Divergent biophysical controls of aquatic CO2 and CH4 in the World’s two largest rivers

Carbon emissions to the atmosphere from inland waters are globally significant and mainly occur at tropical latitudes. However, processes controlling the intensity of CO2 and CH4 emissions from tropical inland waters remain poorly understood. Here, we report a data-set of concurrent measurements of the partial pressure of CO2 (pCO2) and dissolved CH4 concentrations in the Amazon (n = 136) and the Congo (n = 280) Rivers. The pCO2 values in the Amazon mainstem were significantly higher than in the Congo, contrasting with CH4 concentrations that were higher in the Congo than in the Amazon. Large-scale patterns in pCO2 across different lowland tropical basins can be apprehended with a relatively simple statistical model related to the extent of wetlands within the basin, showing that, in addition to non-flooded vegetation, wetlands also contribute to CO2 in river channels. On the other hand, dynamics of dissolved CH4 in river channels are less straightforward to predict, and are related to the way hydrology modulates the connectivity between wetlands and river channels.

The C emissions from inland waters result from complex interactions between hydrology, biogeochemical processing within the aquatic environment and connectivity with riparian zones and the watershed. The CO 2 emissions from inland waters have been traditionally interpreted as mainly resulting from the in-situ degradation of organic C from non-flooded land (that is, terra firme) [7][8][9][10][11][12][13][14][15] . Yet, other sources of CO 2 could also contribute to CO 2 emissions from inland waters. In lakes, there is an increasing recognition of the role of hydrological inputs of CO 2 (rivers and groundwaters) in sustaining CO 2 emissions to the atmosphere [16][17][18][19][20] . In rivers, the contribution of groundwater inputs of CO 2 to riverine CO 2 emissions is also recognized as particularly important in headwaters 21,22 . There is also an increasing recognition of the inputs of C from wetlands in sustaining CO 2 and CH 4 emissions to the atmosphere from rivers and lakes. Wetlands contribute to CO 2 emissions through the respiration from flooded roots of vegetation and by providing labile organic C to sustain bacterial degradation 23,24 . In the Central Amazon basin, CO 2 and CH 4 emissions from floodplain lakes 23,25 and from river channels 24,26 have been attributed to C from wetlands (flooded forest and macrophytes) in addition to non-flooded terrestrial organic C. This was established with a mass balance approach of organic C 23,26 , high-resolution pCO 2 distributions 24 , and stable-isotope signatures of organic C. In African rivers, spatial patterns of pCO 2 and CH 4 relate to the distribution of the fraction of wetland in the catchment within a given system (Congo and Zambezi) and across different basins 27,28 . However, both non-flooded terrestrial biomass and wetlands contribute to CO 2 emissions from inland waters and their relative importance remains uncertain and has not yet been quantitatively resolved 27,29 . This is in part due to the absence of specific molecular tracers for terrestrial organic matter, since numerous plants are common in flooded and non-flooded forests 30 . On the other hand, stable isotopes allow to trace organic matter from floating macrophytes that frequently have a C 4 signature 31 , while non-flooded C 4 grasslands have been found to contribute little to organic matter transported by rivers even in catchments where they occupy extensive areas 32 . The relative contribution of flooded and non-flooded biomes to riverine CO 2 emissions will vary from one basin to another as a function of climate 27 . It will also vary within a given basin with a dominance of non-flooded terrestrial inputs in headwaters and highlands and an increased contribution of wetlands in lowlands 24,27,29,31 . In the Amazon basin, wetlands have been conclusively shown to be hotspots of CH 4 emission compared to river channels 25,33,34 .
Here, we compare the CO 2 and CH 4 distributions in lowland river channels of the two largest rivers in the World and in the tropics, the Amazon and the Congo (Table 1), using a data-set of concurrent pCO 2 and CH 4 concentration measurements in river channels (Fig. 1, Table 2). We acknowledge that there are several other data-sets of pCO 2 and CH 4 in Amazonian aquatic systems 29 but we focus on direct measurements of pCO 2 (not calculated from pH and TA that are highly biased in acid waters 35 ) concurrent with dissolved CH 4 measurements (most other studies are based on either one dissolved gas or the other, but not both). The aim of this study is to determine the extent to which the patterns of CO 2 and CH 4 differ or converge in these two tropical giant water bodies.

Results
The pCO 2 values spanned two orders of magnitude in the Amazon (70 to 16,880 ppm) and one order of magnitude in the Congo (1090 to 22,900 ppm) (Fig. 2a). The CH 4 concentrations spanned four orders Precipitation (mm) 64 2,147 1,527 Air temperature (°C) 64 24.6 23.7 River-stream surface area (km 2 ) 1 74,904 26,517 Wetland surface area (%) 11,58 14 10 Above ground biomass (Mg km −2 ) 65 909 748 Land cover 60,61 Dense Forest (%) 83 49 Mosaic Forest (%) 4 18 Woodland and shrubland (%) 4 27 Grassland (%) 5 3 Cropland/Bare soil (%) 4 2 of magnitude in the Amazon (11 to 189,100 nmol L −1 ) and three orders of magnitude in the Congo (22 to 71,430 nmol L −1 ) (Fig. 2b). Data were aggregated into mainstem (MS), large and small tributaries (T > 100 m and T < 100 m width, respectively 11,36 (Fig. 2a,b) were distinctly above atmospheric equilibrium of ~390 ppm and ~2 nmol L −1 , respectively. The pCO 2 in the Amazon mainstem was significantly higher than in the Congo mainstem, but pCO 2 values were not significantly different in large and small tributaries (Fig. 2a). The CH 4 in the mainstem, large and small tributaries were significantly higher in the Congo than in the Amazon (Fig. 2b). The median CH 4 in the Congo was three to four times higher than in the Amazon, for mainstem/small tributaries and large tributaries, respectively. For a given pCO 2 value, CH 4 concentrations were systematically higher in the Congo than in the Amazon (Fig. 3a-c).

Discussion
The contribution of wetlands to CO 2 emissions in the Amazon, Congo and across tropical rivers. The pattern of higher pCO 2 values in streams compared to rivers in the Amazon and the Congo (Fig. 2) is consistent with an analysis of global averages 36 and also with the regional studies in part of the Congolese "Cuvette Centrale" 37 and in the Oubangui sub-catchment 38 . Higher CH 4 and CO 2 concentrations in tributaries than in the mainstem were also reported in the Paraguay River 39 . The , and a zoom overlain on the main rivers (d,e). Maps were generated with ArcGIS using publically available spatial datasets 60,61 . MS = mainstem. T > 100m = large tributaries. T < 100m = small tributaries.
higher pCO 2 in the mainstem of the Amazon than in the Congo in their lowland regions could be due to the higher wetland coverage (Table 1), since organic and inorganic C from wetlands has been shown to partly sustain the CO 2 emission from the Central Amazon mainstem and floodplains 24,26 . In order to expand the range of wetland coverage, we included pCO 2 data acquired in four other African rivers 27 (Fig. 4). In the small and large tributaries and mainstem, pCO 2 was positively correlated to wetland coverage across these six tropical rivers, confirming the contribution of wetland C in partly sustaining CO 2 emissions from lowland tropical river channels 24,26,27 . These positive correlations between pCO 2 and wetland coverage do not necessarily imply that wetlands are the sole drivers of CO 2 in river channels. As previously noted, semi-arid rivers such as the Tana that are virtually devoid of wetlands are CO 2 sources to the atmosphere, although less intense than other tropical rivers, implying that non-flooded land also    sustains CO 2 emissions from river channels 27 . The relative importance of non-flooded land and wetlands in sustaining riverine CO 2 emissions remains uncertain and has not yet been quantitatively resolved 29 .

Several hypotheses can explain the different behavior of CH 4 in the Amazon and Congo river channels.
Although in African rivers average CH 4 concentrations correlate with wetland coverage 27 , CH 4 concentrations were significantly higher in the Congo than in the Amazon river channels (Fig. 2), despite the fact that the Amazon has a higher wetland coverage (Table 1). Further, the correlations of CH 4 and pCO 2 are different in the Amazon and Congo river channels (Fig 3). In small streams (T < 100 m), the strong positive relationship between CH 4 and pCO 2 in both rivers indicates a common origin. It might indicate a stronger contribution of CO 2 production from anaerobic organic matter degradation compared to aerobic respiration, and that both CO 2 and CH 4 production are related to C processing within wetlands. Small streams receive higher contributions from groundwater that are rich in CO 2 21,22 . However, data in African rivers show that groundwater had an extremely low CH 4 content 27,40 . While groundwater input certainly contributes to high CO 2 in small streams it cannot explain the extremely high CH 4 in small streams. Consequently, the strong correlation between pCO 2 and CH 4 in small streams (Fig. 3b) indicates that groundwater inputs are probably not the major drivers of the high pCO 2 values at our sampling sites in lowland regions. In the mainstem, CH 4 is only weakly positively correlated to pCO 2 in the Congo, while a weak negative relation is observed in the Amazon. This might indicate that in the well mixed and well oxygenated Amazon mainstem, there is a stronger contribution to CO 2 production of aerobic respiration fueled by both non-flooded and wetland organic matter 41 , while CH 4 is lost by emission to the atmosphere and bacterial oxidation. In large tributaries (T > 100 m) an intermediary situation is observed in the Amazon, while in the Congo, CH 4 and pCO 2 remain strongly correlated. These fundamental differences in the dynamics of CH 4 in these two rivers can be further examined by invoking several hypotheses.
First, the Congo flooded wetland is in majority flooded forest 42 and there are no temporary floodplain lakes but only a handful of relatively large permanent lakes (Mai-Ndombe (2,300 km 2 ), Tumba (765 km 2 )). In the Central Amazon, on the other hand, flooded forest accounts for 80% of the maximum flooded wetland extent, and the remaining 20% corresponds to temporary and permanent lakes (7% of open water and 13% of floating macrophytes). There are 6,500 floodplain lakes from 52.5°W to 70.5°W along the floodplain fringing the Amazon mainstem plus 2,300 lakes on the major tributaries, totaling a surface area of 10,400 km 2 43 . Floodplain lakes are abundant downstream of the confluence of the Negro and Solimões Rivers, while upstream wetland is dominated by flooded forest. Floodplain lakes are characterized by high gas transfer velocity (k) values 44,45 , that promote the evasion of CH 4 to the atmosphere and water oxygenation that will favor bacterial CH 4 oxidation. In the Congolese and Amazonian flooded forest, k values should be low due to wind shielding and moderate diurnal water and air temperature variations below the dense canopy, and the release by the flooded plants of hydrophobic organic matter, which might behave as surfactants. This limits CH 4 loss by evasion to the atmosphere and by bacterial oxidation (low oxygen levels).
Second, local upland runoff is the main source of the wetland water in the Congo, and not flooding by riverine overflow as in the Amazon 46 . This unidirectional flow pattern will promote the transport of the CH 4 produced in the flooded forest towards the small and large river channels of the Congo, unlike in the Central Amazon where during rising water and high water, the water transport is from the river channels towards the wetlands. It is during rising water and high water that floating macrophytes grow and their biomass peaks 47 . This corresponds to the period of highest CH 4 emissions 33 , and presumably also highest CH 4 production, when the water transfer from wetlands to the river channels is blocked by flooding. The same applies to flooded forest where CH 4 emissions were also found to be highest during high water 33 .
Third, the Congo wetlands are mostly permanently flooded unlike the Amazon floodplains that are seasonally flooded. Permanently flooded wetlands are known to be stronger CH 4 emitters and presumably CH 4 producers than seasonal flooded wetlands 48,49 .
Fourth, in the Congo, floating macrophytes (mainly Vossia cuspidata) commonly occur along channel edges and within channels, and form large meadows in streams, rivers and mainstem, in all types of waters (white and black). Floating macrophytes are known to host high CH 4 production and emission 25,33,34 that will be directly delivered into the Congo river channels. This does not occur in the Amazon where macrophytes are mainly present in floodplain lakes and do not occur in large tributaries and the mainstem due to important depth and strong currents. This is consistent with the higher CH 4 concentrations in the Congo than in the Amazon mainstem for pCO 2 values > 7000 ppm (Fig. 3a). The CH 4 released by floating macrophytes in the Amazonian wetland lakes will be lost locally by evasion to the atmosphere and CH 4 oxidation (see above), and little dissolved CH 4 will be transported to the river channels.
All these differences are related to the smaller water height variations in the Congo mainstem (3-4 m) compared to the Amazon (10-12 m). The Congo basin straddles on the equator, and the dry season on the Northern part of the basin is compensated by the rainy season on the Southern part of the basin, and vice-versa, leading to a regulation of seasonal water height variations 50 . These different hypotheses need to be tested and verified although this would require a detailed investigation of the hydrology and wetland habitat mapping that are lacking in the Congo where research on aquatic biogeochemistry and ecology was largely abandoned since the early 1960's compared to the Amazon that has been the subject of continued investigations for more than five decades.

Re-evaluation of CO 2 emissions from tropical rivers and streams.
The total CO 2 emission from river and streams estimated by Raymond et al. 1 of 1.8 PgC yr −1 is mostly related to tropical areas that account for 1.4 PgC yr −1 (78%). However, the data coverage in the tropics was lower than for temperate and boreal regions, and data in several basins (including the Congo) were derived from interpolation from adjacent basins rather actual measurements. Furthermore, only one value of pCO 2 was used for the whole watershed while pCO 2 values increase in lower order streams as shown here (Fig. 2) and across the United States 15 . For African rivers we have previously shown that the Raymond et al. 1 dataset underestimated CO 2 fluxes in five basins where new direct pCO 2 measurements were recently made 27 . Although based on a limited number of river basins, we used the regressions in Fig. 4 as a first attempt to re-evaluate CO 2 emissions from tropical rivers and streams globally. The river basins shown in Fig. 4 cover a large range of size, climate, and land and wetland cover typical of those encountered in tropical areas. The resulting flux for the tropics is 1.8 ± 0.4 PgC yr −1 , i.e. 25% higher than the value originally computed by Raymond et al. 1 . While additional data will be required to further refine global estimates, this exercise confirms the importance of CO 2 emissions from rivers in tropical areas.
In conclusion, the analysis of data in river channels in six tropical rivers including the two largest ones (Amazon and Congo) reported here demonstrates that large-scale patterns in pCO 2 across different basins can be apprehended with a relatively simple statistical model related to the extent of wetlands within the basin. Dynamics of dissolved CH 4 in river channels are less straightforward to predict, and appear to be related to the way hydrology modulates the connectivity between wetlands and river channels. The differences we have highlighted in CH 4 concentration in the river channels of the Amazon and Congo should translate into same differences in CH 4 emissions, since in river channels the diffusive CH 4 emission is much higher than CH 4 ebullition flux in both rivers 27,51 . This is not the case in wetlands where ebullition represents the majority of the CH 4 emission to the atmosphere 25,52 . In the Amazon basin, overall aquatic CH 4 emissions are dominated by wetlands 25 , while equivalent estimates are unavailable for the Congo basin.

Methods
Study site characteristics. The Amazon and Congo are the first and second largest rivers in the World, respectively, in terms of catchment area and freshwater discharge ( Table 1). The Amazon basin is on average ~1 °C warmer and has an annual precipitation about two times higher than in the Congo. This leads to a specific discharge that is also much higher in the Amazon than in the Congo. The higher precipitation can also explain the higher coverage of the basin by evergreen forest (dense and mosaic) in the Amazon (87%) than in the Congo (67%), where conversely savannah (shrubland and grassland) is more abundant (30%), in particular in the northern and southern rims of the catchment (Fig. 1). Consequently, average above ground biomass is higher in the Amazon than in the Congo. The Amazon and Congo basins include the largest tropical wetlands in the World, with annual mean flooded area of 730,000 and 360,000 km 2 , respectively 25,42 . Field data collection. Data were acquired during 5 cruises in the Amazon and 6 cruises in the Congo covering different stages of the annual flood cycle ( Table 2). The pCO 2 in the Amazon was measured with an equilibrator 53 coupled to an infra-red gas analyzer (IRGA), as described in detail by Abril et al. 24 . The pCO 2 in the Congo was measured with both an equilibrator (in the mainstem and largest tributaries) and with a syringe headspace technique (in the mainstem and large and small tributaries) with an IRGA, as described in detail by Borges et al. 27 . Both approaches were inter-calibrated and compared very well 35 . Only the data acquired with a syringe headspace technique in the Congo are presented here. Samples for the determination of CH 4 , were conditioned in 50 ml serum borosilicate vials, poisoned with a saturated solution of HgCl 2 (100 μ L) and sealed with gas tight butyl stoppers until analysis by gas chromatography (GC) 54 . The CH 4 partial pressure was measured in a 1 mL subsample of the headspace of 20 mL of N 2 that was allowed to equilibrate about 12h after initial vigorous shaking. The CH 4 concentrations in the Amazon were measured with a flame ionization detector (FID) with a Hewlett Packard 5890A GC calibrated with certified CH 4 :N 2 mixtures (Air Liquide France) of 10 ppm and 200 ppm CH 4 . The CH 4 concentrations in the Congo were measured with a SRI 8610C GC-FID calibrated with certified CH 4 :CO 2 :N 2 O:N 2 mixtures (Air Liquide Belgium) of 1, 10, 30 and 509 ppm CH 4 . The overall precision of measurements was ± 2% and ± 4% for pCO 2 and CH 4 , respectively. Additional data in the Amazon were digitalized with PlotDigitizer© from the plots of Richey et al. 55 . Data presented in Richey et al. 55 were obtained by headspace technique and GC analysis, from April 1982 to August 1985 during 9 cruises upstream of Manaus, while data reported in the present study were acquired downstream of Manaus.
Computation of tropical river CO 2 efflux and error propagation. The air-water CO 2 flux (F) was computed according to: where α is the CO 2 solubility coefficient, k is the gas transfer velocity and Δ pCO 2 is the pCO 2 air-water gradient, whereby a positive value corresponds by convention to an emission of CO 2 from the water to the atmosphere. We used the geographical information system (GIS) of Raymond et al. 1 . The GIS provides k values, surface areas and width for streams and rivers globally, and the data are structured by stream order into COSCATs (coastal segmentation and its related catchment 56 ). The k values themselves are derived from a parameterization as a function of slope and stream velocity 57 included in the GIS. For each of the COSCAT units we derived wetland cover from another GIS, the global database of lakes, reservoirs and wetlands 58 . Based on the wetland coverage and the equations of the regressions in Fig. 4, we computed the water pCO 2 in MS, T > 100m and T < 100m. Since river/stream surface areas in the GIS are structured by stream order it is not possible to distinguish the surface areas corresponding to MS and tributaries. So, the pCO 2 of MS and T > 100m computed from the regressions for each COSCAT were averaged, and computations were further carried for T < 100m and for MS and T > 100m lumped together. The F values were then computed from the k values derived from the GIS for streams/rivers narrower and wider than 100 m, a constant water temperature of 25 °C to compute α 59 and a constant atmospheric pCO 2 of 390 ppm. The F areal values per COSCAT were scaled to the respective stream/river surface area and the data between 30°N and 30°S were summed to provide a total flux value for tropical areas.
An error analysis on the CO 2 flux computation and upscaling was carried out by error propagation of the pCO 2 computation, the k value estimates, and the estimate of surface areas of river channels to scale the areal fluxes, using a Monte Carlo simulation with 1000 iterations. The uncertainty on the pCO 2 computation was derived from the errors on the slope and Y-intercept of the linear regressions in Fig. 4. The uncertainty on k values from the GIS was estimated to be ± 10.0% based on the errors on slope and constant of the parameterization 57 . The river/stream surface areas in the GIS were estimated using two different hydraulic equations, that allow to estimate an uncertainty of ± 31.0%. Statistical analysis. The statistical tests were done with GraphPad Prism ® Version 6.05 for Windows.
Original data-set. The timestamped and geo-referenced data-set of pCO 2 and CH 4 concentrations ( Table 2) are available as a supplementary table.