Aquatic and soil CO₂ emissions from forested wetlands of Congo's Cuvette Centrale

Within tropical forest ecosystems, wetlands such as swamp forests are an important interface between the terrestrial and aquatic landscape. Despite this assumed importance, there is a paucity of carbon flux data from wetlands in tropical Africa. Therefore, the magnitude and source of carbon dioxide (CO₂) fluxes, carbon isotopic ratios, and environmental conditions were measured for 3 years between 2019 and 2022 in a seasonally flooded forest and a perennially flooded forest in the Cuvette Centrale of the Congo Basin. The mean surface fluxes for the seasonally flooded site and the perennially flooded site were 2.36±0.51 and 4.38±0.64 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$, respectively. The time series data revealed no marked seasonal pattern in CO₂ fluxes. As for the environmental drivers, the fluxes at the seasonally flooded site exhibited a positive correlation with soil temperature and soil moisture. Additionally, the water level appeared to be a significant factor, demonstrating a quadratic relationship with the soil fluxes at the seasonally flooded site. δ¹³C values showed a progressive increase across the carbon pools, from aboveground biomass to leaf litter and then to soil organic carbon (SOC). However, there was no significant difference in δ¹³C enrichment between SOC and soil-respired CO₂. This lack of enrichment can be attributed to either a significant contribution from the autotrophic component of soil respiration or closed system dynamics.

An in-situ-derived gas transfer velocity (k₆₀₀=2.95 cm h⁻¹) was used to calculate the aquatic CO₂ fluxes at the perennially flooded site. Despite the low k₆₀₀, relatively high CO₂ surface fluxes were found due to very high partial pressure of CO₂ (pCO₂) values measured in the flooding waters. Overall, these results offer a quantification of the CO₂ fluxes from forested wetlands and provide insights into the temporal variability of these fluxes and their sensitivity to environmental drivers.

1 Introduction

Along with the oceans and Northern Hemisphere forests, tropical forests represent one of the three main components of the global carbon sink (Mitchard, 2018). However, due to relatively high gross primary productivity, temperature, and soil moisture, tropical forest soils also constitute a large terrestrial source of carbon dioxide (CO₂). Indeed, tropical regions are estimated to contribute up to 64 % of global soil respiration, rendering it the largest flux of CO₂ from terrestrial ecosystems to the atmosphere (Hashimoto et al., 2015; Huang et al., 2020).

Wetland cover in the tropical Congo Basin is estimated to range between 332 620 and 359 556 km² (Bwangoy et al., 2010; Fatras et al., 2021). This area includes the Cuvette Centrale, which spans approximately 167 600 km² and hosts lowland and swamp forests, including the largest peatland complex across the tropics (Crezee et al., 2022). With catchment drainage from north to south of the Equator, as well as sustained rainfall at the centre of the basin (Breitengroß, 1972; Runge, 2007), the Cuvette Centrale shows near-permanent inundation. Characterizing CO₂ fluxes in this extensive region is especially important since inland waters are increasingly recognized as significant sources of greenhouse gases (GHGs) within the terrestrial landscape (Bastviken et al., 2011; Drake et al., 2018; Borges et al., 2015b; Rosentreter et al., 2021) and notably in global carbon dioxide emissions (Raymond et al., 2013). Recent data additionally suggest that the Congo Basin's inland waters might emit more carbon (C) per area than their counterparts in the Amazon Basin (Alsdorf et al., 2016). Profound hydrological (Alsdorf et al., 2016), structural (Lewis et al., 2013), ecological (Parmentier et al., 2007; Slik et al., 2015), aquatic biogeochemical (Borges et al., 2015a), and terrestrial biogeochemical (Hubau et al., 2020) differences indicate that GHG flux estimates cannot simply be transferred from the Neotropics to the Afrotropics. However, while recent research on GHG emission from the Congo Basin has focused on either riverine (Borges et al., 2019; Bouillon et al., 2012; Mann et al., 2014; Upstill-Goddard et al., 2017) or terrestrial fluxes (Barthel et al., 2022; Baumgartner et al., 2020; Daelman et al., 2025; Gallarotti et al., 2021), direct measurements from forested wetlands are still lacking. Despite its immense global importance, only two studies, to the best of our knowledge, have looked into GHG emissions from Congo's wetlands (Tathy et al., 1992; Barthel et al., 2022).

Forested wetlands/swamp forests are located at the transition zone between the terrestrial and the aquatic realm. The duration and seasonality of flooding in the forests will constrain the contribution from and to the river system. While flooded, the swamp forests are connected to the river system and receive and/or discharge materials from and to the river network (Aufdenkampe et al., 2011). Variations in riverine greenhouse gas concentrations have been shown to be driven by fluvial–wetland connectivity for the Cuvette Centrale based on data from 10 expeditions across the Congo River network (Borges et al., 2019). Furthermore, streams and rivers draining Congo's flooded forests were found to have the highest dissolved concentrations of CO₂ among different land cover types in the basin, indicating the substantial contribution of forested wetlands to the overall inland water GHG budget (Mann et al., 2014).

Here, we report 3 years of carbon dioxide (CO₂) fluxes measured from two sites situated within the Cuvette Centrale: a seasonally flooded forest site and a perennially flooded forest site. During the observation period, surface CO₂ fluxes – whether soil or aquatic – were measured fortnightly to capture seasonal and interannual variation in the fluxes. Hence, these results provide insights into the temporal dynamics of CO₂ fluxes in forested wetlands across two different flooding regimes.

2 Materials and methods

2.1 Study sites

The study sites are located near the city of Mbandaka (Democratic Republic of the Congo, Équateur Province), which is located at the Ruki–Congo confluence within the Cuvette Centrale (Fig. 2). The mean annual precipitation and mean annual temperature of the sampling area are 1588 mm and 25 °C, respectively (see the measurements detailed below). The long dry season in Mbandaka typically lasts from July to August, while the short dry season occurs between January and February. Here, surface CO₂ fluxes were measured at two different sites across two different hydrological regimes, one in a seasonally flooded forest (0.06335° N, 18.31054° E; 300 m a.s.l.) – referred to as the SFF site – and the other in a perennially flooded forest (0.03135° S, 18.3102° E; 305 m a.s.l.) – referred to as the PFF site (Fig. 1).

Figure 1 Diagram showing the location of the two experimental sites (PFF and SFF) relative to the hydrological gradient.

The SFF site investigated is located within a botanical garden 7 km from the centre of Mbandaka (Jardin Botanique d'Eala, operated by the Institut Congolais pour la Conservation de la Nature (ICCN)). The botanical garden comprises 371 ha of land consisting of 35 % dense swamp forest, 14 % forest on firm ground, and 32 % open forest, with the remaining area consisting of secondary forest, grassland, and deforested land, of which 189 ha is protected forest area. There are 3500 different trees and herbaceous plant species, with the main tree species being Hevea brasiliensis, Ouratea arnoldiana, Pentaclethra eetveldeana, Strombosia tetrandra, and Daniella pynaertii. The soil at the site, covered by a thick litter layer, is characterized as Eutric Gleysols (texture: $\mathrm{42}/\mathrm{50}/\mathrm{8}$, sand $/$ silt $/$ clay in percent; bulk density: 1.27 g cm⁻³). The litter layer harbours a dense mesh of fine roots, whereas almost no roots are found to penetrate the upper mineral soil layer (0–30 cm). The SFF site is seasonally flooded from about December to January (∼2 months).

At the SFF site, combined soil moisture and temperature sensors (ECH₂O 5TM, Meter Group, Inc. USA) connected to loggers (Em50, Meter Group, Inc., USA) were installed at 10 and 30 cm depths, respectively. The data were recorded every 6 h. Unfortunately, one logger was stolen and the other logger stopped working during deployment; thus, data are only available from November 2019 to July 2020 (Fig. 3). Afterward, TMS-4 data loggers (TOMST, Czechia) were installed in December 2020 to record surface volumetric soil water content (0–14 cm) and soil temperature at 8 cm depth in 15 min intervals. Raw data (soil moisture count) retrieved from TMS-4 data loggers were converted into volumetric soil water content with calibration curves, following Wild et al. (2019), using site-specific soil properties (soil texture: $\mathrm{42}/\mathrm{50}/\mathrm{8}$, sand $/$ silt $/$ clay in percent; bulk density: 1.27 g cm⁻³) and measured soil temperatures. The volumetric soil water content values from the ECH₂O 5TM sensors showed a systematic offset compared to those obtained from the TMS-4 data loggers. This was attributed to instrument artefact and corrected by using the difference between maximum values. Furthermore, precipitation, air temperature, relative humidity, solar radiation, and wind speed data were retrieved for the observation period from the Trans-African Hydro-Meteorological Observatory (TAHMO) monitoring station located in close vicinity to the forest site (ATMOS 41, Meter Group, Inc., USA).

The PFF site is located about 8 km upstream of the Congo–Ruki confluence, following a small side tributary named Lolifa. The headwater stream area is completely flooded for most of the year, making the streambed channel indistinguishable. This creates a continuous wetland area where the PFF site is located. While the water is mostly stagnant at the site, a small drainage flow appears during the dry season (late June to early September). The site was accessed with a motorized dugout canoe, and sampling was done fortnightly from the side of the canoe. The main tree species at the PFF site are Uapaca sp., Irvingia smithii, and Daniella pynaertii De Wild. In addition to the surface CO₂ fluxes, water samples were collected on the same day to measure pH, dissolved organic carbon (DOC), and total dissolved nitrogen (TDN). The presented C:N ratio was thus calculated using TDN rather than dissolved organic nitrogen (DON). Previous analyses (Drake et al., 2023) have shown that TDN consistently comprised an average of 90 % of DON and thus reflected the relative changes in DON concentrations well. The specific methods used for sample processing and analysis, as well as the calculations, are described in Drake et al. (2023).

Figure 2 Map presenting the two sampling sites in the vicinity of Mbandaka (Democratic Republic of the Congo). The boundaries of the Jardin Botanique d'Eala are highlighted. Map data: © 2020–2023 Impact Observatory Inc. and GADM.

2.2 CO₂ fluxes

2.2.1 CO₂ surface fluxes at the SFF site

A total of six polyvinyl chloride soil flux chambers (height: 0.3 m, diameter: 0.3 m) were installed in November 2019 at the SFF site. The SFF site was chosen as representative of the surrounding forest. The six chambers were spaced about 20 m apart, randomly distributed across the site but accounting for variations in local microtopography. The chamber bases were inserted approximately 5 cm into the ground and remained in place throughout the measurement period, with the chambers left open except during sampling. Chambers affected by seasonal flooding were measured until they were completely submerged, at which point floating chambers (V=17 L) were used instead. Sampling with the soil chambers was conducted fortnightly for 3 consecutive years (November 2019–December 2022), totalling 403 flux measurements. Sampling was interrupted once for about 6 months due to logistical constraints (first half of 2022).

Each chamber lid was equipped with a thermocouple to measure headspace temperature, a vent tube to avoid pressure changes, and a sampling port. The sampling port had a three-way luer valve attached to it, connecting the syringe, needle, and chamber. Before withdrawing each gas sample from the headspace, chamber air was mixed by moving the syringe plunger several times; for soil GHG flux determination, gas samples were taken at time steps of 20 min throughout 1 h (t1=0 min, t2=20 min, t3=40 min, t4=60 min). A longer chamber closure time than recommended (Pavelka et al., 2018) was used to obtain robust Keeling plots along with the flux measurements. At each time step, 20 mL of gas sample was stored in 12 mL pre-evacuated vials (Labco, UK) using a gas-tight disposable plastic syringe (20 mL). The resulting vial overpressure prevents air ingress due to temperature and pressure changes potentially occurring during transport and is required for sample withdrawal by the gas chromatograph (GC) autosampler. To aid vacuum and sample preservation, each evacuated vial was sealed with an additional silicone layer (Dow Corning 734, Dow Silicones Corporation, USA). Soil CO₂ fluxes were calculated via linear concentration increase over time using the ideal gas law PV=nRT:

$$\begin{array}{}\text{(1)}& n={\displaystyle \frac{PV}{RT}}\end{array}$$

and

$$\begin{array}{}\text{(2)}& F={\displaystyle \frac{\mathrm{\Delta }n}{\mathrm{\Delta }t}}{S}^{-\mathrm{1}},\end{array}$$

where n is the moles of gas [mol], P is the partial pressure of trace gas [atm µmol mol⁻¹], R is the gas constant of 0.08206 [L atm K⁻¹ mol⁻¹], T is the headspace temperature [K], F is the flux of gas [$\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$], $\frac{\mathrm{\Delta }n}{\mathrm{\Delta }t}$ is the rate of change in concentration [mol s⁻¹], V is the chamber volume [L], and S is the surface area enclosed by a chamber [m²]. The coefficient of determination (r²) for the linear regression of CO₂ yielded r²>0.95 for 95 % of the data (Fig. S1). All data with r²>0.1 were kept for the statistical analyses. Such a low r² threshold was maintained because fluxes with low r² values are typically the result of low flux rates rather than due to methodological or technical issues. Increasing the threshold would introduce a bias toward higher fluxes in the data.

2.2.2 Aquatic surface fluxes at the PFF site

The aquatic surface flux to the atmosphere ($F_{{\mathrm{CO}}_{\mathrm{2}}}$, $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$) from the PFF site was estimated according to a simple gas transfer model (Mann et al., 2014):

$$\begin{array}{}\text{(3)}& F_{{\mathrm{CO}}_{\mathrm{2}}}=k_x\cdot K_{\mathrm{H}}\cdot \left({\mathrm{pCO}}_{\mathrm{2}\mathrm{w}}-{\mathrm{pCO}}_{\mathrm{2}\mathrm{a}}\right),\end{array}$$

where k_x is the freshwater gas transfer velocity of CO₂ [m s⁻¹]; K_H is Henry's constant for CO₂ [mol m⁻³ atm⁻¹]; and pCO_2w,a is the partial pressure of CO₂ in water and the atmosphere, respectively [µatm].

Since the magnitude of the gas transfer velocity is governed by numerous factors (e.g. wind speed, water current velocity, slope), an in situ gas transfer velocity k was calculated as 3.5 cm h⁻¹ using the aquatic fluxes from the SFF site sampled between July 2022 and December 2022 with the above-mentioned floating chamber (V=17 L) and the corresponding dissolved CO₂ concentrations of the inundation water at the same site. The value of 3.5 cm h⁻¹ was then applied to the perennially flooded forest dataset where no floating chamber measurements existed. Hence, fluxes from the PFF site were derived using the measured gas transfer velocity from the SFF site (3.5 cm h⁻¹).

In order to compare the in-situ-derived velocity k_x with the temperature-normalized transfer velocity (k₆₀₀) for tropical wetlands of 2.4 cm h⁻¹ (Aufdenkampe et al., 2011), we used the equation from Pelletier et al. (2014) to convert k_x to k₆₀₀:

$$\begin{array}{}\text{(4)}& k_x=k_{\mathrm{600}}{\left({\displaystyle \frac{\mathit{Sc}}{\mathrm{600}}}\right)}^{-b},\end{array}$$

where Sc is the gas-specific Schmid number, and b is derived from the literature (0.66 for wind speed ≤3 m s⁻¹; Pelletier et al., 2014).

The gas-specific Schmid number is a function of water temperature (T in °C), as defined by Wanninkhof (2014):

$$\begin{array}{}\text{(5)}& \begin{array}{rl}{\mathit{Sc}}_{{\mathrm{CO}}_{\mathrm{2}}}& =\mathrm{1923.6}-\mathrm{125.06}T+\mathrm{4.3773}{T}^{\mathrm{2}}\\ & -\mathrm{0.085681}{T}^{\mathrm{3}}+\mathrm{0.00070284}{T}^{\mathrm{4}}.\end{array}\end{array}$$

For pCO_2a, the tropospheric mean value from the year 2020 (400 µatm) was used, while pCO_2w was determined using the headspace equilibration technique. That is, 6 mL of a bubble-free water sample was injected with a syringe into a 12 mL N₂-pre-flushed vial (Exetainer^®, Labco, UK) pre-poisoned with 50 µL of 50 % ZnCl₂ to stop microbial activity. After sufficient equilibration time, the remaining headspace was analysed for CO₂ concentrations using a gas chromatograph (see section below), and total dissolved concentrations were calculated based on Henry's law (for a detailed method, see the Supplement).

For each date, pCO_2w samples were taken in triplicate with an average coefficient of variation (CV) of 8 %.

2.3 Gas chromatography

Gas samples were analysed at ETH Zurich using a gas chromatograph (GC; Bruker, 456-GC, Scion Instruments, Livingston, UK) separating CO₂ from residual air. After separation, the concentration of CO₂ was measured on a thermal conductivity detector. GC calibration was done with a suite of three standards (Carbagas AG, Switzerland; PanGas AG, Switzerland) across a concentration range from 249 to 3040 ppm CO₂. Each standard was analysed 10 times at the start, middle, and end of each set of 140–180 samples. Moreover, because of occasional high CO₂ sample concentrations, an entire system flush was done between each sample measurement to avoid any carry-over effects. The same GC setup was used for both flux samples and dissolved CO₂ samples.

2.4 δ¹³C of soil-derived CO₂ fluxes and dissolved CO₂

The carbon isotopic composition of the CO₂ samples was analysed for one SFF CO₂ flux sample set of each month. That is, after CO₂ concentration measurement with the GC, the same samples were analysed for δ¹³C of CO₂ with a modified Gasbench II periphery (Finnigan MAT, Bremen, Germany) coupled to an isotope ratio mass spectrometer (IRMS; Delta^plusXP, Finnigan MAT), as described in Baumgartner et al. (2020). Post-run offline calculation and drift correction for assigning the final δ¹³C values on the Vienna Pee Dee Belemnite (V-PDB) scale were done following the “IT principle” (Werner and Brand, 2001). The δ¹³C values of the laboratory air standards were determined at the Max Planck Institute for Biogeochemistry (Jena, Germany), according to Werner et al. (2001). The final soil CO₂ δ¹³C values were calculated using the Keeling plot approach (Keeling, 1958) (Fig. S2).

δ¹³C of dissolved riverine CO₂ was determined using the headspace equilibration technique, as described in Sect. 2.2.2. Instead of concentration, δ¹³C of the headspace was analysed via IRMS, as described above. Samples were taken each month from the Ruki between October 2022 and June 2023 with two to three replicates per sampling (Fig. S4).

2.5 δ¹³C of leaves, litter, and soils

Fresh leaf samples were taken from a range of the most representative tree species at two different time points (November 2019 and November 2023). In addition, litter samples were collected at the same time, and both were used to analyse the carbon isotopic composition (δ¹³C). Before analysis, samples were dried, homogenized, and ground. Soil samples were taken in November 2019, February 2020, and November 2023 at 0–30 cm depth and air dried, sieved, and milled. All samples were analysed using an elemental analyser (Flash EA 1112 Series, Thermo Italy, formerly CE Instruments, Rodano, Italy), interfaced with an IRMS (Finnigan MAT Delta^plus XP, Bremen, Germany) via a six-port valve (Brooks et al., 2003) and the ConFlo III (Werner et al., 1999). Soil samples are subsequently referred to as soil organic carbon (SOC) samples. Calibration of laboratory standards (acetanilide, caffeine, tyrosine) was done by comparison to the corresponding international reference materials provided by the IAEA (Vienna, Austria).

2.6 Water level

Direct measurements of the water level were not available for the whole observation period. Previous work has shown a linear relationship between the water level of the Congo River and the Ruki (unpublished, Fig. S3). Additionally, the rainfall and/or the hydrological dynamics of the river influence the water levels in the wetlands. In the Cuvette Centrale, Georgiou et al. (2023) determined that the water levels of riverine locations in the Democratic Republic of the Congo (DRC) correlate more with the hydrological dynamics of the river system than with the rainfall input. Hence, available daily measurements of the water level of the Congo River in Mbandaka were used as a proxy of the water level below- and aboveground at the SFF site (Fig. S6). These data were extracted from an almost-continuous record of water gauge readings, collected in the vicinity of the SFF site (∼4 km) by the Congolese public institution Régie des Voies Fluviales since 1913.

2.7 Statistical analyses

Daily environmental conditions were used to explain variability in the measured soil CO₂ fluxes (n=403) at the SFF site. For this, a linear mixed-effects model was fitted using soil temperature, volumetric soil water content, and river level as fixed effects. River level showed a non-linear relationship with surface fluxes. Hence, a quadratic term was added to account for the non-linear effect. The predictor variables were standardized before fitting the models. All models were controlled for repeated measurements in the same chambers by adding chamber ID as a random intercept. Models were fitted by the restricted maximum likelihood method using lme4 (Bates et al., 2015). Full and reduced models were compared using a likelihood ratio test and adjusted r² values using the MuMin package (Bartoń, 2023). Furthermore, a backward stepwise regression analysis was conducted on the full model, incorporating all effects and interaction terms, to identify the most cost-effective model with the highest explanatory power (Kuznetsova et al., 2017). The resulting model included an additional interaction term between soil moisture and river level. However, this term was subsequently removed due to multicollinearity and its lack of practical significance. Marginal and conditional r² values for mixed effects were calculated following Nakagawa et al. (2017), with inclusive r² estimated with the partR2 package (Stoffel et al., 2021) and p values using Satterthwaite's approximation with the lmerTest package (Kuznetsova et al., 2017). Additionally, confidence intervals for the effect estimates were computed to confirm the interpretation of the estimated parameters. The assumptions of the model were validated by verifying the linearity, normality, and homoscedasticity of the residuals. Multicollinearity between the predictor variables was also estimated (variance inflation factor (VIF) inferior to 3). Statistical differences between δ¹³C values measured across the different carbon pools were tested with the Kruskal–Wallis test, followed by a pairwise Wilcoxon comparison. Significance was established when the Bonferroni-adjusted p values were inferior to 0.05. Statistical and graphical data analyses were done in R v.4.3.2 (R Core Team, 2023) via RStudio v.2023.12.0 (Posit team, 2025), using the packages tidyverse v.2.0.0, tydr v1.3.0, dplyr v.1.1.4 (Wickham et al., 2023), ggplot2 v. 3.4.4 (Wickham, 2009), sjPlot (Lüdecke, 2013), and lubridate v.1.9.3 (Grolemund and Wickham, 2011). QGIS v.3.16 was used to compile the map of the sampling locations.

3 Results

3.1 Environmental conditions

The long dry season in Mbandaka is considered from July to August, whereas the short dry season spans January and February. However, frequent rainfall, as shown in Fig. 3, causes the region to be relatively wet throughout the entire year. Annual precipitation was the highest in 2020, with 1855 mm, and the lowest in 2022, with 1417 mm (self-measured; Fig. 3). The flooding period at the study site is typically centred around December and January. The highest weekly precipitation occurred in July and September of each year, with 120–182 mm (Fig. 3a). Overall, the weekly precipitation ranged from 0–182 mm, with a monthly average of 31 mm (Fig. 3a).

Volumetric soil water content, hereinafter referred to as soil moisture, averaged 0.60±0.09 m³ m⁻³, ranging between 0.35 and 0.76 m³ m⁻³ for the observation period (Fig. 3a). In general, soil moisture showed strong seasonality, with an increase observed in November and peak values seen in January. Thereafter, soil moisture decreased before stabilizing until the following wet season. This pattern was less pronounced over the 2021–2022 season (Fig. 3a).

Soil and air temperatures were stable throughout the observation period (Fig. 3b). During the observation period, the recorded mean air temperature at the weather station was 25.0 °C (±0.7 °C), and mean soil temperature at the SFF site was 24.7 °C (±0.3 °C).

3.2 Soil and aquatic CO₂ fluxes

Over the observation period, CO₂ fluxes from the PFF site were higher than those from the SFF site (Fig. 3f). At both sites, CO₂ fluxes exhibited intra-annual variability. However, distinct seasonal patterns were not clear. Notably, at the SFF site, the onset of flooding appeared to induce a decline in fluxes. Furthermore, among the environmental variables, CO₂ fluxes exhibited significant correlations with soil moisture, soil temperature, and river level (Table 1). At the PFF site, the highest fluxes were recorded in June and August of 2020, with 5.71 and 5.76 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$, whereas the lowest values were observed in September and October 2020, with 3.35 and 3.42 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$. Mean weekly surface fluxes ($F_{{\mathrm{CO}}_{\mathrm{2}}}$) from the PFF site ranged from 3.35 to 5.76 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ with an average flux of 4.38±0.64 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$, using the in-situ-derived gas transfer velocity of 3.5 cm h⁻¹ (Fig. 3e). Mean weekly surface fluxes ($F_{{\mathrm{CO}}_{\mathrm{2}}}$) from the SFF site ranged from 0.87 to 3.64 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ with an average of 2.36±0.51 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$. Here, the lowest flux was observed in July 2022 with 0.87 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$, a period corresponding to the lowest soil moisture recorded (0.35 m³ m⁻³), while the flux peaked in May 2020 with 3.64 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ (Fig. 3e).

Figure 3 Weekly precipitation, volumetric soil water content, temperature, and CO₂ fluxes. (a) The sum of the weekly precipitation [mm] (blue) obtained from the Trans-African Hydro-Meteorological Observatory and mean volumetric soil water content [m³ m⁻³] measured with soil moisture sensors (ECH₂O 5TM: solid line, TMS-4 data loggers: dotted line). (b) Distribution of volumetric soil water content [m³ m⁻³], with both sensor types combined. (c) Mean weekly air temperature [°C] (gold) was obtained from the Trans-African Hydro-Meteorological Observatory. The mean weekly soil temperature [°C] was measured with soil temperature sensors (ECH₂O 5TM: solid grey line, TMS-4 data loggers: dotted grey line). (d) Distribution of air and soil temperatures [°C], with both sensor types combined. (e) Measured surface CO₂ fluxes (cross) [$\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$] from the SFF site (brown) and calculated CO₂ fluxes from the PFF site with a K of 3.5 cm h⁻¹ (pink). The calculated weekly means (line) and the standard error of the mean are displayed. Blue shading represents river levels (see Sect. 2.6), while grey bands indicate flooding periods (December–January) at the SFF site. The displayed time series are discontinuous due to fieldwork constraints (see Sect. 2.2). (f) Distribution of surface CO₂ fluxes at the PFF and SFF sites.

3.2.1 Controls on surface CO₂ fluxes at the SFF site

The linear mixed-effects model (n=324) explained 43.0 % of the total variability, of which 35.4 % is allocated to fixed effects (river level, soil moisture, and soil temperature; Table 1). Soil temperature and soil moisture are positively correlated with surface CO₂ fluxes. The river level, used as a proxy for the water level, exhibited a quadratic relationship with the CO₂ fluxes measured at the SFF site (Table 1; Fig. 4c). The non-linear component exhibited a negative sign, describing an inverse U-shaped curve (Fig. 4c). Initially, the relationship had a positive slope at lower river levels, reaching a maximum point before transitioning to a negative slope. As the river level is used as a proxy for the on-site water level, a short-term campaign was conducted during the wet season (2023–2024) to confirm the influence of the water level with direct measurements (unpublished; Fig. S7). Finally, the significant positive interaction term between temperature and river level suggests a synergistic effect where the combined influence of these two variables on surface fluxes is greater than the addition of their respective individual effects (Fig. 4d).

Table 1 Fixed-effect estimates for surface CO₂ fluxes at the SFF site, including river level, soil temperature, and soil moisture as standardized predictors, allowing for comparison of their relative importance. For each effect, standard error and p values (Sattherhwaite's method) are estimated, as well as the marginal (m) and conditional (c) $R_{\mathrm{adfj}}^{\mathrm{2}}$ (Nakagawa et al., 2017).

^* Interaction term between soil temperature and river level.

For a deeper understanding of the LMER outputs (Table 1), the individual relationships between surface CO₂ fluxes and the different predictors (soil temperature, river level, and soil moisture), as well as the effect of the interaction between the soil temperature and river level, are visualized in Fig. 4. The inclusive r² (IR²) of each predictor is also presented, offering a measure of the proportion of variance explained by each predictor, including both its direct effects and its interactions with other predictors (Stoffel et al., 2021). In this context, the soil temperature (IR²=0.225), the soil moisture (IR²=0.126), and the quadratic component of the river level (IR²=0.097) appear to be the primary factors explaining the variance of surface CO₂ fluxes, whereas the interaction between soil temperature and river level (IR²<0.001), along with the linear component of the river level (IR²=0.001), makes no meaningful contribution (Fig. 4).

Figure 4 Individual relationships between soil CO₂ fluxes and environmental parameters (soil moisture (VWC) (a), soil temperature (b), and river level (c)). Measurements taken while the soil chamber was partially flooded are represented in blue. Regression lines are displayed as dashed black lines. The interaction between soil temperature and river level is illustrated. Values were predicted based on the LMER model (Table 1) (d). Inclusive r² values (a, b, c) were estimated based on the LMER model (Table 1; Stoffel et al., 2021).

3.2.2 Controls on surface CO₂ fluxes at the PFF site

At the PFF site, surface CO₂ fluxes did not exhibit statistically significant relationships with pH, the river level, the carbon to nitrogen ratio (C:N), dissolved organic carbon, and biodegradable dissolved organic carbon. Trends were observed, such as an increase in CO₂ fluxes with rising river levels and a decrease in CO₂ fluxes with increasing pH (Fig. S7). However, these are just visual tendencies and not statistically significant findings.

3.3 δ¹³C of leaves, litter, soils, soil CO₂ flux, and riverine dissolved CO₂

The measured δ¹³C values increased from leaves and litter to SOC and soil CO₂ fluxes and became more positive along this cascade of organic matter transformation (p values <0.05; Fig. 5). δ¹³C of leaves ranged from −37.1 ‰ to −28.9 ‰ with a mean of $-\mathrm{33.8}\pm \mathrm{2.1}$ ‰. The δ¹³C signature of litter was between −32.6 ‰ and −28.7 ‰ with an average of $-\mathrm{30.5}\pm \mathrm{1.0}$ ‰. SOC had δ¹³C values of −30.1 ‰ to −22.3 ‰, while the mean was $-\mathrm{27.4}\pm \mathrm{1.9}$ ‰. The δ¹³C of soil-derived CO₂ ($F_{{\mathrm{CO}}_{\mathrm{2}}}$) was in the range of SOC values for the SFF site (Fig. 5) and very stable throughout the measurement period (Fig. S4). Here, measured δ¹³C values were −30.2 ‰ to −26.5 ‰ with a mean of $-\mathrm{28.5}\pm \mathrm{0.8}$ ‰. In contrast, the carbon isotopic composition of CO₂ fluxes from the SFF site during flooding was strongly ¹³C enriched with −24.8 ‰ to −13.3 ‰ and an average of $-\mathrm{20.4}\pm \mathrm{3.4}$ ‰ (p values <0.01). The δ¹³C of the inundated soil CO₂ fluxes was higher throughout the whole measurement period (Fig. S4). The δ¹³C value of dissolved CO₂ from the adjacent Ruki was highly stable throughout the measurement period from October 2022 to June 2023, ranging from −24.9 ‰ to −23.3 ‰ with a mean of $-\mathrm{24.3}\pm \mathrm{0.5}$ ‰ (Fig. S4).

Figure 5 δ¹³C values of leaves, litter, soil organic carbon (SOC), and soil CO₂ flux (F) at the SFF site, as well as riverine dissolved CO₂ (Ruki). Surface CO₂ flux (F) is further separated into dry and inundated based on the chamber type (floating, static). The non-significant difference between SOC and dry $F_{{\mathrm{CO}}_{\mathrm{2}}}$ is indicated by n.s.

4 Discussion

4.1 CO₂ fluxes

The surface CO₂ flux dataset from the SFF site, measured for 3 consecutive years, showed intra-seasonal and interannual variability. However, no clear seasonal patterns were observed (Fig. 3e). Baumgartner et al. (2020) showed a similarly low seasonality in lowland forests of the Congo Basin, attributing it to the limited rainfall variation between dry and wet seasons. The unclear seasonal pattern of the CO₂ fluxes at the SFF site could be attributed to a lasting effect of the flooding on soil moisture (Docherty and Thomas, 2021) and/or consistent rain events during the whole year (Fig. 3a–b). These factors, along with the brief duration of both dry seasons, may lead to soil moisture contents remaining near optimal conditions for vegetation and soil microbes to thrive. Such uniform environmental conditions may maintain autotrophic and heterotrophic respiration at a steady level, despite undergoing a discernible dry and wet season cycle. The reported mean flux of 2.36±0.51 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ from the SFF site was lower compared to that in previous studies in the Congo Basin. These studies found mean values of 3.13±1.22 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ and 3.45±1.14 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ in montane and lowland forests, respectively (Baumgartner et al., 2020), and 4.07±0.90 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$ in a lowland secondary forest of Cameroon bordering the Congo Basin (Verchot et al., 2020). Compared to similar tropical forest studies, our values are at the low end of the range reported across the pantropical forest realm (Table 2).

The perennially flooded forest (PFF) site, located at the interface between terrestrial (forest) and aquatic (stream) ecosystems, showed relatively high emissions (4.38±0.64 $\mathrm{\mu }\mathrm{mol}{\mathrm{m}}^{-\mathrm{2}}{\mathrm{s}}^{-\mathrm{1}}$) compared to other tropical flooded forests (Scofield et al., 2016; Table 2) or streams draining catchments dominated by seasonally or continually inundated swamp forests (Mann et al., 2014; Alin et al., 2011; Table 2). The elevated CO₂ fluxes at the PFF site resulted in higher fluxes relative to the SFF site. Further research is needed to determine whether greater water-depth-integrated respiration (Amaral et al., 2020), a positive correlation with a larger inundated area (Amaral et al., 2020), prolonged river interactions, or other factors explain such differences. In contrast, the SFF site presented reduced CO₂ fluxes during the onset of flooding, speculatively due to the inhibitory effect of excessive soil moisture on soil respiration (Courtois et al., 2018; Nissan et al., 2023). A non-significant positive trend between water level and the aquatic CO₂ fluxes was visually discernible (Fig. S7), which is in line with a positive relationship between pCO₂ and discharge measured on the adjacent Ruki (Drake et al., 2023). As a constant gas transfer velocity was used in the present study, short-term changes in aquatic CO₂ fluxes reflect the variations in carbon dioxide concentrations (pCO₂) in the water. Moreover, the generally low gas transfer velocity (3.5 cm h⁻¹) further reflects the very high pCO₂ concentrations (10 197–17 260 ppm) measured at the PFF site. These values are significantly higher than the range (3069–9088 ppm) found by Drake et al. (2023). However, the adjacent Ruki water has a long transit time compared to a swamp and a stronger current, which in turn results in higher CO₂ outgassing. Generally, the pCO₂ concentration itself is driven by factors such as terrestrial inputs, gas exchange with the atmosphere, water temperature (gas solubility), water chemistry (pH, alkalinity), and in-stream metabolism (Battin et al., 2023; Hotchkiss et al., 2015; Rocher-Ros et al., 2019).

Finally, the in-situ-derived gas transfer velocity (k_x) expressed as normalized k₆₀₀ (2.95 cm h⁻¹) was higher than the global normalized estimate (k₆₀₀) for tropical wetlands (2.4 cm h⁻¹; Aufdenkampe et al., 2011). The gas transfer velocity itself changes by factors influencing the near-surface water turbulence (wind speed, water current velocity). Generally, assuming a constant gas transfer velocity (k_x), as applied in this study, has its limitations since it likely varies throughout the year, with increased values during the dry season when the water flows in the streambed channel (Alin et al., 2011).

Table 2 Reported mean values of surface CO₂ fluxes across various tropical forested environments.

^a Here, terra firma forests refer to non-flooding forests. ^b Measurements done only during the inundated period. ^c Mean soil respiration for the wet season.

4.2 Temperature, soil moisture, and water level controls

While the observed CO₂ fluxes at the SFF site showed no clear seasonal pattern, soil temperature, soil moisture, and the river level as a proxy for the water level emerged as significant controls. While the positive effect of temperature and soil moisture on soil CO₂ fluxes is well known and used to model soil CO₂ fluxes (Nissan et al., 2023), the effect of the water level is less well understood. The observed quadratic relationship with the water level suggests an optimal water level beyond which further increases lead to reduced CO₂ fluxes. This optimal point speculatively corresponds to the shift to water-saturated conditions in the organic-rich surface soil transitioning from oxic to anoxic conditions. A negative effect of the water level beyond a critical threshold aligns well with the results of Goodrick et al. (2016), who found maximal soil CO₂ fluxes associated with a water level between 1.5 and 2 m below the ground and minimal fluxes when the water level was within 0.15 m of the surface for a tropical riparian swamp forest in Australia. Similarly, Rubio and Detto (2017) found a quadratic relationship between CO₂ fluxes and soil water content in the Amazonian Basin. CO₂ fluxes can be reduced in both high and low soil water content, and fluctuations in water level introduce additional factors beyond its influence on soil saturation. Both heterotrophic soil respiration and autotrophic soil respiration are reduced under dry conditions due to limited microbial activity and reduced photosynthetic activity through stomatal closure (Baumgartner et al., 2020). In our study, the lowest-flux event recorded in July 2022 coincided with a marked decrease in soil moisture. This suggests that, during this event, the reduced soil moisture levels became a limiting factor for supporting soil respiration. Conversely, increased soil moisture generally enhances respiration. This was generally the case during our study period, as evidenced by the positive correlation between soil moisture and surface CO₂ fluxes (p value <0.05; Table 1). However, excessively high moisture conditions (due to strong-rain events or high water levels during flooding) can also hinder substrate decomposition by physically impeding the diffusion of atmospheric oxygen and respired CO₂ through the soil pores, thereby limiting both the production and the diffusion of CO₂ (Courtois et al., 2018; Nissan et al., 2023). This could explain the temporary decrease in CO₂ fluxes observed at the onset of the flooding period (Fig. 3). Furthermore, fluctuations in the water level can influence soil respiration through physical processes, like flushing out soil CO₂ during rising phases, enhanced lateral movement of dissolved CO₂, air ingress, and redistribution of organic material during receding phases (Dalmagro et al., 2018; Goodrick et al., 2016). Finally, the positive interaction between soil temperature and water level (p value <0.05; Table 1) suggests that higher temperatures will reinforce the effect of the water level and shift the maximum soil flux towards higher water levels, delaying its inhibitive effect (Fig. 4).

Nevertheless, it is important to note that both the water level and the soil moisture measurements exhibit seasonal patterns but do not effectively capture the short-term changes in surface CO₂ fluxes at the SFF site. Furthermore, the CO₂ fluxes exhibit unclear seasonal patterns (Fig. S5b). This suggests that other factors, such as aboveground inputs from vegetation, river sediment deposition, and rain-induced events, may significantly influence surface CO₂ fluxes, both in the short term and at seasonal timescales. Additionally, it is important to stress that using river level as a proxy for water level at the SFF site presents limitations, such as neglecting local topography or soil characteristics. Thus, fortnightly variations in soil CO₂ fluxes may not be fully captured by this proxy, as local hydrological dynamics might differ from those of the broader river system. Hence, this method may not fully capture the dynamics of the water level and its influence on surface CO₂ fluxes.

Overall, while soil moisture content and temperature are often considered primary drivers of soil CO₂ fluxes (Courtois et al., 2018; Nissan et al., 2023; Oertel et al., 2016), our findings also indicate that incorporating water level can help to unravel the variability of the fluxes for lowland forests with shallow water tables.

At the PFF site, on the other hand, we did not find any statistically significant relationships between potential drivers (DOC, BDOC, river level, pH, C:N) and pCO_2w. This suggests that the chemical composition of the water is relatively homogenous throughout the year and that allochthonous rather than autochthonous processes determine pCO_2w concentrations.

4.3 Isotopic indicators

The general carbon isotopic composition of plant tissue is determined by the degree of ¹³C discrimination at the leaf level (Brüggemann et al., 2011). Due to the high photosynthetic activity of tropical plants, ¹³C discrimination is also high, resulting in very negative δ¹³C values at the leaf level, as observed in this study (−37.06 ‰ to −28.89 ‰). As C moves across the various ecosystem C pools, the substrate becomes gradually enriched in ¹³C due to kinetic isotope fractionation. In the case of the studied SFF site, a total ¹³C enrichment of 5.27 ‰ was observed when moving down the cascade from leaves and litter to SOC and respired CO₂ under dry conditions (p values <0.05; Fig. 5). Particularly interesting here is the absence of ¹³C fractionation between SOC and soil-respired CO₂, which might initially be interpreted as a result of closed-system dynamics where the substrate is limited and organic decomposition tends to be complete. However, soil-respired CO₂ is a two-component flux, comprised of heterotrophic and autotrophic respiration. In other words, SOC is not the sole factor governing soil-respired CO₂. Indeed, autotrophic respiration is to a large degree fuelled by recently photosynthesized ¹³C-depleted carbon (Ottosson-Löfvenius and Read, 2001; Barthel et al., 2011), which in turn can decrease the overall soil-respired δ¹³C value relative to SOC (depending on the relative contribution of autotrophic vs. heterotrophic soil respiration). Transport rates from above to below the ground can reach up to 0.5 m h⁻¹ (Kuzyakov and Gavrichkova, 2010). Thus, whether the similar δ¹³C values between SOC and respired CO₂ are driven by substrate limitation or a strong influence of autotrophic respiration requires further investigation.

The highest ¹³C enrichment observed was from CO₂ emitted during flooding at the SFF site (−20.4 ‰; p values <0.05). These δ¹³C values were even higher than the δ¹³C values measured in the adjacent Ruki (−24.30 ‰; p value <0.05). The reason for such highly ¹³C-enriched CO₂ outgassing during inundation remains unclear, but given that the water in the inundated forest likely experiences relatively long residence times compared to the river, the outgassed CO₂ might become this heavily ¹³C enriched due to extensive outgassing. Moreover, the standing water allows for the growth of methanogenic archaea, which use simple carbon compounds, such as acetate, as electron donors (Conrad et al., 2021). The CO₂ molecules obtained from acetate cleavage is another fractionation process that potentially influences the overall isotopic composition of outgassed CO₂. Lastly, as the inundation of the SFF site is mainly driven by backflow from the river system, the dissolved CO₂ in the inundated water could be a mix of riverine and locally soil-respired CO₂ that undergoes further in situ ¹³C enrichment.

5 Conclusion

This study presents a multi-year dataset of CO₂ fluxes from two forested wetland sites along a flooding gradient: a seasonally flooded forest (SFF) and a perennially flooded forest (PFF). While exhibiting short-term and interannual fluctuations, CO₂ fluxes showed limited seasonal patterns. At the SFF site, surface emissions increased with rising soil moisture and temperature, while the water level demonstrated a significant quadratic relationship. Despite the significant sensitivity to environmental conditions over the observation period, the short-term variability observed at both sites and the interannual variability at the SFF site were incompletely explained, suggesting the influence of additional factors in regulating emissions.

Our results emphasize that water level, alongside soil temperature and soil moisture, significantly affects surface CO₂ fluxes in lowland areas with shallow, fluctuating water tables. Future research should include direct measurements of the water level over the entire hydrological year to elucidate the temporal dynamics of this relationship. Overall, the reported measurements contribute to filling the data gap for soil respiration rates of tropical forests in the Congo Basin and provide baseline fluxes for parameterizing Earth system models.