Climate changes have been considered an essential factor controlling the shaping of the recent alluvial landscapes in central Amazonia, with implications for explaining the biogeographic patterns in the region. This landscape is characterized by wide floodplains and various terrace levels at different elevations. A set of older terraces with ages between 50 and
The lowlands of the central Amazonia host one of the largest systems of alluvial deposits on Earth, composed of extensive Quaternary fluvial terraces with wide incised valleys where modern rivers flow (Fig. 1) (Rossetti et al., 2005; Pupim et al., 2019; Sioli, 1984). These deposits are of great relevance for the reconstruction of both the evolution of the physical landscape and the patterns of biotic diversity in the region (Ribas et al., 2012; Ribas and Aleixo, 2019; Pupim et al., 2019; Bicudo et al., 2019). The areas that currently display terraces were part of a large depositional system in which rivers in this region had high deposition and avulsion rates in an aggradational pattern (Hoorn et al., 2010; Pupim et al., 2019; Wilkinson et al., 2010). During this stage, flooded environments covered much larger areas, because they were not restricted to incised valleys as is currently the case. During the Late Pleistocene, most likely because of an adjustment of the river's long profile to a lowering of the base level or a change in discharge, the main rivers started to incise to reach their modern position. This phase of incision was associated with a widening of the valleys, which was accomplished through channel migration and lateral bank erosion (Merritts et al., 1994). This long-term period of fluvial erosion was accompanied, at a shorter timescale, by multiple cycles of incision and aggradation, which resulted in the formation of both cut terraces and cut-and-fill sequences with various depositional ages and elevations (Pupim et al., 2019). Sea-level changes have been proposed as the controlling mechanisms for the formation of these cycles (Irion and Kalliola, 2010); however, the depositional ages show that the cycles of aggradation and incision occurred at a different frequency or phase than the cycles of sea-level changes during the Pleistocene (Pupim et al., 2019). In fact, there is ample evidence for the occurrence of Cretaceous bedrock exposed at the bottom of the Amazon River next to the confluence with the Negro River (Ianniruberto et al., 2018; Gualtieri et al., 2020), implying no net aggradation since the last glacial maximum. Therefore, the Holocene sea-level rise cannot be used to explain the occurrence of incised floodplains upstream of the confluence with the Negro River. Climate-driven discharge variations is another mechanism that has been used to explain the multiple phases of aggradation and erosion because such controls have the potential to impose changes on the course of the main and tributary rivers through changes in the sediment supply and the rivers' capacity to evacuate the supplied material (e.g., Tucker and Slingerland, 1997). There is indeed ample evidence that rainfall rates have varied in central Amazonia during the last 200 kyr (Cheng et al., 2013; Mertes and Dunne, 2007; Fritz et al., 2004). In addition, the pattern of terrace ages suggests that the related cyclicity of sediment accumulation and erosion has been quite similar to that of rainfall variations (Pupim et al., 2019). As yet, no quantitative model has been employed to test the hypothesis of a climate-driven control on the formation of the various terraces in central Amazonia.
Topography of central Amazonia with key geographical designations such as cities, rivers and geological and geomorphic features including terrace levels (lowlands), floodplains (wetlands) and drainage network. The red square illustrates the location of the city of Manaus. Dotted red rectangle represents the region that is reproduced in the model and that represents the model grid. This illustration is based on a digital elevation model from the Shuttle Radar Topography Mission (SRTM).
Here, we develop a landscape evolution model (LEM), which is tailored to
reproduce alluvial processes at a 10
The study area (Fig. 1) is situated in central Amazonia to the west of the
city of Manaus, which is near the confluence between the Negro and the
Solimões rivers. The study region, which is approximately 300 000 km
The morphology of the study region is characterized by the occurrence of
multiple terrace levels and floodplains. Conventionally, these are referred to as wetlands for the floodplains and lowlands for the higher-elevated
terraces (see Fig. 1 and Räsänen et al., 1990). Estimates of the
elevation differences between the highest terrace level (e.g., green colored
lowland surfaces in Fig. 1) and the floodplain (blue colored wetlands) range
from approximately 10 to
Chronology of sediment deposition representing the formation of
terraces in Amazonian lowlands compared with Quaternary climate records from
Pupim et al. (2019).
Ages for some of these terrace levels have been established with optically
luminescence dating techniques and measurements of
Following the scope of this paper, we developed a landscape evolution model
which allows us to simulate fluvial processes of erosion, transport and
deposition of sediments in alluvial landscapes (e.g., Tucker and
Slingerland, 1997; Braun, 2006; Sacek, 2014). This model, which is referred to as SPASE (Sedimentary Processes and Alluvial Systems Evolution) is implemented in Python. The source code is openly accessible and can be downloaded from the GitHub repository (
To simulate fluvial erosion and mass transport, previously developed LEMs, such as CAESAR, SIBERIA and others (Beaumont et al., 1992; Braun, 2006; Hancock et al., 2010; Sacek, 2011, 2014), considered both fluvial processes, such as the long-distance advection of material by rivers and slope processes including local mass transport by hillslope diffusion for slopes flatter than a threshold and advection for steep slopes (Tucker and Slingerland, 1997; Braun, 2006; Sacek, 2014). Since the Amazon alluvial system has very gentle slopes and because the area bordering the floodplains and terraces is occupied by a dense vegetation cover, we do not consider that hillslope processes largely contribute to the redistribution of sediment in the channel network inside the model grid. In fact, all material in the model domain is considered to result from either accumulation and/or erosion by fluvial and channelized processes only. However, the sediment supply from outside of the model grid could indeed be produced by hillslope processes in a steeper landscape, e.g., the Andes, which is considered as a boundary condition (see Sect. 3.3).
Upon modeling fluvial dynamics, we proceed following Beaumont et al. (1992)
and employ a stream power law where, for a given channel, the carrying
capacity of sediment,
Fixed parameters in all simulated scenarios.
Similar to Braun (2006), the entrainment, deposition and transport of
sediment in a channel (in suspension or as bedload) is computed by Eq. (2),
where the erosion/deposition rate is considered proportional to the difference between the carrying capacity (
To simulate lateral erosion we infer that there is no bedrock confinement for the channels and that the eroded material is homogeneous in composition and strength. Therefore, the volume of sediment
Because one model cell can potentially contain both a part of a channel
belt floodplain and a portion of a bordering terrace, it is necessary to
record two elevation values upon modeling lateral erosion (Eqs. 3 and 4). One elevation variable, referred to as
We use the modern topography as the initial condition. This surface was
reconstructed over a regular mesh with
The initial landscape displays six main rivers entering the grid. These are
the Solimões, Japurá, Juruá, Jutaí, Purus and Negro rivers
(Figs. 1 and 3b). For these streams, water flux values were obtained by the
HydroRivers model for the locations where they enter the study area (Lehner
and Grill, 2013). Also for these locations, the sediment discharge of these
streams was then estimated by applying Eq. (1), thereby using the calibrated
parameters of
Current discharge, slope and sediment flux of the rivers outside of the grid. The sediment flux was calculated using the Eq. (1) and the parameters in Table 1.
In order to make the initial topography more stable upon modeling, the numerical model was run using the smoothed DEM as initial topography with the model parameters of Tables 1 and 2 during 10 kyr inside the model. The output topography for this scenario (Fig. 3b) was then used as initial topography for all the climatic and base-level tests in this work.
For the model scenarios where different climate conditions are taken into
account, it is necessary to consider the variation of the rainfall rate
outside of the model grid, which induces a change of the water discharge of
the six main rivers entering the model grid and consequently impacts the
sediment flux. We used the following relationship in order to estimate the
sediment flux,
Additionally, all cells on the right edge of the grid are considered as sink cells. They also control the elevation of the base level of the grid. For the generation of the initial topography, the sink cells are kept at a fixed position. This corresponds to the elevation of the bedrock at Manaus, which we set to 0 m (see above).
The controls of climatic variations on the formation of terraces are explored in the framework of two scenarios, the results of which are detailed in the following sections. In scenario 1 (Fig. 4), a dryer period is simulated by a 30 % decrease of the current water input during the first 10 kyr, which is then followed, until 20 ka, by an instantaneous increase in the water supply up to the same level as the current water input. Scenario 2 is similar but considers first a 30 % increase of the water input during the first 10 kyr (Fig. 5), thereby simulating a wetter period, which is superseded by a 10 kyr long model run during which the water supply is instantaneously reduced to the current level. The 30 % increase or decrease in rainfall on central Amazonia during the Holocene is based on estimates proposed by van Breukelen et al. (2008) and Cheng et al. (2013). In both scenarios, after shifting to a drier or wetter period, the model returns to the modern climate setting to evaluate what would be preserved in the landscape from such climatic shifts. In these models, a change in precipitation rates will affect the water input into the model, which is the sum of the water flux where the stream enters the grid and the runoff that is generated through rain on the grid itself. Both scenarios start with the same initial topography (Fig. 3b), and the grid base level is locked at a bedrock elevation of 0 m, thereby considering the occurrence of a bedrock in the Amazon trunk stream at the downstream end of our model. The consequence of such a fixed base level is that sediment does not get eroded or deposited after the confluence of the Solimões and Negro rivers although that the high water supply of the Negro River lowers the sediment to water discharge ratio, which, in turn, would actually promote a phase of erosion. Following these model runs, we also explore a third scenario where we allow the base level first to increase and then to decrease to the modern situation.
Results of the landscape evolution of scenario 1. The left figures
show the difference of the model topography in relation to the initial
topography, with colors saturated in red for values in elevation differences
higher than 10 m and in blue for values below
Results of the landscape evolution of scenario 2. The left figures
show the difference of the model topography in relation to the initial
topography, with colors saturated in red for values in elevation differences
higher than 10 m and in blue for values below
The model results of scenario 1, which commences with a dry period, show that sediment starts to accumulate in the upstream part, and the wave of accumulation progresses from upstream to downstream. As a consequence, the average slope of the Solimões River and its main tributaries increases in pace with the accumulation of sediment. This change towards steeper channel gradients occurs during the first 5 kyr, after which a nearly steady-state situation is established with constant topographic gradients. At this stage, a total of approximately 13 m of sediment has accumulated along the Solimões River upstream of the Juruá River confluence (Fig. 4, at 2, 5 and 10 ka). Also during the first 10 kyr with drier conditions, the course of the Solimões trunk river is deflected to the north at the confluence with the Purus River. The Purus River flows from the south and experiences, similar to the other main tributaries in the model, a phase of sediment accumulation in the upstream part. In the confluence area, this northward-directed shift of the Solimões River initiates a wave of erosion along the northern banks of this stream (Fig. 4). The model simulates the occurrence of lateral bank erosion also along those river segments where sediment accumulation occurs (Fig. 4). This is mainly caused by the meandering of the channel's course to the lateral floodplain boundaries, which then initiates a phase of lateral erosion as long as higher terrace levels form cut banks. These results document that the SPASE model reproduces one of the first order features of a meandering stream such as cut banks despite the simple model architecture.
During the subsequent 10 kyr long period of scenario 1, the modeled increase in water input causes a wave of erosional recycling of the previously deposited sediments. An example can be seen in Fig. 4 where new terrace levels are formed, e.g., around the Japurá River (see location A in figure). This phase of downcutting is associated with a phase of valley widening mainly along the main rivers, thus forming cut terraces (Fig. 4, at 15 and 20 ka). At the end of the model run, just a fraction of the previously deposited sedimentary material is preserved (red cells on Fig. 4 at 20 ka). The model results predict that such terrace fragments mainly occur in the upstream part of the Solimões River and next to the Japurá, Juruá, Jutaí and Purus rivers.
In the second model run, which corresponds to a scenario where water supply is increased during the first 10 kyr and then decreased to modern conditions, the landscape response is different. In particular, the increase in water runoff initiates a wave of erosion from upstream to downstream, which is also associated with a widening of the floodplain through lateral bank erosion. Erosion and valley lowering results in a decrease of the energy gradient and in the formation of new floodplains at a lower elevation (Fig. 5, at 2, 5 and 10 ka) of approximately 10 m. During the last 10 kyr when the model water input is the same as the modern one, the model rivers adapt an aggradational pattern, but the rivers' thalwegs do not reach the same initial elevations. At the end, no new terrace levels are formed (Fig. 5, at 15 and 20 ka). Finally, during the evolution of this model scenario, the main rivers do not show any preferential direction of lateral bank erosion. Similar to the previous model run, the topography response to each climate change occurs mainly in the first 5 kyr, after which nearly steady-state conditions are reached. Note that in comparison to scenario 1 (Fig. 4), the changes in the landscape are much fewer (Fig. 5).
To simulate the consequences of a hypothetical base-level variation on the topography of central Amazonia, we consider a scenario where the base level gradually increases until 35 m during the first 10 kyr, which is followed by a rapid base-level drop to the elevation of the bedrock at 0 m (Fig. 6). The water input is equal to the current water input during the entire scenario.
Model results showing the landscape evolution of scenario 3. The
model water input is constant and equal to the current water input. The left
panels show the difference of the model topography in relation to the
initial topography, with colors saturated in red for values in elevation
differences higher than 10 m and in blue for values below
During the first 10 kyr, the main rivers turn to an aggradational pattern and thus adjust their long-stream profile to the increasing base level, even accumulating new sedimentary material above older terraces near the interfluve area of the Solimões and Negro rivers (red cells on Fig. 6 at 2, 5 and 10 ka). The smaller tributaries, with a lower sediment discharge, are not able to accumulate sediment in response to the base-level rise at the same pace as the main rivers, thus forming shallow lakes within their own valleys (inundated cells on Fig. 6 at 2, 5 and 10 ka). During the second period of the model run the main rivers initiate a phase of erosion, which occurs in response to the drop in the base level. This downwearing is associated with a period of valley widening. In addition, the model predicts that streams can adapt to a new course. This is exemplified by a northward shift of the Solimões River that takes the course of the Negro River towards the end of the model grid (Fig. 6 at 15 and 20 ka). Such shifts in the course of the major streams is due to the random position of the avulsive rivers during the aggradational stage and the preservation of their last configuration during the incision stage.
The numerical model (SPASE) presented in this work simulates the evolution of alluvial landscapes. It is capable of reproducing the formation of floodplains and terraces at conditions similar to central Amazonia. The tested scenarios allow us to quantitatively explore the responses of the central Amazonia topography to climatic and base-level changes.
On the floodplains, the occurrence of erosion or sediment accumulation in
the modeled scenarios are inherently related to the difference between the
sediment transport capacity, which depends on water discharge and slope (Eq. 1 in Sect. 3), and the sediment discharge (Eq. 2 in Sect. 3). Accordingly, as we start our calculations with the conditions where sediment discharge in the model streams is at capacity, then any changes in either water or sediment discharge will initiate a phase of channel steepening or flattening in order to re-establish the at-capacity conditions. Accordingly, by applying Eq. (5), we infer that the rivers in our model are at capacity upstream the model grid during the model runs. Since most of these rivers have their sediment sources either in the high Andes or in the elevated plateaus with a generally thick regolith cover (an exception is the Negro River), we anticipate that sufficient sediment is available to be eroded by overland flow erosion so that sediment flux of the major streams is at capacity. Accordingly, if we maintain sediment discharge at a constant value and only change water supply, then an increase or a decrease in water discharge will respond in a flatter or steeper channel gradient, respectively, in order to reach a condition where sediment discharge equals the sediment transport capacity. In addition, because data on modern sediment and water discharge in the Amazonian streams (Filizola and Guyot, 2009) suggest that sediment discharge relates to water discharge with an exponential factor of
The model shows that terrace levels form not only where the sediment transport capacity outpaces the sediment discharge but also where the meanders' cut banks approach the lateral margin of a floodplain. A terrace level might thus even form during a period when sediment accumulation occurs, particularly where the model streams start to erode the lateral banks of a floodplain (Fig. 4). The combination of cut bank erosion and sediment accumulation and/or erosion on the floodplains result in a pattern where the terrace levels appear as randomly distributed within the Amazonian landscape and at various elevations. As outlined in the following section, the model results thus offer an explanation for why published ages and elevation of terrace levels record a pattern which is rather difficult to correlate with specific changes in external forcings.
We also conduct model runs where we change the initial precipitation conditions (see Figs. S11 to S19). The model predicts that the erosion of the elevated plateaus adjacent to the trunk rivers becomes larger as we increase the precipitation rates (1000 to 3000 mm as initial condition). However, the patterns at which the low-elevated terraces form will not change. Therefore, a change of the initial precipitation rates will not alter the main conclusions.
The model outputs predict what appears to be a random distribution of middle–lower terrace levels, particularly if the dry-to-wet climate change scenario is considered. This pattern is caused by lateral shifts of the main rivers during both the aggradation and incision phases. These lateral shifts occur where the sediment flux exceeds the sediment transport capacity, which forces the streams to accumulate a fraction of the transported sediment. The models show that such conditions occur randomly along most of the main streams. Exceptions are observed at the confluence of streams where the tributary stream is in an aggradational state. Such aggradation locally increases the elevation in the tributary floodplain relative to that of the trunk stream, with the consequence that it reduces the possibility and thus the probability of the trunk stream to laterally shift towards the tributary river. In this sense, the trunk river is pushed towards the opposite floodplain margin where erosion can form a cut terrace.
The model results also show that climatic variations that occur at the scale of thousands of years in central Amazonia (Cheng et al., 2013; Fritz et al., 2004; Mertes and Dunne, 2007) can have a large influence on the cycles of sediment aggradation and incision. The main differences in the topographic responses between the two climatic conditions are mainly observed within the floodplains, because the channels steepen during dryer periods in response to sediment accumulation, and they flatten during wetter periods as sediment is entrained in the floodplains. Both scenarios result in the formation of terrace levels at different locations and elevations. Whereas the drier-to-wetter climate change scenario results in the formation of new middle–lower terrace levels that have quite a large preservation potential and that could eventually be mapped over a larger area, the wetter-to-drier climate change scenario returns model results where new terrace levels are poorly preserved. Following these model results, we thus propose that the middle–lower terrace systems in central Amazonia mainly reflect the response to climate cycles from drier to wetter in comparison to modern conditions. These results could offer a solution of why published terrace ages neither show at distinct clustering of ages nor a particular spatial pattern. We explain this by the short response times in combination with the high frequency of climate change and the rather stochastic pattern at which channels migrate laterally.
It is important to note that the initial topography that we used for our model was based on the SRTM digital elevation model (DEM) where the elevation of trees was maintained upon calculating the DEM. This introduces a bias in the sense that the difference in elevations between the terraces and the floodplains are larger in the DEM than they actually should be. This implies that the spatial extents where sediment accumulations during the drier periods could be even larger than our model suggests. This implies that our modeling results can be considered as conservative scenarios of landscape change.
The model results show that climatic variations that occur at the scale of thousands of years in central Amazonia (Cheng et al., 2013; Fritz et al., 2004; Mertes and Dunne, 2007) can have a large influence on the cycles of sediment aggradation and incision, and that the middle–lower terrace systems in central Amazonia mainly reflect the response to climate cycles from drier to wetter in comparison to modern conditions. This is different from the cyclicity recorded by terrace systems, e.g., in the Andean mountain belt valleys where terrace levels have been related to wetter-to-drier climate cycles, also relative to the modern climate (e.g., Steffen et al., 2009; Veit et al., 2016). The main difference between these environments is that in mountain belts, hillslopes have been considered as an additional landscape component where sediment can be stored during dry periods, whereas the material gets mobilized and supplied to the drainage network during wet climates, resulting in a temporary accumulation of material in the valley floor. During dry conditions, the hillslopes are considered as stable, which allows the trunk streams to recycle the previously deposited sediment in the valley floor, thereby forming a terrace level. Such a mechanism at work in mountainous areas has also been reproduced with numerical models (e.g., Tucker and Slingerland, 1997; Norton et al., 2016). This is, however, different from our case where the buffering effect of hillslopes is missing, since the central Amazonian landscape is mostly flat and far from the Andean mountain belt. Therefore, we mainly see the channel's responses to climate change, which thus have an aggradation–incision cyclicity and which thus contrasts to that in mountain valleys.
The scenario where we model the landscape response to an increase and then a decrease in the local base level (Fig. 6) provides an explanation why the topography of central Amazonia has preserved terrace levels at a high elevation (Fig. 5 at 10 ka). Such a scenario has the potential to form wide floodplains at higher elevations. As the base level drops, the rivers start to recycle previously deposited material. In addition, the rivers can take new courses. The model results thus predict that major changes in the channel network can preferentially be initiated following a phase of major sediment accumulation, possibly controlled by a higher base level at the downstream end of the model. Published ages of the higher terraces suggest that these geomorphic features are also made up of multiple terrace sequences with a stochastic pattern of ages and elevations, which is a feature that is similar to the middle–lower terrace levels but at a higher elevation (Fig. 2). Accordingly, using the model results of scenarios 1 and 2, we suggest that the formation of the higher terrace levels, possibly conditioned by a higher local base level, was superimposed by high-frequency climate changes. Accordingly, it is very likely that at such a stage of a higher base level, climate cycles would also cause the formation of cut-and-fill terrace sequences with a similar stochastic distribution of ages as those of the middle–lower terraces. This suggests that we can expect a superposition of low-level, high-frequency terraces on terrace deposits occurring at high levels, yielding a complex pattern in the landscape architecture.
According to the scenarios described herein, the current configuration of the terraces and floodplains in central Amazonia could be explained by aggradational and incisional processes controlled by high-frequency climatic variations and local base-level changes, which however occurred at a lower frequency. In particular, the higher terraces were deposited in a condition of a higher base level for the basins upstream of the confluence between the Solimões and Negro rivers. The subsequent decrease in the base level initiated a phase of gradual incision, thereby resulting in the current fluvial architecture. The high-frequency climate changes then yielded in the construction of middle–low terraces at various elevations, which however, are all situated at a lower elevation than the higher terrace levels. Our model shows that dry-to-wet shifts in climate, in relation to the modern situation, returns a landscape architecture where middle–lower terrace levels are better preserved than wet-to-dry changes in climate, again if the current situation is considered as reference.
Our results also show that fast and widespread landscape changes possibly occurred in response to high-frequency climate changes in central Amazonia, at least since the Late Pleistocene, with great implications for the distribution and the connectivity of different biotic environments in the region. Because of this short timescale of response to external perturbations, we suggest that the central Amazonian streams in the study area, and also at a broader scale, possibly respond in a rapid and sensitive way to human perturbations. This could be, for instance, the ongoing deforestation with implications for sediment flux and hydrology, and particularly the planned construction of multiple hydropower dams (Latrubesse et al., 2017; Best, 2019) with major implications on water discharge and local base levels as we have modeled in our contribution.
The open-source landscape evolution model SPASE is freely available at
SRTM data were provided by
The supplement related to this article is available online at:
RPdA designed the study together with AHdP. AHdP developed the model with support by VS and RPdA. CPG prepared the geological framework and synthesized data on water discharge, channel widths and sediment flux. AHdP analyzed the model results together with RPdA, CPG, VS and FS. AHdP wrote the manuscript with support from FS and RPdA. All authors approved the current version.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
The first author wishes to thank Beatriz Cardoso Silveira for comments made on earlier versions of this paper.
This research has been supported by the São Paulo Research Foundation (FAPESP, grant nos. 2016/03091-5, 2018/23899-2, 2017/06874-3 and 2018/02197-0), the National Research Foundation of Korea (NRF, grant no. 2019R1A6A1A10073437), the University of Bern, Innovative Training Network S2S (grant no. 860383), CNPq (grant nos. 305218/2009-3 and 426654/2018-8), the Royal Society – Newton Advanced Fellowship (NAF/R2/192188) and CAPES-PPGG (demanda social).
This paper was edited by Greg Hancock and reviewed by two anonymous referees.