Rivers are no longer viewed as passive conduits from land to sea but, rather, play an active role in processing organic carbon derived from land and returning it to the atmosphere as carbon dioxide (CO₂) (Cole et al., 2007; Battin et al., 2009). The remaining organic and inorganic carbon that is exported to the coastal ocean is further processed, released to the atmosphere, or stored in marine waters and sediments (Medeiros et al., 2015; Ibanhez et al., 2016). Streams, rivers, and lakes have most recently been estimated to emit 2.1 Pg C year⁻¹ to the atmosphere (Raymond et al., 2013), increasing from past estimates of 1.4 Pg C year⁻¹ (Tranvik et al., 2009), and 0.8 Pg C year⁻¹ (Cole et al., 2007). Although data coverage is more sparse, wetlands, which were not included in estimates by Raymond et al. (2013), emit another ~2.1 Pg C year⁻¹ (Aufdenkampe et al., 2011). These combined estimates, along with storage and export terms, imply that roughly 5.7 Pg C year⁻¹ is transported through inland waters, with nearly 75% of this carbon being returned to the atmosphere (Le Quéré et al., 2015). Tropical regions have been identified as hotspots for aquatic CO₂ outgassing, representing ~75% of global emissions, yet they are under-represented in global datasets, particularly with respect to direct measurements of fluxes and concentrations, which allows quantification of gas transfer velocity values that are used in regional and global models (Regnier et al., 2013; Wehrli, 2013).

The Amazon River is the largest river system in the world, responsible for 20% of the fresh water discharge to world's oceans and 25% of the emissions of CO₂ from inland waters to the atmosphere, globally (Richey et al., 2002; Raymond et al., 2013). The influence of the Amazon River on primary productivity in the Atlantic Ocean can be seen from space, driving a net uptake of CO₂ in the plume (Subramaniam et al., 2008). The source of dissolved CO₂ in large river systems shifts from headwaters to higher order rivers/streams. In small headwater streams the primary source is subsurface flow from riparian soils (Johnson et al., 2008). The relative contribution from soil respiration decreases compared to in situ production via microbial respiration as stream order increases (Butman and Raymond, 2011). The breakdown of young terrestrially-derived organic carbon (OC) by heterotrophic river microbes is thought to be the primary source of CO₂ in the Amazon River mainstem (Mayorga et al., 2005; Ward et al., 2013, 2016), although plant respiration and OC decomposition in floodplains also contribute to CO₂ supersaturation (Abril et al., 2014).

The majority of geochemical studies in the Amazon River have focused on the central Amazon, which represents about 30% of the 6 million km² drainage basin (Hedges et al., 1986, 2000; Moreira-Turcq et al., 2003). For example, the current basin-scale CO₂ budget for the Amazon River is based on aereal outgassing rates determined for this corridor, and outgassing rates were conservatively assumed to be 50% less in unstudied regions outside of the central corridor (Richey et al., 2002). The lower reaches of the Amazon River, between the historic gauging station, Óbidos, and ~800 km downstream to the mouth, have not been included in current basin-scale budgets. This represents ~13% of the basin's total surface area (in terms of land, not water surfaces) and is characterized by expansive floodplains and flooded forests, which likely provide large inputs of OC and CO₂ to the river. In fact, 75% of the particulate OC load is lost between Óbidos and the mouth largely due to degradation, while dissolved OC concentrations slightly increase due to constant inputs from the watershed and floodplains that balance OC degradation (Seidel et al., 2015; Ward et al., 2015). Tidal effects can be detected more than halfway upstream to Óbidos with flow completely reversing near the mouth. These forces increase water residence time and along with strong winds and wide channels (2–15 km) with a long fetch, create rough water surface conditions that likely promote CO₂ degassing. Including a quantitative evaluation of CO₂ emissions in this unique reach of the river is critical for constraining the basin scale carbon budget, which directly influences global estimates.

This study provides the first detailed evaluation of CO₂ concentrations and fluxes along the lower Amazon River and its major tributaries, the Xingu and Tapajós rivers. Direct measurements of CO₂ outgassing were made with floating domes for each hydrologic period (i.e., low, rising, high, and falling water) from 2014 to 2016 along with measurements of CO₂ concentrations and calculations of gas transfer velocities. Total CO₂ evasion was estimated for three discreet sections of the lower river: (1) the Amazon River main channel from Óbidos to the downstream study boundaries near Macapá, (2) the lower regions of the Tapajós and Xingu tributaries, and (3) the extended region from Macapá to the actual geographic river mouth. These estimates were used to calculate a range of updated basin scale CO₂ outgassing budgets based on previous estimates (Richey et al., 2002; Rasera et al., 2013), which were compared with global budgets.

In this study we measured pCO₂ and CO₂ fluxes from the water to atmosphere along with discharge, water current velocity and wind speed in the Lower Amazon River channel, from the last gauging station at Óbidos to the mouth. The two major clear water tributaries of the Amazon River had lower pCO₂ and FCO₂ values than the Amazon River mainstem, which is consistent with other studies (Alin et al., 2011; Rasera et al., 2013). This was expected based on the lower suspended sediment loads, which enables high rates of primary productivity as indicated by enhanced levels of Chlorophyll-a (Sioli, 1985; Mayorga and Aufdenkampe, 2002; Ward et al., 2015).

Previous pCO₂ measurements for sites within our study boundaries ranged from 1600 to 6037 μatm in the mainstem and from 70 to 1070 μatm in the tributaries, agreeing well with our observations (Alin et al., 2011; Borges et al., 2015). Direct FCO₂ measurements in the lower Amazon River mainstem and Tapajós River have only been reported by Alin et al. (2011) and ranged from 1.87 to 10.62 μmol m⁻² s⁻¹ in the mainstem and from 0.04 to 6.36 μmol m⁻² s⁻¹ in the Tapajós River, which is also similar to our measurements. The seasonal trend observed in pCO₂ in the lower Amazon River follows the hydrologic cycle as observed by Richey et al. (2002), with maximum CO₂ concentrations during high water and minimal concentrations during the lower water season. FCO₂ measurements followed the same trend since it was strongly correlated with pCO₂ (Table 2).

Despite the similar ranges of pCO₂ and FCO₂ our range of k₆₀₀ was considerably higher than Alin et al. (2011). This can mostly be attributed to the higher wind speeds recorded in the river during our measurements compared with Alin et al. (2011). The furthest downstream station sampled by Alin et al. (2011) was near Santarém, which is roughly 500 km upstream of Macapá. We observed a downstream increase in wind speeds (Figure 6), which should lead to higher k₆₀₀ values considering the typical correlation between wind speed and k₆₀₀ in large rivers (Borges et al., 2004b; Alin et al., 2011; Rasera et al., 2013). Although we did not find any correlation between direct k₆₀₀ calculations and the hydrological parameters or wind, the average binned k₆₀₀ for U₁₀ bins of 0.5 m s⁻¹ presents a stronger positive correlation here and in the aforementioned studies (Figure 8). Channel width and the fetch length (i.e., the distance traveled by wind or waves across open water) also increases downstream, which amplifies the effects of wind, waves, and currents on surface water texture and turbulence. For example, the Amazon River is characterized by large sweeping curves and relatively narrow channels upstream of Santarém, which limits wave formation, whereas the main channel remains fairly straight for ~250 km between Santarém and Almeirim, and remains straight again after a slight turn to the northwest for 2,250 km between Almeirim and Macapá.

The gas exchange coefficient and its variability within a system is among the most important factors controlling CO₂ emissions from different parts of a large basin (Raymond and Cole, 2001; Borges et al., 2004a,b; Alin et al., 2011; Striegl et al., 2012). In shallower streams and rivers where turbulence is high due to bottom friction, k₆₀₀ can be expressed as a function of the water flow speed and depth, where shallower and faster streams tend to have higher k₆₀₀-values than slower streams (Raymond and Cole, 2001). In the Yukon basin, tributaries had higher k-values for CO₂ than the mainstem, and in the mainstem k increased downstream (Striegl et al., 2012). In the Amazon, a similar trend was observed where k₆₀₀ was higher in rivers with narrower channels (< 100 m wide) (Rasera et al., 2008; Alin et al., 2011).

In larger rivers and estuaries the main driving factors controlling k₆₀₀ is wind and water currents, which in turn are attributed to a balance between several factors (e.g., discharge, tidal range and fetch) (Alin et al., 2011). In the Scheldt estuary water current is an important factor controlling k₆₀₀ but its significance decreases with increasing wind speed (Borges et al., 2004b). In the Amazon and Mekong rivers, water currents are generally higher than in estuaries but with a more limited fetch and lower wind speeds (Alin et al., 2011). However, as previously mentioned, fetch dramatically increases along the lower Amazon River (Figure 1). The factors controlling k in large river basins is a mixture of all those described above. Wind are generally higher than those in headwater streams and current velocities are typically faster than in estuaries (Beaulieu et al., 2012). Additionally, the effect of wind on the water turbulence can be related to the water body orientation, shape, size, and direction of wind and water current. When wind and water currents are directionally opposed they can interact synergistically to produce unusually high k-values for any given wind speed (Zappa et al., 2007; Beaulieu et al., 2012). Prevailing winds in the lower Amazon are from the NE, which is the opposite direction of river outflow from Almeirim and Macapá (during an outgoing tide when net discharge is positive).

In large rivers and estuaries, the simple parametrization of k as a function of wind speed tends to be site specific due to local climatological, geomorphological, and hydrological characteristics, implying substantial errors in flux estimates using generic wind based functions from different systems (Borges et al., 2004a; Zappa et al., 2007). Rasera et al. (2013) used more detailed estimates of k₆₀₀, taking into account spatial variability including the difference between k₆₀₀ for small (< 100 m wide) and large (>100 m) rivers in the Amazon River, which nearly doubled estimates of basin-scale CO₂ emission from the Amazon basin estimated by Richey et al. (2002).

The lower Amazon River contains many of the features that would lead to very high gas transfer velocities and overall emissions. It is characterized by extremely wide channel(s) that flow in a convergent direction with the Intertropical Convergence Zone (ITCZ) creating a large fetch where the stronger winds can substantially increase turbulence, resulting in higher k₆₀₀ as observed in this study. Not only are k-values high, but the lower river also has a very high amount of wetted surface area for CO₂ to escape from, particularly in wide channels near the mouth.

Here we considered two different areas for upscaling annual CO₂ emissions from the lower Amazon River. Our most conservative estimate, including the area from Óbidos to Macapá (Area 1), was 20 Tg C year⁻¹. Area 2, which extends to the geographic terminus of the river, is a relatively short distance compared to Area 1, but covers 58% more surface area than the Óbidos to Macapá reach (Table 3). Applying the average FCO₂ observed near Macapá, we estimate Area 2 to emit 28 Tg C year⁻¹, which combined with the upstream section totalize 48 Tg C year⁻¹. Emissions from the Xingu and Tapajós tributaries contribute an additional 2.3 Tg C year⁻¹, resulting in a total flux of 51 Tg C year⁻¹ from the lower Amazon River basin. This estimate for the lower Amazon River, alone, is roughly 53% in magnitude compared to CO₂ emissions from all rivers in the conterminous United States (97 Tg C year⁻¹; Butman and Raymond, 2011) and ~12% of past estimates of basin-scale emissions from the Amazon (0.47 Pg C year⁻¹; Richey et al., 2002).

Adding our estimations of the fluxes estimated for Area 1 and the sum of Area 1 and 2 to estimations by Richey et al. (2002) increases basin-scale CO₂ outgassing to 0.49 and 0.52 Pg C year⁻¹, respectively. A recent re-evaluation of basin-wide outgassing estimates upstream of Óbidos was done using a combination of direct flux measurements and more detailed k-values calculations along with observations by Alin et al. (2011) for tributaries and streams. It was estimated that basin scale CO₂ outgassing upstream from Óbidos was roughly 0.8 Pg C year⁻¹ (Rasera et al., 2013). First order streams add an additional 0.1 Pg C year⁻¹ to basin scale CO₂ fluxes in the Amazon basin (Johnson et al., 2008). Adding our estimates for the lower river to these values results in basin-wide budgets of 0.92 Pg C year⁻¹ and 0.95 Pg C year⁻¹ for Area 1 and the sum of Area 1 and 2, respectively (Table 4). Replacing these new estimates for the Amazon in the global CO₂ budget by Raymond et al. (2013) increases the global budget by as much as 18% for a total of 2.48 Pg C year⁻¹.

In the case of this updated global budget, the Amazon River represents 38% of global CO₂ emissions. However, the contribution of other tropical rivers to the global budget are likely also underestimated considering that most tropical rivers are even less well-characterized than the Amazon, particularly in the lower reaches, where we've demonstrated that emissions can be high relative to the rest of the basin. Furthermore, we have not included the entirety of the Amazon in our newest budgets. For example, the Tapajós and Xingu rivers were not included due to their large spatial expanse and minimal data coverage. The lower portion of these tributaries, alone (Figure 1), emit roughly 2.3 Tg C year⁻¹, and these estimates do not encompass their entire surface area nor potentially elevated CO₂ concentrations closer to their headwaters.

Another factor that can lead to an underestimation of basin-wide budgets is not including Amazon River water that travels further offshore from Area 2 and along the coastline. For example, the Amazon River can remain unmixed with the ocean as far as 60 km offshore from Area 2 (Molinas et al., 2014). Abril et al. (2014) estimated that only 18% of the CO₂ from a point source would be degassed in a stretch of approximately 150 km downstream in the Amazon River taking into account a k-value of 15 cm h⁻¹ and water current of 150 cm s⁻¹. Thus, it is reasonable to assume that the mouth of the Amazon is the last point source of CO₂ to the Amazon plume, sustaining significant emissions for a significant distance offshore. This region, along with near-shore coastal waters, is not included in any studies of CO₂ cycling in the Amazon River plume in the Atlantic Ocean due to a lack of sampling and terrestrial contamination of remote sensing products by adjacency effect near-shore (Cooley et al., 2007; Subramaniam et al., 2008). We estimate that Area 3 may emit up to an additional 31 Tg C year⁻¹, but note that this is a simple calculation based on measurements at Macapá. Further exploration of this offshore area is essential for constraining the total basin-wide CO₂ flux. Although it is too early to confidently incorporate this offshore freshwater region (Area 3) into basin-wide budgets, this rough estimation highlights that expansive regions of offshore freshwater plumes may be an important missing gap in aquatic carbon budgets.

Numerous studies in the Amazon River have demonstrated its importance on global carbon budgets (Richey et al., 2002; Johnson et al., 2008; Rasera et al., 2013; Scofield et al., 2016). However, a missing gap in the continuum that connects the river network to the ocean has been neglected. Here we show that the lower reaches of the Amazon River are an active area in terms of freshwater CO₂ emissions from the Amazon, and perhaps the world, although, lower river reaches have yet to be adequately studied in other comparable tropical systems. A mixture of vast floodplains, tidally-flooded forests, and the surrounding watershed supplies the lower river with both CO₂ (Abril et al., 2014) and organic carbon, which is broken down extensively by heterotrophic bacteria (Ward et al., 2013, 2016). The downstream expansion of channel width and the alignment of a long fetch opposing the prevailing winds, which also increase toward the mouth, leads to high CO₂ emissions attributed to higher k-values in the lower river.

The enormous surface area of the lower Amazon River (18,379 km²) is slightly less than half of the area of rivers and stream in the conterminous United States (Butman and Raymond, 2011), and similarly emits roughly half as much CO₂ to the atmosphere. This area alone releases an amount of CO₂ in the same order of magnitude than the uptake by the Amazon River plume in the Atlantic Ocean (Kortzinger, 2003; Cooley et al., 2007; Subramaniam et al., 2008). We argue that the region where freshwater extends offshore should be included in basin-scale budgets considering that studies in the plume do not cover this area, however, there is no available data in this region to date. There is a pressing need to perform measurements of pCO₂ and FCO₂ along lower rivers and near-shore plume waters, which are currently large missing gaps in our coverage of regional and global scale carbon budgets.

JR, AK, HS, and NW responsible for the conception and design of the work. HS, VN, AV, WG, DL, and JD executed in situ measurements. HS, VN developed the data calculation and interpretation. MB conceived the areal estimations. JR, AK, HS, and NW organized overall project logistics. AC, DL, JD, and DB organized local logistics. All authors critically revised the manuscript and approved the final submission. All authors also agreed to be accountable for all aspects of the work related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

This study was supported by FAPESP Grants 12/51187-0, 2014/21564-2, and 2015/09187-1, and NSF DEB Grant #1256724, and NSF IGERT grant DGE-125848.