The Amazon River drains the largest basin of the planet, with an estimated area of 6.1 × 10⁶ km² (Goulding et al., 2003) and is formed at the confluence of the Ucayali and Marañón Rivers in Peruvian Amazonia (Fig. 1A). The Ucayali River drains the southern part of the Peruvian Andean Cordillera and flows into the Marañón River, draining 360,000 km², with a mean annual discharge of approximately 11,200 m³·s⁻¹. The Marañón River drains 350,000 km² from the northern and central parts of the Peruvian Andean Cordillera to its confluence with the Ucayali, with a mean annual discharge of 16,200 m³·s⁻¹ (Espinoza-Villar et al., 2012).

The suspended sediment load varies according to the season. The annual average concentration of suspended sediment is 395 mg·L⁻¹ at the Requena station (Ucayali River) and 173 mg·L⁻¹ at the San Regis station (Marañón River) (Armijos et al., 2013). The width of the Ucayali River varies from 500 m to 1250 m. The average sinuosity is 1.94, the belt meandering current is 30 km wide and is characterized by multiple abandoned channels (Dumont, 1996). The width of the Marañón River varies from 1000 to 2500 m, with numerous islands in the straight sections. The average sinuosity is 1.33. The Marañón's flow is 14% greater than that of the Ucayali River. Average annual rainfall between 1946 and 1994 was of about 3080 mm·yr⁻¹ (Marengo et al., 1998). Precipitation is most intense between January and May. The hydrological cycle is very similar in the Marañón and Ucayali rivers, with flood peaks occurring between March and May, and low waters between August and October.

The study area is in Peruvian Amazonia, in the department of Loreto between 5°00'S–74°40’W and 4°00'S–73°20’W (Fig. 1B). This study analysed three lakes in the Ucayali River floodplain: Lakes Hubos and La Moringa (Fig. 1C1), Lake Carmen (Fig. 1C2), and one lake in the Marañón River floodplain, Lake Lagarto (Fig. 1C3), which is located 200 km from the confluence with the Ucayali River.

Four cores were collected in the upper Amazon Basin (Fig. 1). The CAR2 (4°35’23“S and 73°28‘.45”W) and HUB1 (4°30’29.19“S and 73°23‘.40”W) cores were collected in Lakes Carmen and Hubos using a “vibro-core.” The MOR1 (4°28‘48.61“S and 73°24’11.04”W) and LAG1 (4°31’04.81“S and 74°34’55.15”W) cores were collected manually in Lakes La Moringa and Lagarto, respectively. All cores were collected using aluminium tubes 7.5 cm in diameter. The CAR2, HUB1, MOR1 and LAG1 cores (measuring 600, 140, 120 and 98 cm, respectively) were opened, described and analysed with SCOPIX X-ray equipment at the EPOC (“Environnements et paléoenvironnements océaniques et continentaux”) laboratory at the University of Bordeaux-1 (France). Scopix uses classical X-ray equipment (X-ray source: 160 kV, 19 mA), coupled with new radioscopy instrumentation (CCD camera 756 Å ∼ 581 resolution), connected to a computer for data acquisition and processing (Migeon et al., 1998).

2.2.1 Water content and bulk density

2.2.2 Radiocarbon dating

2.2.3 Particle size analysis

2.2.4 Elemental carbon composition

2.2.5 Digital Elevation Model (DEM)

Eighteen samples along the CAR2 core were dated and presented an older age of 4612 cal yr BP (Table 1). Eight radiocarbon dates were obtained from the HUB1 core, with an age of 3215 cal yr BP for the basal section (Table 1). Twelve TOC AMS radiocarbon dates were obtained for the MOR1 core, with an age of 2961 cal yrs BP (Table 1). In the MOR1 and CAR2 cores collected in the Ucayali floodplain, the age model (Fig. 2) was based on a cubic-spline curve to the calibrated ¹⁴C ages and in the HUB1 core, the age model was based on a polynomial regression. A gap in sedimentation records was observed in the three sediment cores. This gap is well dated in the HUB1 record and occurs between 2950 and 719 cal yr BP. A sediment package of remobilized sediment, likely due to reworking of older sediment, occurred at the top of the three cores and was not considered in the age model. Four TOC AMS dates defined the sedimentary chronology of the LAG1 core, with a basal age of 630 cal yr BP and a probably modern top age (Fig. 2). No upper package of reworked sediment is observed in this core.

The CAR2, MOR1 and HUB1 cores are comprised of sediments that are poor in organic matter, with a predominance of silt and sand in the CAR2 core and silt and clay in the MOR1 and HUB1 cores (Fig. 3). Nevertheless, low carbon content, organic-rich layers containing plant remains can be identified along the three cores. These layers can be observed:

• • in the CAR2 core between 449–420 cm and 76–56 cm;
• • in the MOR1 core between 81–64 cm, 45–32 cm and 20–0 cm;
• • in the HUB1 at 45–44 cm.

The LAG1 core is composed of organic-rich clay with plant remains.

The sedimentary bulk density in the CAR2, HUB1 and MOR1 cores averages was 1.01 g·cm⁻³, 0.78 g·cm⁻³ and 0.73 g·cm⁻³, respectively, ranging from 0.31 g·cm⁻³ to 1.65 g·cm⁻³, 0.40 g·cm⁻³ to 1.53 g·cm⁻³ and 0.39 to 1.26 g·cm⁻³, respectively (Fig. 3). The LAG1 core presented lower bulk density values than the Ucayali cores, with an average of 0.19 g·cm⁻³, ranging from 0.05 g·cm⁻³ to 0.45 g·cm⁻³ (Fig. 3).

The sediment grain size analysis of the CAR2 core was characterized by a predominance of coarser sediments. Mean values for clay and fine silt are 3% and 20%, respectively, with values for medium silt, coarser silt, and fine sand of 19%, 40% and 18%, respectively (Fig. 3). Analysis of the MOR1 and HUB1 cores showed predominance of clay and fine silt. The mean percentages of clay, fine silt, medium silt and coarse silt, were 33%, 51%, 15% and 1% in MOR1 and 32%, 46%, 17% and 5% in HUB1, without traces of sand (Fig. 3). In the LAG1 core, a predominance of clay and fine silt can be observed, with mean values of 53% and 42%, respectively, and with low values for medium silt, with an average of 5% (Fig. 3).

The carbon content of the CAR2 core averaged 1.01 wt%, ranging from 0.25 wt% to 11.7 wt%, with a significant increase between 256 cm and 288 cm (Fig. 3). The HUB1 core had relatively low TOC levels, between 0.56 wt% and 5.38 wt%, with an average of 1.94 wt% (Fig. 3). The MOR1 core had also low TOC levels, between 1.09 and 5.4 wt%, with a mean content of 2.40 wt% (Fig. 3). The LAG1 core TOC values were very high, with mean values of 25.7 wt%, ranging from 8.06 wt% to 50 wt% (Fig. 3).

The Ucayali, Marañón and Beni Rivers are responsible for nearly all of the sediments transported by the Amazon (Guyot et al., 2007). More than 540 million tons of sediments are transported each year in the upstream part of the Amazon River (Filizola and Guyot, 2009; Filizola et al., 2011; Gibbs, 1967; Guyot et al., 2007; Martinez et al., 2009; Meade et al., 1979). The transport and deposition of this large amount of sediment significantly influences the fluvial dynamics of the Amazonian rivers (Constantine et al., 2014). Historical geomorphological data from the Maranon and Ucayali rivers show that the confluence of the two rivers has migrated from SW to NE during the last century (Dumont, 1996). There are few studies dedicated to the recent dynamics of the Ucayali and Marañón Rivers (Dumont, 1994; Quintana, 2015). Through the study of LANDSAT images, Quintana (2015) estimated an average lateral migration of 140 m·yr⁻¹ for the Ucayali River, reaching a peak of 2500 m·yr–¹ between 1975 and 2011. In contrast, the Marañón River presented a low migration rate during this period, as shown in Fig. 4. Changes in climate and environmental conditions may trigger a range of alterations in the fluvial system (May et al., 2015).

During the late Holocene, sedimentation in the Ucayali and Marañón floodplain lakes was characterized by different depositional processes. The chronological models obtained from the three sediment cores collected in the Ucayali floodplain lakes show that sedimentation was not gradual and continuous. The CAR2 core, from Lake Carmen, is the only one with expressive sandy sedimentation, mainly at its base and its top, indicating stronger hydrodynamics in this lake. The base of the core is characterized by sandy silt (Fig. 3) deposited between cal 5000 and 3540 ± 80 cal yr BP (0.06 g·cm⁻²·yr⁻¹). At 3540 ± 80 yr BP, a 90-cm sequence of coarsening-up silts (480–390 cm) settled very quickly. A rapid sedimentation event is also observed at 3000 cal yr BP in the HUB1 core from Lake Hubos (Fig. 2), but it does not present a granulometric trend. Periods characterized by abrupt accumulation (i.e. sediment packages of the same age) were also reported for the late Holocene by Moreira-Turcq et al. (2014), and have been interpreted as rapid deposition of sediment packages that probably correspond to one or several successive events of extreme flooding of the Amazon River. This type of deposition by sediment packages has also been evidenced in modern sediments from the Beni floodplain (Aalto et al., 2003), where they correspond to extreme floods associated with La Niña events. The CAR2 480–390-cm and HUB1 140–120-cm layers also seem to correspond to that kind of sedimentary process, related to overbank deposition and associated with the erosion of inlet channels and levees during very high floods. These flood occurrences are in agreement with a relative humid episode evidenced between 3500 and 2800 cal yr BP in a speleothem from Huagapo Cave, in the central Peruvian Andes (Kanner et al., 2013). After this event, the record of CAR2 shows the deposition of fine silt sediments, with a lower accumulation rate of 0.04 g·cm⁻²·yr⁻¹, while in HUB1 the sedimentation stopped at 3000 cal yr BP. Both reflect a decrease in river sediment supply to the lakes, which may be due to a reduction in river discharge or, more likely, considering the high dynamics of river migration in the Ucayali floodplain (see Fig. 1), a migration of the main river course away from the lakes. The trend of increasing precipitation recorded in the central Andes between ca 6500 and 2500 cal yr BP (Bird et al., 2011; Kanner et al., 2013; Seltzer et al., 2000; Thompson et al., 1995) reinforced the hypothesis of a change in river position rather than a decrease in river discharge. In Lake La Moringa, core MOR1, the accumulation rate is also low (0.013 g·cm⁻²·yr⁻¹) between 2960 and 660 cal yr BP, reflecting a weak influence of the river during this period. At 720 ± 45 cal yr BP, deposition resumed in HUB1 with a rate of 0.03 g·cm⁻²·yr⁻¹, while at 620 ± 60 cal yr BP in MOR1, the accumulation rate increased to 0.09 g·cm⁻²·yr⁻¹, evidencing renewed influence of river supply, culminating with a 15-cm sediment package at 390 cal yr BP. This deposition increase was probably due to the return of the river's main channel to a closer location to Lakes Hubos and La Moringa, until it reached its present-day position.

In contrast, sedimentation in the LAG1 core from Lagarto Lake in the Marañón floodplain is very different, although this core only reached the last 600 cal yr BP. The core is characterized by fine sediments with high TOC values and a rapid constant sedimentation rate of 0.22 cm·yr⁻¹ from the base to the top of the core. Hydrodynamics are weaker in that lake, which is surrounded by palms and characterized by low-density sediment, resulting in an accumulation rate between 0.01 to 0.09 g·cm⁻²·yr⁻¹, which is greatly supported by accumulation of organic matter (20% to 50% TOC). Organic sedimentation in Lake Lagarto reflects significant inputs from the local watershed (dominated by Mauritia flexuosa) and probably in situ primary production. Lake Quistococha, in the Iquitos region, also presents an organic sedimentation due to Mauritia flexuosa and algae production during the last 2600 yr cal BP (Aniceto et al., 2014). We estimated the mean river sediment supply to the lake at about 0.06 g·cm⁻²·yr⁻¹ between 630 and 500 cal yr BP, which however lowered to 0.02 g·cm⁻²·yr⁻¹ after 400 cal yr BP. This decrease in sediment supply may be due to lateral migration of the river's main channel, although there is no evidence of recent meander migrations in that section of the Marañón River (Fig. 1), or to a decrease in river discharge since 600 cal yr BP. The analysis of Landsat images confirms the relative stability of the Marañón meanders (Dumont, 1996; Quintana, 2015). Lower precipitation is observed in the Andes since 1600 AD (Apaéstegui et al., 2014; Bird et al., 2011; Kanner et al., 2013; Thompson et al., 1995) and would explain the reduced discharge of the Marañón River during that period.

Channel migration is a natural process that remobilizes old alluvial sediments, which mix with modern sediments transported by the river from actively eroding hill slopes (Wittmann et al., 2011). Our data show that sediment stratigraphy in Ucayali floodplain lakes over the past 4000 years is characterized by the presence of packages of very high sedimentation rates (abrupt accumulation) in the three Ucayali cores (Fig. 2). The very short duration of this deposition is below the ¹⁴C resolution, and the age of these packages varies between 400 and 3500 cal yr BP. These packages are interpreted as the rapid deposition of sediment that probably corresponds to one or several successive events of extreme flooding of the Ucayali River (Moreira-Turcq et al., 2014). These are characterized by the presence of older ¹⁴C age values in the sediment layer, which are interpreted as a contribution of older reworked alluvium. We also observed reworked sediments in the top unit of the Ucayali floodplain cores (Fig. 2), whose ages vary between 600 and 2800 yr BP, deposited upon lake sediment younger than 1950 AD (post-bomb). These sediment packages have a thickness of 250, 30, and 15 cm in Lakes Carmen, Hubos, and La Moringa, respectively. These packages were thicker in the lake nearest the river main channel, as observed in the Carmen sedimentary records. They record a modern period of very high floods that reworked past alluvial sediments of the Ucayali floodplain.

In the case of Lake Carmen, this modern deposition is directly linked to the invasion of the lake by a meander of the Ucayali River (Fig. 4). Due to lateral meander migration, the river levee migrated into Carmen Lake and rebuilt itself over past lacustrine deposits. Based on several satellite images (Fig. 4), we can determine that, in 2004, the river levee reached the lake and cut almost half of its surface. In the case of Lakes La Moringa and Hubos, however, the river remains farther away, and the modern deposition of reworked sediments implies an occurrence of extreme flooding during the past 60 years, at most.

Extreme hydroclimatic events can multiply the effects of the natural channel migration process, and Carmen Lake modern deposition, related to meander migration, may also be a consequence of an increase in flood magnitude. This migration began in 1989, after a stability of several millennia. The river had migrated to the south 3500 years ago, leaving the proximity of Lake Carmen, but never went north of its position of 1973, as shown by the pattern of older paleochannels on the left bank of the river (Fig. 4). If the resumption of meander migration is linked to the extreme floods registered in Lakes Hubos and La Moringa, it means that these extreme events occurred after 1989. Based on hydrography data, Espinoza et al. (2012) have shown an increase in flood levels in Amazonian rivers after 1980, probably due to current anthropic-driven climate change. Our data confirm that floods from the last 60 years, and probably from the last decade (Espinoza et al., 2009, 2011, 2012), have had a higher impact, in term of alluvial sediment reworking, than any other floods in the past 3000 years.