## Main

Earth’s habitability is regulated by atmospheric CO2 concentrations. Over geologic timescales, silicate weathering and terrestrial organic carbon burial drive atmospheric CO2 drawdown, while solid Earth degassing and organic carbon oxidation release CO2 to the atmosphere. Terrestrial organic carbon burial is facilitated by geomorphic processes such as erosion, which removes carbon from the terrestrial biosphere, and fluvial transit, which transfers biospheric particulate organic carbon (POC) to downstream depositional basins1,2. The amount of POC that can be buried long term is modulated by the balance between oxidation and preservation during fluvial transit, including time spent in transient storage in floodplains3,4. Predicting how fluvial transit regulates long-term POC burial, and ultimately atmospheric CO2 concentrations, requires an understanding of the feedbacks between geomorphic and geochemical processes governing sediment routing and POC decomposition5,6.

We hypothesize that the balance between fluvial POC oxidation and preservation (‘POC budget’) is controlled primarily by two factors: (1) fluvial transit time and (2) mineral protection of POC. First, fluvial transit time determines the residence time of POC in river systems3,7,8,9, and longer residence times facilitate POC oxidation. In coastal mountains with short (101–102 km), steep rivers linking to the ocean, fluvial POC may be efficiently transferred to depocentres8,10. In continental settings, fluvial transit lengths and times can exceed 103 km and 20 kyr11,12,13, respectively, and many large rivers have been characterized as net CO2 sources due to degassing from channels and floodplains14,15,16,17,18. Where measured, 14C ages of exported fluvial POC are an order of magnitude younger than clastic sediment transit times, indicating substantial POC turnover during lowland transit19,20. As POC oxidation is probably limited during in-river transport21, turnover must occur primarily during transient floodplain storage. However, sediment transit time data are scarce22, making it difficult to quantify the effect of sediment routing on the fluvial carbon budget.

A second possible control on POC turnover is the degree of coupling between clastic sediment and POC as stronger coupling reduces decomposition rates. Fluvial POC may travel as discrete particles or bound to minerals9,23,24. Reactive mineral phases, such as aluminium, iron and manganese oxyhydroxides, may protect organic carbon from microbial degradation25,26,27 via cation-bridging reactions with phyllosilicates28 or ligand exchange on hydroxylated mineral surfaces29. This mineral association reduces POC turnover, and evidence of POC that has aged up to 104 yr30,31,32 suggests that this mechanism can be volumetrically important in river systems. However, the relative importance of organo–mineral complexes for preserving fluvial POC remains unclear33.

In this article, we trace organic matter through a river system with no tributaries for nearly 1,300 km to quantify the effects of sediment transit time and organo–mineral interactions on the fate of fluvial POC. Using existing fluvial sediment transit time estimates34, POC isotope measurements and secondary mineral data for actively transported sediment, we show that POC oxidation is regulated by both transit time and mineral protection. Our analysis allows us to develop a simple model to estimate POC turnover during fluvial transit, and we use this model to explore the geomorphic and geochemical drivers of fluvial organic carbon cycling.

## A natural flume experiment

The Rio Bermejo in northern Argentina (Fig. 1) drains eastward from the Andes, delivering ~103 Mt yr–1 of sediment to the lowland basin34, ~92% of which is transported during the South American monsoon season (December–May) (Extended Data Fig. 1). The river flows ~1,270 km from the last tributary confluence at the mountain front (Rio San Francisco, river km 0) to the Rio Paraguay, with no notable tributaries, limited anthropogenic activity and negligible aquatic productivity due to high flow velocity and turbidity35 (Extended Data Fig. 2).

From river km 0 to km 265, the river is braided and perched above the flood basin. The remaining ~1,000 km is a single-thread, meandering channel that migrates up to 30 m yr–1 (ref. 34) (Fig. 1). While water can transit this system in ~14 days, sediment requires on average ~8,500 years, as determined by the net accumulation of 10Be in river sediment between river km 0 and 1,220 (ref 34). As a result of lateral channel migration, sediment experiences on average ~4.5 deposition–erosion episodes during lowland transit, each taking ~1,900 years (ref. 34). Averaged across all particle sizes, the downstream sediment transit velocity (vsed) is ~0.15 km yr–1. Due to local differences in channel morphodynamics, vsed varies along the channel, from ~0.59 km yr–1 in the braided reach to ~0.1 km yr–1 in the lower meandering reach (Fig. 1). In the following, we show that vsed is a first-order control on POC oxidation during fluvial transit.

## Downstream evolution of POC composition

To evaluate how POC evolves during fluvial transit, we measured POC concentrations, stable (13C/12C) and radioactive (14C) carbon isotopes and mineral specific surface area (SSA) of suspended sediment sampled from river depth profiles at six stations during monsoon season (Methods). To account for hydrodynamic sorting and characterize the bulk composition of the suspended load, we depth-integrated each profile, weighting individual sample measurements by suspended sediment and POC concentrations (Supplementary Table 1). At river km 0, the depth-integrated POC concentration (POCDI) was 0.18 ± 0.06%. Assuming POCDI is constant during monsoon season and dry season changes are negligible, the mean annual headwater POC flux (Qhw) is 1.85 ± 0.62 × 105 tC yr–1 (Methods). At river km 1,220, POCDI increased to 0.28 ± 0.02%, yielding a mean annual POC export of 2.24 ± 0.16 × 105 tC yr–1 (Qout). The difference between Qhw and Qout is +0.39 × 105 tC yr–1, suggesting that, under modern conditions, lateral POC inputs exceed POC lost during transit.

POCDI increased along the river’s braided reach to 0.30 ± 0.04%, while OC loading (mass of POC per unit SSA) remained constant at 0.21–0.23 mgC m–2 (Fig. 2 and Supplementary Table 1). These diverging trends probably arise from selective deposition of coarse sediment with low POC (Extended Data Fig. 3). The OC loading decreased significantly in the meandering reach to ~0.13 mgC m–2 due to lateral erosion of weathered floodplain sediment with high SSA. We further explored downstream POC transformations using stable and radiocarbon isotope measurements.

The POC stable carbon isotope composition (δ13CPOC) for individual samples ranged from –27.3‰ to –24.9‰, and radiocarbon content (expressed as fraction modern, F14CPOC) ranged from 0.78 to 0.94 (Fig. 2). All δ13CPOC and F14CPOC values were within the compositional range of regional topsoil, vegetation and floodplain sediment (Fig. 2 and Extended Data Table 1), indicating that fluvial POC comprises a mixture of compounds ranging in age and turnover time. This is supported by F14C values of refractory terrestrial leaf wax n-alkanes (F14Calk) in river sediment. F14Calk ranged from 0.74 to 0.86, significantly lower than F14CPOC values for the same samples, suggesting that a fraction of fluvial POC derives from an older, preferentially preserved POC pool36.

Within depth profiles, surface-water samples generally had higher POC concentrations, more positive δ13CPOC values and lower F14CPOC values than samples collected >0.5 m below the surface (Fig. 2). In agreement, F14Calk values were generally lower for surface samples. These compositional differences reflect hydrodynamic sorting33,37, where water-logged plant debris dominated POC at depth, and surface-water samples were concentrated in 13C-enriched POC associated with low-density mineral phases25,26,38. With increasing transit distance along the meandering segment, surface-water POC became consistently more 13C enriched and 14C depleted, reflecting progressive addition of aged, mineral-associated POC. Bedload samples had low POC concentrations (Supplementary Table 1), suggesting that vertical exchange between bedload and suspended load has negligible effects on depth-integrated POC composition.

From river km 135 to 420, where vsed was highest, δ13CDI remained constant, but F14CDI increased from 0.83 to 0.92 (14C-age decrease of ~780 years) (Fig. 2). This may reflect mass loss of 14C-depleted sediment through selective deposition of coarse sand (including lithic fragments and petrogenic POC) and/or fresh organic matter addition. From river km 420 to 1,220, where vsed is low, δ13CDI values increased from –26.4 ± 0.3‰ to –25.5 ± 0.3‰, and F14CDI values decreased from 0.92 to 0.86 (~510 years 14C-age increase). These changes probably reflect entrainment of aged, 13C-enriched POC from floodplain deposits. These data are consistent with the idea that the time allowed for POC aging and transformation, as captured by sediment transit time, exerts a strong control over POC composition.

Complete POC turnover during floodplain storage would cause F14C values to increase downstream. The 13C-enrichment and apparent POC aging trends along the meandering reach (Fig. 2) suggest that some POC is preserved during floodplain storage, surviving multiple episodes of deposition and erosion. Below, we test the hypothesis that mineral protection limits POC turnover during fluvial transit24,28.

## Organo–metal complexation slows POC decomposition

To test whether organo–metal complexation slows POC turnover during fluvial transit, we measured acid-extractable metal concentrations in suspended sediment (Methods) as proxies for the abundance of reactive oxyhydroxide phases39. Extractable [Al], [Fe], [Mg], [Mn] and SSA showed statistically significant correlations with F14CPOC (negative), F14Calk (negative) and δ13CPOC (positive) (Fig. 3 and Extended Data Figs. 4 and 5). Samples with higher metal concentrations contained relatively older, 13C-enriched POC. All correlations between F14CPOC and metal concentrations are stronger than the correlation between F14CPOC and SSA, suggesting that organic compounds are not only adsorbed onto mineral surfaces but also bound to secondary minerals via complexation reactions29. The 13C enrichment of mineral-stabilized POC may result from decomposition in the floodplain40 and/or preferential reaction of minerals with 13C-enriched organic compounds25,41. These data suggest that organo–mineral complexes developed during transient floodplain storage help stabilize biospheric POC (POCbio), in agreement with our hypothesis, and can therefore increase the probability of long-term POCbio burial.

## A model for POC oxidation during fluvial transit

Our data provide evidence for POC oxidation, fluvial recruitment of young biomass via lateral channel migration and POC preservation by organo–mineral complexation during fluvial transit. In the following we use these findings to estimate the POC budget of the Rio Bermejo, as regulated by sediment transit time and mineral protection.

We calculated the annual Rio Bermejo POC oxidation flux by defining a relationship between vsed and POC decomposition rate. We assume that POC travels with clastic sediment at vsed because we found water-logged plant organic matter at all water depths, and POC reactivity is described by a decomposition rate constant (k, 1 yr–1) (Methods). Linking vsed and k, the characteristic channel length (xc) over which the fluvial POC load is turned over (depleted) is:

$$x_{\mathrm{c}} = \frac{{v_{{\mathrm{sed}}}}}{k}$$
(1)

Then the number of POC turnover cycles (nc) during transit along a river system of length L is:

$$n_{\mathrm{c}} = \frac{L}{{x_{\mathrm{c}}}}$$
(2)

Finally, the annual POC oxidation flux (Qox) resulting from fluvial transit of the mean annual headwater POC load (POCin, tC) over transit time t is:

$$Q_{{\mathrm{ox}}} = \frac{{n_{\mathrm{c}} \times {\mathrm{POC}}_{{\mathrm{in}}}}}{t}$$
(3)

We assume geomorphic steady state, where mean POCin and vsed are constant over ~104 yr transit timescales. This also assumes POCin is spatially constant, such that we underestimate POC oxidation if the POC load increases in downstream reaches. Fluvial POC contains numerous carbon pools ranging in source, age and decomposition rate4,23,42,43. For simplicity, we divided fluvial POC into two pools: (1) fast-cycling discrete organic particles, POCfast, and (2) slow-cycling mineral-associated POC, POCslow22,32. We assume each POC pool is homogeneous and decomposition rates are constant, although this is rarely the case in nature44.

Annually, the Rio Bermejo receives ~1.85 × 105 tC POC via erosion in the headwaters (POCin). According to a Bayesian endmember isotope mixing model, 47% ± 10% of this load behaves as POCslow and 53% ± 17% as POCfast (Methods and Extended Data Table 2). For these two pools, we estimated k using radiocarbon data, following Torn et al.45 (Methods). For POCfast, we set the turnover time to ~17 years, which is the modelled ecosystem turnover time in the subtropical Rio Bermejo region46, yielding kfast ≈ 6 × 10–2 yr–1 (Methods and Extended Data Table 3). For kslow, we utilized n-alkane F14Calk data because of their recalcitrance and association with secondary minerals36. As a conservative estimate of POCslow oxidation, we used the lowest F14Calk value (0.74), yielding kslow ≈ 3.4 × 10–4 yr–1 (turnover time of ~2,900 years) (Methods).

Using equations (1–3), we estimated separate oxidation fluxes for POCslow and POCfast during an 8,500 yr fluvial transit along the ~1,270 km lowland Rio Bermejo (Methods). POC turnover varies significantly along the Rio Bermejo (Extended Data Table 4). With lower transit velocity (increased storage time), the meandering reach is the locus of turnover, with a POC turnover length scale (xc) nearly an order of magnitude shorter than in the braided reach (Fig. 4). This highlights the control of channel morphodynamics on the fate of POC in lowland rivers.

Along the full channel length, xc for POCfast is 2 ± 1 km, resulting in nc of 520 ± 160 (Fig. 4). By contrast, xc for POCslow is 430 ± 190 km, resulting in nc of 3.0 ± 0.9. Since the Rio Bermejo sediment load experiences ~4.5 deposition–erosion cycles along this pathway34, POCslow must be coupled to the clastic sediment trajectory, while POCfast is decoupled from it. During ~8,500 yr transit between the mountain front and Rio Paraguay, turnover results in Qox of 30.4 (+23.8/–17.0) tC yr–1 for POCslow and 6,050 (+4410/–3210) tC yr–1 for POCfast, suggesting that ~2–6% of POCin is oxidized annually during fluvial transit.

## Controls on the fluvial POC budget

Our data show that sediment transit time and mineral protection are primary controls on the fate of fluvial POC during source-to-sink transit. Our model for POC turnover allows quantification of the respective roles of these two governing mechanisms. Here we define a transit time-dependent fluvial POC budget:

$$\begin{array}{l}Q_{{\mathrm{out}}} - Q_{{\mathrm{hw}}} = Q_{{\mathrm{lat}}}\\\quad\ \ \, - \frac{{\left( {n_{\mathrm{c}} \times f_{{\mathrm{slow}}} \times {\mathrm{POC}_{\mathrm{in}}}} \right)_{{\mathrm{slow}}} + \left( {n_{\mathrm{c}} \times f_{{\mathrm{fast}}} \times {\mathrm{POC}_{\mathrm{in}}}} \right)_{{\mathrm{fast}}}}}{t}\end{array}$$
(4)

The last term on the right side of equation (4) represents the total annual POC Qox, encompassing POCslow and POCfast turnover during transit, where f is the fraction of POCslow or POCfast contributing to POCin. Qlat represents POC delivered via tributaries or lateral erosion. Whether Qout – Qhw is positive or negative determines whether fluvial transit results in net CO2 drawdown or release, respectively. Our estimates for the Rio Bermejo, Qout (~2.24 × 105 tC yr–1) > Qhw (~1.85 × 105 tC yr–1), suggest that additional POC is sequestered from the floodplain during transit. Given Qox is estimated at ~6.08 × 103 tC yr–1, equation (4) can be solved for Qlat ≈ 4.58 × 104 tC yr–1, resulting from lateral erosion into floodplain forests. This agrees with estimates from measured channel migration rates and net primary productivity (Methods). Qlat is nearly an order of magnitude greater than Qox, more than offsetting oxidative POC loss.

Using equation (4), we test the sensitivity of the fluvial POC budget by varying environmental boundary conditions for the Rio Bermejo. Qox is linearly proportional to vsed and k (equations (2–4)). Holding all else constant, decreasing vsed by an order of magnitude (mean transit time of ~85 kyr) would increase Qox by an order of magnitude, but such a long transit time is unrealistic for this system. Removing mineral protection (100% POCfast) would increase Qox by one order of magnitude, reducing the potential for long-term burial. Doubling k for POCfast would result in net CO2 release, but such high decomposition rates are realistic only under warmer and wetter conditions. These tests suggest that modern conditions in the Rio Bermejo do not cause significant oxidation relative to the amount of POC exported annually. Drastic changes in boundary conditions are required to oxidize more POC than is preserved. However, the key variables here are linked in complex ways by climate, hydrology, vegetation, tectonics and even anthropogenic disturbance47, and associated feedbacks may be important.

At a global scale, vsed and k vary across different climatic and tectonic settings. Using equations (1–4), we estimated carbon budgets for the Amazon (tropics) and Mackenzie (Arctic) rivers, where Qout, k and vsed are known. In the Amazon, where POCfast turnover is fast, <10 years (refs. 18,46), and vsed is ~0.21 km yr–1 (ref. 48), ~13% of the POC load is oxidized, but Qlat > 1 MtC yr–1 augments POC export to the ocean. In the Mackenzie River, where POCfast turnover is slow, >70 years (ref. 46), and vsed is ~0.09 km yr–1, <1% of the POC load is oxidized, indicating efficient source-to-sink transit and potential CO2 drawdown on burial in the Beaufort Sea49. These analyses suggest that lateral erosion into vegetated floodplains enhances CO2 drawdown if POC recruited from the lowlands is buried long term. River engineering, particularly the construction of artificial levees and groins that reduce lateral mobility, can significantly reduce Qlat (ref. 47), thereby decreasing potential CO2 drawdown. Dams may also increase POC oxidation by increasing source-to-sink transit times. Applying this model to rivers globally may yield more robust estimates of river–atmosphere carbon fluxes, advancing our understanding of the global carbon cycle.

## Methods

### River sediment sampling

We collected 24 river suspended sediment samples from the Rio Bermejo between 13 and 25 March 2017, during the peak of the South American monsoon season (December to May), with 75th–85th percentile water discharge conditions (over a 50 yr gauging record) (673 to 1,079 m3 s–1) (Argentina National System of Hydrologic Information, https://snih.hidricosargentina.gob.ar/). Suspended sediment flux data indicate that 92% of the annual suspended sediment load is transported during monsoon season, suggesting our samples are representative of the majority of POC transported by the Rio Bermejo. We sampled river water from depth profiles (two to five sampling depths depending on total water depth) at four locations along the mainstem Rio Bermejo, one location on the Rio Bermejo 10 km upstream of the Rio San Francisco confluence, and one location on the Rio San Francisco 15 km upstream from the confluence (Fig. 1). River water was sampled by boat using a weighted eight-litre horizontal sampling bottle. Sediment was recovered by filtering the water under pressure through 0.22 μm polyethersulfone filter paper in a custom filtration device. Four bedload samples were collected with a weighted mesh net. Sediment was dried in an oven at 40 °C and subsequently disaggregated and homogenized with mortar and pestle. Suspended sediment concentrations were calculated as dry sediment mass normalized by the sampled water volume.

### POC concentrations and isotope measurements

For bulk organic carbon analyses, sediment samples were powdered in a disc mill, and inorganic carbon was removed following Galy et al.50. Sediment was decarbonated by leaching in 4% HCl solution, discarding the supernatant, rinsing with deionized water and drying before measurement. Total organic carbon (TOCPOC) and δ13CPOC were measured in duplicate at Durham University using a Costech elemental analyser (EA) coupled to a CONFLO III and Thermo Scientific Delta V Advantage isotope ratio mass spectrometer. Radiocarbon content was measured using an EA coupled to an accelerator mass spectrometer (EA-AMS) at ETH Zürich51. We report 14C content as F14CPOC by normalizing measurements to 95% of the 1950 NBS Oxalic Acid II standard (δ13C = –17.8‰) and correcting for mass-dependent fractionation using a common δ13C value of –25‰.

### Grain size and SSA measurements

Grain size distributions were measured on ~10 mg aliquots of river sediment using a Horiba LA-950 laser particle size analyser. Before measurement, we added 1.5 ml sodium metaphosphate dispersion agent to each sample and shook samples on an overhead shaker for ~24 hours.

SSA was measured on ~1–4 g aliquots of river sediment using a Quantachrome NOVAtouch LX gas sorption analyser. Before measurement, samples were combusted at 350 °C to remove organic matter then degassed to 40 mtorr at 350 °C on a Quantachrome FLOVAC degasser to remove excess water. SSA values were calculated from the N2 adsorption isotherm, following the Brunauer, Emmett and Teller theory52.

### Acid-extractable metals

To extract the Al, Fe, Mg and Mn ions from the reactive grain coatings of the suspended sediment samples, we used a two-step leaching procedure adapted from Wittmann et al.53. We first dried 0.5–1.0 g sediment aliquots at 110 °C overnight. After drying, samples were sealed and weighed immediately upon cooling. Reactive amorphous oxyhydroxide phases were leached with 10 ml 0.5 M HCl solution, with mild shaking at room temperature for 24 hr. Crystalline oxide phases were leached in 10 ml 1 M hydroxylamine–hydrochloride solution (NH2OH × HCl in 1 M HCl) in an ultrasonic bath at 80 °C for 4 hr, with shaking every 10 min. The leachates were dried completely, treated with aqua regia to destroy matrices and then diluted in 3 M HNO3 for measurement. Al, Fe, Mg and Mn concentrations were measured via inductively coupled plasma–optical emission spectroscopy. Concentrations were normalized by initial dry sample mass. Measurements of amorphous and crystalline phases were combined to obtain total reactive metal concentrations.

### Depth integration and POC fluxes

For each river depth profile, we calculated depth-integrated δ13CPOC and F14CPOC values as the weighted means of the measured values for individual samples. Depth-integrated values were weighted by suspended sediment concentration and POC concentrations (TOCPOC) measured for each sampling depth (Supplementary Table 1). We estimated uncertainty as the standard error of the weighted mean.

We quantified the annual fluxes of POC delivered to (Qhw) and exported from (Qout) the mainstem Rio Bermejo as the product of mean annual suspended sediment flux and depth-integrated POC concentration of the suspended sediment. For these calculations, we used the long-term suspended sediment fluxes measured at gauging stations at river km 0 and km 1,086 reported by Repasch et al.34 and depth-integrated POC concentrations measured at downstream km 0 and downstream km 1,220 (this study).

We measured compound-specific 14C content of leaf wax n-alkanes to trace the radiocarbon signature of vascular plant-derived POC. Lipid compounds were extracted from bulk sediment using a Dionex accelerated solvent extraction system (ASE 350), and the n-alkanes were further isolated by solid-phase extraction over silica gel columns following the manual procedure of Rach et al.54. The C27, C29, C31 and C33 n-alkanes were then purified from the saturated lipid fraction using preparative capillary gas chromatography following Eglinton et al.55. The purified long-chain n-alkanes were transferred to tin capsules using dichloromethane and placed in combusted glass vials. Radiocarbon measurements were made by EA-AMS at ETH Zürich following Haghipour et al.56. Sample sizes ranged from 21 to 63 μg. The 14C measurements were corrected using the measured 14C content of process blanks and empty tin capsules.

### Bayesian isotope mixing model—MixSIAR

To determine the relative contributions of different OC sources to riverine POC, we used a three-endmember Bayesian isotope mixing model constrained by bulk POC δ13C and F14C data for the endmembers. Input to the model included the mean and standard deviation of measured isotopic values for sample sets of leaf litter (n = 6), floodplain sediment40 (n = 51) and topsoil (n = 17) (Extended Data Table 2). We calculated the fractional contribution of each endmember to all suspended sediment sample ‘mixtures’ using the R package MixSIAR57. We ran the MixSIAR model with an uninformative prior, making no initial assumptions about the source contributions to the samples. The Markov chain Monte Carlo method was performed with 1 × 106 iterations, burn-in of 5 × 105 iterations, thinning factor of 500 and three chains. For each river sediment sample, this analysis yielded three posterior distributions, which contained the full range of possible fractional contributions of the three OC sources. We calculated the mean and standard deviation of the posterior distribution for each source to obtain the most probable values (Extended Data Table 3). We combined the leaf litter and topsoil values to represent the proportion of POCfast in each sample and used the floodplain sediment values to represent POCslow. We estimated the proportions of these two pools for all suspended sediment samples, and then calculated the depth-integrated mean contributions of POCfast and POCslow at the headwaters, resulting in 0.53 ± 0.17 and 0.47 ± 0.10, respectively. We calculated the mean annual Qhw of POCfast and POCslow as the product of these values and the mean annual headwater POC flux.

### Calculating decomposition rate constants

We calculated a characteristic decomposition rate, k, for POCfast and POCslow. We estimated k for POCfast and POCslow following the method of Torn et al.45:

$$k = \frac{{\lambda \times {\mathrm{F}}^{14}{\mathrm{C}}}}{{1 - {\mathrm{F}}^{14}{\mathrm{C}}}}$$
(5)

where λ is the 14C decay constant (1.21 × 104 yr–1) and F14C is a representative fraction modern value for the POC pool of interest. The turnover time corresponding to the POC pool is simply 1/k, resulting in short turnover times for fast decomposition rates and long turnover times for slow decomposition rates45. By selecting just one characteristic F14C value to input into equation (5), we assume that the entire POC pool being modelled decomposes at the same rate through time.

To determine the POCfast turnover time, we used radiocarbon data from the Rio Bermejo catchment. Leaf litter samples had a mean F14C value of 1.00 (Extended Data Table 2), indicating modern carbon and suggesting a turnover time <50 years. Total ecosystem turnover times were estimated at a global scale by Carvalhais et al.46, and the ecosystem turnover time for the subtropical Rio Bermejo region was modelled at ~17 years. We used this turnover time to determine a characteristic decomposition rate of 0.06 yr–1 for POCfast. This is six times faster than, but of the same magnitude as, other estimates made for POCbio in the literature (for example, Torres et al.22). Where turnover times for POCbio are estimated, mineral-bound POCbio is included in this estimate, while we consider only free, labile POC in the fast-cycling pool.

### Validating the modelled lateral POC influx

Balancing the Rio Bermejo carbon budget (equation (4)) revealed that the lateral influx of floodplain POC significantly influences the strength of the carbon sink. We estimated the possible lateral influx of POC into the Rio Bermejo from the floodplain due to lateral channel migration (Qlat) as:

$$Q_{{\mathrm{lat}}} = m_{{\mathrm{lat}}} \times L \times {\mathrm{NPP}}$$
(6)

where mlat is the mean lateral channel migration rate (m yr–1), L is the total channel length (m) and NPP is the net primary productivity of floodplain biomass (kgC). Lateral channel migration rates for the Rio Bermejo range from 0 to 30 m yr–1, averaging 14.9 ± 6.2 m yr–1. The length of the lowland segment of the Rio Bermejo is 1,267 km. We calculated the mean NPP for the lowland portion of the Rio Bermejo catchment (east of the mountain front) using the MODIS MOD17A3H V6 annual NPP product for the year 2014 (500 m pixel resolution). The mean NPP for the ~70,000 km2 area was 3,623 ± 4,373 kgC m–2. Applying equation (6) gives Qlat = 6.84 ± 2.85 × 104 tC yr–1, which agrees well with our model estimate of 4.58 × 104 tC yr–1 derived from the equation (4).