The Andes leave an indelible imprint on the geochemistry and ecology of the Amazon River system, the largest in the world. The Madeira River, the second largest tributary of the Amazon, drains the southwestern Amazon basin, which has a great geographical, biological, and climatic diversity (Molina-Carpio et al., 2017). At its confluence with the Amazon, the Madeira River delivers 26,580 m³·s⁻¹ and provides 426 Mt·year⁻¹ of suspended sediment load (SSL; nearly 50% of the Amazon's SSL at Óbidos for the 2002–2011 period; Vauchel et al., 2017). At the Porto Velho station (Figure 1) a mean annual discharge of 18,550 m³·s⁻¹ and SSL of 441 Mt·year⁻¹ were recorded for the 2002–2011 period (Vauchel et al., 2017).

The complexity of the Madeira River system is evident from the presence of four geomorphological units (Figure 1): The Andes, the Fitzcarrald Arch, the Llanos de Moxos floodplain, and the Brazilian shield. Steep terrain and young lithologies make the Andes the main source of sediments and solutes to the lower reaches of the Madeira River (McClain and Naiman, 2008). The average sediment production of the Andean stretch of the Upper Madeira was estimated at 640 Mt·year⁻¹ (2002–2011; Vauchel et al., 2017), implying that over 30% of the Andean sediment is deposited in the Andean piedmont and the Llanos de Moxos floodplain. Depending on which geomorphological unit the tributaries are born, contrasting characteristics are observed: white-water tributaries (the Beni, Mamoré and Madre de Dios rivers that rise in the Andes), the clear-water Guaporé River that comes from the Brazilian shield and several black-water lowland tributaries. In fact, the large differences in SSL between the Mamoré and Beni rivers (Figures 2B,D) are mainly explained by the different geomorphology of their watersheds. Although both rivers rise in the Andes, no sedimentation process is observed in the Beni River floodplain, resulting in the Andean Beni River at Rurrenabaque (Figure 1) to provide 45% of the Madeira SSL with a watershed area of only 7% of the Upper Madeira basin (Vauchel et al., 2017). Hence, the Beni River provides nearly 80% of the SSL at Porto Velho: 47% from the Beni River itself and 32% from its main tributary, the Madre de Dios River (Vauchel et al., 2017). On the other hand, significant sedimentation occurs in the Andean piedmont and the entry to the floodplain in the Mamoré River basin. This is explained by Vauchel et al. (2017) by the Mamoré River slope break at the piedmont, that reduces the energy available in the river channel to transport sediment. Furthermore, the Guaporé River, the main tributary of the Mamoré, has most of its basin on the geomorphic stable Brazilian shield, resulting in very low sediment yield and denudation rate (0.02 mm·year⁻¹; Guyot et al., 2007; Withmann, 2008). As a consequence, the Mamoré and Guaporé rivers provide only 15% of the Madeira SSL at Porto Velho (Vauchel et al., 2017).

The tropical climate of the Madeira basin, and especially the intensity, magnitude, and seasonality of the rainfall in sediment-source regions as the Andes greatly influences the sediment dynamics at different time scales (e.g., Pepin et al., 2013). A significant rainfall spatial variability has been identified in the Upper Madeira basin. From north to south mean annual rainfall is 2600 mm for the Madre de Dios sub-basin, 1800 mm for the Beni sub-basin and 1600 mm for the Mamoré-Guaporé (Molina-Carpio et al., 2017). High rainfall in the Madre de Dios sub-basin is explained by its northernmost position and the presence of an important “rainfall hotspot” in the Sub-Andes and piedmont, where up to 6600 mm·year⁻¹ is recorded (Espinoza et al., 2015). On a seasonal timescale, the entire Upper Madeira basin is characterized by a south tropical rainfall regime (Molina-Carpio et al., 2017), with a marked wet season during the austral summer, related to the mature phase of the South American Monsoon System (Berbery and Collini, 2000). On average, over half of the annual rainfall is recorded between December and March, with January and February being the wettest months (Espinoza et al., 2019a).

The Madeira River intra-annual discharge seasonality at Porto Velho, which peaks between March and April (Figure 2A), is the result of the rainfall and hydrological regimes of the main tributaries. Figure 2B shows that the mean annual discharge peak of the Beni River at Cachuela Esperanza comes at about the end of February and the beginning of March, while the peak at the station of Guayaramerin on the Mamoré River usually occurs later in mid-April (Figure 2D). Surface suspended sediment concentration (SSSC) at Porto Velho starts to increase in October with the beginning of the rainy season. It peaks between January and March (Figure 2A), in response to the Beni River SSSC annual cycle (Figures 2B,C).

Information on the Madeira River sediment dynamics is even more essential as the consequences of human activities (e.g., deforestation, sedimentation induced by dams, climate change) on the wellbeing of the riverine and indigenous population and ecosystems become more evident. The Andean-Amazon connectivity has been compromised for years. Nobre et al. (2016) suggest that deforestation over the southwestern Amazon could result in savannization in the upcoming decades. Deforestation in the Upper Madeira might increase soil erosion and sediment production as observed in the Andes of Colombia (Restrepo et al., 2015) and in Ecuador (Molina et al., 2008). Additionally, land use changes can modify the local hydroclimatic conditions as observed in the Upper Madeira Basin (Gutierrez-Cori et al., 2021). There is no doubt that the effects of land use changes need to be evaluated over a large timeframe. Changes in the spatio-temporal rainfall variability over the Andes-Amazon transition region as reported by Espinoza et al. (2019a) and Segura et al. (2020) has been related to large scale atmospheric conditions and could modify the sediment dynamics as well over different temporal scales. Indeed, Armijos et al. (2020) found that the temporal rainfall variability is well-related (R² = 0.89) to the fine suspended sediment concentration (FSC; >0.45 μm and <63 μm) at Porto Velho, while the coarse sediment (>63 μm) is better related to the water discharge.

Hydropower facilities already in operation or been planned in the Madeira basin (Anderson et al., 2018) can significantly reduce the SSL reaching the Amazon River. Santo Antonio and Jirau, two hydropower dams in the river stretch upstream from Porto Velho (Figure 1) and completed in 2011 and 2012, respectively, have already caused serious social and environmental impacts (Almeida et al., 2019, 2020; Van Damme et al., 2019; Lima et al., 2020). These impacts had a profound effect on downstream and upstream riverine areas in Brazil and Bolivia and the communities who depend on the river resources. The dams were also credited by Latrubesse et al. (2017) to have caused an ~20% decrease in the mean surface suspended sediment concentration (SSSC) of the Madeira River at Porto Velho, located just 7 km downstream of Santo Antonio dam. Li et al. (2020) also detected a negative trend of SSSC for the 1995–2017 period at Fazenda Vista Alegre, the station furthest downstream on the Madeira River before its confluence with the Amazon River. In addition, Ayes et al. (2019a) found an ever-greater fall (30%) in FSC at Porto Velho in the post-dam period. However, a concomitant 36% decline in the FSC observed at Cachuela Esperanza, located upstream from the dams on the Beni River, which provides about 80% of the Madeira SSL, suggests that most of the fall in FSC could not be attributed to the dams. Indeed, the study of Vauchel (2014) demonstrated that the dams do not generate a backwater effect at Cachuela Esperanza. On the other side the already observed sedimentation of coarser particles (mostly fine and medium sands) in the Santo Antonio and Jirau reservoirs has been directly associated with the river impounding by the dams (International Hydropower Association (IHA), 2017).

While negative trends from 2003 to 2017 and a downward break in 2010 were detected at Porto Velho for the annual and December-February (DJF) series of FSC, trends were not detected for the discharge series during this period (Ayes et al., 2019a). For the DJF sediment peak season at Cachuela Esperanza, the study identified a FSC negative trend and a significant break in 2010. No FSC trend or break were found for the Mamoré River at Guayaramerin for the same period in the previous study. The causes of the FSC downward break just before the dams were completed and the negative trend during DJF for the 2003–2017 period in the Beni River, merit being identified and explained.

Given the importance the Madeira River sediment dynamics has on the geochemistry and ecology of the Amazon River system and with a view to improve the knowledge of the different processes (natural and anthropogenic) that are occurring within the Upper Madeira River basin, this study aims to investigate the role of the rainfall variability on the SSSC downward break in 2010 and declining trend identified for the 2003–2017 period at the Porto Velho station.

Figure 3 shows that the mean SSSC (590 mg·L⁻¹ for the DJF season) at Porto Velho from 2010 to 2017 was lower than during the 2003–2009 period (SSSC = 750 mg·L⁻¹). Following the SSSC changes detected in the Beni River at Cachuela Esperanza (Ayes et al., 2019a) and as nearly all of the fine suspended sediment transported by the Beni River comes from the Andean part of its basin (see Introduction), the temporal relationship between Rurrenabaque and Cachuela Esperanza SSSC series was evaluated at a monthly time-step from November to February. From 2003 to 2017 a significant positive correlation between Rurrenabaque and Cachuela Esperanza SSSC series is found (τ = 0.43, p < 0.001), which even improves when the extreme 2014–flood event is removed (τ = 0.48, p < 0.001). Both the Rurrenabaque sediment contribution and the significant correlation between Rurrenabaque and Cachuela Esperanza SSSC indicate that the temporal variability of the Andean Beni River at Rurrenabaque is key to understand the fine sediment dynamics at Cachuela Esperanza.

The SSSC, Q and rainfall time evolution were evaluated at Porto Velho and Rurrenabaque gauge stations and the watersheds they monitor, for each month separately (Figure 4). In December, a negative SSSC trend (MKτ = −0.48, Sen = −24.7, p < 0.05; Figure 4A), was identified at Porto Velho for 2003–2017, accompanied by a downward break in 2010 (p < 0.10) and a change of distribution after the Mann-Whitney test (w = 50; p < 0.05). There is no significant change in the December SSSC at Rurrenabaque or in discharge for both stations (Figures 4A,D). Rainfall in December for the whole Upper Madeira basin monitored at Porto Velho does not exhibit a significant trend either (Figure 4A).

In January, a negative trend for the SSSC time-series (MKτ = −0.43, Sen = −29.0, p < 0.05; Figure 4B) in 2003–2017 was detected at Porto Velho. At Rurrenabaque a significant negative January SSSC trend (MKτ = −0.33, Sen = −120.6, p < 0.10) for the 2003–2017 period and a downward break (p < 0.10) in 2008 were detected (Figure 4E). A noticeable change in SSSC occurred in 2008, from a mean concentration of 4070 mg·L⁻¹ (2003–2008) to 2060 mg·L⁻¹ (2009–2013), which was followed by an upward change from 2014 to 2016 (3760 mg·L⁻¹; Figure 4E). In spite of the good correlation between SSSC and rainfall at Rurrenabaque (τ = 0.67, p < 0.001), neither a significant break nor a trend was detected for January rainfall and discharge at the 90% confidence level for this sub-basin.

A negative SSSC trend (MKτ = −0.35, Sen = −26.8, p < 0.10; Figure 4C) was identified again in February at Porto Velho but no discharge and rainfall trends were detected. In contrast to January, a positive rainfall trend (MKτ = 0.63, Sen = 10.0, p < 0.01) was identified at Rurrenabaque for February (Figure 4F). This suggests a shift in the monthly rainfall peak from January to February from 2003 to 2017 for the Rurrenabaque sub-basin. Indeed, CHIRPS.v2 data shows that between 2003 and 2008 the mean rainfall was 240 mm in January and 160 mm in February. This changed to 175 mm (−27%) and 240 mm (+50%), respectively, for the 2009–2013 period, while a mean rainfall of 240 mm has been recorded for both months since 2014 (Figures 4E,F). The shift in the rainfall between January and February was also detected for the observed rainfall in most of the Andean stations of the Beni and Madre de Dios River basins, when comparing the 2003–2008 and 2009–2013 periods (Figure 5). Supplementary Table 1 summarize trends and breaks tests during these 3 months.

As spatio-temporal rainfall variability seems to differ along the Upper Madeira basin and no trend was detected for the basin mean rainfall, we evaluated the temporal evolution for each month from December to February during the 2003–2017 period (Figure 6). In December, significant negative rainfall trends (p < 0.10) were observed in the nearby areas of Porto Velho and to the southern part of the Mamoré-Guaporé sub-basins (Figure 6A). A negative but not significant trend was also observed in the upper Madre de Dios River and most of the northern part of the Guaporé sub-basins and a positive not significant elsewhere. Negative but mostly not significant rainfall trends were observed in January in the Andean part of the Beni and Madre de Dios watersheds and in the Guaporé sub-basin. Only a few regions in the Mamoré-Guaporé sub-basin exhibited a significant negative trend (p < 0.10). Positive trends were detected in the lowlands to the north of the Upper Madeira basin (Figure 6B). A significant positive trend (p < 0.10) was detected for the rainfall in February for the Andean parts of the Beni and Madre de Dios sub-basins and adjacent lowlands (Figure 6C). Because of these spatial differences, the basin-averaged rainfall at Porto Velho did not present a significant change in the December to February period (Figures 4A–C).

Figures 6D–F confirms that for December the 2010–2017 time period was significantly drier than the 2003–2009 near Porto Velho, most of the Madre de Dios and in the southern part of the Mamoré-Guaporé sub-basins. For January, the 2010–2017 period was drier than the 2003–2009 one in the Andean Beni River sub-basin and the Guaporé watershed. In February the 2010–2017 period was significantly wetter than the 2003–2009 period in the Andes, particularly in the Andean Beni River watershed at Rurrenabaque.

Being December a transition month to the SSSC peak at Porto Velho and since dry conditions were observed in the north region after 2010 (Figures 6A,D), the onset of the rainy season for the four main tributaries sub-basins was evaluated. Figure 7B shows that the rainfall season starts earlier (mid of September) in the Madre de Dios sub-basin than in the other three sub-basins. This is accompanied by a much higher (almost twice) daily rainfall than in the remaining Upper Madeira basin. Because of this and considering the importance the rainfall has on soil erosion at the start of the rainfall season, the precipitation trend for October to December (OND) was evaluated (Figure 7A) for 2003–2017. A significant negative trend (p < 0.05) over the Madre de Dios sub-basin was found. In parallel, the OND intensity as measured by the basin mean daily rainfall (mm.day⁻¹) in this sub-basin exhibits a significant negative trend (p < 0.05) for the same period. In fact, the highest OND daily rainfall (>50 mm.day⁻¹) occurred at the beginning (2003–2005) of the study period (Figure 7C), compared to <40 mm.day⁻¹ for eight of the remaining years.

The results of this study indicate that for a relatively short period (2003–2017) significant diminution in the surface suspended sediment concentration (SSSC) time-series of the Upper Madeira basin are associated with the spatiotemporal rainfall variability in the basin. The negative rainfall trend (p < 0.05) from October to December in most of the Madre de Dios River sub-basin, and in the northeastern part of the Upper Madeira basin near Porto Velho, resulted in a probable decline in the Madeira River SSSC at the beginning of the SSSC annual cycle. Whereas, a decline in the SSSC contribution of the Andean regions of the Beni and Madre de Dios rivers is the likely cause of the January SSSC negative trend (p < 0.10) at Porto Velho. This was related to a shift of the rainfall peak from January to February during the study period. During 2009–2013 the mean January rainfall in those Andean regions dropped by more than 20% from the 2003–2008 period and with it the mean SSSC fell nearly 50% at Rurrenabaque. Conversely, the mean February rainfall increased nearly 50% after 2008 in the Andean part of the Beni and Madre de Dios sub-basins. The positive rainfall trend (p < 0.05) during February rainfall did not generate a positive SSSC trend as the sediment dynamics in this month is already influenced by the dilution associated with base flow at Rurrenabaque (as shown in Ayes et al., 2019b) and the Llanos de Moxos floodplain. It is also likely that the diminution in the interannual DJF rainfall variability, including rainfall extremes from 2002 onwards (e.g., Mamani, 2019; Segura et al., 2020) in the semiarid and subhumid regions that provides most of the fine sediment at Rurrenabaque had a relevant influence on the erosion processes and SSSC trend at this station.

These changes in the rainfall and SSSC coincided with the construction of the Jirau and Santo Antonio dams and the time when they started to operate. Although the hydroelectric facilities could have an influence, the sediment trapping effect of the dams is probably of minor importance for the SSSC dynamics. However, it probably was very relevant for coarser particles (fine and medium sands) transported by the Madeira River. This study shows that the decline of SSSC in the Madeira River should not be evaluated just on the basis of the SSSC data downstream from the hydroelectric facilities, but also on the basis of the natural processes upstream from the dams.

These results illustrate the complexity of the Madeira sediment dynamics and its relationship with changes in the regional hydrological cycle. Changes in the rainfall peak over the Andean Beni and Madre de Dios region, which covers <1.8% of the Amazon Basin, were related to a decline of fine suspended sediments in the middle stretch of the Madeira River. The changes in the rainfall patterns occur at a time when the Upper Madeira Basin is undergoing a major biophysical transition (e.g., deforestation and construction of major infrastructure projects) that usually is expected to act in the opposite way, increasing soil erosion and sediment transport. These results highlight the need to identify the complex relationships acting on the several components of the hydrological cycle in the Amazon-Andes transition zone, where erosion and sediment transport are key processes for the preservation of the Amazon biodiversity. Further insights and knowledge could emerge as more information becomes available.

Publicly available datasets were analyzed in this study. This data can be found at: https://hybam.obs-mip.fr/, https://www.geography.unibe.ch/research/climatology_group/research_projects/decade/index_eng.html, and https://data.chc.ucsb.edu/products/CHIRPS-2.0/.

IA: conceptualization, methodology, formal analysis, investigation, writing-original draft, and visualization. JM-C: conceptualization, methodology, formal analysis, investigation, validation, and writing-original draft. JE: conceptualization, methodology, formal analysis, validation, and writing-original draft. OG-C: methodology, formal analysis, validation, visualization, and writing-review, and editing. WC: writing-review, editing, and visualization. FF: validation, writing-review, and editing. EA, RE-V, and NF: writing-review and editing. JA: methodology, writing-review, and editing. All authors contributed to the article and approved the submitted version.

JM-C was supported by the Universidad Mayor de San Andres (UMSA) within the framework provided by the HYdrogéochimie du Bassin AMazonien (HYBAM) program and PHYBAAM (Processus Hydrologiques des Bassins Andins Amazoniens) project. JE was supported by the French AMANECER-MOPGA project funded by ANR and IRD (ref. ANR-18-MPGA-0008). OG-C was supported by French MOPGA program funded by Sorbonne Université. WC was supported by the Universidad del Valle (Cali-Colombia).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.