## Introduction

Methane (CH4) is a potent greenhouse gas that contributes to climate change with a global warming potential 34 times greater than carbon dioxide (CO2) at a 100 year time scale1. Methane fluxes from hydroelectric reservoirs, especially from tropical reservoirs, have been debated over the past years2,3,4. Even though hydropower is often thought of as a ‘green’ source of energy, studies suggest that reservoirs are not carbon neutral and in extreme cases might have even higher carbon footprint than fossil fuel energy, particularly those situated in the tropics5,6. Tropical reservoirs have been estimated to emit about 3.0 Tg CH4-C year−1 2 and to contribute 64% of the total reservoir CH4 emission3. Conversely, latitude was not a strong predictor of reservoir CH4 emission in a more recent global analysis4. Apparently, the CH4 emission from reservoirs is currently not well understood.

At an estimated global organic carbon (OC) burial rate of 0.06 Pg C year−1 in reservoir sediments7, there is apparently a large supply of organic substrate to the methanogenic microbes that live in anoxic sediments of reservoirs. Important factors that influence rates of CH4 formation (net production, i.e. production minus oxidation) in sediment are temperature as well as the organic matter (OM) characteristics (i.e. its reactivity or bioavailability) and OM supply rate. Strong exponential relationships between temperature and CH4 formation rates were found in sediments of lakes8,9,10,11 and rivers12, and ecosystem-scale CH4 emission strongly depends on temperature13. In addition, methanogenesis relies on OM characteristics and supply rate. There is evidence that more CH4 is produced from autochthonous OM (derived from aquatic plants and phytoplankton) than from allochthonous OM (derived from land plants and soils)14. Also, a high OM supply rate can stimulate high sediment CH4 formation and emission, particularly if the OM is biologically reactive15. However, allochthonous OM might degrade slower but produce the same amount of CH4 over longer period of time16, since allochthonous OM contains more support tissues that are processed more slowly16,17. When the OM is being decomposed at steady conditions, its CH4 formation decreases over time due to rapid initial decay of labile substances16. However, at timescales relevant in sediments (years), the effect of decreasing reactivity during OM decomposition on sediment CH4 formation is currently not clear. While it has been shown that CH4 formation decreases over time18, the ageing effect on sediment OM reactivity and CH4 formation was not quantified. Therefore, sediment CH4 formation can at present not be predicted at environmentally relevant timescales (years), severely limiting our understanding of the globally important function of reservoir sediments as both sinks of OC and sources of atmospheric CH4.

Here, we investigate the impact of OM source, characteristics and age on CH4 formation rates in sediment of three widely different tropical reservoirs. We performed a long-term incubation experiment (ca 739 days) of sediment varying in age between 1 and 48 years, in order to understand CH4 formation in reservoir sediments over the typical lifetime of reservoirs. We selected 3 reservoirs of different productivity and located in different biomes and, therefore, of a range of sediment OM characteristics. We expected a decrease of CH4 formation potential with increasing sediment age, and lower CH4 formation in the sediment of a reservoir with little autochthonous OM production than in a reservoir with high autochthonous OM production.

## Materials and Methods

### Sampling

Sediment samples were obtained from three Brazilian reservoirs of different trophic status: Chapéu D’ Uvas (CDU), Curuá-Una (CUN) and Funil (FUN). CDU is an oligotrophic reservoir (mean total phosphorus (TP) concentration, 12 µg L−1) that was constructed in 1994 for water supply19. CUN is a mesotrophic hydroelectric reservoir (mean TP, 19 µg L−1) that was constructed in 197719. FUN is a eutrophic hydroelectric reservoir (mean TP 34 µg L−1) that was built in 196920. CDU and FUN are located in the Atlantic Forest biome, and CUN is located in the Amazon.

In each reservoir, coring sites (6 in CDU, 4 in FUN and 7 in CUN) were distributed across the entire reservoir (Supplementary Figs 13), from the river inflow areas to the dam, to cover gradients in sediment characteristics. Sediment cores (one per site) were taken with a UWITEC gravity corer equipped with a hammer. Tubes of various length (0.6 to 3 m) were used depending on the expected sediment depth. In 14 of 17 sediment cores, we reached the pre-flooding soil. The depth of the interface between pre-flooding soil and reservoir sediment (sediment depth) was visually detected and recorded. If the pre-flooding soil was not reached; sediment depth was estimated from other cores sampled in the same area. Samples from CDU were taken on the 7th of March 2016, from FUN on the 14th of March 2016, and from CUN on the 20th–28th of February 2016.

### Sample preparation

Sediment cores were stored in dark at room temperature, i.e. similar to in-situ temperatures in these tropical systems, and sliced the day after sampling. Samples were sliced at 4 cm intervals in order to ensure sufficient amount of material for the experiment. The sub-surface (2–6 cm) layer of all sediment cores was used for the experiment to represent rather fresh but already anoxic sediment. We considered 2 cm sediment depth to be anoxic because oxygen penetration depths of 0.3–1.2 cm were reported from a Brazilian reservoir21. If possible the 4 cm layer of sediment just above the soil-sediment interface was used and up to 3 additional samples, distributed over the sediment core depth down to the soil-sediment interface, were obtained; for example, the longest core was FUN_48 (204 cm), and the soil-sediment interface was at 192 cm. This core was sampled at 2–6 cm, 32–36 cm, 60–64 cm, 116–120 cm, 172–176 cm. The total number of sediment samples per core varied between one and four, depending on the total length of each sediment core. In total, 17 samples were obtained from 2–6 cm deep sediment, and 25 samples were from deeper sediment layers. After slicing, the headspace in the sample containers (PP jars with LDPE snap lock) was filled with N2, and samples were placed in a N2-filled glove bag within 1–2 hours after slicing. The glove bag was filled with N2 and evacuated using a vacuum pump 2 times to minimize the oxygen exposure of the samples.

Each sample was homogenized in the glove bag, and a sub-sample of 10 mL of homogenized sediment was transferred into pre-weighted 60 mL serum vials. 2.5 or 5 mL (depending on the sediment density) of sterile-filtered (0.2 µm GF filter), N2-bubbled water from the respective reservoir was added to the samples. 3 replicates of each sample were prepared. The samples were sealed and flushed with N2 after preparation. Vials were weighted before and after filling to know the exact mass of sediment in each vial. As a control treatment, 3 replicates of 10 mL of sterile water (filtered with 0.2 µm GF filter) from each reservoir were prepared in the same vials as sediment sample and sampled in the same way as sediment samples. Dissolved oxygen concentration was checked in those controls at the beginning of incubation with optical sensors (PreSens) and was below detection level (0.1 mg/L detection level).

The rest of the sediment was dried in a custom made oven at about 55 °C, and water content was gravimetrically determined.

### Incubation

Samples were incubated anoxically at 25 °C in the dark throughout the whole experiment time (739 days). The lag phase until the onset of growth of methanogens in similarly handled sediment samples is very short (a few days;16) in comparison to the experiment duration, and thus considered negligible. The CH4 formation rate was measured in the samples at 7 sampling occasions. Each sampling occasion contained 2 measurements: start and end of incubation, which lasted for about 2 weeks. Exact days of sampling occasions varied for each sample, and were on average day 1–9 for the 1st sampling, day 9–28 the 2nd sampling, day 72–91 for the 3rd sampling, day 185–200 for the 4th sampling, day 275–290 for the 5th sampling, day 423–442 for the 6th sampling and day 719–736 for the 7th sampling (Supplementary Table 3). Before and after each sampling occasion, samples were flushed with N2 to prevent any possible inhibition of methanogenesis by CH4, CO2 or other volatile and potentially inhibitory substances22,23. The thermodynamics of metabolic reactions imply a change in free energy in dependence of product concentration24, and the resulting negative effect on methanogenic metabolism has been observed in incubation and modelling studies23,25,26.

To measure initial gas concentration in the vials (at start of sampling occasions), 8 mL of N2 was added to each sample vial, and then pressure in the vial was measured with a needle barometer. Samples were vigorously shaken for 1 min to allow equilibration of CH4 between the headspace and the pore-water. Then 7 mL of the headspace gas was extracted in a 10 mL syringe. At the end of each sampling occasion (after about 2 weeks) the procedure was repeated. 8 mL of N2 was added to the samples and the gas was extracted after shaking the vial for 1 min. Sample dilution after addition of N2 was calculated from pressure difference that was measured before and after the N2 addition. Samples were injected the day they were obtained into a GC equipped with a flame ionization detector (FID) (Agilent Technologies, 7890 A GC system). Concentrations of CH4 and CO2 in the sediment pore water were calculated according to the ideal gas law and Henry’s law, where coefficients for CH4 solubility in water at varying temperatures were obtained from Wiesenburg and Guinasso Jr27, and coefficients for CO2 were obtained from Weiss28.

At the end of the experiment, after the last CH4 measurement was performed, the sediment was extracted from the serum vials. The sediment of replicates was pooled, mixed and dried in an oven at 60 °C; we decided to pool the replicates because no significant differences were detected in their CH4 formation.

The total C (TC) and total N (TN) content in the sediment were measured in each sample at the beginning and end of the incubation with a Costech Elemental Analyzer. We found no significant contribution of inorganic C in our samples.

### Data analysis and calculations

The rates of CH4 and CO2 formation were obtained by the difference in CH4 concentration between two measurements and divided the time interval. The obtained rates represent net CH4 formation rates, including both production and any potential consumption of CH4 (e.g. by anaerobic CH4 oxidation29,30) in the sediment.

The age of individual sediment layers was estimated from the multi-year average sedimentation rate, which can be precisely calculated from sediment thickness, and the time elapsed since dam closure, according to the formula:

$$sample\,age=\frac{sediment\,layer\,depth}{total\,sediment\,depth}\ast reservoir\,age+incubation\,length$$
(1)

where sediment layer depth is the average depth of a slice, total sediment depth is the depth between the sediment surface and the depth of the interface between pre-flooding soil and reservoir sediment. Reservoir age is the time since damming at the sampling day, and incubation length is the time between sediment sampling and the day when CH4 formation rate was obtained. Sediment age here represents, then, a measure of time since sediment deposition (and not since OM fixation by photosynthesis or since sediment formation in the aquatic system) assuming constant sedimentation rate. We acknowledge that there probably is year-to-year variability in the sedimentation rate, e.g. due to hydrological extremes, which adds uncertainty to the age estimate of individual sediment layers. However, since there is no reason to assume any systematic change in sedimentation rate with reservoir age, the uncertainty in the estimated age of individual sediment layers is unlikely to suffer from any systematic bias, and is therefore considered to add random error to the age estimates.

Because TC and TN contents were only measured at the start and end of the experiment, we assumed a linear change in TC and TN through the experiment to estimate TC and TN at every CH4 measurement occasion. Although it has been shown that the decrease of sediment TC and TN is exponential over timescales of several years-decades31, a linear approximation of TC and TN decrease over time is reasonable for relatively short timescales as used in this experiment. Accordingly, when we compared linear TN decrease with exponential TN decrease, modelled using the TN decay rate of 0.16 year−1 given by Gälman et al.31, the difference between TN estimates at any point of time was <0.002% TN, i.e. the potential error of assuming linearity is below analytical precision (data not shown). Since N is preferentially lost over C during sediment organic matter degradation31, the effect of interpolation approach on estimated TC will be even smaller than for TN.

In some samples of the deepest sediment layers that had the lowest CH4 formation rates, the first 1–2 measurements were excluded because of possible oxygen contamination during sample preparation that delayed the establishment of methanogenic conditions (low microbial activity in old sediment delays the consumption of trace oxygen).

To estimate sediment CH4 formation rates over reservoir lifetime we fitted an exponential decay model with 3 OM pools of different reactivity with core as a factor (gnls(CH4~a * exp(−b * Age) + c)32,33. The significance of the model parameters was assessed with the ANOVA function and the quality of the model was checked by visual examination of the residuals and predicted against measured data. We have also tested other non-linear models (exponential decay model with 2 OM pools, and reactivity continuum model) but they did not converge.

From the exponential decay models we defined a “transition age“ beyond which microbial OM degradation is not quantitatively significant for the standing stock of OC but at a low background level, i.e. the age beyond which sediment OM may be considered “buried” even though microbial activity does not cease entirely. In order to model sediment CH4 formation over the typical lifetime of reservoirs (100 years), and in order to estimate the “transition age” to low background CH4 formation, we used exponential decay models (described above). We define the transition age as the age at which the slope of the modelled curve is 179°, i.e. almost flat and not distinguishable from 180°, and thus indicative of a constantly low background rate of CH4 formation that no longer changes over time (Supplementary Fig. 4). We then integrated CH4 formation over the assumed typical lifetime of reservoir of 100 years considering yearly increase in sediment thickness, and calculated the total amount of CH4 that would be formed in each sediment core layer over 100 years. We calculated the average and standard deviation of CH4 formed for each sampled sediment core separately.

The effect of sample age and TN on the CH4 formation rates was modeled with a linear model in R (lm) including the interaction between age and TN (lm (ln (CH4) ~ ln(Age) * TN). In the model CH4 formation rate per sediment dry weight (µmol g(dw)−1 d−1) was used because TC and TN are dependent on each other. The CH4 formation rates and sample age were ln transformed. The model assumptions were checked through model residual plots.

Correlation between transition age and C:N ratio was checked with a Pearson correlation. The difference between fresh and old sediment CH4 formation was checked with repeated measures ANOVA. Both model assumptions were checked through model residual plots. We used R (R 3.4.334) for data analysis, modelling and plotting.

## Results

### Sediment characteristics

The sediment samples used in our study differed in sediment characteristics (Table 1). The Amazonian CUN reservoir had the highest mean TC content (6.6%) as well as the highest mean C:N ratio (15.2), while the eutrophic FUN reservoir showed the lowest values (2.5% TC and C:N 10.6, Table 1). TN content was in general similar among the reservoirs, with slightly higher values in CUN (Table 1).

### CH4 formation rates measured during the incubation experiment

CH4 formation per sediment dry weight (µmol g(dw)−1 d−1) significantly correlated with the amount of TC in the sediment (R2 = 0.56, p < 2.2−16, data not shown) therefore CH4 formation rates were normalized for TC (µmol gC−1 d−1). There was a significant difference between CH4 production in sub-surface (2–6 cm) and deeper sediment layers (p < 2−16, Fig. 1).

CH4 formation rates decreased over time of the 739 d incubation experiment in sub-surface sediment layers (2–6 cm), but remained rather stable in deeper sediment layers (Fig. 1). CO2 formation rates also decreased over time for sub-surface sediment layers and also for deeper sediment for FUN (Supplementary Fig. 5). The highest CH4 formation rate was measured in a FUN sub-surface sediment sample (core FUN_46 at the first measurement occasion at day 7: 30.4 ± 2. µmol gC−1 d−1 (mean ± sd)). At the last measurement at day 736, the same sample had CH4 formation rate of 2.4 ± 0.1 µmol gC−1 d−1. On average FUN experienced highest CH4 formation rates in sub-surface sediment over the whole incubation period (8.8 ± 7.0 µmol gC−1 d−1, (mean ± sd)), while in CUN and CDU mean CH4 formation rates in sub-surface layers were lower over the whole incubation period (2.7 ± 2.3 µmol gC−1 d−1 and 3.9 ± 1.8 µmol gC−1 d−1 respectively). CH4 formation rates decreased most rapidly in FUN sub-surface sediment over the time of the incubation. The final CH4 formation rates (day 739) in sub-surface sediment in FUN were 24 ± 12% of the initial rates, while in CUN and CDU the CH4 formation rates decreased less and were 50 ± 16% of the initial rates in CUN and 50 ± 15% in CDU.

CH4 formation rates of deepest sediment were significantly different between reservoirs (p < 0.001, Kruskal – Wallis test). CDU had higher mean CH4 formation rates over the whole incubation period in bottom layers (1.07 ± 0.49 µmol gC−1 d−1) than FUN (0.44 ± 0.14 µmol gC−1 d−1) and CUN (0.44 ± 0.16 µmol gC−1 d−1).

### CH4 formation rates over sediment age

CH4 formation rates decreased exponentially with sediment age (Fig. 2). In none of the samples, the CH4 formation rate was zero.

The decrease of CH4 formation rate over time was most pronounced in FUN (Fig. 2). The highest CH4 formation in FUN was measured in 1 year old sediment (30.4 ± 2.7 µmol gC−1 d−1) and the lowest in 42.9 years old sediment (0.2 ± 0.005 µmol gC−1 d−1). We could also observe clear exponential decrease in sediment CH4 formation in CUN, where the estimated age of the samples varied from 2 to 38 years. In CUN, the highest CH4 formation of 11.4 ± 0.6 µmol gC−1 d−1 was measured at 7 years old sediment and the lowest CH4 formation rate was measured in 36 years old sediment (0.2 ± 0.01 µmol gC−1 d−1). CDU is the youngest of the studied reservoirs. Estimated sediment age in CDU varied from 1 to 22 years. The highest CH4 formation in CDU was measured at 6 years old sediment (8.0 ± 0.2 µmol gC−1 d−1) and the lowest was 0.1 ± 0.03 2 µmol gC−1 d−1 at 18 years old sediment.

Models for CH4 formation over time achieved best fit on deeper sediment cores where a wide age gradient was covered by many samples. In cases where the age gradient was very short, model fits were sometimes poor. Model statistics and coefficients can be found in Supplementary Table 1.

### CH4 formation rates as a function of TN and sediment age

Merging all 764 measured CH4 formation rates in these three widely different reservoir sediments (Table 1) into one dataset, we found that CH4 formation rates could be predicted from the sediment TN (%) and the age of the sediment (R2 = 0.81, p < 2−16; Fig. 3; see Supplementary Table 2 for model statistics):

$$\mathrm{ln}({{\rm{CH}}}_{{\rm{4}}}\,{\rm{formation}})=-\,0.59\ast \mathrm{ln}({\rm{Age}})+6.46\ast TN-0.99\ast \mathrm{ln}({\rm{Age}})\ast {\rm{TN}}-{\rm{3.12}}$$
(2)

where CH4 formation is expressed in µmol g(dw)−1 d−1, TN is expressed in mass %, and age is expressed in years.

The back-transformation of ln-transformed predictions renders residuals that are heavily skewed, leading to underestimation of predicted mean CH4 formation rate35. Therefore we corrected the predicted CH4 formation rate following36:

$${{\rm{CH}}}_{4}\,{{\rm{formation}}}_{{\rm{corr}}}=\exp (\mathrm{ln}\,{{\rm{CH}}}_{4}\,{\rm{formation}}+0.5\ast {{\rm{s}}}^{2})$$
(3)
$${{{\rm{s}}}^{2}}_{{\rm{corr}}}={({{\rm{CH}}}_{4{\rm{corr}}}{\rm{formation}})}^{2}\ast \exp ({{\rm{s}}}^{2})-1)$$
(4)

where CH4 formation is the ln transformed predicted CH4 formation rate and s2 = 0.28 is a residual variance of the model. CH4 formation corr is the corrected CH4 formation rate and s2 corr is the corrected variance.

## Discussion

We report here the first model of CH4 formation in freshwater sediment that is applicable at relevant timescales (Fig. 3). Accordingly, CH4 formation rates stretching over more than 2 orders of magnitude could be predicted for sediments that cover a wide gradient in sediment characteristics (Table 1) and span over decades in age. Sediment age (i.e. time since sediment deposition) was an important predictor of CH4 formation (Supplementary Table 2), which can be seen in the exponential decrease of CH4 formation as the sediment got older, both in sub-surface sediment samples during our 739 days of experiment (Fig. 1), and when comparing sediment layers along a decadal age gradient (Fig. 2). We note that age was important in spite of the random noise inflicted by the unknown year-to year variability in sediment accumulation on the age estimate of individual sediment layers. The sediment TN concentration was the other important predictor of the CH4 formation model, which explained 81% of the variability in CH4 formation rates (Fig. 3). Even though a significant relationship between TN and CH4 formation in freshwater sediment was previously reported18, no relation to sediment age has been made beyond the timescales of incubation experiments (e.g.18). Contrarily, our analysis revealed that not only the TN concentration was important for CH4 formation, but also the change of TN over sediment age (i.e. the interaction term age * TN; Eq. 2). We suggest that the importance of TN may be related to the amount, source and diagenetic state of the sediment OM. First, we found that in our dataset, the sediment TN concentration was also positively related to the TC concentration (TN = 0.067 * TC + 0.077, p < 2.2−16, R2 = 0.90, not shown). Second, the TN concentration in the sediment is representative of the OM source37: autochthonous OM is comparatively rich in protein and poor in cellulose (i.e. high in TN), while allochthonous OM is protein-poor and rich in lignocellulose (i.e. low in TN). It was previously shown that N-rich autochthonous OM is more biodegradable38 and produces more CH4 than the degradation of allochthonous OM14 (even if very fresh allochthonous OM can also give rise to substantial CH4 production16), explaining the relationship between sediment TN for CH4 production (Eq. 2;18). Third, N is preferentially consumed during microbial degradation of sediment OM (e.g.31), meaning that over time, the sediment OM becomes poorer in TN and less reactive, in accordance with our finding that the interactive term (age * TN) was significant in the model. In fact, also sediment age is a proxy of OM reactivity, given the steep exponential decrease of the sediment OC decay with increasing sediment age39. The model presented here for the first time quantifies the influence of OM input (amount and source) and its diagenetic change, approximated by age, TN concentration and their interaction, on CH4 formation rates from sediment OM over decal timescales. In addition, the model is derived from tropical reservoirs of different trophic status and biomes, and in spite of the considerable variability of CH4 formation between reservoirs, all predicted CH4 formation rates for all cores from all reservoirs converge on one line (Fig. 3). For this reason, it seems likely that this relationship is valid in tropical reservoirs in general, and that CH4 formation can be predicted from TN and age, factors that can be relatively easily measured or approximated. Importantly, the rates presented here were derived at a standardized temperature of 25 °C. In order to derive in-situ CH4 rates, our model needs to be used in conjunction with the well-documented temperature dependence of CH4 production40. Also, before applying it to other systems, or beyond the data domain from which it was constructed, the model first needs to be verified.

CH4 formation decreased over age in all sediment cores, however, the CH4 formation never reached zero, not even at over 40 years of age. Despite that CH4 formation slows down rapidly after sediment is deposited, CH4 continues to be produced in older sediment layers. For this reason, it is strictly speaking not warranted to speak of OC burial in reservoir sediment, simply because it is in many cases relatively young and thus still being degraded. However, we may determine the age at which CH4 formation asymptotically reaches a stable and low ‘background’ rate. Sediment younger than that may still be considered to degrade significantly, while sediment older than this threshold may be considered largely stabilized, i.e characterized by low and stable CH4 formation (see Methods). This age may therefore serve as an operationally defined age beyond which sediment OC may be considered “buried”, i.e. not degraded to an extent that is quantitatively significant for the standing stock of OC. The age of transition to low background CH4 formation calculated by this approach was about 11 years in CDU and 12 years in CUN and 6 years in FUN (Table 2), which corresponds to 13, 9 and 23 cm of sediment depth in CDU, CUN and FUN respectively. Assuming a 100 year lifetime for these reservoirs, around 38% (Table 2) of the total time-integrated sediment CH4 formation would take place in sediment layers that are beyond the transition age threshold and that have reached a low and stable background CH4 formation. In other words, sediment layers that we often consider biologically inactive might significantly contribute to the overall sediment CH4 production.

Despite a significant contribution of older sediment layers to CH4 formation (with an average measured CH4 formation in sediment older than the transition age for 179 degree criteria 1.2 ± 0.6 µmol gC−1 d−1 in CDU, 0.5 ± 0.2 µmol gC−1 d−1 in CUN and 0.8 ± 0.1 µmol gC−1 d−1 in FUN), the amount of C that is degraded through the CH4 formation is quantitatively negligible compared to typical carbon burial rates. According to our measured CH4 formation rates in sediment older than the transition age, on average 0.3 ± 0.2% of the C contained in a sediment layer is transformed into CH4 in a year.

We used the C:N ratio of the surface sediment to evaluate if the source of the sediment deposited onto the sediment surface has an effect on the transition age (Fig. 4). We found a significant increase in the transition age with increase in C:N ratio (p = 0.012, R2 = 0.34, n = 17). This indicates that autochthonous OM (low C:N) is relatively easily decomposed and readily transformed into CH416, but once the labile substances are decomposed the degradation rate declines steeply (Fig. 2). Allochthonous OM (high C:N) on the other side breaks down more slowly than autochthonous OM, and the slow anaerobic decay of refractory fraction of allochthonous OM can provide a continuous source of CH4 in reservoir sediment over the reservoir lifetime.

### Implications

Our study shows that the CH4 formation rate of reservoir sediments is predictable although the fluxes may further vary with sediment temperature and dissolved oxygen concentration40. As CH4 emission from reservoirs is very difficult to measure representatively due to the strong spatial and temporal variability of CH4 emission via diffusion19, and particularly via ebullition41,42, the sediment CH4 formation model (Eq. 2, Fig. 3) represents the first step towards constraining reservoir CH4 emission from sediment characteristics.

Further, our study indicates that the high CH4 emission from eutrophic reservoirs4 may stem primarily from CH4 formation in young sediment layers (e.g. FUN, <6 years old; Table 2). Hence, reservoir management strategies that decrease nutrient inputs to limit autochthonous primary production could within less than a decade lead to a reduction in reservoir CH4 emission to the atmosphere.

Lastly, the current hydropower boom in tropical areas43 carries a risk of creating new reservoirs that are large CH4 emitters: the high potential for soil erosion in the tropics44,45 implies that new tropical reservoirs may have high sedimentation rates and thus high supply of young and reactive OM that stimulates CH4 formation in the reservoir sediment (Fig. 2). This effect is aggravated in case nutrient management strategies are lacking and excess nutrient loads stimulate productivity, since the transformation of autochthonous OC to CH4 in reservoir sediment results in an anthropogenic enhancement of atmospheric radiative forcing46.