Improving the accuracy of electricity carbon footprint: Estimation of hydroelectric reservoir greenhouse gas emissions

location, age of impoundment, and watershed properties. The objective of this article is to compare different approaches to estimate hydropower reservoir emissions in LCA, to select the most appropriate one, and to apply it to the calculation of the carbon footprint of electricity distributed in the Canadian province of Qu ´ ebec. Net biogenic GHG emissions of all hydropower reservoirs in the province (with 2.5 and 97.5% confidence intervals), as estimated using the G-res model, are 16.5 (14.7 – 18.6) gCO 2 ∙ kWh (cid:0) 1 and 0.29 (0.23 – 0.35) gCH 4 ∙ kWh (cid:0) 1 . Combined to ecoinvent data for other life cycle emissions, the carbon footprint of electricity distributed in the province in 2017 is 34.5 gCO 2 eq ∙ kWh (cid:0) 1 .


Introduction
Electricity and heat production were responsible for 30% of the world's anthropogenic greenhouse gas (GHG) emissions in 2016, mainly due to fossil fuel combustion [1]. Producing electricity from low-carbon energy sources is thus seen as a solution having a high climate change mitigation potential. For instance, life cycle GHG emissions for wind, hydropower, concentrating solar power, and solar photovoltaics range between 5 and 50 gCO 2 eq•kWh − 1 according to a review study performed by the National Renewable Energy Laboratory, while they range from 480 to 1000 gCO 2 eq•kWh − 1 for thermal natural gas, oil, and coal [2]. Hydropower is currently the largest source of renewable energy and is a low-carbon electricity source. It contributed to 16.4% of the world's electricity production in 2017, and a steady growth is projected in the future [3,4]. Hydroelectricity is produced from the energy of flowing water, which does not lead to direct GHG emissions, unlike producing electricity from fossil fuel combustion. Although, when dam construction causes the flooding of land, the overall carbon balance will be affected, generally resulting in net biogenic carbon dioxide (CO 2 ), methane (CH 4 ) and nitrous oxide (N 2 O) emissions from the degradation of biomass found in these newly created reservoirs [5][6][7]. To estimate biogenic GHG emissions associated with hydroelectricity production, reliable measurements of CO 2 , CH 4 and N 2 O emissions from reservoirs over space and time are needed, as well as robust estimates of carbon sinks and sources from the terrestrial and aquatic ecosystems prior to flooding [7]. Net biogenic carbon emissions (commonly called Net GHG Emissions) to be attributed to hydroelectricity production are estimated by the difference between pre-and post-impoundment carbon fluxes, representing respectively the emissions of the landscape before impoundment and the new emissions associated with the reservoir [8].
Several mechanisms are involved in the carbon cycle of freshwater ecosystems. Indeed, freshwater ecosystems receive carbon from terrestrial ecosystems through drainage, sequester carbon through primary production, bury carbon in sediments, emit carbon from biomass degradation and respiration, and transport carbon downstream up to oceans [9,10]. Human activities in the land surrounding the reservoir may also result in additional GHG emissions from freshwater ecosystems through sewage and agricultural pollution [11]. Dams can affect the natural carbon cycle of freshwater ecosystems through the flooding of terrestrial vegetation and soils, which could result into additional carbon emissions, especially during the early years following the creation of the reservoir. Over time, flooded organic matter will slowly decompose according to local conditions, and emissions will tend to decrease [8,12]. The impoundment may also increase sedimentation and decay in reservoirs due to longer water residence times, potentially leading to higher CO 2 and CH 4 emissions [13]. However not yet fully documented and rarely included in studies, the change in hydrology regime will also displace where the carbon is processed, leading to high emissions observed in the reservoir that would have occurred anyway further down the water continuum [8]. N 2 O is another greenhouse gas that can be emitted from reservoirs. However, studies have shown no difference in terms of N 2 O emissions for boreal reservoirs compared to natural aquatic ecosystems [11,12,14]; they will therefore be excluded from this study. Moreover, the Intergovernmental Panel on Climate Change (IPCC) considers that large sources of nitrogen are related to human activities taking place in the watershed upstream of the reservoir. Therefore, to avoid double counting, they are not considered for hydropower reservoirs.
Ideally, the estimation of pre-and post-impoundment GHG emissions would consider all these mechanisms occurring in the reservoir, as well as downstream. As these emissions have strong spatial and temporal variability, which makes measurement challenging [8,12,15,16], an international project has been conducted to provide consensual guidelines to help scientists estimating net GHG emissions from reservoirs in a standardized and robust way [17]. Different types of GHG fluxes must be estimated or measured according to these guidelines: 1) bubbling (ebullition) emissions, 2) diffusive emissions from the reservoir water surface, 3) diffusion through aquatic plant stems, 4) degassing at the reservoir outlet (immediately after water passes through turbines), and 5) diffusive emissions further downstream. Depending on the location (e.g. climate) and characteristics of the reservoir (e.g. depth, vegetation cover), these types of emission can be more or less important. For instance, diffusion through aquatic plant stems is not really observed in reservoirs located in the province of Québec as most of their shorelines are eroded by the combined action of wave and wind to mineral horizon and bed rock [12].
Bubbling emissions come mainly from CH 4 accumulating in sediments following anaerobic degradation, and usually occur in shallow parts of the reservoirs where the hydrostatic pressure is lower [8,12,15]. Bubbling emissions are intermittent and more important in warm waters [18] containing high levels of organic matter [15]. They can be quantified using different techniques, such as inverted-funnel method or acoustic techniques [17,19]. Diffusive CO 2 and CH 4 fluxes at the reservoir water surface can be quantified using surface floating chambers, or calculated based on the partial pressure gradient and using the thin boundary layer diffusive process model [8,12,16,20]. Degassing emissions are caused by the important pressure change at the outlet of turbines and spillways. They can be quantified using gas concentrations directly upstream and downstream the dam [17]. Downstream diffusive fluxes are more difficult to quantify because of currents and rapid flowing waters, as opposed to reservoirs. The main technique used consists in calculating these fluxes based on measurements of gas concentrations [8,12,16,20]. Finally, pre-impoundment fluxes from the terrestrial ecosystem can be measured using chambers, soil core sampling, or eddy covariance towers [7,21].
The life cycle assessment (LCA) methodology [22,23] is widely used to calculate the carbon footprint of different electricity production pathways, while considering all life cycle stages, such as construction of infrastructures, fuel production, or electricity generation. Life cycle carbon footprint results can be used to compare the climate change impacts associated with different electricity sources. They can also be integrated in other LCA studies in which electricity is consumed. As electricity is part of most products' life cycle, the reliability of electricity carbon footprint will inadvertently affect the reliability of most LCA results, and of any decisions made on their basis.
Biogenic GHG emissions from hydropower reservoirs have been poorly considered for a long time in LCA because of the scarcity of data [11,24]. Moreover, it is difficult to extrapolate observations from one reservoir to another since emissions of biogenic CO 2 and CH 4 vary greatly among them, depending on different parameters, such as climate, geographic location, age of impoundment, and watershed properties [6,7,12,25]. Published reviews of LCA studies on hydropower plants have shown great variability in results, from 0.2 to 152 gCO 2 eq•kWh − 1 for [26], and from 1.2 to 3000 gCO 2 eq•kWh − 1 for the more recent [24], mainly due to biogenic reservoir emissions. LCA studies that included these emissions usually estimated only gross emissions (i.e. post-impoundment only) as pre-impoundment data were not available [27]. For instance, pre-impoundment emissions have been found for two hydroelectric reservoirs only, i.e. the Eastmain-1 reservoir in the province of Québec [7], and the Three Gorges reservoir in China [28]. Recent studies have addressed this issue and proposed emission factors to quantify biogenic GHG emissions from hydropower reservoirs in LCA [29,30]. In addition to its contribution to academic research projects, Hydro-Québec, the public utility that manages electricity generation, transmission and distribution in the province, has carried out several field measurements in the past. However, quantification of reservoir emissions is still challenging, as results vary substantially from reservoir to reservoir and need to be better documented. Recently, the G-res model has been developed by the International Hydropower Association and the UNESCO Chair in Global Environmental Change in order to more accurately estimate GHG emissions from hydropower reservoirs [31]. The objective of this article is to compare different approaches to estimate hydropower reservoir emissions in LCA, to perform a critical analysis in order to select the most appropriate one, and to apply the selected approach to the LCA of hydroelectricity production in the Canadian province of Québec in order to calculate the carbon footprint of the electricity mix distributed in the province.

Overview of electricity produced and distributed in Québec
In 2017, hydropower represented 99.8% of the total amount of electricity produced by Hydro-Québec, relying on 63 power plants for a total installed capacity of 36,767 MW [32]. From this installed capacity, 62.4% (22,959 MW) was composed of 20 reservoir-type power plants, while the remaining 37.6% was composed of 43 run-of-the-river-type power plants. The composition of electricity mix distributed in the province is slightly different because of electricity purchase from local producers, as well as imports and exports. In 2017, as shown in Table 1, 94.47% of distributed electricity was from hydropower, while 0.04% was from thermal, 0.18% from nuclear, and 5.31% from other renewable sources [33].
Run-of-the-river power plants are defined as power stations fed directly by a river. They have little to no storage capacity, meaning that their generating output entirely depends on the flow of the river. However, some of the run-of-the river-type power plants might have small associated flooded lands. Table A1 in appendix presents a detailed list of reservoir and run-of-the-river with flooded lands hydropower plants in Québec, while Table A2 presents run-of-the-river hydropower plants without any flooded lands. In addition, Hydro-Québec's hydropower fleet also includes reservoirs (included in Table A1) and other flooded lands that are not directly associated with a power plant (Table A3 in appendix). As an example, Rupert downstream and upstream diversion bays have been created to divert part of the Rupert's flow toward the Eastmain reservoir.
For each approach presented in section 2.2, reservoir CO 2 and CH 4 emissions were calculated in g•kWh − 1 for two different bases: i) per kWh of electricity produced by reservoir and run-of-the-river with flooded lands power plants (first column in Table 5), and ii) per kWh of electricity produced by all hydropower plants (second column in Table 5). For the first case, the denominator is the sum of annual electricity production for all the reservoir and run-of-the-river with flooded lands power plants (Table A.1, 1.42 × 10 11 kWh•yr − 1 ). For the second case, it is the total annual hydroelectricity production from all hydropower plants owned by Hydro-Québec (Table A1 and Table A.2, 1.78 × 10 11 kWh•yr − 1 ). The values of annual electricity production are averages over a 5-year period, i.e. from 2011 to 2015 inclusive, except for three power plants that became operational after 2011 (see Table A1).

Comparison of different approaches to estimate reservoir emissions
The state-of-the-art approach to estimate reservoir emissions is to calculate the cumulative net emission over 100 years, which is considered as a good estimate for the lifetime of a reservoir [34][35][36]. The net emission is the difference between post-impoundment emissions (from bubbling, diffusion at the reservoir surface, degassing, and downstream diffusion) and pre-impoundment emissions. The study on Eastmain-1 reservoir (Québec, Canada) is the only one that has integrated all emission pathways to estimate the net GHG emission [7]. A similar approach is actually taking place at the Romaine complex (Québec, Canada) and the results should be available by 2020-2021. There is a clear need to assess more reservoirs using this approach, as explained in section 3.
However, this approach can rarely be fully applied because of missing data. For instance, pre-impoundment measurements are rarely available, as no measures have been taken prior to impoundment. Data are also often lacking for some types of post-impoundment emissions, such as downstream diffusive emissions, because they are difficult to measure. Another issue is that some natural emissions that were occurring downstream from the dam prior to impoundment might now be observed at the reservoir surface as the residence time in the reservoir has considerably increased [8]. If those natural displaced emissions are ignored because of missing data, there is a risk of overestimating GHG emissions caused by the creation of the dam. One solution often proposed to overcome this issue is to calculate emissions over 10 years, assuming that after this period, emissions are all from natural processes, so that the net emission would be zero. A new approach proposes to use predicted emissions at 100 years as the natural baseline emissions, and to remove them from the post-impoundment emissions [8]. When pre-impoundment emissions are not available, gross emissions (i.e. only post-impoundment emissions) for a period of 10 years could be used as a proxy in the absence of alternatives. The net emission obtained is then divided by total electricity generation over the lifetime to get GHG emissions per kWh of electricity produced.

Gross emissions (10 and 100 years) based on a set of historic measurements
Average CO 2 emissions per kWh, based on a 10-year gross emissions approach, were calculated using Equation (1), as inspired from the 2006 IPCC guidelines, volume 4 (Agriculture, Forestry and Other Land Use), Appendix 2, Equation 2A.1 [37]. Average CO 2 daily diffusive emissions come from historical data, as estimated from more than 11,000 measurements on 24 reservoirs and natural water bodies over the past decades by Hydro-Québec ( Table 2). The total reservoir area is the sum of average areas for all water bodies included in Tables A1 and A3 (1.93 × 10 10 m 2 ). As a sensitivity analysis, gross emissions were also calculated using the sum of minimum and maximum reservoir areas (instead of average) for water bodies in Table A1. For these cases, total reservoir areas were 1.60 × 10 10 m 2 and 2.26 × 10 10 m 2 respectively, and results of this sensitivity analysis are shown in parenthesis in Table 5.
where E gross 10  To calculate 100-year gross emissions, emissions from years 10-100 were added to the brackets at the numerator in Equation (1). To obtain these emissions, the average daily emission as shown on third row in Table 2 (>10 years after flooding) was simply multiplied by 90. Table 2 shows that emissions after 10 years are of the same order of magnitude than emissions from natural lakes and rivers, supporting the hypothesis that a 100-year gross emissions approach probably leads to an overestimation of emissions.

Gross emissions (10 years) based on reservoir-specific measurements
Biogenic GHG emissions vary considerably depending on reservoir characteristics, such as depth or type of soil flooded [12,25]. Therefore, a set of historic reservoir-specific measurements from Hydro-Québec for CO 2 and CH 4 diffusive emissions were used to calculate emissions per kWh for each reservoir using Equation (2). Run-of-the river power plants may also alter the biogenic carbon balance as they may cause some flooding [8]. Therefore, Equation (2) was used for both reservoir and run-of-the-river with flooded land dams. Table A1 in appendix provides raw data and results for 21 water bodies (associated with 25 power plants) for which diffusive emission data were available (no data were available for the remaining 7 power plants). Emissions per kWh were calculated for the average reservoir area, as well as maximum and minimum areas (in parenthesis) as a sensitivity analysis, in order to get results that cover the full operation range.
where E gross 10 specific = average emissions per kWh [gCO 2 or Prod = average annual electricity production [GWh•yr − 1 ] LT = assumed lifetime of reservoirs [100 yr] The weighted average over the 21 water bodies for which data were available (Table A1) was then calculated. Reservoirs and other impoundments that have no direct associated power plants were let out of the calculation, because no reservoir-specific emission data were available. Finally, to get reservoir emissions per kWh of hydroelectricity produced by all hydropower plants (second column in Table 5), the weighted average was multiplied by annual electricity production from reservoir and run-of-the-river with flooded land hydropower plants (1.42 × 10 11 kWh•yr − 1 ), and then divided by annual electricity production from all hydropower plants (1.78 × 10 11 kWh•yr − 1 ).

Net emissions (100 years) from Eastmain-1 reservoir extrapolated to all reservoirs
Teodoru and colleagues performed one of the most comprehensive large-scale assessment of CO 2 and CH 4 emissions associated with the creation of a reservoir (Eastmain-1), including pre-and postimpoundment phases [7]. Therefore, net emissions over a period of 100 years were estimated from this study, and extrapolated to all reservoirs to calculate average emissions per kWh.
Teodoru and colleagues estimated pre-impoundment carbon fluxes (diffusive CO 2 and CH 4 emissions and carbon storage from sedimentation) for three major components of the landscape, i.e. terrestrial (forests and soils), wetlands (fens, bogs, swamps/marshes), and aquatic systems (streams, rivers, lakes). They calculated post-impoundment fluxes for the first four years following impoundment (2006-2009 inclusive) based on measurements, and then proposed an empirical relationship to estimate longer-term carbon fluxes (for 2010 and beyond). Diffusive CO 2 and CH 4 , bubbling CH 4 , carbon storage from sedimentation, and degassing CO 2 and CH 4 fluxes were included. The empirical relationship provides net emissions. Therefore, pre-impoundment emissions should not be removed from the values obtained for 2010 and beyond. Table 3 presents CO 2 and CH 4 fluxes used for the calculation.
Net emissions per surface area were calculated for CO 2 and CH 4 over a period of 100 years using Equation (3).
where E net 100 = net emissions (100-year) per surface area [gCO 2 or E i = emissions of CO 2 or CH 4 for year i as per Table 3 [  Table 3 Data used for the calculation of net emissions (100-year) from Eastmain-1 reservoir extrapolated to all reservoirs (data from Ref. [7]). Finally, net emissions per surface area were multiplied by the sum of average areas for all water bodies included in Tables A1 and A3 (1.93 × 10 10 m 2 ), and divided by the sum of average annual electricity production.

Net emissions (100 years) from a generic approach
Net emissions were calculated over a period of 100 years using the generic approach proposed in the Product Category Rules (PCR) on electricity, steam, and hot/cold water generation published by the International EPD System [38] as per Equations (4) and (5). The PCR provides generic values for S CH4 , C degr and D degr depending on ecosystem type, latitude (<or >30 • ) and average reservoir depth (<or >5 m). Emissions for three sets of assumptions were calculated to account for uncertainty: best case, average case, and worst case. All reservoirs are located in three different ecosystem types according to the map provided (i.e. main taiga, southern taiga, and open boreal woodland). C degr depends on ecosystem types. Therefore, the ecosystem type with the lowest C degr value was used for the best case (i.e. open boreal woodland), the one with the highest value for the worst case (i.e. main taiga), and the one with the middle value for the average case (i.e. southern taiga). Moreover, S CH4 can be 0 or 1% for latitudes over 30 • depending on the average reservoir depth (<or >5 m). For the best case, 0% was used, while 1% and 0.5% were used for the worst and average cases respectively. The sum of average areas for all water bodies included in Tables A1 and A3 (1.93 × 10 10 m 2 ) was used for A inund . Total emissions, as provided by Equations (4) and (5), were then divided by 100 years, and by the sum of average annual electricity production.

Generic values from the literature
A few recent publications propose average values for CO 2 and CH 4 emissions from hydropower reservoirs based on existing literature. Hertwich [29] supplemented emission data from Barros and colleagues [6] with information on electricity generation from various sources to get average emissions of 85 gCO 2 •kWh − 1 and 3 gCH 4 •kWh − 1 for electricity produced by reservoir power plants. To get reservoir emissions per kWh total hydroelectricity produced, these values were multiplied by the average annual production for all the reservoir and run-of-the-river with flooded land power plants as per Table A1 (1.42 × 10 11 kWh•yr − 1 ), and then divided by the total average annual hydroelectricity production from all hydropower plants owned by Hydro--Québec (1.78 × 10 11 kWh•yr − 1 ).
Scherer & Pfister [30] also proposed average reservoir emissions from a statistical analysis performed among 1500 hydropower plants, leading to average emissions of 173 gCO 2 •kWh − 1 and 2.95 gCH 4 •kWh − 1 for electricity produced by reservoir power plants. To get reservoir emissions per kWh total hydroelectricity produced, these values were multiplied by the average annual production for all the reservoir and run-of-the-river with flooded land power plants as per  4 . These values were multiplied by the ratio of molecular weights, and by the sum of average areas for all water bodies included in Tables A1 and A3 (1.93 × 10 10 m 2 ), and then divided by the sum of average annual electricity production.

Net emissions (100 years) from the G-res model
Net emissions over a period of 100 years were also calculated using the G-res model, a publicly available web-based tool developed by an international team of researchers supported by the International Hydropower Association (IHA) and the UNESCO Chair in Global Environmental Change [31]. The G-res tool is based on several statistical relationships, derived from the global analysis of published measured GHG fluxes (diffusive, ebullitive and degassing) as functions of site-specific climate variables, reservoir age and shape, and flooded soil carbon content (see Ref. [31] for details). It also accounts for pre-impoundment GHG emissions, simulates the long-term evolution of GHG emissions after impoundment, and following the approach outlined in Ref. [8], accounts for the CO 2 emissions that would have occurred even in the absence of the reservoir. The G-res tool also provides an estimate of the reservoir emissions that are fueled by human activities occurring in the catchment. However, since hydropower reservoirs in Quebec are in isolated locations where no other significant human activities occur, this last element is not relevant for our case and will not be accounted for.
Post-impoundment emissions from this approach account for bubbling CH 4 fluxes, diffusive CO 2 and CH 4 fluxes at the reservoir surface, and degassing CH 4 fluxes. They are estimated based on measured GHG fluxes data from the literature for different reservoir characteristics, such as age, size, carbon contained in the flooded land, and adapted for temperature. Since there is a limited number of available publications regarding CH 4 bubbling and degassing emissions, as well as how they vary over time, the values calculated by the G-res model are probably conservative for boreal regions.
Pre-impoundment fluxes are estimated for nine potential types of land cover, i.e. wetland, forest, cropland, water bodies, grassland, bare areas, permanent snow and ice, and settlements, and for different climate zones (boreal, temperate, subtropical and tropical) and soil types (organic or mineral), using emission factors such as those published by the IPCC [37]. Total CO 2 and CH 4 emissions (gCO 2 or CH 4 •yr − 1 ) were calculated summing total net emissions of all water bodies (Table A4, 2.5% and 97.5% confidence intervals are included in parenthesis to account for uncertainty).

Carbon footprint of electricity distributed in the province of Québec
In 2014, the International Reference Centre for the Life Cycle of Products, Processes and Services (CIRAIG) conducted for Hydro-Québec an LCA of the electricity generated, purchased, transmitted, and distributed in the province of Québec [39]. The functional unit for this study was "the generation or purchase, transmission and distribution of 1 kWh of electricity in Québec through Hydro-Québec's main power system in 2012".
Electricity is generated by Hydro-Québec's power plants, or purchased from independent producers in the province or from power systems in adjacent provinces or U.S. states. Power from all these facilities is then brought to the load centers by the transmission system, operated by the division Hydro-Québec TransÉnergie. The transmission system includes lines and substations. Lines comprise support structures (towers), equipment and conductors. Substations perform switching operations, and maintain or transform the voltage. In 2012, the transmission system had 33,911 km of lines and 516 transformer substations. The distribution system, operated by the division Hydro-Québec Distribution, includes all the facilities needed to distribute power from the transformer substations to the customer connection points. Most of the distribution system is overhead (more than 2,700,000 poles and 114,649 km of lines), but some is underground (3900 km).
Data collection was performed through sampling; a representative sample of the various activities of each division was first defined. Primary data, obtained directly from Hydro-Québec and its suppliers, were collected for this sample, and extrapolated to cover the rest of the power system. Overall, all life cycle phases were included. For end-of-life, dismantling and waste management were included for transmission and distribution equipment, but not for dams. Secondary data and assumptions were gathered to complement the information supplied by Hydro-Québec divisions, consisting of the ecoinvent database [40], the CIRAIG in-house database, available public databases, a literature review, and the contribution of a number of experts. Inventory data from this LCA have since been integrated to the ecoinvent version 3 database [40] to model electricity produced, transmitted and distributed in the province of Québec.
Since the first version of the study, published in 2014 [39], updates have occurred and been integrated to ecoinvent to take into account the changing composition of the generation mix (the amount of electricity produced and purchased by the different sources), the new infrastructures built (new generation plants, transmission lines, transformation posts and distribution lines), and new operation data based on the state-of-the-art research (e.g. GHG emissions from reservoirs).
For this study, the process 'market for electricity, low voltage, CA-QC' from the latest version of the ecoinvent database (version 3.5) was used and adapted. The grid composition was changed for that of 2017 in the associated 'market for electricity, high voltage, CA-QC' process, reservoir emissions were changed for those of the method selected in section 2.2, and the carbon footprint of 1 kWh of electricity distributed on the grid was calculated. Table 4 presents the grid composition for 2017 as per Hydro-Québec data with which we adapted the 'market for electricity, high voltage, CA-QC' process.
Three processes from Table 4 include reservoir emissions: 1) 'electricity production, hydro, reservoir, non-alpine region, CA-QC', 2) 'electricity, high voltage, hydro, import from CA-ON, CA-QC', and 3) 'electricity, high voltage, import from CA-NF, CA-QC'. The values of reservoir CO 2 and CH 4 emissions were changed for these three processes using the G-res model approach, following our analysis as presented in section 3. For the 'electricity, high voltage, import from CA-NF, CA-QC' process, net emissions in gCO 2 •yr − 1 and gCH 4 •yr − 1 for the Churchill reservoir in Table A4 were used. Indeed, 100% of the electricity imported from Newfoundland is produced by the Churchill Falls power plant. These values were then divided by the total amount of electricity produced by Churchill Falls in 2017, i.e. 30,927 GWh [41], to get emissions per kWh. For the 'electricity production, hydro, reservoir, non-alpine region, CA-QC' process, total net emissions, without Churchill reservoir, were used as per Table A4, and divided by the total amount of electricity produced by reservoir power plants in 2017 according to Hydro-Québec, i.e. 105,264 GWh. Hydropower produced and imported from Ontario is represented by the 'electricity, high voltage, hydro, import from CA-ON, CA-QC' process. As data for Ontario reservoirs are not available, and as hydropower from Ontario represents only 0.014% of the grid, values for Québec were used as a proxy for reservoir emissions.
Finally, the carbon footprint of 1 kWh electricity as distributed in the province of Québec was calculated from the 'market for electricity, low voltage, CA-QC' process using the ecoinvent v.3.1 IPCC2013 GWP100 method in the OpenLCA 1.7.0 software (https://openlca.org). Table 5 presents reservoir emissions per kWh of electricity from reservoir and run-of-the-river with flooded land hydropower plants (first column), and per kWh of electricity from all hydropower plants (second column), for the different approaches presented in section 2.2. Results in gCO 2 eq/kWh have been calculated using GWP100 and GTP100, as recommended by the UNEP/SETAC Life Cycle Initiative following a consensus-building workshop [42]. Both indicators are published by the IPCC [43]. While GWP100 represents the ratio of cumulative radiative forcing over 100 years caused by a unit-mass pulse emission of a given GHG relative to that of CO 2 , GTP100 represents the absolute change in global mean surface temperature 100 years following a unit-mass pulse emission of a given GHG relative to that of CO 2 . Fig. 1 presents in a bar chart the results for GHG emissions per unit electricity produced by reservoir and run-of-the-river with flooded land power plants using GWP100 (fifth column of Table 5).

Results and discussion
First four results (average gross emission over 10 and 100 years, Table 4 Grid composition for 2017 as integrated in the 'market for electricity, high voltage, CA-QC' ecoinvent process (data from Hydro-Québec).
Production Process % electricity production, hydro, reservoir, non-alpine region, CA-QC 47.5255 electricity production, hydro, run-of-river, CA-QC 34.6761 electricity, high voltage, import from CA-NF, CA-QC 11.8371 electricity, high voltage, hydro, import from CA-ON, CA-QC 0.0140 electricity production, wind, >3 MW turbine, onshore, CA-QC 3.0665 electricity production, wind, 1-3 MW turbine, onshore, CA-QC 1.3939 electricity production, wind, <1 MW turbine, onshore, CA-QC 0.1858 electricity production, oil, CA-QC 0.00073 heat and power co-generation, biogas, gas engine, CA-QC 0.0813 heat and power co-generation, wood chips, 6667 kW, state-of-theart 2014, CA-QC 0.8849 electricity, high voltage, import from CA-NB, CA-QC 0.0104 electricity, high voltage, import from CA-ON 0.2677 electricity, high voltage, import from NPCC, US only, CA-QC 0.0560 weighted average of reservoir-specific gross emissions over 10 years, and net emission from Eastmain-1 reservoir over 100 years extrapolated to all reservoirs) are based on measurements taken on Hydro-Québec reservoirs. These results are therefore more representative of the given context than those obtained from a generic approach (such as in section 2.2.4) or a world average value (such as in section 2.2.5). The best approach is to calculate the net emission over 100 years as done for the Eastmain-1 reservoir, following an intensive measurement campaign of pre-and post-impoundment fluxes. Indeed, this approach includes all types of fluxes (diffusion, bubbling, degassing, carbon storage), and distinguishes between natural and anthropogenic fluxes, as it calculates a net emission over 100 years. However, such comprehensive studies are rare as they are very expensive and require considerable sampling efforts. For the province of Québec, the Eastmain-1 reservoir is the only one for which results are currently available. Results for a second set of reservoirs (Romaine complex) will be available in the coming years.
Extrapolating the results of one study to other reservoirs is problematic, as they are all very different, and emissions highly depend on the reservoir characteristics. Eastmain-1 is a relatively shallow reservoir, impounding high carbon content soil, which usually leads to higher emissions. Emissions are therefore overestimated when these results are extrapolated to all reservoirs, as done in section 2.2.3. As shown in Table 5, emissions of CO 2 and CH 4 are higher for this approach (respectively 51.0 and 0.26 g/kWh) than for the weighted average of reservoir-specific gross emission over 10 years (respectively 5.4 and 0.02 g/kWh) and the average gross emission over 10 years (7.8 gCO 2 / kWh) and over 100 years (37.4 gCO 2 /kWh).
As reservoir emissions depend on reservoir characteristics, the reservoir-specific approach (section 2.2.2) is probably better than the average approach (section 2.2.1). However, data were not available for all reservoirs, which could bias the results. In particular, reservoirs and flooded lands that are not associated directly to a power plant (Table A3) were left out because of missing data. This could explain the lower value obtained with this approach compared with that obtained with the average gross emission over 10 years approach (5.4 gCO 2 /kWh versus 7.8 gCO 2 /kWh). Moreover, reservoir-specific measures have been taken at different ages, which might lead to an under-or over-estimation of the results. Therefore, the reservoir-specific approach has the advantage of considering the characteristics of reservoirs, but not the age. Both these approaches (sections 2.2.1 and 2.2.2) are based on gross emissions, because no data about pre-impoundment emissions are available, and include only CO 2 diffusive emissions.
The generic approach proposed by a PCR (section 2.2.4) is very simple to use. However, uncertainties are very high as it is not based on any measurements or modelling. The calculation includes parameters in order to consider some reservoir characteristics, such as the latitude and water depth, but this approach is not recommended if data based on measurements or modelling are available. Average values from the literature (section 2.2.5) all show very high results compared with other approaches (85 gCO 2 /kWh and 3.0 gCH 4 /kWh for [29], 173 gCO 2 /kWh and 3.0 gCH 4 /kWh for [30], 34.6-59.2 gCO 2 /kWh and 0.8-3.7 gCH 4 /kWh for [25]). These results are based on data published in the literature from different regions of the world. They include reservoirs in tropical zones, for which emissions can be very high compared with those in boreal zones, especially for methane. As they are not specific to the context of the province of Québec, they are less representative. A good way to improve this type of approach would be to divide the data collected into categories that would better reflect, at least, the climate zone of the reservoir. However, even this approach can generate inappropriate results, since some reservoirs in tropical regions are less emitting because of the climate, soil and design characteristics.
Finally, the results from the last approach (section 2.2.6) are the net emissions over 100 years as calculated using the G-res model. The advantages of this approach are that it takes into account the specific characteristics of the reservoirs (e.g. size, carbon contained in the flooded land, temperature) in the estimation, and that all types of fluxes (diffusion, bubbling, degassing) are included. Moreover, it is based on a net 100-year emission approach, as pre-and post-impoundment fluxes are estimated, and it allows including all reservoirs and water bodies. The model could be used for any geographical location, as it can be calibrated using local parameters. Moreover, uncertainty can be quantified and expressed using 2.5% and 97.5% confidence intervals. The results from this approach (16.5 gCO 2 /kWh and 0.29 gCH 4 /kWh) are in between that of average gross emission over 10 years (7.8 gCO 2 /kWh) Table 5 Reservoir emissions estimated using different approaches per unit electricity produced by reservoir and run-of-the-river with flooded land power plants (first column) and per unit electricity produced by all hydropower plants owned by Hydro-Québec (second column). a Emissions for average area, emissions for minimum and maximum area in parenthesis. b Result from G-res, 2.5% and 97.5% confidence interval in parenthesis. Modelling and weighted average of reservoir-specific gross emissions over 10 years (5.4 gCO 2 /kWh and 0.02 gCH 4 /kWh) and that of net emissions over 100 years from Eastmain-1 extrapolated to all reservoirs (51.0 gCO 2 /kWh and 0.26 gCH 4 /kWh). It might be difficult to select the best approach since they all have different limitations (e.g. type of flux considered, based on measurement versus modelling, reservoir-specific versus generic). Table 6 presents a summary of the comparative analysis performed on all the approaches. In the absence of site-specific data to calculate the net GHG emission based on the five types of fluxes as per Eastmain-1, the use of G-res is considered the most reliable and comprehensive approach. G-res has therefore been used to model all the reservoirs and flooded lands according to their specific characteristics, and to calculate their net emissions over 100 years to be included in the carbon footprint of electricity distributed in the province of Québec.
The results from G-res, as per Table A4, were used to calculate net biogenic GHG emissions per kWh produced for each hydropower complex. To do so, all power plants (Tables A1 and A2) and water bodies that are not directly associated to a given power plant (Table A3), that are in the same watershed, were grouped together. Total annual GHG emissions from all water bodies were then divided by total annual electricity production from all power plants situated in the watershed. Results, as per Table A5, show that emissions vary from one hydropower complex to another, from 0 (for watersheds without any reservoirs or flooded lands) to 73.2 gCO 2 eq•kWh − 1 .
As per section 2.3, the carbon footprint of electricity distributed in Québec, using the 2017 grid composition and reservoir emissions from the G-res model approach, is 34.5 gCO 2 eq•kWh − 1 . It represents an increase of 42% compared with the carbon footprint calculated from the process currently available in ecoinvent, which is 24.3 gCO 2 eq•kWh − 1 . This increase is caused by biogenic carbon emissions from flooded lands that are higher using G-res than the values found in ecoinvent, which were calculated in 2014 using a 10-year gross emission approach based on a set of measurements (such as in section 2.2.1).
Biogenic emissions from flooded lands contribute to about 70% of the carbon footprint of the electricity distributed in the province of Québec in 2017. Emissions associated with the construction of hydropower infrastructures in Québec and Newfoundland (Churchill complex) contribute to about 12% of the carbon footprint of the electricity distributed in the province. Emissions from electricity production from non-hydro sources in Québec (wind, biomass, oil) represent about 5% of the carbon footprint, while those associated with the generation of imported electricity from other jurisdictions than Newfoundland represent about 1.4%. Finally, transmission, distribution, and operations (including losses) contribute to about 10% of the carbon footprint. Regarding hydroelectricity production only, i.e. without transmission and distribution, the contribution of the construction of infrastructures to GHG emissions is much higher for run-of-the-river power plants (96%) than for reservoir power plants (7.5%) as they do not cause any biogenic GHG emissions.

Conclusions
This paper presents a comparative analysis of different approaches to quantify biogenic CO 2 and CH 4 emissions from hydropower reservoirs in the province of Québec (Canada). These approaches differ according to various aspects, such as the type (diffusion, bubbling, degassing) and nature (CO 2 and CH 4 ) of emissions considered, the use of reservoir-and/ or geography-specific data, the number of impoundments covered, and the technique used to estimate emissions (measurements, modelling, generic). In the absence of a comprehensive set of data for all reservoirs, that include pre-and post-impoundments diffusive, bubbling and degassing CO 2 and CH 4 emissions, the G-res model, supported by the IHA and the UNESCO Chair in Global Environmental Change as the result of an international multi-stakeholder research project [31], has been used. Since it can be calibrated for any geographical location using local parameters, this approach takes into account all types of emissions, as well as the specific characteristics of each reservoir, which highly influence the results, as shown in this paper.
Using the G-res model, biogenic emissions were estimated for all the hydropower reservoirs in Québec, as well as for the Churchill reservoir for hydropower imported from Newfoundland-Labrador, a neighbouring province. These results were then used to calculate an updated value for the carbon footprint of electricity distributed in the province of Québec in 2017. The result obtained is 34.5 gCO 2 eq•kWh − 1 , which is 42% higher than the value currently available in the ecoinvent database. The carbon footprint has increased because G-res considers all flooded lands and all types of flux, which was not the case for the approach used before.
In this paper, biogenic emissions from reservoirs were calculated as average emissions over 100 years, meaning that the temporal profile has been ignored. However, the largest share of these emissions occur during the first 10-15 years following creation of the reservoir [8,12]. This is very different from a thermal power plant, which releases more or less constant GHG emissions over its lifetime. Some hydropower plants in Québec are several decades old, meaning that their reservoirs currently release very little biogenic carbon emissions. By contrast, recent hydropower complexes, such as Eastmain or Romaine, probably still releases high amounts of biogenic carbon, more than their average emission calculated over 100 years. To better represent the current and future carbon footprint of electricity produced in the province, a dynamic LCA approach [44] could be used to consider the temporal profile of GHG emissions for each power plant according to the year of its construction. This would provide a better assessment of the potential contribution of electricity production to current and future global warming.

Funding
This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sector.

Declaration of competing interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.   To calculate the sum of minimum or maximum surfaces, we used the average surface for this reservoir.
i The McCormick power plant is owned by Hydro-Québec at 60% (40% is private and bought by Hydro-Québec).
A. Levasseur et al.