the viability of the planned highway across the National Park Territorio Indígena y Parque Nacional Isiboro Secure (TIPNIS) in Bolivia.

The SAFB is a largely pristine environment, where rivers move freely across the alluvial plains. The SAFB is drained by three large rivers: the Beni, the Mamoré and the Iténez (o Guaporé). It comprises two regions, the seasonally flooded savannah of the Llanos de Moxos (LM), where nine out of the 12 tributaries of the Mamoré are located, and the northern part of the Department of Santa Cruz, where the remaining three rivers are located (Fig. 1). These rivers drain the Andean catchment of the Mamoré, which includes the second most important rainfall hotspot of the southern tropical Andes (Espinoza et al., 2015). Several paleocourses of the Río Beni have been identified, these seem to be the result of avulsions caused by a fault located a few kilometers from the Andes (Dumont and Fournier, 1994). The Mamoré avulsed during the mid-to-late Holocene (Plotzki et al., 2013) and occupied one of the Río Beni paleocurses (Lombardo, 2014). Stratigraphic cores performed across the alluvial plain have shown that, since the mid-Holocene, distributary fluvial systems formed by the Mamoré's tributaries (Fig. 1) have deposited thick layers of sediments over the southern and central part of the LM (Lombardo, 2014; Plotzki et al., 2015). This region hosts one of the most important collections of pre-Columbian earthworks in Amazonia, including monumental mounds, raised fields, ring ditches, fish weirs, canals and causeways (Lombardo et al., 2011; Prümers and Jaimes Betancourt, 2014). Throughout the Holocene, river avulsions have played a central role in both causing the abandonment and burial of early Holocene archaeological sites (Lombardo et al., 2013) and later favouring the development of pre-Columbian complex societies through the deposition of fertile and relatively well drained sediments (Lombardo et al., 2015, 2012). The LM is largely covered by savannahs, crisscrossed by strips and patches of forest that grow on slightly elevated fluvial deposits, mostly river levees and crevasse splays. This forest-savannah patchwork is key for the survival of its rich biodiversity, which includes several endemic, rare and threatened species (Herzog et al., 2012; Wallace et al., 2013; Langstroth, 2011). The recent designation of three new protected areas in the LM has made it the world's largest Ramsar site, which is a protection scheme for wetlands of worldwide ecological importance (http://www.worldwildlife.org/press-releases/ last access: 15 March 2016). The LM constitutes the southern border of the Amazonian rainforest, hence a preferential area to study forest–savannah dynamics (Carson et al., 2014; Mayle et al., 2000; Whitney et al., 2011).

All 12 tributaries of the Río Mamoré with a catchment in the Andes have been included in this study. Crevasse splays and avulsions since 1984 have been identified using the Landsat Annual timelapse in Google Earth Engine (https://earthengine.google.org/#intro/Amazon). Subsets of Landsat imagery have been downloaded from the USGS service LandsatLook (http://landsatlook.usgs.gov/viewer.html), these include all the river reaches identified for all the years where high-quality coverage is available (Table 1). Images have been transformed into 2 bit (black and white) data sets and channel centrelines have been digitalized using the ArcScan extension of ArcGis software. Meander migration rates have been calculated as in Micheli et al. (2004) and Constantine et al. (2014). Values of the Multivariate ENSO Index (MEI) (Wolter and Timlin, 2011) have been downloaded from http://www.esrl.noaa.gov/psd/enso/mei/rank.html. As in Aalto et al. (2003), only the ranks of the early rainy season months for Bolivia have been included in the analysis.

During the 30-year period for which images are available, the Mamoré's tributaries show extremely high activity: 41 crevasses opened up along seven of the 12 tributaries, 29 of which initiated an avulsion process (Table 2, 3). Only 8 out of the 41 crevasses for which the exact year of formation has been identified coincide with La Niña years, while 12 coincide with El Niño years and 21 crevasses opened up in years of no moderate to strong ENSO activity (Table 4). No crevasses initiated during El Niño years 1987, 1992, 1998 and 2003 or during La Niña in 1989 and 1999. Except for the Río Grande, crevasses are found at an average distance of 68 km (σ23) from the point in which the rivers enter the alluvial plains. In the case of the Río Grande this distance is 276 km (σ63.4). All the modern crevasses are closer to the Andes than the mid- to late Holocene terminal splays of the distributary systems formed by these rivers (Fig. 1), particularly in the case of the Río Grande, which during the mid- to late Holocene deposited a sedimentary lobe 280 km further away from the Andes (Lombardo et al., 2012).

Based on their behaviour (Table 3), the tributaries of the Mamoré can be grouped into three categories (Fig. 2): rivers that do not show evidence of avulsions in the last 30 years, with a multi-thread channel along the piedmont that becomes a single-thread meandering channel once it enters the alluvial plains (Chipiriri, Chapare, Chimoré, Ichilo, Sacta); rivers avulsing on a multi-decadal timescale (Sécure, Isiboro, Moleto, Yapacaní); and rivers avulsing on a sub-decadal timescale (Maniqui, Piraí, Grande).

In aggrading alluvial plains characterized by the presence of well-developed paleosols within fine-grained alluvium, as it is the case in the SAFB (Lombardo et al., 2012; Lombardo, 2014), crevasse splays and river avulsions are the most important depositional processes in alluvial plains (Slingerland and Smith, 2004; Smith et al., 1989). Despite a large body of studies, the exact mechanisms controlling crevasse splays and river avulsions are not entirely understood (Hajek and Edmonds, 2014; Stouthamer and Berendsen, 2007; Ashworth et al., 2004). When the various processes that push the river towards the avulsion threshold proceed at a faster pace than those that act as triggers, the latter control the frequency of crevasses and, eventually, avulsions (Jones and Schumm, 1999). It is generally accepted that in southern Amazonia the trigger behind the formation of crevasses in large rivers is the sudden increase in river discharge that follows extreme precipitation events linked to La Niña (Aalto et al., 2003). In the SAFB, research suggests that the frequency of river crevasse formation increases during la Niña events (Aalto et al., 2003), because higher precipitation in the eastern flanks of the Andes is accompanied by reduced precipitation in the lowlands. This increased precipitation towards the Andes causes an important rise in the rivers' discharge, whilst the floodplain water table remains relatively low. Under these conditions, the formation of crevasses becomes more likely because the water level inside the river channel rises faster than the water level in the surrounding floodplain (Aalto et al., 2003). The thick deposits of sediment in the Mamoré and Beni floodplains are believed to be the result of crevasse splays and sheet sand deposits that formed in this way (Aalto et al., 2003; Aalto and Nittrouer, 2012). Hence, Aalto et al. (2003) conclude that these deposits, triggered by La Niña events, cause most of the flood-plain sediment accumulation across the lowland plains. However, the validity of this hypothesis has been challenged by Gautier et al. (2007), who did not find any visible evidence of crevasse-splay formations along the Beni River, although their field coring efforts were limited to a single transect on the lower limb of an active meander.

The new data here presented challenges both the importance of large rivers in controlling alluvial plain dynamics in the lowland plains of the SAFB and the role of La Niña in controlling the timing of crevasse splays.

The results of this study, combined with other published data (Guyot et al., 1996; Lombardo, 2014; do Nascimento Jr. et al., 2015; Plotzki et al., 2015), suggest that these small rivers are highly active and play a dominant role in shaping the SAFB alluvial plains. These rivers, and in particular the Sécure, Isiboro, Moleto, Maniqui, Piraí and Grande rivers, show extremely reduced meander migration rates downstream from where the crevasses opened up (for example Fig. 6), probably as a consequence of a decrease in their sedimentary load, following the same pattern described for rivers in other geologic and climatic settings (Smith et al., 1989; Buehler et al., 2011). Most of the sediments, along with associated nutrients and carbon, eroded from the Andean catchment of the Mamoré are therefore sequestered in the flood plains of its tributaries through the formation of crevasse splays and avulsions. This can explain why about half of the total sediment flux discharged from the Bolivian Andes is deposited in the SAFB (Guyot et al., 1996), including most of the sand fraction (do Nascimento Jr. et al., 2015). It also explains why the Río Beni, which only has one tributary with a catchment in the Andes (the Río Madidi), brings to the Madeira three times more sediments than the Mamoré (Guyot et al., 2007; Aalto et al., 2002), despite the fact that the Beni has a smaller catchment and an average water discharge of 3070 m3s-1, versus the 5080 m3s-1 of the Mamoré (Guyot et al., 1996). This reinforces the observation that, in the mid Holocene, the tributaries of the Mamoré deposited thick layers of sediments over the southern and central part of the LM (Lombardo, 2014). This research adds new evidence to the idea that most of the modern continental sedimentary basins are filled primarily by DFS (Weissmann et al., 2013; Hartley et al., 2010) and shows that the SAFB is an excellent natural laboratory for the study of river processes in sedimentary basins.

The study of the tributaries of the Mamoré over a period of 30 years shows no link between the timing of the crevasses and La Niña events (Table 2). The behaviour of the rivers Maniqui, Piraí and Grande seems to be controlled by the seasonal lowering of the water table, below stream level, during the dry season. This causes the river to leak water from its channel into the ground beneath, causing a reduction in the rivers' sediment transport capacity, increased channel infilling and likelihood of logjam formations. However, as described in the case of Río Pilcomayo in the Chaco plains, which shows a similar seasonal behaviour (Martín-Vide et al., 2014), increased sediment discharge due to modern land use changes in the Andes could also contribute to the channel infilling. The behaviour of the studied rivers suggests that, on an annual to decade timescale, the activity of southern Amazonian small rivers is controlled by bed aggradation and logjams. These are likely caused by the rivers' high sedimentary load combined with a perched channel during the dry season and an extremely low along-valley slope, which not only bring the rivers to the threshold conditions for the formation of crevasse splays, but also trigger the crevasses. In this setting, the decrease in average precipitation over the SAFB experienced in recent years (Espinoza Villar et al., 2009) and the lengthening of the dry season (Fu et al., 2013) increase the frequency of river crevasses and move their formation closer to the Andes. The fact that all the modern crevasses are closer to the Andes than the mid- to late Holocene distributary systems formed by the rivers in groups 2 and 3 suggests, on a millennial scale, a common climatic (Mayle et al., 2000; Baker, 1977) and/or neo-tectonic (Lombardo, 2014; Dunne et al., 1998) control over the shifting of these rivers' depozone. A lack of discrete deposition events has been reported along the Mamoré floodplain after 1971, which could have been caused by a change in regional climate that took place around this time (Aalto et al., 2003). Thus, further research is needed in order to assess whether and how this change could have affected the dynamics of the Mamoré tributaries. Further research is also needed in order to better understand the exact mechanisms behind the formation of crevasses; the contribution of La Niña driven sheet sand deposits to the total floodplain sediment deposition of the Mamoré's tributaries; and the shift of the tributaries' sedimentary depozones.

The evolution of the fluvial network and the constant and frequent changes in river connectivity can have important effects on forest disturbance, aquatic ecosystems and the indigenous populations that live along these rivers. The topographic changes caused by the deposition of fluvial sediments have immediate effects on the local forest-savannah ecotone, which is largely controlled by topography (Mayle et al., 2007). Crevasse splays form on the lower part of the landscape, which is normally covered with savannah vegetation. But, as the sediments are deposited, an elevated area is created that eventually becomes forested (Figs. 7, 8). Likewise, crevasses and avulsions cause the flooding, and hence die-off, of large areas of forest (see for example the reddish area in the upper part of Fig. 9). Aquatic animals, especially migratory fish, must continuously adapt to the frequent changes in the fluvial network. The regime of continuous river fragmentation and forest disturbance could explain the unusually high fish and plant biodiversity found in this area (Pouilly et al., 2004; Thomas, 2009).

It is important that Bolivian policy makers take into account the dynamics of the SAFB's river network in order to mitigate future risks to the local population and to better assess the feasibility of new development plans. In particular, river avulsions can have catastrophic consequences on indigenous communities, as shown in the case of the Sécure. If the trend in the formation of new crevasses continues in the future, it is likely that the rivers Maniqui, Moleto and Isiboro will change their course in the next decade or two. When this happens, the indigenous communities settled along these rivers will have to be relocated. The frequent avulsions of the Piraí following the 1988, 1993 and 1994 crevasses suggest that it could avulse again in the near future in this same diversion site, flooding the city of Montero, with important economic and human costs. The planned highway cutting through the Territory and National Park Territorio Indígena y Parque Nacional Isiboro-Secure (TIPNIS) has received strong protests, for and against it (see for example http://www.bbc.com/news/world-latin-america-15138784 and http://www.bbc.com/news/world-latin-america-16804399). Nevertheless, the technical feasibility of the project has received little attention. The road, linking Villa Tunari to San Igancio de Moxos, will go through areas that are under constant threat of major floods resulting from the frequent formation of crevasse splays and the avulsion of the rivers Moleto and Sécure (Figs. 5, 6). It is likely that, if built, the road will require costly and continuous maintenance. Moreover, the road will dam these rivers, with unpredictable effects on the aquatic ecosystems of this still largely pristine environment.

This paper analyses the behaviour of 12 southern Amazonian small rivers and their role in the formation of the SAFB alluvial plains. Several studies about alluvial plain dynamics in Amazonia have focused on large rivers, concluding that they cause most of the alluvial plain sediment accumulation. It has been proposed that this sedimentation is primarily the result of crevasse splays and sheet sand deposits triggered by large, rapid-rise ENSO floods. The analysis of the 12 tributaries of the Río Mamoré over a period of 30 years shows that these rivers are extremely active, continuously reshaping the landscape, with immediate effects on the local topography, the forest-savannah ecotone and the region's biodiversity. Most of the sediments that these rivers bring from the Andes are sequestered in the alluvial plains before they reach the Mamoré. In contrast with what has been reported for the Mamoré and Beni rivers in previous studies, in the case of the smaller tributaries no correlation emerges between the frequency of crevasse splays and ENSO events. In the case of the southern Amazonian small rivers, the frequency of crevasse splays and avulsions is controlled by intrabasinal processes on a yearly to decadal timescale, while their location, i.e. the average down-valley distance from the Andes where crevasses form, is controlled by climate and tectonic activity on a millennial scale. Small rivers' fluvial activity greatly affects the livelihoods of rural communities, particularly indigenous people who are often settled along these rivers and dependent on their resources. The study has shown how river avulsions can have a catastrophic impact on communities settled on the reach of the river that is cut-off. On the other hand, these highly active rivers have also favoured agricultural development in some areas, through the deposition of fertile sediments. It is important that alluvial plain dynamics are taken into account by policy makers and development organisations in Bolivia, particularly when planning major infrastructure projects in the area. In light of the study's results, it should be advisable that the technical feasibility of the planned road linking Villa Tunari to San Ignacio de Moxos is re-assessed.