River logjams cause frequent large-scale forest die-off events in southwestern Amazonia

This paper investigates the dynamics of logjam-induced floods and alluvial deposition in the Bolivian Amazon and the effects these have on forest disturbance and recovery cycles. It expands on previous work by Gullison et al. (1996) who reported a case of catastrophic floods triggered by logjams in the Chimane Forest in the Bolivian Amazon. No further studies have followed up on this observation and no research has been published on the effects of large wood in tropical lowland rivers. The study is based on the analysis of a time series of Landsat imagery (1984–2016) and field evidence. Results show that logjam-induced floods are a major driver of forest disturbance along the Andean piedmont in the Bolivian Amazon. New logjams form on an almost yearly basis, always further upriver, until an avulsion takes place. Logjam-induced floods are characterized here by the sudden deposition of a thick sand layer and the death of forest in a V-shaped area. The Bolivian Amazon offers a unique opportunity for further research on how large wood affects river behavior in lowland tropical settings and how large and frequent forest disturbance events resulting from river logjams affect forest biodiversity and community successions.


Introduction 20
Understanding the spatial and temporal frequency of disturbance events is of great relevance to forest ecologists, due to the importance of disturbance in shaping forest ecological processes (Asner, 2013).
Forest disturbance and recovery cycles affect forest tree species distribution, community composition, ecosystem processes, biodiversity patterns, nutrient cycles and the carbon balance (Chambers et al., 2013;Lewis et al., 2004;Phillips et al., 2004;White and Jentsch, 2001). Disturbance events are commonly 25 classified along a continuum that goes from small scale/frequent events, such as a tree fall that creates an open space, to large scale/more rare events, such as large fires (White and Jentsch, 2001). The study of the geography of forest disturbance is particularly important in Amazonia, as it is home to about 10% of world's biodiversity (Lewinsohn and Prado, 2005) and among the most important terrestrial carbon sinks (Pan et al., 2011). 30 In Amazonia, where large forest fires are uncommon and there are no hurricanes, large scale disturbance (>30 Ha) is mostly caused by blowdowns (Nelson et al., 1994) that result from convective cloud downdrafts (Garstang et al., 1998). However, at least in Western Amazonia, studies have shown that river activity can also be an important driver of forest disturbance and landscape reshaping via lateral erosion, overbank deposition, crevasse formation and avulsions (Aalto et al., 2002;Kalliola et al., 1992;35 Lombardo, 2016;Salo et al., 1986).
Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. In 1996, Gullison et al. (1996 observed that 48000 hectares of forest in the Bolivian Amazon died-off and were replaced by savannah due to scouring, flooding and the deposition of alluvial sediments. These processes were caused by the overbank flow of the Cuberene River triggered by a logjam. Here, the term logjam refers to a partial or complete obstruction of a river channel caused by large wood (LW). Logjams 40 can affect rivers and floodplains in several ways. They can decrease stream flow to an extent that it leads to bank erosion and overbank flow, influencing channel-floodplain interactions and shaping the evolution of floodplain topography (Montgomery and Abbe, 2006;Sear et al., 2010;Wohl, 2013). Logjams are also an important component of river ecosystems, as they influence the transfer of solutes, mineral sediment and organic material within the river channel system and between river and floodplain (Coe et al., 2009;45 Gurnell et al., 2002;Jochner et al., 2015;Wohl and Beckman, 2014). Despite the fact that logjams can have a dramatic impact on fluvial dynamics and fluvial-floodplain exchanges, their systematic study is relatively recent and has mostly been carried out in temperate zones (Dixon, 2016;Ruiz-Villanueva et al., 2016;Wohl, 2017Wohl, , 2013. Very few studies have looked at the wood-river interactions in the neotropics and these have mostly focused on headwater rivers (Cadol andWohl, 2011, 2010;Iroumé et al., 2015;50 Wohl et al., 2012). Wohl (2017) reports that no field-based studies of LW in Amazonian rivers has ever been published in the English-language literature. Nevertheless, the observations of Gullison et al. (1996) do suggest that logjams are an important factor driving forest disturbance and biodiversity patterns in the Bolivian Amazon. For example, it has been shown that the increase in light that follows forest die-off events caused by logjams creates optimal conditions for the regeneration of Mahogany (Swietenia 55 macrophylla King), influencing its population dynamics (Gullison et al., 2003;Snook, 1996). Gullison et al. (1996) also reported the transformation of large areas of forest into savannahs after being flooded.
Despite these indications of the importance of logjams in shaping Bolivian forests, no further studies have followed up on Gullison et al. (1996). As of today, very little is known about the spatial extent and recurrence of these logjam-induced floods. 60 In this paper, I use a time series of Landsat imagery and field evidence to study logjam-induced floods and alluvium deposition in the Bolivian Amazon and map their extent and recurrence. I discuss why logjams form in these rivers and how they affect modern forest disturbance/recovery cycles and pre-Columbian landuse in the past.
The Bolivian Amazon is largely covered by a seasonally flooded savannah known as the Llanos de 65 Moxos. The area focus of this study is located between the flat Llanos and the Bolivian Andes, where lowland forests grow on relatively well drained fluvial sediments. This area of lowland forests is approximately 60 km wide and 400 km long and runs parallel to the Bolivian Andes (Fig. 1).

Methods
The phenomena described by Gullison et al. (1996) -formation of logjam, forest die-off and 70 transformation of forest into savannah -can be studied through the visual analysis of remote sensing Imagery (Gullison et al., 2003). In this paper, rivers affected by logjam-induced floods have been identified by analyzing the changes in river path and vegetation cover visible on the Timelapse application of Google Earth Engine (https://earthengine.google.com/timelapse/). In the study area, the forested lowland bordering the Bolivian Andes piedmont, there are a total of 22 rivers that experience 75 Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. logjam-induced flooding (Fig. 1). These include all the small rivers between the River Secure and the Beni and all but three of the rivers north of the Beni River. A subset of eight of these rivers has been selected and their evolution analyzed throughout the period for which cloud-free landsat images are available: from 1984 to 2016. These have been chosen in order to cover the whole area where the logjaminduced floods take place and in order to have a representation of rivers of different size. These eight 80 rivers are: the Tequeje, the Tacuaral, the Colorado and the Cuberene and other four rivers, here referred to as Y, Z, W and Q, for which the toponyms could not be found (Fig. 1). The location of the logjams (shown in Fig. 5) has been measured as the distance between the logjam and an arbitrary line parallel to the Andean piedmont. The Landsat coverage of these eight rivers during the period 1984-2016 has been downloaded through the USGS Landsat look viewer (https://landsatlook.usgs.gov/). Field work was 85 carried out during the dry season of 2016. A survey was conducted along the Tacuaral and Colorado rivers ( Fig.1), because of their easier access and the availability of local guides in these areas. The analysis of Google Earth Engine Timelapse identified twenty-two rivers in this area of the Bolivian 95 Amazon that are affected by logjam-induced floods. The river logjams occur within a 400 km long belt along the eastern Andean piedmont, from Lat -15.9 to Lat -13.5. All but two, of the 22 rivers affected by logjam-induced floods, have their headwaters on the eastern side of two small pre-Andean mountain chains (Fig. 1). The northern one, between the Beni River and the Maniqui River, has a maximum elevation of about 1100 m. a. s. l.; the southern mountain chain, between the Maniqui River and the 100 Secure River, has a maximum elevation of 650 m. a. s. l. (Fig. 1B). Rivers originating in the southern part of the study area cross the Chimane Forest, where forest die-offs induced by logjams were first reported by Gullison et al. in the floodplain of the Cuberene River ( Fig. 2A and Fig. 3). These small pre-Andean mountain chains act as a water divide between the basin of large rivers on the west (i.e. Maniqui and Secure Rivers in Fig. 1B) and smaller rivers to the east. 105

Results
In the 8 rivers analyzed in greater detail, from 1984 to 2016, 174 logjams causing river collapse and forest die-off have been identified and mapped (Fig. 5). The eight rivers show a similar pattern: a new logjam forms upriver from the previous one almost every year, unless an avulsion takes place and the sequence restarts at a greater distance from the Andes (i.e. rivers Colorado, Cuberene, Tacuaral, Tequeje in Fig. 5).
The average distance between successive logjams ranges from 1170 meters, in the case of the Cuberene, 110 to 224 meters, in the case of the Colorado.
The river Tequeje, a left hand tributary of the Beni River, is the second largest among the 8 studied. where the forest died off in 1987 is forested again. In 2016, the logjam is closer to the Andes than at any other moment during the 29 years studied. The sediments deposited between 2014 and 2016 have dammed two small tributaries that have been transformed into small lakes (inset E Fig. 2). Two small areas, which were forested in 1987, have been transformed into savannah after being constantly flooded between 2011 and 2016 (insets A, D, and E in Fig. 6). A total of about 8900 hectares of land have been 130 flooded at least once by the Tequeje River during the 29 years analyzed. Of these, in 1987, 7100 hectares were forested and 1800 hectares were savannah (Table 1). Tequeje's channel shows a meandering pattern from the point at which it enters the alluvial plains till about 4 km upriver from the 2016 logjam (Fig 7). Colorado. A few tens of meters upriver from the 2016 logjam in the river Colorado ( Fig. 9 and 10) sand was deposited by overbank flow and the area immediately colonized by Parajobobo (Tessaria integrifolia Ruiz & Pav.). Parajobobo is a pioneer, fast growing species that tends to colonize sand deposits along fluvial channels by forming monospecific stands of the same age (Neiff, 2004). Fig. 10 shows a two meter thick sand layer, which has been deposited overnight at the end of January 2016 (Macario Huanca 145 Quispe, community of Inka Agropecuaria, personal communication). The Parajobobo trees behind the hut in Fig. 10 are seven months old. The plant communities that colonize the clearance that follow the forest die-off (Fig. 11) are the same as the communities that colonize fluvial deposits along the Mamoré River.
(Patucú) and Gynerium sagittatum (Maldonado and Beck, 2004). After the sand is deposited, the water 150 moves across the forested areas eroding the topsoil and exposing tree roots. Scouring forms many small channels (Fig. 12). The survey along the Colorado River revealed a paleosol associated to pre-Columbian pottery. The paleosol, located about 4 meters below the top of the river bank, has been radiocarbon dated 1390 ± 43 cal yrs BP. Pre-Columbian raised fields, ancient elevated platforms that were used for Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. agriculture (Lombardo et al., 2011;Rodrigues et al., 2016), are visible in Google Earth imagery a few 155 kms downriver from the areas affected by the logjam-induced floods of the Cuberene River (Fig. 13). Fig.   13 shows that pre-Columbian earthworks (raised fields and causeways) are found within the area crossed by paleochannels of the Cuberene River and other small rivers to the south.

Discussion
Logjam-induced floods have been largely overlooked as important agents of forest disturbance in 160 Amazonia. The analysis of time series of Landsat imagery shows that, in a vast area of lowland forest (approx 60 km x 400 km) that runs parallel to the eastern Andean piedmont in Bolivia, logjam-induced floods are a major driver of forest disturbance. Logjam-induced floods drive a characteristic form of forest disturbance, as they affect hundreds of hectares of forest recurrently, in most cases on a yearly basis. This is a type of large scale disturbance that can be classified as intense and frequent (Turner et al., 165 1998).
In the Chimane forest and its surroundings, Paneque-Gálvez et al. (2013) have estimated that, during the period 1986 to 2009, a total of 26000 hectares of old-growth forest was lost due to human activities.
During the same period, the Cuberene's logjam induced floods affected 18500 hectares. The rate of forest disturbance caused by rivers here is, therefore, comparable to the current rates of deforestation driven by 170 the expansion of the agricultural frontier.
Forest ecological processes, and in particular tropical forest carbon sinks, are assessed by monitoring relatively small plots of forest during relatively long periods of time (~ 30 yrs) (Lewis et al., 2004;Phillips et al., 2004). However, to what extent the long-term monitoring of small plots can take into account large scale events is controversial (Chambers et al., 2009;Espírito-Santo et al., 2014). This study 175 shows an example of an area where the monitoring of small plots can lead to critically skewed results if logjam-induced forest disturbance is not taken into account. The high rate of logjam-induced forest disturbance could be the reason for the yet unexplained low tree biodiversity observed in the Chimane Forest (Gullison et al., 1996); and can also explain why current models are unable to capture a large part of the floristic variation here (Guèze et al., 2013). 180 It is known that two of the most important factors controlling the formation of logjams are the recruitment of LW and the relation between the size of the LW and the width of the river channel (Gurnell et al., 2002;Wohl, 2013). The rivers studied here have very high meander migration rates (see Fig. 7) and hence can recruit a large amount of wood because of the lateral erosion of forested areas. They have very small mountain catchments; hence they enter the alluvial plains when they are still not very wide. Moreover, 185 due also to uplift events that have occurred north of the study area (Lombardo, 2014), the study area is extremely flat and rivers have a relatively low transport capacity and a propensity to river channel siltation (Lombardo, 2016). All these conditions contribute to the formation of channel spanning logjams.
Although the collapsing of the river on an almost yearly basis is restricted to the small rivers here analyzed, channel spanning logjams causing river avulsions have been reported in larger Bolivian rivers 190 such as the Maniqui and the Secure too (Lombardo, 2016). The presence of pre-Columbian earthworks in an area crossed by the Cuberene River paleochannels (Fig.   13) suggests, on the one hand, that the preferential location for logjam formation and deposition of alluvium has moved westward, at least since the construction of the earthworks, otherwise these would have been destroyed or, more likely, never built in an area under such a high risk of catastrophic floods in the first place. On the other hand, the presence of Cuberene paleochannels in the very same area as the 200 pre-Columbian earthworks suggests that, prior to the construction of the raised fields, logjams were less frequent or happened further to the east than their modern location. Raised fields were built to mitigate the risk posed by extreme floods (Lombardo et al., 2011) and their geometry, size and location responded to the local hydrology (Rodrigues et al., 2015;Rodrigues et al., 2016). The fact that raised fields in the Cuberene area were built along the whole slope that goes from the former levees of the Cuberene river to 205 the former backswamps (Fig. 13) suggests that the local hydrology was highly variable in pre-Columbian times, as it can be expected in an area subject to frequent river avulsions occurring just a few kilometers upriver. The paleosol in the riverbank of the Colorado River containing pre-Columbian pottery dated 1390 ± 43 cal yrs BP is buried under a 4 meter layer of sediment. This shows the high depositional rate of these rivers, leading to the burial of many archaeological sites in the region. This example highlights the 210 importance of taking into account river dynamics when analyzing the spatial patterns of pre-Columbian settlements in the region. In recent years, the Andean piedmont has been increasingly occupied by small communities of campesinos from the Andean region that clear the forest to practice agriculture (Paneque- Gálvez et al., 2013). The agricultural frontier is now expanding eastward from the road that links the towns of Yucumo and Rurenabaque, running along the part of the Andean piedmont north of the Maniqui 215 River (Fig. 1). This is the area where LW is recruited by lateral erosion of forested river banks; therefore, the modern land-use practices here will probably reduce forest recruitment and cause the reduction and eastward migration of the logjam formations in the near future. Most of the studies on logjams have been carried out in rivers of temperate regions, where fluvial dynamics have been impacted by human activity throughout history and logjam dynamics differ from tropical regions (Wohl, 2013). The almost pristine 220 environmental conditions of the Bolivian Amazon and the recent change in land-use here offer an excellent natural laboratory to study the dynamics of wood in lowland tropical rivers, the process of forest disturbance and community successions and how changes in LW recruitment affect both river dynamics and tree communities.

Conclusions 225
This paper analyses the dynamics of logjam formation in tropical meandering rivers in Southwestern Amazonia. The study focuses on rivers that cause logjam-induced floods in an area (approx. 60 x 400 km) of lowland forest that stretches parallel to the Andean piedmont in Bolivia. The analysis of remote sensing imagery shows the existence of 22 such rivers, representing practically the totality of the smaller rivers in the area. These rivers are characterized by i) being relatively small when they enter the alluvial 230 plains, hence their width is similar to the height of the trees they transport; ii) having a high meandering rate that causes lateral erosion of forested areas and a high recruitment rate of LW; and iii) they flow Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. across a very gentle slope, which reduces their capacity to transport LW. The study shows that logjaminduced floods are a major driver of forest disturbance in the Bolivian Amazon. Large logjam-induced floods follow a pattern that is consistent in all the rivers studied: channel spanning logjams form on an 235 almost yearly basis and migrate upriver until an avulsion takes place and the upriver migration of the logjam re-starts in a new location within the alluvial plains, at a greater distance from the Andes. Each logjam-induced flood is characterized by i) a sudden deposition of a thick sand layer a few tens of meters upriver from the logjam and ii) a V shaped area where forest dies off and is replaced by pioneer plants or transformed into a savannah grassland if repeatedly flooded. This study shows that large and frequent 240 floods triggered by logjams have a major impact on forest disturbance/recovery cycles and can potentially explain local floristic variations. This case study offers a unique opportunity to further research on how LW affects river behavior in lowland tropical settings and how large and frequent forest disturbance events resulting from logjams affect forest biodiversity and community successions.

Acknowledgment 245
The present study has been funded by the Swiss National Science Foundation (SNSF) grant no  Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. Figure 2: Examples of forest die-off events as the one described in Gullison et al. (1996). Reddish plumes 370 diverging from the river channel indicate dead forest. Here, trees have lost their leaves and this causes a fall in near infrared reflectance, which is shown in green in the RGB composition of the Landsat imagery (insets A-F), and an increase in mid-infrared reflectance (shown in red) which is mostly due to sediments. Non forested, light green areas are covered by pioneer species; forested, light green areas are young secondary forests which have re-grown after disturbance; forested, dark green areas are old-growth forests. Non-forested, reddish 375 areas are savannah. Blue/black areas are water.

420
Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. Earth Syst. Dynam. Discuss., doi:10.5194/esd-2017-19, 2017 Manuscript under review for journal Earth Syst. Dynam. Discussion started: 24 March 2017 c Author(s) 2017. CC-BY 3.0 License. Figure 14: Location of raised fields in the proximity of a Cuberene River paleochannel (for location see red box in Fig. 12). In a) raised fields built in the forested upland that grows over the fluvial levee; in b) raised fields built across the current forest-savannah ecotone; in c) fields built in the seasonally flooded savannah. Table 1: Overview of the dynamics of the logjams and their effect on the forest. "Yrs covered" is the number of years for which cloud-free Landsat imagery was available; "Starting" is the year of the oldest cloud-free 430 Landsat image available; "Width" in the width of the river in meters; "AFA" stands for "average flooded area" and is the average number of hectares that are flooded when there is a logjam induced flood in a given river. TFA stands for "total flooded area" and is the number of hectares that have been flooded in total during the period 1984-2016. TFA is smaller than the sum of the AFA for all the years because large extents of land are often repeatedly flooded, so that there is a significant overlap between areas flooded in different years. "% 435 Savannah" and "%Forest" indicate the type of vegetation cover in 1984 (percentage of the TFA).

River
Yrs covered