Global warming is accelerating and the world urgently needs a shift to clean and renewable energy. Hydropower is currently the largest renewable source of electricity, but its contribution to climate change mitigation is not yet fully understood. Hydroelectric reservoirs are a source of biogenic greenhouse gases and in individual cases can reach the same emission rates as thermal power plants. Little is known about the severity of their emissions at the global scale. Here we show that the carbon footprint of hydropower is far higher than previously assumed, with a global average of 173 kg CO2 and 2.95 kg CH4 emitted per MWh of electricity produced. This results in a combined average carbon footprint of 273 kg CO2e/MWh when using the global warming potential over a time horizon of 100 years (GWP100). Nonetheless, this is still below that of fossil energy sources without the use of carbon capture and sequestration technologies. We identified the dams most promising for capturing methane for use as alternative energy source. The spread among the ~1500 hydropower plants analysed in this study is large and highlights the importance of case-by-case examinations.
Citation: Scherer L, Pfister S (2016) Hydropower's Biogenic Carbon Footprint. PLoS ONE 11(9): e0161947. https://doi.org/10.1371/journal.pone.0161947
Editor: Juergen P. Kropp, Potsdam Institute for Climate Impact Research, GERMANY
Received: March 24, 2016; Accepted: August 15, 2016; Published: September 14, 2016
Copyright: © 2016 Scherer, Pfister. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Data are available as supporting information of the manuscript.
Funding: The authors received no specific funding for this work.
Competing interests: The authors have declared that no competing interests exist.
The annual emission rates of carbon dioxide (CO2) and methane (CH4), and the consequent global temperature increase, are accelerating rapidly . To meet the target of limiting the average temperature increase to 2°C, emission reductions are urgently needed. Even in the case that emissions are halved in 2050 compared to 1990, there is still a 29% chance that the target is missed . A portfolio of strategies, coined the stabilization wedges, is available in order to limit emissions using existing technologies. These strategies include energy efficiency increases, carbon capture and storage (CSS), alternative (nuclear and renewable) energy sources and forest and agricultural conservation . Among these, hydropower is considered a low-carbon technology helping to mitigate climate change . However, it was ascertained that the carbon footprint of some hydropower plants is even larger than that of thermal power plants . Emissions are particularly high in tropical regions [6,7] and at high area-to-electricity ratios [5,8].
The emissions from hydroelectric reservoirs arise from the decomposition of organic matter that was either flooded during reservoir construction, transferred to the reservoir by river runoff, grown in the reservoir such as by algal production , stems from dead trees protruding from the water , or was grown in newly created marshes in the drawdown area . Besides reservoirs, also other ecosystems influence the greenhouse gas (GHG) fluxes. While rivers and lakes emit GHGs, forests, peatlands and wetlands rather bind them. The real impact of the reservoir is therefore the difference between GHG emissions before and after flooding. The net emissions can either be larger than the gross emissions when carbon sinks are lost or smaller when emissions have already occurred [6,11]. In the investigated cases, the sink  or source  prior to flooding is small compared to the emissions afterwards so that, when also taking into account the uncertainties, it can be considered as carbon-neutral in non-tropical  as well as tropical regions .
Barros et al.  compiled data of about 100 CO2 and CH4 emission measurements, as well as some reservoir and site characteristics, from published literature. Based on this data, he estimated total emissions of hydroelectric reservoirs and analysed the relationship with explanatory variables. Hertwich  supplemented the dataset with electricity information and an additional predictor variable to carry out similar analyses. This study sets up a new statistical model, including additional environmental variables, and applies this model to a global dataset of ~1500 plants (~43% of global hydropower generation) in order to get a better estimate of hydropower’s total climate change impact and its range among individual plants.
Materials and Methods
We analysed the carbon footprints of 1473 hydroelectric reservoirs across 104 countries, under the assumption that the emissions prior to flooding are negligible and do not significantly change the net footprint . Reservoir locations, its purpose and key characteristics were given in the Global Reservoir and Dam (GRanD) database . This information was linked to annual electricity generation provided by the CARMA database . The analyses represent the year 2009.
CO2 and CH4 emissions, electricity information and site characteristics as potential predictors of about 100 hydroelectric reservoirs were compiled by Barros et al.  and Hertwich  (S1 File). This data serves as a training dataset for statistical modelling from which we estimate the emissions of the above mentioned 1473 reservoirs.
We tested various variables as predictors including area, area-to-electricity ratio, age (derived from the year when the dam construction was completed), volume, volume-to-area ratio (as proxy for depth), mean, minimum and maximum temperature , net primary productivity , topsoil organic carbon content  and erosion rate  as well as their logarithms. All predictors are available in global datasets; however, the dissolved organic carbon (DOC) used as a predictor by Barros et al.  was not available as a global dataset and was therefore excluded from our analysis. The erosion rate serves as a proxy indicator thereof. If the year of dam completion was missing, various online sources were consulted including the AQUASTAT dam database .
Generalized linear modelling
We estimated carbon emissions per energy unit (kg/MWh, Model 1) and as areal fluxes (mg C m-2 d-1, Model 2) using generalized linear models (GLMs) and combined the two models by averaging (Model A). A Gaussian distribution was assumed for CO2 and a logarithmic distribution for CH4. In a first step of selecting predictors, explanatory variables were excluded if they correlated with other predictors (correlation coefficient R ≥ 0.5), but have a weaker effect on the model output than the alternative predictor. In a second step, we applied multi-model inference  during which competing models with different predictor combinations and numbers of predictors were compared. The analysis was carried out in R using the package MuMIn  in which models are compared based on Akaike weights, a transformation of Akaike’s information criterion (AIC) . The final models were evaluated by multiple criteria in order to ensure a more robust assessment . The criteria used were deviance explained (DEX), also called adjusted R2, AIC, the percent bias (PBIAS), the normalized root mean square error (NRMSE) and the normalized mean absolute error (NMAE). The DEX and AIC reward a good model fit, but penalize a large number of predictors, which potentially leads to overfitting. Similar to the coefficient of determination (R2), a higher value of DEX closer to 1 indicates superior model performance. By contrast, a smaller value of AIC implies better model performance. The remaining criteria only evaluate the performance. Previous models using the same training dataset [7,8] were rebuilt and the performance was compared to our own efforts.
The measurements in the training dataset are likely to contain errors: 1) for CO2 due to negligence of carbon burial and 2) for CH4 due to negligence of methane ebullition (bubbling) or sampling errors. Therefore, correction factors were derived and it was assumed that 1) the predicted carbon footprints should be 13% lower and these excess emissions should be deducted from CO2, and 2) predicted CH4 emissions are underestimated and should be increased by a factor of 1.4 (Model AC, S2 File).
The GLMs were subsequently applied to the global dataset of 1473 hydroelectric reservoirs for which the predictors were extracted from the same global datasets as the training dataset. CO2 and CH4 emissions were aggregated to carbon dioxide equivalents (CO2e) by assuming a global warming potential of 34 kg CO2e/kg CH4 over a time horizon of 100 years (GWP100) and for sensitivity analyses also assuming a global warming potential of 86 kg CO2e/kg CH4 over a time horizon of 20 years (GWP20) . Although the global warming potential of biogenic CO2 can differ from fossil CO2, they were assumed to be equal, as the flooded biomass is not replanted and therefore no significant removal of emissions during regrowth can take place . Potential sequestration of CO2 in algae and fish biomass of dams is supposed to be marginal compared to terrestrial vegetation (S3 File), but might be evaluated more specifically in future research. First, we calculated production-weighted average emissions per energy unit and then we derived total emissions by multiplying these with the global hydroelectricity generation in 2009, which amounted to 3551 TWh . In addition, we calculated the median and maximum of the reservoirs’ carbon emissions. Plants with a high potential for energy recovery from methane emissions (Model AC) were defined as plants that produce at least 10 kg CH4/MWh and contribute to at least 50% to the carbon footprint. All results are presented in more detail in the supporting information (S1 Table).
Since reservoirs often fulfil multiple purposes, the environmental impacts should not solely be attributed to hydropower. Instead, the responsibility should be shared among the purposes of the reservoir. Therefore, allocation factors (fA) were applied. These factors are based on the ranking of hydropower among all of the n purposes that the reservoir fulfils : (1)
Model set-up and evaluation
We set up generalized linear models (GLMs) for estimating CO2 and CH4 emissions based on a training dataset of ~100 hydroelectric reservoirs [7,8]. The best model (Model 1) for CO2 emissions per energy unit (kg CO2/MWh) includes only two predictors: area-to-electricity ratio (ATE, km2/GWh) and area (A, km2), while the best model for CH4 emissions (kg CH4/MWh) used age (AGE, years), ATE and maximum temperature (TMX, °C): (2) (3)
As an alternative, we calculate areal fluxes (Model 2) of GHG emissions (mg C m-2 d-1), which are independent of the energy production and can also be applied to reservoirs with purposes other than hydropower. In the optimal GLM, the model includes the predictors AGE and the erosion rate (ERR, t ha-1 a-1) for CO2, and additionally A and TMX for CH4: (4) (5)
As already pointed out by Hertwich , the ATE is the single most important predictor for emissions per energy unit. Besides river discharge, the hydroelectricity output is largely determined by the dam height , which, in turn, depends strongly on the topography. Mountainous regions with narrow deep river channels require less land and are therefore preferable sites for reservoirs [29,30].
TMX can be interpreted as a proxy indicator to distinguish tropical from non-tropical reservoirs. The higher the TMX, the more likely the reservoir is situated in a tropical region and therefore the higher the CH4emissions. This finding is in accordance to previous studies that reported greater emissions, especially of CH4, from tropical reservoirs [6,7].
While CH4 emissions decline with reservoir age, no significant relationship was found between CO2 emissions and age when setting up the model for GHG emissions per MWh. However, Barros et al.  and Hertwich  included an inverse relationship between age and both types of emissions and it has been included in our alternative approach, with higher weight for CO2 (Eq 4). Some previous studies confirmed the decrease with age for CO2 [6,7,11] and CH4 . By contrast, other studies could not ascertain a relationship between CH4 and age  or even found an increasing trend during initial years with a levelling-off at a later stage . It has to be noted that our results reflect the analysis across hydropower plants as a result of the training dataset and is not resulting from time series analyses of specific power plants. Since the production of CH4 strongly depends on the flooded ecosystem type and measurement procedures can deviate (especially with regards to the inclusion of bubbling in CH4 measurements) , it is difficult to predict the relationship of CH4 emissions to age, and this might explain the differences among studies.
The erosion rate was shown to be a significant predictor for modelling areal fluxes of CO2. It is a proxy indicator for biomass continually transported by the river discharge to the reservoir as opposed to the biomass from the initial flooding. How erosion influences CH4 fluxes is unclear and it remains to be investigated if the small weight of the predictor is a statistical artefact or if there is a weak causal relationship on a global level.
The model of emissions per energy unit better predicts the measurements than the model of areal fluxes. The same could be observed by comparing the model performance of Barros et al.  for areal fluxes and of Hertwich  for emissions per energy unit. As indicated by multiple performance criteria, the newly developed models outperform those of Barros et al.  and Hertwich  with the exception of the model predicting areal CO2 fluxes, which performs comparably to the previous model (Table 1). However, by including the ERR and TMX, the models take into account additional environmental conditions relevant for the generation of greenhouse gases, whereas the latitude used as predictor by Barros et al.  is not a physical parameter in itself, but might represent the temperature gradient.
Model application to a global dataset
We applied the calibrated model to a dataset of 1473 hydropower plants across the world (Fig 1). Our estimate of total carbon emissions exceeds previous estimates by far, due to large CO2 emissions. By contrast, our CH4 emissions are close to previous estimates. They contribute ~42% to the carbon footprint (Table 2) and therefore do not dominate the footprint results unless a shorter time horizon of 20 years is considered for calculating the GWP. The higher global CO2 emissions cannot be explained by outliers, as our maximum value is in the same order of magnitude as the measured maximum (Table 3). Despite the overall higher emission estimates, some plants represent small carbon sinks (Fig 1). Allocation of carbon footprints between multiple purposes of a reservoir leads, on average, to a reduction of ~30% in climate change impacts associated with hydropower (Table 3).
Carbon footprints of hydropower plants across the world (a) and hydropower plants with high methane emissions (≥ 10 kg CH4/MWh) and a large share of methane emissions (≥ 50% of the carbon footprint) (b). Country boundaries are obtained from Natural Earth (http://www.naturalearthdata.com/).
When allocating the impacts between hydropower and other uses, plants with the largest hydroelectricity generation of each continent are below the production-weighted global average emissions, with the exception of the Churchill Falls plant. The high (although not extreme) footprint of the Churchill Fall plant can be attributed partly to the high area-to-electricity ratio and partly to being a single purpose reservoir, not allowing for allocation. Still, each of these largest plants, except Sysenvatnet in Norway, exceed the median (Table 4), indicating that the average footprint is driven by a few hydroelectric power plants with very high emissions.
Climate change impact or renewable energy source?
Although the CH4 stocks in tropical reservoirs might contribute to global warming and climate change, they also constitute a potential renewable energy source. The gas can be oxidized in situ or recovered for later energy production with a recovery efficiency of 60% by using low-cost technologies . Using biogenic CH4 otherwise emitted to the atmosphere as an energy source is not only cleaner than using the equivalent volume of fossil natural gas, but it also reduces the need for additional hydroelectric reservoirs. This entails far more benefits than reducing the climate change impacts modelled here, as it avoids construction of new reservoirs that lead to resettlements, habitat destruction , evaporation and flow alterations . Since the biogeochemical dynamics differ among reservoirs, the potential for climate change mitigation or energy recovery has to be examined case by case . Where energy technologies other than hydropower are more likely to be phased out, different benefits are provided, which might reduce climate change impacts even more.
While on a global level CH4 recovery does not seem to be a promising strategy considering the relatively low efficiency of CH4 capture combined with a low contribution of CH4 to the total carbon footprint of reservoirs, it might be suitable in some locations. We identified 187 dams with a high CH4 recovery potential, located mainly in the United States, India and West Africa (Fig 1). If methane emissions from these dams would be captured with the assumed efficiency of 60% , 19% of total methane emissions could be saved, which would reduce the overall carbon footprint of global hydropower by 8%.
Biogenic carbon emissions from hydropower reservoirs are far higher than previously assumed. Consequently, our results question the sustainability that is often associated with hydropower. Although the carbon footprint of hydropower exceeds that of all other renewable energy sources and that of fossil energy sources combined with carbon capture and storage (CSS), it is on average about half the footprint reported for conventional fossil energy sources [32,33] (Fig 2). The emissions vary greatly among plants and the relationship with the reservoir age is not yet well understood, as demonstrated by the contradicting reports addressed above. In addition, uncertainties of estimates remain high, as the comparison of the two approaches (per energy unit and areal fluxes) reveals, with a production-weighted average coefficient of variation (CV) of 57%, 43% and 29% for CO2, CH4 and CO2e emissions, respectively. This highlights the need for a more extensive monitoring network covering diverse ecosystems, repeated measurements over a longer observation period (at least a decade) as well as standardized measurement procedures taking into account carbon burial , drawdown areas  and methane bubbles . The dam construction is typically not relevant for the total carbon footprint with emissions of ~19 CO2e/MWh .
The lower and upper value of the dark bar for hydropower are the lower and upper quartiles for the corrected model average (Model AC). The light extensions represent the 10 and 90% quantiles and the red diamond marks the median.
Besides potential environmental damages, reservoirs also have economic and social impacts. Therefore, multiple criteria have to be examined in order to determine the optimal location for a reservoir . Climate change is, although crucial, only one aspect in such a set of criteria. The statistical model set up in this study serves to estimate the carbon footprint of potential new sites prior to reservoir construction. It also identifies those dams where methane emissions are relevant and biogas electricity could be combined with hydropower. Finally, while hydropower has significant GHG emissions, it can also serve as highly efficient energy storage (pumped storage power plant) of alternative energy sources, such as solar or wind power, which can reduce overall emissions if properly combined.
S1 File. Representation of climate zones in the model training dataset.
S2 File. Derivation of correction factors (methane ebullition and carbon burial).
S3 File. Comparison of biomass density in forests and lakes / reservoirs.
We thank Stefanie Hellweg for helpful comments on the manuscript and Christie Walker for proofreading it.
- Conceptualization: LS SP.
- Formal analysis: LS.
- Methodology: LS.
- Writing – original draft: LS.
- Writing – review & editing: LS SP.
- 1. Smith SJ, Edmonds J, Hartin CA, Mundra A, Calvin K. Near-term acceleration in the rate of temperature change. Nat Clim Change. 2015; 5: 333–336.
- 2. Meinshausen M, Meinshausen N, Hare W, Raper SCB, Frieler K, Knutti R, et al. Greenhouse-gas emission targets for limiting global warming to 2°C. Nature. 2009; 458: 1158–1162. pmid:19407799
- 3. Pacala S, Socolow R. Stabilization Wedges: Solving the Climate Problem for the Next 50 Years with Current Technologies. Science. 2004; 305: 968–972. pmid:15310891
- 4. Yüksel I. Hydropower for sustainable water and energy development. Renew Sustain Energy Rev. 2010; 14: 462–469.
- 5. Santos MA dos, Rosa LP, Sikar B, Sikar E, Santos EO dos. Gross greenhouse gas fluxes from hydro-power reservoir compared to thermo-power plants. Energy Policy. 2006; 34: 481–488.
- 6. Louis VL St., Kelly CA, Duchemin É, Rudd JWM, Rosenberg DM. Reservoir Surfaces as Sources of Greenhouse Gases to the Atmosphere. A Global Estimate. BioScience. 2000; 50: 766–775.
- 7. Barros N, Cole JJ, Tranvik LJ, Prairie YT, Bastviken D, Huszar VLM, et al. Carbon emission from hydroelectric reservoirs linked to reservoir age and latitude. Nat Geosci. 2011; 4: 593–596.
- 8. Hertwich EG. Addressing Biogenic Greenhouse Gas Emissions from Hydropower in LCA. Environ Sci Technol. 2013; 47: 9604–9611. pmid:23909506
- 9. Fearnside PM, Pueyo S. Greenhouse-gas emissions from tropical dams. Nat Clim Change. 2012; 2: 382–384.
- 10. Chen H, Wu Y, Yuan X, Gao Y, Wu N, Zhu D. Methane emissions from newly created marshes in the drawdown area of the Three Gorges Reservoir. J Geophys Res. 2009; 114: n/a-n/a.
- 11. Teodoru CR, Bastien J, Bonneville M, Giorgio PA, Demarty M, Garneau M, et al. The net carbon footprint of a newly created boreal hydroelectric reservoir. Glob Biogeochem Cycles. 2012; 26.
- 12. Richey JE, Melack JM, Aufdenkampe AK, Ballester VM, Hess LL. Outgassing from Amazonian rivers and wetlands as a large tropical source of atmospheric CO2. Nature. 2002; 416: 617–620. pmid:11948346
- 13. Lehner B, Reidy Liermann C, Revenga C, Vorosmarty C, Fekete B, Crouzet P, et al. High-Resolution Mapping of the World's Reservoirs and Dams for Sustainable River-Flow Management. Front Ecol Environ. 2011; 9: 494–502.
- 14. CARMA. CO2 emissions, energy and intensity of power plants; 2015. Available: http://carma.org/dig.
- 15. Hijmans RJ, Cameron SE, Parra JL, Jones PG, Jarvis A. Very high resolution interpolated climate surfaces for global land areas. Int J Climatol. 2005; 25: 1965–1978.
- 16. Haberl H, Erb KH, Krausmann F, Gaube V, Bondeau A, Plutzar C, et al. Quantifying and mapping the human appropriation of net primary production in earth's terrestrial ecosystems. Proc Natl Acad Sci USA. 2007; 104: 12942–12947. pmid:17616580
- 17. Nachtergaele F, Velthuizen H van, Verelst L, Wiberg D. Harmonized world soil database. Version 1.2. Rome: FAO; 2012.
- 18. Scherer L, Pfister S. Modelling spatially explicit impacts from phosphorus emissions in agriculture. Int J Life Cycle Assess. 2015; 20: 785–795.
- 19. FAO. AQUASTAT dams; 2016. Available: http://www.fao.org/nr/water/aquastat/dams/index.stm. Accessed 23 February 2016.
- 20. Johnson JB, Omland KS. Model selection in ecology and evolution. Trends Ecol Evol. 2004; 19: 101–108. pmid:16701236
- 21. Barton K. MuMIn: Multi-Model Inference. R package version 1.15.6; 2016. Available: https://CRAN.R-project.org/package=MuMIn.
- 22. Wagenmakers E, Farrell S. AIC model selection using Akaike weights. Psychon B Rev. 2004; 11: 192–196.
- 23. Scherer L, Venkatesh A, Karuppiah R, Pfister S. Large-Scale Hydrological Modeling for Calculating Water Stress Indices: Implications of Improved Spatiotemporal Resolution, Surface-Groundwater Differentiation, and Uncertainty Characterization. Environ Sci Technol. 2015; 49: 4971–4979. pmid:25822378
- 24. Myhre G, Shindell D, Bréon F, Collins W, Fuglestvedt J, Huang J, et al. Anthropogenic and natural radiative forcing. In: Stocker TF, Qin D, Plattner G, Tignor M, Allen SK, et al., editors. Climate Change 2013. The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press; 2013.
- 25. Cherubini F, Peters GP, Berntsen T, Strømman AH, Hertwich E. CO2 emissions from biomass combustion for bioenergy. Atmospheric decay and contribution to global warming. Glob Chang Biol Bioenergy. 2011; 3: 413–426.
- 26. Kumar A, Schei T, Ahenkorah A, Rodriguez RC, Devernay J, Freitas M, et al. Hydropower. In: Edenhofer O, Pichs-Madruga R, Sokona Y, Seyboth K, Matschoss P, et al., editors. IPCC Special Report on Renewable Energy Sources and Climate Change Mitigation. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press; 2011.
- 27. Scherer L, Pfister S. Global water footprint assessment of hydropower. Renew Energy. 2016; 99: 711–720.
- 28. Paish O. Small hydro power. Technology and current status. Renew Sustain Energy Rev. 2002; 6: 537–556.
- 29. Takeuchi K. On the scale diseconomy of large reservoirs in land occupation. In: Rosbjerg D, editor. Sustainability of water resources under increasing uncertainty. IAHS Publ. no. 240: IAHS; 1997.
- 30. Baban SMJ, Wan-Yusof K. Modelling Optimum Sites for Locating Reservoirs in Tropical Environments. Water Resour Manage; 17: 1–17.
- 31. Bambace LAW, Ramos FM, Lima I, Rosa RR. Mitigation and recovery of methane emissions from tropical hydroelectric dams. Energy. 2007; 32: 1038–1046.
- 32. Turconi R, Boldrin A, Astrup T. Life cycle assessment (LCA) of electricity generation technologies: Overview, comparability and limitations. Renew Sustain Energy Rev. 2013; 28: 555–565.
- 33. Schlömer S, Bruckner T, Fulton L, Hertwich E, McKinnon A, Perczyk D, et al. Annex III. Technology-specific cost and performance parameters. In: Edenhofer O, Pichs-Madruga R, Sokona Y, Farahani E, Kadner S, et al., editors. Climate Change 2014. Mitigation of Climate Change. Contribution of Working Group III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge, United Kingdom and New York, NY, USA: Cambridge University Press; 2014.
- 34. Mendonca R, Kosten S, Sobek S, Barros N, Cole JJ, Tranvik L, et al. Hydroelectric carbon sequestration. Nat Geosci. 2012; 5: 838–840.
- 35. DelSontro T, Kunz MJ, Kempter T, Wüest A, Wehrli B, Senn DB. Spatial Heterogeneity of Methane Ebullition in a Large Tropical Reservoir. Environ Sci Technol. 2011; 45: 9866–9873. pmid:21985534