Partitioning carbon sources in a tropical watershed ( Nyong River , Cameroon ) between wetlands and terrestrial ecosystems – Do CO 2 emissions from tropical rivers offset the terrestrial carbon sink ?

We characterized the spatio-temporal dynamics of carbon (C) in rivers of the tropical Nyong catchment (South Cameroon). In 2016, we measured fortnightly at 6 locations along an upstream-downstream gradient from groundwater to the 20 main stream of order 6, total alkalinity, dissolved inorganic C (DIC) used together with pH to compute pCO2, dissolved and particulate organic C (DOC and POC) and total suspended matter. Forest, groundwater had low DOC content (<1 mg L-1) as its leaching was probably prevented in the overlaying lateritic soils. Forest groundwater was supersaturated in CO2 (~50 times the atmospheric value) because of the solubilisation of the CO2 originating from soil respiration. Wetlands water exhibited higher DOC (>14 mg L-1) and similar DIC concentrations than the forest groundwater. Surface runoff was considered 25 negligible in the basin due to low slopes and high infiltration capacity of the soils, making wetlands and forest groundwater the two main sources of C for surface waters. The influence of wetlands on C dynamics in rivers was significant during periods of high waters when the hydrological connectivity between surface waters and wetlands was enhanced. On annual scale, wetlands exported 60% (15.4±7.2 t C km-2 yr-1) of the total amount of C transferred laterally to surface waters, the remaining 40% (12.1±5.8 t C km-2 yr-1) being transferred from forest groundwater. Heterotrophic respiration in rivers averaged 89 mmol 30 m-2 d-1 whereas CO2 degassing was 1260 mmol m-2 d-1, which shows that it is unlikely that the river heterotrophic respiration was the main process sustaining CO2 emission. The comparison of the hydrological export of terrestrial C via forest groundwater with the net terrestrial C sink in the Nyong watershed shows that only ~4% of the net terrestrial C sink reach the aquatic ecosystem. The carbon mass balance of the Nyong watershed highlights that attributing to a unique terrestrial source https://doi.org/10.5194/bg-2021-69 Preprint. Discussion started: 7 April 2021 c © Author(s) 2021. CC BY 4.0 License.

emissions estimates from inland waters and the hydrological export of C from terrestrial well-drained systems could be actually attributed to carbon transfer from wetlands to inland waters.
There is now consensus that tropical regions are hotspots of riverine C export and degassing (Borges et al., 2015aSawakuchi et al., 2014). Indeed, recent estimates of CO2 emissions from tropical rivers (latitude < 25°) (Borges et al., 2015b) 70 based on results in several African rivers (Borges et al., 2015a) and in the Amazon (Abril et al., 2014) is about 1.8±0.4 Pg C yr −1 . Noteworthy, the estimate of Borges et al. (2015a) is conservative since it does not account for lakes and extensive tropical wetlands that act as hotspots of biological productivity, which could fuel C emissions to the atmosphere from rivers to which they are connected (e.g., Abril et al., 2014;Abril and Borges, 2019;Borges et al., 2015). C global emissions estimates from inland waters are thus probably underestimated, due to the scarcity of data in the tropics, where inland waters exhibit higher 75 emission rate per unit area compared to temperate ecosystems (Bastviken et al., 2011;Borges et al., 2015a).
The Nyong River basin (South Cameroon) has been subject to multidisciplinary hydro-biogeochemical monitoring since 1993.
This basin belongs to the Critical Zone Observatories' network named Multiscale TROPIcal CatchmentS (M-TROPICS; https://mtropics.obs-mip.fr/), a long-term monitoring program of hydrobiogeochemical cycles in the tropics. The main objectives of this study are the description of the spatial (from groundwater to increasing stream order) and temporal (through 80 the hydrological cycle) variations in C concentrations in the waters of the Nyong basin; and to decipher the impact of wetlands on these variations through the establishment of the riverine C budget. In the study, the estimated C fluxes are lateral hydrological inputs from land (i.e., from forest groundwater) and from wetlands to the river network, C degassing from the rivers and total C export to the ocean. The different C fluxes are estimated independently, allowing the quantitative determination of both C inputs from terrestrial ecosystems and from wetlands at the catchment scale. To the best of our 85 knowledge, our study is the first to estimate lateral hydrological export of C both from wetland and from well-drained terrestrial ecosystem (i.e., from forest groundwater) in a tropical catchment. In lines with recent studies in tropical rivers (Abril et al., 2014;Borges et al., 2015;, we expected that lateral inputs of C from wetlands to the river network are significant in comparison with C exported laterally from forest groundwater. We therefore hypothesized that the contribution of the net terrestrial C sink to the riverine carbon budget was minor. 90

Study site
The Nyong catchment (27800 km 2 , Cameroon) is located between 2.8 and 4.5 ° N and 9.9 and 13.5 ° E in the Southern Cameroon Plateau (Fig. 1). The main stem (Nyong River, stream order 6) is 690 km long and flows into the Atlantic Ocean ( Fig.1). The catchment lithology is mainly composed of granitic rocks with the absence of carbonate minerals (Maurizot et al., 95 1986;Toteu et al., 2001). Slopes and hills are recovered by a thick lateritic profile (20-40 m) poor in C, whereas in the depressions and the swamps the upper part of the hydromorphic soils shows an enrichment in organic matter (OM) (Boeglin et al., 2003;Nyeck et al., 1999). Most of the catchment area is covered by a dense permanent forest whereas semi-aquatic https://doi.org/10.5194/bg-2021-69 Preprint. Discussion started: 7 April 2021 c Author(s) 2021. CC BY 4.0 License. plants such as raffia or palm tress predominates in the swampy depressions. The Nyong catchment experiences an equatorial climate with four seasons of unequal importance with two maxima and minima: a short rainy season (Apr-June), a short dry 100 season (July-August), a long rainy season (Sept.-Nov) and a long dry season (Dec-March) (Suchel, 1987). In 2016, the average annual precipitation in the basin measured in the Mengong catchment was ~1610 mm (Fig. 2). In 2016, the hydrological cycle, based on the hydrograph, can be separated into three four-months periods: base (Jan-Apr), medium (May-Aug) and high (Sep-Dec) flows (Fig. 2).
In the Nyong catchment, 6 sampling sites were selected: 2 located on the main stem (Mbalmayo and Olama) and 4 in the So'o 105 sub-catchment (Mengong source, Mengong outlet, Awout and So 'o) (Fig. 1). The sampling sites cover a wide range of stream orders: from groundwater to order 6 (  (Fig. 2) and export or riverine C at this site is considered as representative of the C epxorted to the 110 Atlantic Ocean by the Nyong basin because the contribution of the tributaries downstream from this station is negligible (Nkoue-ndondo, 2008).
The Mengong catchment is considered representative of the tropical lateritic environments of the South Cameroon plateau (Braun et al., 2012). Therefore, hydro-biogeochemical processes occuring in the Mengong catchment are considered as representative of hydro-biogeochmical processes ocurring in the other first-order basins of the Nyong catchment. Noteworthy, 115 at the Mengong source, groundwater seeps out from the hillside of the Mengong catchment and eventually converge to the stream (i.e., Mengong stream). Then, the Mengong flows donwstream trough a swamp located in the lowland part of the Mengong basin to eventually reach the Mengong outlet 850 m downstream of the seepage point (Maréchal et al., 2011). In the lowland part of the Mengong catchment, the Mengong stream is indeed adjacent to the swamp whose extent is on average through the year ~20% of the total surface area of the catchment (Braun et al., 2005). Accordingly, water at the Mengong 120 outlet is fed by two different sources: the forest groundwater that seeps from the hillside of the Mengong catchment and by the swamp located in the lowland part of the basin (Maréchal et al., 2011). In the Mengong catchment, surface runoff on the slopes is negligible due to the high infiltration capacity of the soils and the relatively low slope (Maréchal et al., 2011). At the Nyong catchment scale, surface runoff is also of small magnitude and might not contribute significantly to the input of terrestrial C to surface waters (Nkoue-ndondo, 2008). The geographical and hydrological characteristics of the different sub-catchments are 125 shown in Table 1.

Sampling and laboratory work
Water in the Nyong, So'o and Awout Rivers was collected from bridges with the Niskin Bottle (3L). At the Mengong source, samples were taken directly from the source where the groundwater seeps through an inert PVC pipe. At the Mengong outlet, because of the shallow depth, the samples were cautiously taken directly in the stream. Water samples were collected from inorganic C (DIC), TA, dissolved organic (DOC) and particulate organic (POC) C and total suspended matter (TSM). At each samplig site, we measured the physico-chemical parameters of the water (temperature, pH, dissolved oxygen as O2, and specific conductivity). In addition, we carried out 14 measurements of pelagic respiration at two sampling sites (Mengong outlet and Nyong at Mbalmayo, which are first-and fifth-order streams, respetively). 135 The water temperature, pH, dissolved oxygen and specific conductivity were measured in situ in 2015 and between January and March 2016 using portable probes (WTW®). Calibration of sensors was carried out prior to the sampling campaigns and regularly checked during the campaigns. The conductivity cell was calibrated with a 1000 µS cm −1 (25°C) standard. The oxygen optical probe was calibrated with humidity saturated ambient air. The pH probe was calibrated using NBS buffer solutions (4 and 7). From April 2016, a multi-parameter probe (YSI ProDSS) was used for in situ measurements of the same 140 physico-chemical parameters. The multi-parmater probe was calibrated before each sampling campaign using standard protocols (YSI Proplus).
For TSM and POC, a first filtration (0.5-1.5 L) was carried out on pre-weighed and pre-combusted GF/F glass fiber filters (porosity of 0.7 µm). The filters were dried at 60 °C and stored in the dark at room temperature for subsequent analysis. TSM was determined by gravimetry with a Sartorius scale (precision was ±0.1 mg). The filters were acidified in crucibles with 2N 145 HCl to remove carbonates and were then dried at 60 °C to remove inorganic C and the remaining acid and water and then analyzed by the Rock Eval pyrolysis method (Lafargue et al., 1998). For DOC analysis, a portion of the POC filtrate was kept in glass bottles (60 mL) previously pyrolyzed in which 3 drops of H3PO4 (85%) were added to convert all the DIC into CO2.
The glass bottles were sealed with septas made of polytetrafluoroethylene (PTFE). The samples were stored at 3-5°C before analysis. DOC concentrations were measured by thermal oxidation after a DIC removal step with a SHIMADZU TOC 500 150 analyser in TOC-IC mode (Sharp, 1993). The repeatability was better than 0.1 mg L -1 .
We stored TA samples in polypropylene bottles after filtration using a syringe equipped acetate cellulose filters (porosity of 0.22 µm). TA was analyzed on filtered samples by automated electro-titration (Titrino Metrohm) on 50 mL-samples with 0.1N HCl as the titrant. The equivalence point was determined from pH between 4 and 3 with the Gran method (Gran, 1952).
The DIC samples were collected using 70 mL glass serum bottles sealed with a rubber stopper and treated with 0.3 mL of 155 HgCl2 at 20 g L -1 to avoid microbial respiration during storage. Vials were carefully sealed such that no air remained in contact with samples and were stored in the dark to prevent photo-oxidation. The DIC was measured with the headspace technique.
The headspace was created with 15 mL of N2 gas, and 100 µL of phosphoric acid (H3PO4, 85%) was added in the serum bottles in order to convert all the DIC species to CO2. After overnight equilibration at constant temperature, a subsample of the headspace (1 mL) was injected with a gastight syringe into a gas chromatograph equipped with a flame ionization detector 160 (SRI 8610C GC-FID). The gas chromatograph was calibrated with CO2 standards of 400, 1000 and 3000 ppm (Air Liquide® France).
Six extra 70 mL serum bottles were collected similarly as DIC and then were used for the determination of pelagic respiration.
Three serum bottles were directly poisoned in the field with 0.3 mL of HgCl2. The three other serum bottles were incubated in a cool box during 24 hours. The cool box was protected from light and filled with water from the river to maintain inside the cool box a water temperature similar to the one in the river. At the end of the incubations, the serum bottles were poisoned with 0.3 mL of HgCl2 and stored in the dark and at room temperature. To estimate pelagic respiration, we measured the increase in CO2 in the incubated serum bottles compared to those poisoned in the field. CO2 was measured similarly as DIC, using a headspace technique but without acidification.
In the surface waters of the Nyong catchment, we estimated the pCO2 from DIC, pH and water temperature measurements 170 using the carbonic acid dissociation constants of Millero (1979) and the CO2 solubility from Weiss (1974); using the CO2SYS software (Lewis et al., 1998).

Determination of catchments surface area, water surface area, slope and gas transfer velocity (k600)
The sub-catchments surface areas were estimated from hydrological modeling tools available in ArcGIS10.5® and the digital elevation model (DEM, 15 sec resolution) conditioned for hydrology (HydroSHEDS; Lehner et al., 2008). Wetlands surface 175 areas were estimated from Gumbricht et al. (2017). In the Nyong catchment, the HydroSHEDS flowline dataset (15 sec resolution) enabled the precise determination of the total length of each stream order (1 to 6). We combined the river flow measurements from the five gauging stations located on stream-orders 1, 3, 4, 5 and 6 (Table 1) and the hydraulic equation described by Raymond et al. (2012) in order to estimate the average river width (W) for all rivers of a given stream orders: where Qmean is the mean river flow in 2016 in the stream orders 1, 3, 4, 5 or 6.
Within a basin, river length, width, and surface area scale exponentially with stream order for all river orders (Strahler, 1957).
Since we did not measure river flow in stream order 2, the average width of stream order 2 was extrapolated from the best exponential regression curve obtained from the relationship between stream order and average river width in the catchment.
We then used the average river width (estimated from hydraulic equation by Raymond et al. 2012) and the total length per 185 stream order (estimated from HydroSHEDS flowline dataset) to calculate the water surface area per stream order. We fused the HydroSHEDS DEM and flowline datasets to assign an altitude to each river point and thus to determine the average slope (S) per stream order. To calculate the average flow velocity (V) per stream order, we used the following hydraulic equation: where Qmean is the mean river flow in 2016 in the stream orders 1, 3, 4, 5 or 6. The average flow velocity in stream order 2 was 190 extrapolated similarly as the the average river width (see above). The gas transfer velocity normalized to a Schmidt number of 600 (k600 in m d -1 ) was derived from the parameterization as a function of S (unitless) and V (m s -1 ) as in the Eq. 5 by Raymond The CO2 degassing at the water-air interface (Fdegas) was estimated as follows: 200 where K0 is the solubility coefficient of CO2 determined from the water temperature (Weiss, 1974), k600 is the gas transfer velocity of CO2 (see section 2.4), pCO2w and pCO2a are the partial pressures of CO2 in the surface water and in the overlying atmosphere (set to 400 ppmv), respectively.
We estimated Fdegas (mmol m -2 d -1 ) in each stream order of the Nyong basin (1 to 6) and we multiplied Fdegas by the respective 205 surface area of each stream order to estimate the total CO2 emission (Gg C yr -1 ) per stream order. We summed the total CO2 emission in each stream order to estimate the total quantity of CO2 degassed in the basin by the entire river network. We normalized the latter flux by the surface area of the Nyong catchment to estimate an average degassing weighed by the surface area of the basin (t C km -2 yr -1 ). Note that we did not measured pCO2 in second-order streams. However, we estimated the average pCO2 in second order streams by averaging the pCO2s observed in the first and third order streams. 210

Lateral inputs of C from land and wetland
In the Nyong basin, we considered that lateral inputs of C in the river network originates from two sources: forest groundwater (i.e., terrestrial C) and wetlands (i.e., semi-aquatic C). In our study, these two lateral inputs were estimated from data acquired in the Mengong catchment. Maréchal et al. (2011) estimated that the recharge rate of the hillside of the Mengong catchment was 332 mm yr -1 (between 1994-1998). The recharge rate of the hillside comprises two hydrological fluxes: one is hillside 215 groundwater outflow and one is baseflow from the hillside to the swamp (Maréchal et al. 2011). The hillside drains a surface area of 0.48 km², thereby the average annual outflow of the terrestrial groundwater (QGW) is ~0.005 m 3 s -1 , from which an hydrological lateral export of C from forest groundwater to surface waters (FGW) can be estimated as the following: GWr is the average annual concentration of DIC or DOC in the Mengong source. FexGW (t C yr -1 ) is normalized by 220 the surface area drained by the hillside (t C km -2 yr -1 ).
In the Mengong catchment, Maréchal et al. (2011)  is normalized by the surface area drained by the swamp (t C km -2 an -1 ). DOC and DIC concentrations in the swamp were estimated from DIC and DOC measured by Nkoue-ndondo et al. (2020) in the soil solution of the swamp at 0.4 meter depth.
In the Mengong catchment, surface runoff is negligigble and there is no particulate C in forest groundwater, therefore, the POC 230 at the Mengong outlet should originates mostly from the drainage of the swamp (which is adjacent to the stream). Accordingly, we considered the lateral hydrological export of POC at the outlet of the Mengong catchment (i.e., Mengong outlet) representative of the POC exported laterally from the swamp (FSWPOC). FSWPOC can be thus estimated as the following: where Qoutlet is the average annual river flow at the Mengong outlet and [POC]outlet is the average annual concentration of POC 235 at the Mengong outlet. FSWPOC (t C yr -1 ) is normalized by the surface area drained by the swamp (t C km -2 an -1 ).

Internal metabolism in the river network
Frespi (mmol m -3 d -1 ) is the average pelagic respiration obtained from the increase in CO2 in the incubated serum bottles over 24h. Frespi was converted in mmol m -2 d -1 by multiplying by the average river depth at the corresponding sampling site.
Noteworthy, Frespi does not represent total respiration in the river since it does not include benthic respiration. A mean benthic 240 respiration measured in various tropical rivers of 21 mmol m -2 d -1 (Cardoso et al., 2014) was therefore added to estimate total Frespi which is used throughout the manuscript.

C export to the ocean
The annual average C exported to the ocean (Focean) was calculated at Olama as the following: where Qolama and [C]olama are the annual average flow and concentrations of POC, DIC or DOC at Olama, respectively. Focean was estimated in Gg C yr -1 and then normalized by the catchment surface area at Olama (t C km -2 year -1 ).

C budget in the waters of Nyong basin
The sum of the C transferred by forest groundwater (FGW) and by wetland (Fsw) represents the total C inputs (Cinputs) transferred laterally to the river network, thus: 250 Cinputs = FSW (DOC, DIC, POC) + FGW (DOC, DIC) The C entering the surface network has two outputs (Coutputs) as this C is degassed at the water-air interafce (Fdegas) or exported to the ocean (Focean), thus: The difference between Cinputs and Coutputs represents the C budget (Cbudget) in inland waters of the Nyong basin, as the following: 255 Cbudget = Cinputs -Coutputs Noteworthy, external Cinputs assumed to be provided by both forest groundwater and wetland should sustain total Frespi. Therefore, total Frespi is not included in the estimation of Cbudget.

Statistical procedures
To assess statistical differences between stream orders (upstream-downstream) or between seasons (related to base, medium 260 and high flows), we used the ordinary one way-ANOVA with Tukey's multiple comparisons test for data normally distributed whereas we used Kruskall-Wallis test with Dunn's multiple comparison test for data that were not normally distributed. For both statistical tests, p>0.05 indicated that samples did not differ significantly, while p<0.05 indicated a statistical difference and p<0.001 an even higher confidence level of statistical difference. To assess correlations between a given parameter and river flow we used a Pearson correlation test for data normally distributed and Spearman correlation test for data that were not 265 normally distributed. For these correlation tests, p>0.05 indicated that samples were samples were not correlated, while p<0.05 indicated a correlation and p<0.001 an even higher confidence level of correlation. Normality was tested with the Agostino & Pearson normality test. All tests were made with Golden Software Prism (version 7).

Seasonal and spatial variations of physico-chemical parameters 270
In forest groundwater and surface waters of the Nyong Basin, water temperature ranged between 21.9 and 29.0 °C, pH between 4.6 and 6.8, O2 between 12 and 81% and specific conductivity between 5.1 and 86.3 µS cm -1 (Table 2; Fig. 3). Minimum values of water temperature and specific conductivity were observed in the Mengong outlet (21.9 °C and 13.1 µS cm -1 , respectively, stream order 1) whereas maximum values were observed in the Nyong River at Mbalmayo (29.0 °C and 86.3 µS cm -1 , stream order 5) (Table 2; Fig. 3). For oxygen saturation, both minimum and maximum values were observed in the Nyong 275 River at Mbalmayo (13-81%) (Table 2; Fig. 3). For pH, minimum values were observed in the Mengong source (4.6, groundwater) whereas maximum values were observed in the Nyong River at Mbalmayo (6.9) (Table 2; Fig. 3).
Spatially, on average throughout the year, water temperature, pH and specific conductivity increase from upstream (i.e., groundwater) to downstream (i.e., order 6) (Table 2; Fig. 3). Along an upstream-downstream gradient, annual-averaged oxygen saturation peaked significantly (p<0.001) in the So'o River (57%, order 4) (Table 2; Fig. 3). In the groundwater and stream 280 orders 1, 3 ,5 and 6 oxygen saturations were significantly lower than in the So'o River but similar (p>0.05) between each other's (Table 2; Fig. 3). In general, temporal variation are related to river flows pH, oxygen saturation and specific conductivity were inversely related to the river flow in the groundwater and all stream orders (Fig. 3). During base flow conditions, values of water temperature increased significantly (p<0.05) in forest groundwater and most of the stream orders ( Fig. 3). Specific conductivity was also significantly higher (p<0.05) during base flow conditions in all stream orders, except 285 in groundwater (Fig. 3).

Seasonal and spatial variations of particulate compounds (i.e., TSM, POC% and POC)
In surface waters of the Nyong Basin, TSM ranged between 1.8 and 27.5 mg L -1 , POC content in the TSM (i.e., POC%) between 9 and 29% and POC between 14 and 360 µmol L -1 (Tables 2-3; Fig. 4). On average throughout the year, TSM was significantly (p<0.05) lower in in the Mengong outlet (stream order 1) than in stream orders 3, 4, 5 and 6 (Table 2; Fig. 4). 290 POC% was significantly (p<0.05) higher in stream order 1 than in stream orders 3, 4, 5 and 6 (Table 3; Fig. 4). In general, on average throughout the year, we observed from upstream (stream order 1) to downstream (stream order 6) an increase in TSM and a concomitant decrease in POC% (Tables 2-3; Fig. 4). In contrast, on average throughout the year, POC was similar (p>0.05) within all stream orders (Table 3; Fig. 4). Irrespective of the stream orders, TSM, POC% and POC were in general higher (sometimes significantly) during medium and high flows than base flow (Fig. 4). Besides, when all data are plotted 295 together, TSM and POC are both strongly correlated (p<0.001) with river flow (data not shown).

Carbon mass balance of the Nyong River
At the basin scale, 12.1±5.8 and 15.5±7.2 t C km -2 yr -1 are drained by forest groundwater and wetlands, respectively (Fig. 6).
POC, DOC and DIC represents 5, 30 and 65%, respectively, of the C exported laterally to surface waters by wetlands (Fig. 6). 320 DOC and DIC represents 3 and 97%, respectively, of the C exported laterally to surface waters by forest groundwater (Fig. 6).
At the most downstream station, 7.2±0.5 t C km -2 yr -1 are exported to the Atlantic Ocean, in which 25, 69 and 6% are in the DIC, DOC and POC forms, respectively (Fig. 6). Total inputs from wetlands and terrestrial ecosystems and total outputs (emissions to the atmosphere and export to the ocean) are fairly balanced as the difference between Coutputs and Cinputs was 16%, and Coutputs and Cinputs were not statistically different from each other's. Given the high temporal variability in streams and 325 rivers, this shows that our methodology was fairly robust.

Dynamics and origin of C in forest groundwater
Considering the granitic lithology (i.e., absence of carbonate minerals) of the Nyong basin (Maurizot et al., 1986), TA in forest groundwater originates from the weathering of silicate minerals as dissolved CO2 can react with silicate minerals to produce 330 bicarbonates (Meybeck, 1987). In forest groundwater, the low TA concentrations and the absence of significant seasonal variations of TA were likely related to low weathering rates in the lateritic soil cover, which exhibits a relatively inert mineralogy (Braun et al., 2005(Braun et al., , 2012. Based on TA concentrations (53±26 µmol L -1 ) in forest groundwater and seepage of the groundwater of ~0.005 m 3 s -1 , we estimated a weathering rate in forest groundwater and overlaying lateritic soil of 0.2±0.1 t C km -2 yr -1 . The low weathering rates are mainly attributed to the thickness of the lateritic cover on hills and slopes, which 335 considerably slows the advance of percolating water in the bedrock (Boeglin et al., 2005). This was confirmed by the low mineral dissolved load in the aquifer of the Nyong basin (Braun et al., 2002) and by the dissolved silica fluxes in rivers that were significantly lower compared to the annual rainfall (e.g., White and Blum, 1995). Accordingly, in the Nyong basin, forest groundwater is therefore a limited source of TA for surface waters, as reported in other monolithic catchments draining only silicate rocks (Meybeck, 1987). 340 The dissolved CO2 in groundwater originates from the solubilization of soil CO2 that originates itself either from soil OM respiration in the unsaturated soil (Raich and Schlesinger, 1992), in the saturated soil or in the groundwater itself, using DOC that has been leached from the soil (Deirmendjian et al., 2018). In the Mengong source, groundwater pCO2 and O2 were not correlated (p>0.05, data not shown) and groundwater was free of DOC (Table 2; Braun et al., 2012;Brunet et al., 2009). In the lateritic soil overlaying the Mengong source, iron oxides represent up to 50% of the average volumetric weight of the soil 345 (Braun et al., 2005). Thus, the low concentrations of dissolved iron (Fe 2+ ) observed in the Mengong source indicated that despite the oxidizing environment above the groundwater, the transfer of Fe 2+ from the soil to the aquifer is not allowed (Braun et al., 2005). OM in the upper organic-rich horizons of the soil is probably well complexed with iron oxides and prevents DOC leaching to groundwater (Deirmendjian et al., 2018;Sauer et al., 2007). Based on this information, we hypothesize that groundwater pCO2 did not originate from respiration in the groundwater itself but originated mainly from soil respiration 350 occurring in the overlaying unsaturated soil (Fig. 6). On average throughout the year, groundwater pCO2 was ~50 times higher than the atmospheric value showing that forest groundwater is a major source of CO2 for the river network of the Nyong basin (Table 3).
In tropical forests, seasonal variations of soil temperature are relatively low throughout the year, and therefore, the soil water content is the main parameter controlling the intensity of soil respiration (Adachi et al., 2006;Hashimoto et al., 2004). During 355 base flow, precipitation was low compared to the two other hydrological periods (Fig. 2), likely decreasing soil moisture and thus soil respiration.. During the following periods of medium and high flows, precipitation increased (Fig. 2), which in turn had likely a positive effect on soil respiration. In addition, during medium and high flows periods the groundwater level has risen closer to surface organic-rich horizons where soil respiration is more intense (Amundson et al., 1998;Maréchal et al., 2011). During high flow conditions, the percolation of rainwater through the soil pores may also facilitated the transport and 360 the dissolution of soil CO2 to the underlying groundwater. Altogether, this may explain the lower groundwater pCO2 and the higher groundwater O2 observed during base flow conditions (Figs. 3, 5).

Dynamics and origins of C in the Mengong catchment (first-order stream)
There are two main types of waters in the Mengong catchment (and in the Nyong catchment): clear waters and colored waters (Boeglin et al., 2005). Clear waters exhibited very low DOC concentrations (<0.1 mg L -1 ) and come from springs (i.e., forest 365 groundwater) that seeps out from hills and slopes, whereas colored waters, constituting the river waters, exhibited higher DOC concentrations (>14 mg L -1 ) and originated from surface waters located in depressions (i.e., from swamp located in the lowland part of the catchment) (Boeglin et al., 2005;Ndam Ngoupayou, 1997). When clear waters flow through swamps, the chemistry of clear waters is significantly affected downstream. Accordingly, significant increases in DOC and POC concentrations were observed between the Mengong source (i.e., clear water) and its outlet (i.e., colored water) because the Mengong stream has 370 flowed through the swamp which occupies 20% of the surface of the basin ( Table 2). The Mengong source is free of both DOC and POC (i.e., below the detection limit) and is oxygenated throughout the year (Table 2). Therefore, POC and DOC at the Mengong outlet originates almost exclusively from the drainage and erosion of the swamp, as observed by Nkoue-Ndondo et al. (2020) based on the stable isotopic composition of DOC and POC in the Mengong basin. Tropical wetlands such as swamps are recognized as productivity hotspots (Borges et al., 2015b;Saunois et al., 2019) and a large fraction (excluding wood) of 375 their biomass is degraded in situ by heterotrophic respiration in the water and sediment which enrich swamp water in CO2 ( Fig. 6; Abril and Borges, 2019). High biological productivity in wetlands also explains significantly higher POC% observed in the Mengong outlet in comparison with the other stream orders, which exhibited lower wetland surface area in their basin, diluting POC% (Table 3; Fig. 4). In addition, wetland vegetation can actively transport oxygen to the root zone via their aeremchyma (Haase and Rätsch, 2010). This creates a complex oxic-anoxic interface that promotes aerobic respiration, but 380 also supplies labile OM to anaerobic degradation and methanogenesis fuelling CO2 and CH4 production and other fermentative organic compounds that are transferred to waters and pore waters (Piedade et al., 2010). Nkoue-Ndondo et al. (2020) measured on average through the year, high concentrations of DIC and DOC in the swamp of the Mengong catchment in the range of 150-3270 and 140-1510 µmol L -1 for DIC and DOC, respectively. Noteworthy, we observed significant higher DOC concentrations during base flow at the Mengong outlet (Fig. 5), which might be explained by longer water residence time in 385 the swamp during this period.
In comparison to the Mengong source, TA concentrations significantly increased (~two times) in the Mengong outlet after the stream has flowed through the swamp (Table 3; Fig. 5). In addition, based on TA concentrations (90±36 µmol L -1 ) and average river flow at the Mengong outlet, we estimated a weathering rate of 0.5±0.2 t C km -2 yr -1 , which is 150% higher than weathering rate in forest groundwater. This weathering rate was comparable to the previous estimate by Boeglin et al. (2005) for the entire 390 Nyong catchment of 0.8-1.1 t C km -2 yr -1 , but was three times lower than those previously estimated for other lateritic basins (see Boeglin et al., 2005). This indicates that swamps in the Nyong river basin are a limited but significant source of TA for surface waters of the watershed. Under reducing conditions usually occurring in swamps, ferro-reducing bacteria can use iron oxides to oxidized OM and this process produces TA (ΔTA = + 2.2 moles of TA produced for 1 mole of iron oxide consumed; Abril and Frankignoulle, 2001). The contribution of Fe(III) reduction to anaerobic carbon metabolism in a given environment 395 is mainly proportional to the availability of labile OM and reducible Fe(III) (Roden and Wetzel, 2002). In the Mengong catchment, these two substrates are abundant in the hydromorphic soils of the swamp (Braun et al., 2005). In the swamp of the Mengong basin, Braun et al. (2005) observed a vertical decrease in Fe(III) in the soil profiles, confirming that the swamp behaves as hotspot of Fe(III) reduction and TA production (Liesack et al., 2000). In addition, weathering rates in the swamp area of the Mengong catchment is enhanced by the leaching of humic acids from the vegetation to the hydromorphic soils 400 (Braun et al., 2005;Nkoue-Ndondo, 2008).

C variations at the scale of the Nyong catchment
At the scale of the Nyong basin, TA increased along an upstream-downstream gradient and TA was inversely related with river flow (Fig. 5). During base flow period, surface waters are fed by deeper groundwater level, which are older and likely exhibited higher TA concentrations than shallower levels. In addition, we suppose that during high flow TA is diluted by forest 405 groundwater and swamp waters with lower TA concentrations.
In contrast to temperate rivers, the pCO2 in the surface waters of the Nyong basin peaked during periods of high waters (Fig.   5). This seasonality can be explained by the fact that soil moisture increases during the rainy season, enhancing the respiration in the soil surface organic-rich horizon. Concomitantly, period of high waters probably increases the hydraulic gradient between forest groundwater and surface waters, increasing the flushing of forest groundwater when groundwater pCO2 is the 410 highest (Fig. 5). In addition, the rise of the water level in the watershed enhances of the connection of surface waters with wetlands during periods of high waters. Therefore, the CO2 resulting from the degradation of the wetland biomass can be transferred to the river main stem. These two processes explain why usually pCO2 is positively correlated with river flow in tropical rivers, (e.g., Borges et al., 2019;Bouillon et al., 2012). In the surface waters of the Nyong catchment, the supersaturation in CO2 in comparison to the overlying atmosphere and the 415 decrease of pCO2 from groundwater to stream order 6 showed a gradual degassing of CO2 at the water-air interface along an upstream-downstream gradient (Table 3; Fig. 5). In acidic waters, CO2 degassing at the water-air interface is related with an the increase of the δ 13 C of the DIC (Deirmendjian and Abril, 2018). Such a pattern was observed by Brunet et al. (2009) in the waters of the Nyong basin confirming the upstream-downstream degassing. It confirms that headwaters are a major source of CO2 for the river system. Groundwaters are obviously a significant source of CO2 for the headwaters. However, in tropical 420 rivers like the Nyong which do not have floodplains, most of the wetlands are actually located in low order streams and therefore fuels small tributaries with CO2 which is mostly degassed further downstream in the main stem.
We also observed an increase in POC, in the surface waters during periods of high waters, concomitant to the increase in pCO2 . Similarly to what we observed in the Mengong catchment, wetlands can also be considered as the main source of POC for surface waters in the entire Nyong basin based on (1) the low slopes in the catchment, (2) the high infiltration capacity 425 of the soil and (3) the probable low pelagic primary production in the surface waters. During the high-water stage, dissolved and particulate C from wetlands are flushed to surface waters, as frequently observed in tropical catchments and confirmed by the increase in POC% in stream orders 4, 5 and 6 during periods of high water stages (Fig. 5) which reflect the high biological productivity of these ecosystems. Indeed, wetlands exhibited higher POC% compared to rivers, as indicated by the highest values of POC% observed in the Mengong outlet, the latter being strongly influenced by wetland (Fig. 5). In addition, the 430 normalized export of POC from wetland to rivers and from the watershed to the ocean is similar (Fig. 6). This confirms that wetlands are the main source of POC for surface waters, in particular during periods of high waters, and that degradation of POC in the riverine water is not significant. At the scale of the Nyong basin, the increase in connectivity between wetlands and surface waters during periods of high waters is also in agreement with a significant drop in O2 saturation during these periods because the flooding of OM-rich wetlands enhance respiration (Fig. 3). We also observed a remobilization of mineral 435 matter from the riverbanks and beds during high waters periods, as evidenced by higher TSM in the Nyong River during these hydrological periods (Fig. 4). When all data are plotted, POC% and TSM were negatively correlated together (p<0.001), showing a dilution of the POC with mineral matter from upstream to downstream.
In surface waters, by contrast to pCO2 and POC, we did not observe a positive correlation between DOC and the river flow, in agreement with Brunet et al. (2009) who showed that DOC was only flushed during a short period of time at the beginning of 440 the rainy season (in September and in April). In the Awout River (order 3), a significant increase in DOC was observed during medium flow period indicating an additional source of DOC (Table 2; Fig. 4). The Awout River was dry during the previous base flow period: its riverbed was completely invested by large macrophytes (up to 2 m tall) and many small pockets of stagnating water remained (visual observations). During base flow period, DOC could accumulate in these stagnating waters and then it is remobilized when the water flows again, as observed in temperate rivers (Deirmendjian et al., 2019;Sanders et 445 al., 2007). As we do not see a general increase of DOC in base flow period in the whole watershed we believe that these local vegetation regrowth in river beds and riverbanks have a minor impact of the whole carbon balance in the Nyong watershed.

C budget in the Nyong basin
We measured each fluxes of the riverine C budget independently and we show that 60% of the carbon entering the riverine system originates from wetlands, which makes them the main source of C in the river, the remaining 40% being transferred by 450 forest groundwater. Therefore, our results are in lines with the growing consensus that tropical wetlands contribute significantly to the riverine C budget in the tropics. (Abril and Borges 2019;Abril et al., 2014;Borges et al., 2015a-b,) A major portion (~80%) of the C entering surface waters returns to the atmosphere trough degassing at the water-air interface.
In the Nyong catchment, in situ respiration (pelagic plus benthic) was on average 89 mmol m -2 d -1 whereas the average degassing was 1260 mmol m -2 d -1 (Table. 4), which implies that only ~7% of the degassing at the water-air interface was 455 supported by in situ respiration. This is coherent with previous observations by Borges et al. (2015a) and (2019) in 12 African watersheds. Accordingly, our observations and measurements in the Nyong basin suggest the same conclusion: net heterotrophy in the river sensu stricto plays a minor role in the riverine CO2 budget. Therefore, we can hypothesize that most of the CO2 emitted by the rivers is actually produced in the wetlands and soils/groundwater and transferred to rivers where it escapes to the atmosphere. 460 In the Nyong catchment, the ratio between the C exported to the ocean and emitted to the atmosphere is 1:0.3, in agreement with ratio 1:0.2 measured by Borges et al., (2015a) in the Congo river but contrasting with the global ratio of 1:1 estimated by Ciais et al. (2013) during the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC). It shows that at least African Rivers but probably all tropical rivers are strong emitters of C and therefore biogeochemical data in these rivers are urgently required to improve accuracy of regional and global C emission estimates from inland waters, and 465 understand how they will respond to climate change (warming, change in hydrological cycle).
The integration of these fluxes at the scale of the whole watershed can be done by comparing this carbon budget with the terrestrial carbon budget. Brunet et al. (2009) estimated the net terrestrial C sink of the mature forest of the Nyong basin at approximately 300 t C km -2 yr -1 . Thus, the hydrological export of terrestrial C (via forest groundwater) represents ~4% of the terrestrial C sink. This conclusion is in agreement with the two local studies in temperate rivers, which have shown that the 470 hydrological export of C from forest ecosystems is ~3% (Deirmendjian et al., 2018b;Kindler et al., 2011). Altogether, this study not only confirms the importance of the C inputs from wetlands for fuelling the riverine C cycle and most importantly CO2 emissions to the atmosphere but it shows that high emissions from inland waters are compatible with a strong terrestrial C sink.

Conclusions
The current paradigm to explain CO2 emissions from inland waters is based on the prevalence of heterotrophy sustained by terrestrial OM inputs from the terrestrial ecosystems to the aquatic systems. However, in the present study, we showed that degassing of riverine C at the water-air interface cannot be supported by river heterotrophy, as river heterotrophy represents ~7% of the average degassing. Thus, we highlighted the importance of lateral inputs of C from soils and groundwater on the 480 one side (40% of the total C inputs), and from wetlands (60%) on the other side, to understand C dynamics in tropical rivers.
Terrestrial groundwater was a significant source of C for surface waters, particularly for CO2, whereas in contrast, DOC and POC in surface waters were mainly provided by the drainage and erosion of wetlands. The C exported laterally from wetlands to surface waters occurs mainly during periods of high waters when the connectivity with surface waters is enhanced. We showed here by determining all the terms of the carbon mass balance independently that attributing the whole amount of carbon 485 emitted to the atmosphere and exported to the ocean to a terrestrial source and ignoring the river-wetland connectivity can lead to the misrepresentation of C dynamics in tropical watersheds.

Acknowledgments
The      Table 4: C degassing at the water-air interface and pelagic respiration. Fluxes in mmol m -2 d -1 are related to the water surface area. a taking into account an additional benthic respiration of 21 mmol m -2 d -1 . b calculated from the sum of C degassing (Gg C yr -1 ) in each stream order. c in t C km -2 yr -1 , which fluxes are weighed by the surface area of the entire Nyong basin.