Year‐2020 Global Distribution and Pathways of Reservoir Methane and Carbon Dioxide Emissions According to the Greenhouse Gas From Reservoirs (G‐res) Model

Abstract

Collectively, reservoirs constitute a significant global source of C‐based greenhouse gases (GHGs). Yet, global estimates of reservoir carbon dioxide (CO2) and methane (CH4) emissions remain uncertain, varying more than four‐fold in recent analyses. Here we present results from a global application of the Greenhouse Gas from Reservoirs (G‐res) model wherein we estimate per‐area and per‐reservoir CO2 and CH4 fluxes, by specific flux pathway and in a spatially and temporally explicit manner, as a function of reservoir characteristics. We show: (a) CH4 fluxes via degassing and ebullition are much larger than previously recognized and diffusive CH4 fluxes are lower than previously estimated, while CO2 emissions are similar to those reported in past work; (b) per‐area reservoir GHG fluxes are >29% higher than suggested by previous studies, due in large part to our novel inclusion of the degassing flux in our global estimate; (c) CO2 flux is the dominant emissions pathway in boreal regions and CH4 degassing and ebullition are dominant in tropical and subtropical regions, with the highest overall reservoir GHG fluxes in the tropics and subtropics; and (d) reservoir GHG fluxes are quite sensitive to input parameters that are both poorly constrained and likely to be strongly influenced by climate change in coming decades (parameters such as temperature and littoral area, where the latter may be expanded by deepening thermoclines expected to accompany warming surface waters). Together these results highlight a critical need to both better understand climate‐related drivers of GHG emission and to better quantify GHG emissions via CH4 ebullition and degassing.

Full text

1. Introduction
Recent work has highlighted the importance of inland waters generally, and reservoirs specifically, as hot-
spots for carbon processing and C-based greenhouse gas (GHG) emission (Butman & Raymond,2011; Cole
etal.,2007; Deemer etal.,2016; Deemer & Holgerson,2021; DelSontro etal.,2018; Maavara etal.,2017,2020;
Raymond etal.,2013; Rosentreter etal.,2021; Tranvik etal.,2009). Yet, global estimates of reservoir GHG
emissions remain highly uncertain, varying by more than four-fold (range: 741–3,380Tg CO2 equivalents yr−1)
in recent analyses (Barros etal.,2011; Bastviken etal.,2011; Deemer etal.,2016; Hertwich,2013; St. Louis
etal.,2000; Rosentreter etal.,2021).
To date, global estimates of reservoir GHG emissions have been derived simply by multiplying global reser-
voir surface area by emission rates averaged from a limited, but growing, set of in situ flux measurements.
While this approach may provide a reasonable first-order approximation of global fluxes, the accuracy of
Abstract Collectively, reservoirs constitute a significant global source of C-based greenhouse
gases (GHGs). Yet, global estimates of reservoir carbon dioxide (CO2) and methane (CH4) emissions
remain uncertain, varying more than four-fold in recent analyses. Here we present results from a global
application of the Greenhouse Gas from Reservoirs (G-res) model wherein we estimate per-area and per-
reservoir CO2 and CH4 fluxes, by specific flux pathway and in a spatially and temporally explicit manner,
as a function of reservoir characteristics. We show: (a) CH4 fluxes via degassing and ebullition are much
larger than previously recognized and diffusive CH4 fluxes are lower than previously estimated, while CO2
emissions are similar to those reported in past work; (b) per-area reservoir GHG fluxes are >29% higher
than suggested by previous studies, due in large part to our novel inclusion of the degassing flux in our
global estimate; (c) CO2 flux is the dominant emissions pathway in boreal regions and CH4 degassing and
ebullition are dominant in tropical and subtropical regions, with the highest overall reservoir GHG fluxes
in the tropics and subtropics; and (d) reservoir GHG fluxes are quite sensitive to input parameters that
are both poorly constrained and likely to be strongly influenced by climate change in coming decades
(parameters such as temperature and littoral area, where the latter may be expanded by deepening
thermoclines expected to accompany warming surface waters). Together these results highlight a critical
need to both better understand climate-related drivers of GHG emission and to better quantify GHG
emissions via CH4 ebullition and degassing.
Plain Language Summary By damming rivers, humans have created millions of reservoirs,
which, collectively, constitute an important greenhouse gas source, especially for methane, a particularly
potent greenhouse gas. Using observed relationships between reservoir characteristics and greenhouse
gas emissions, we show that much more methane either bubbles out of reservoirs or is emitted just
downstream from reservoirs than was previously known. This is important because it may be possible
to reduce methane emissions from downstream of reservoirs by selectively withdrawing water from
near the surface of reservoirs, which tends to be methane-poor, rather than from greater depths, where
methane often accumulates. We also found that on a per-area basis reservoirs are a more potent source of
greenhouse gases than previously recognized, and that the highest rates of emissions occur in the tropics
and subtropics. Finally, we show that estimates of reservoir greenhouse gas emissions are quite sensitive
to climate-related factors like temperature.
HARRISON ET AL.
© 2021. American Geophysical Union.
All Rights Reserved.
Year-2020 Global Distribution and Pathways of Reservoir
Methane and Carbon Dioxide Emissions According to the
Greenhouse Gas From Reservoirs (G-res) Model
John A. Harrison1 , Yves T. Prairie2 , Sara Mercier-Blais2 , and Cynthia Soued2
1School of the Environment, Washington State University, Vancouver, WA, USA, 2Department of Biological Sciences,
University of Quebec at Montreal (UQAM), Montréal, QC, Canada
Key Points:
• This is the most comprehensive
global analysis of reservoir methane
and CO2 emissions to-date, and the
first to estimate methane degassing
• Although diffusive CH4 fluxes are
somewhat lower than previously
believed, CH4 fluxes via degassing
and ebullition are much larger
• The highest reservoir greenhouse
gas emissions globally occur in
the tropics and subtropics, with
CH4 degassing and ebullition as
dominant flux paths
Supporting Information:
Supporting Information may be found
in the online version of this article.
Correspondence to:
J. A. Harrison,
john_harrison@wsu.edu
Citation:
Harrison, J. A., Prairie, Y. T., Mercier-
Blais, S., & Soued, C. (2021). Year-2020
global distribution and pathways
of reservoir methane and carbon
dioxide emissions according to the
greenhouse gas from reservoirs (G-res)
model. Global Biogeochemical Cycles,
35, e2020GB006888. https://doi.
org/10.1029/2020GB006888
Received 13 NOV 2020
Accepted 14 MAY 2021
10.1029/2020GB006888
RESEARCH ARTICLE
1 of 14

Global Biogeochemical Cycles
such a method relies upon a number of assumptions, all of which may have a large but heretofore unknown
and largely unexplored impact on global flux estimates. Notably, the generally applied approach implicitly
assumes that sampling of reservoirs accurately reflects the existing age distribution of reservoirs (because
GHG emissions from reservoirs tend to decline over time [Abril etal.,2005; Teodoru etal.,2012]). It also
implicitly assumes that sampling adequately captures and represents the natural spatial variation in GHG
flux rates both within and between reservoirs, and temporal variation over seasonal to multi-year scales,
all of which are often substantial, necessitating adequate spatial and temporal coverage of measurements
(Wik etal.,2013,2016). The importance of such assumptions to estimates of aquatic GHG fluxes at large-
scales is, to-date, largely unknown and untested. In addition, due to the lack of any continentally or globally
applicable models for predicting aquatic GHG fluxes, it has not been possible to estimate the geographic
distribution of reservoir GHG fluxes at the global scale beyond broad generalizations about latitudinal pat-
terns (e.g., Deemer etal.,2016). Nor have spatial distributions of various flux pathways for CH4 been quan-
tified or evaluated at large scales. In fact, efforts to estimate reservoir GHG fluxes have generally focused
solely on diffusive gas fluxes (Barros etal.,2011; St. Louis etal.,2000), and it is only recently that ebullition
fluxes have been incorporated into global reservoir GHG flux estimates (Deemer etal., 2016; delSontro
etal.,2018). It has been shown that CH4 emissions due to turbine degassing can be the dominant pathway
in individual reservoirs (Abril etal.,2005; Soued & Prairie,2020) and could therefore be substantial at the
global scale, but these fluxes have not previously been estimated (Deemer etal.,2016).
In previous work, we reported on the development of the G-res model, which is an open and globally
consistent predictive framework for estimating the anthropogenic GHG footprint of individual reservoirs
(Prairie etal.,2017,2018). Here, our main objective is to apply the underlying emission models of G-res to
provide the first-ever spatially and temporally explicit global estimate of reservoir CO2 and CH4 fluxes and
the first-ever global estimate of reservoir degassing fluxes. In addition, we explore the relative contribution
of the various emission pathways and how they vary across regions of the world. Lastly, we conducted effi-
ciency and sensitivity analysis of the models to help identify important research needs and facilitate better
estimates of global and regional GHG emissions from reservoirs going forward.
2. Methods
2.1. Overview
Broadly, our approach to developing a spatially explicit estimate of the global GHG flux from reservoirs in-
volved three steps: (a) extraction of information required to apply G-res models to a larger set of reservoirs
using globally consistent GIS layers, (b) application of the predictive models globally, including estimates of
associated uncertainty, and (c) assessment of model sensitivity and efficiency in order to understand how
changes in drivers are likely to affect global GHG fluxes and to identify especially important and promising
avenues for additional research and refinement. The development of the emission models is described in
detail in publicly available model documentation (Prairie etal.,2017), and model equations, a description
of model input datasets, and model calibration and validation data are provided in TablesS1 andS2, and in
a downloadable data set (Prairie etal.,2021), respectively. However, the models are described briefly below
for completeness (Sections2.2 and2.3), and steps 2 and 3 are described in Sections2.4 and2.5.
2.2. Model Development/Calibration Data Set
Using data compiled from 223 globally distributed reservoirs for which published measurements were avail-
able (Figure1, Prairie etal., 2021), the G-res framework was developed from multiple linear regression
models to predict per-area and per-reservoir CO2 and CH4 emissions. Prior to regression analysis, per-area
flux estimates were processed in several ways to improve internal consistency. First, flux measurements
were annualized to account for the fact that measurements are often made during summer months when
fluxes might be higher than at other times of the year. Briefly, this annualization was accomplished for lit-
erature-reported diffusive fluxes of CO2 and CH4, and CH4 bubbling fluxes by combining the mean monthly
air temperature for the month(s) in which fluxes were measured with the known temperature dependencies
for CO2 and CH4 production (Q10=2, and 4 for CO2 and CH4, respectively [Inglett etal.,2012; Yvon-Duro-
cher etal.,2014]), and averaging flux estimates across all 12months in a year to achieve an annual flux rate.
HARRISON ET AL.
10.1029/2020GB006888
2 of 14

Global Biogeochemical Cycles
This process was carried out for each reservoir. In addition, when direct CH4 ebullition data were only avail-
able from littoral sites within reservoirs (17 of the reservoirs used to develop the G-res ebullition model), we
accounted for the fact that bubbling tends to decline with increasing water column depth (due to increased
hydrostatic pressure) by multiplying littoral ebullition rates by the ratio of littoral: total reservoir area to
achieve reservoir-wide estimates. The extent of littoral area used was either the littoral area reported in the
original publication (used when available) or the littoral area estimated using the maximum depth sampled
for ebullition provided in the original publication (used when a direct estimate of littoral area was not avail-
able). Although this approach may underestimate lake-wide ebullition fluxes of CH4 in some systems, we
view it as an improvement over the historical approach, which simply assumes that these littoral flux rates
are representative of whole-lake fluxes, thereby potentially overstating their importance. Once annualized
and area-corrected, flux estimates were log10 transformed in order to achieve approximate normal distribu-
tion of fluxes. For CH4 degassing emissions, the empirical model uses the change in concentration between
the water intake depth and directly out of the outlet as the predicted response variable.
In addition to pre-processing flux measurements, we also collected information on reservoirs and associ-
ated catchment characteristics likely to influence CO2 and CH4 production and emission from a variety of
sources. These potential drivers are described in detail in Prairie etal.(2017,2018). Briefly, they included
reservoir characteristics such as latitude, surface area, volume, maximum depth, mean depth, littoral area
(defined as area with depth less than 3m) as a fraction of total lake area, water residence time, mean month-
ly and annual air temperature, mean annual wind speed, mean global horizontal irradiance, estimated
phosphorus loading and concentration, soil carbon content prior to flooding and catchment characteristics
such as catchment area, annual precipitation, mean annual runoff, population density, annual reservoir
inflow, and catchment land cover (Prairie etal.,2017). When possible, these data were taken directly from
peer-reviewed studies of reservoirs, but when it was not possible to mine primary literature for reservoir
or catchment characteristics, we relied on attributes provided by the Global Reservoirs and Lakes (GRanD)
database (Lehner etal.,2011) or from published GIS coverages (Prairie etal.,2017,2021). In some cases (no-
tably average depth, thermocline depth, littoral area, water residence time, and phosphorus concentrations)
we estimated values using peer-reviewed approaches (Prairie etal.,2017). Model selection was determined
based on best fit, reasonableness (i.e., drivers had to have a reasonable biophysical explanation to be includ-
ed), and global availability of model drivers. Flux estimates from 107, 102, 27, and 38 reservoirs were used
to develop CO2 diffusion, CH4 diffusion, CH4 ebullition, and CH4 degassing models, respectively (Figure1,
Prairie etal.,2021). In a subset of reservoirs, flux estimates were available from multiple years, such that a
total of 169, 160, 46, and 38 individual flux estimates were used to develop the CO2 diffusion, CH4 diffusion,
CH4 ebullition, and CH4 degassing models, respectively. Significant input parameters (P<0.05 by multi-
ple linear regression) for each GHG flux pathway are listed in Table1, and complete models and model
input variables are included in TablesS1 andS2, respectively. Whereas diffusive and ebullitive flux models
HARRISON ET AL.
10.1029/2020GB006888
3 of 14
Figure 1. Locations of greenhouse gas flux measurements used to develop G-res submodels; symbols represent the pathway for which flux estimates were
available and the size of each symbol is proportional to the estimated flux rate (g CO2 eq. m−2 yr−1).

Global Biogeochemical Cycles
were developed using per-area flux measurements, the degassing sub-model was developed to predict the
difference in CH4 concentrations upstream and downstream of dams. When multiplied by water flow and
divided by a reservoir's surface area, this concentration difference was considered the per-area degassing
flux of CH4.
2.3. Emission Pathway Sub-Models
Each emission pathway (CO2 diffusion, CH4 diffusion, ebullition, and degassing) was modeled as a multi-
variate relationship between the annualized per-area emission rates and potential predictor variables de-
scribed above, following suitable transformation. Flux rates are reported in units of CO2 equivalents, assum-
ing a per-mass 100-year warming potential for CH4 34-fold that of CO2 (IPCC 2013). Variable selection was
carried out through the elastic net regression procedure which is particularly well-suited to “short and fat”
datasets (small number of observations relative to the number of potential predictors) with colinear predic-
tor variables. The exact form of the resulting models is described in more detail in publicly available G-res
documentation (Prairie etal.,2017,2021) but the variables retained, the sign of their coefficients and their
relative importance are summarized for each sub-model in Table1. For completeness, the sub-model equa-
tions (G-res V.3) and variable definitions are reproduced in Supplementary Material (TablesS1 &S2). In
order to facilitate attribution of reservoir GHG fluxes, G-res partitions CO2 flux from each reservoir into an-
thropogenic and non-anthropogenic fractions. For the purpose of this global GHG flux estimation and com-
parison with previous estimates, we report here only total CO2 emissions. Similarly, the G-res framework
estimates the degassing pathway only for reservoirs with deep water intakes (deeper than the thermocline),
where CH4 can accumulate to very high concentrations. Information on water intake depth was largely
unavailable. However, due to the greater operational flexibility it offers, the deep-water withdrawal config-
uration is known to be more common among hydropower reservoirs. Thus, we used this (hydroelectricity
as a reported usage in the GRanD database) as a criterion to consider degassing in the modeled emissions
of a given reservoir. This assumption is clearly a simplification of reality, since deep water intake is neither
systematic in, nor limited to, hydropower reservoirs. However, it remains the most sensible approach for
a first-order estimate of degassing on a global scale given currently available information. Hydroelectricity
was a reported use for 33.7% of the reservoirs in our global reservoir database, but these reservoirs tended to
be large, such that cumulatively they accounted for 83% of the total global reservoir surface area.
HARRISON ET AL.
10.1029/2020GB006888
4 of 14
CO2 diffusion CH4 diffusion CH4 ebullition CH4 degassing
Age − −
Temperature + +
Soil C content +
Total phosphorus +
Reservoir area +
% Littoral area + +
Cumulative radiancea+
Hypolimnetic release (Y/N) x
Water residence time +
Diffusive CH4 emissions +
aCumulative radiance is estimated as the mean global horizontal radiance for the ice-free period multiplied by the
number of ice-free months.
Notes. Minus (−) or plus (+) signs indicate a negative or positive relationship between the emission pathway and the
parameter respectively. “x” indicates a binary decision whether to include a given reservoir in the global total based on
its characteristics. See TableS1 for model equations, TableS2 for coefficient definitions and values, Prairie etal.(2021)
for flux estimates used in model development and evaluation, and Prairie etal.(2017) for a more complete description
of the models and their development.
Table 1
Influence of Each Input Parameter on the Four Pathways Estimated by the G-res Model

Global Biogeochemical Cycles
2.4. Spatial Extrapolation
G-res submodels were applied to 4,563 reservoirs either in the GRanD database with the necessary driver
data available (Lehner etal.,2011) or added by us (164 reservoirs). These reservoirs had a worldwide dis-
tribution and collectively represent 68% of estimated global reservoir surface area. The four largest reser-
voirs in the GRanD database (Lakes Victoria, Baikal, Ontario, and Onega) collectively occupy more than
128,000km2. As these systems are large natural lakes where damming has resulted in negligible increases
in surface area and depth, these large lakes were excluded from our global estimate of reservoir GHG fluxes.
The GRanD database underestimates the surface area of small reservoirs since it is a compilation of the
world largest reservoirs (reservoirs exceeding 0.1Mm3 in volumetric capacity). Because the G-res database
used in this analysis (Prairie etal., 2021) relies heavily upon the GRanD database, it was subject to this
same limitation. The extent of this underestimation is difficult to assess, but can be examined using the
approach developed by Downing etal.(2006) and applied by others (e.g., Lehner etal.,2011), which relies
on the statistical properties of a canonical set of reservoirs, extrapolated to a reasonable lower size limit.
Using this approach, and discounting the four large lakes, yielded an estimated total global surface area for
all reservoirs greater than 0.1km2, of approximately 350,000km2 (comprised of 87,800 reservoirs), which
is close to estimates used in other recent global analyses (e.g., Deemer etal.,2016). To account for the dif-
ference in surface area between reservoirs represented in the G-res database and our estimated global total,
we multiplied the total global fluxes from G-res database reservoirs by a factor of 1.47 (the ratio between
an estimated global reservoir surface area of 350,000km2 and the total surface area of reservoirs globally
for which it was possible to use G-res to calculate all four G-res-estimated GHG flux pathways) to attain
global flux estimates, and we distributed additional GHG fluxes spatially in proportion to the occurrence
of the G-res database reservoir surface area. As noted above, following a necessary, albeit oversimplifying,
assumption that only hydroelectric reservoirs have deep water intakes, we calculated a degassing flux only
for those systems.
2.5. Estimating Global Total Gas Fluxes and Associated Uncertainty
To avoid influence of extreme outliers on error estimates, we removed outliers using Cook's distance (cri-
terion for removal: Cook's distance >3*µ; Cook 1977). This resulted in removal of 3, 15, 3, and 2 reservoirs
from G-res CO2 diffusion, CH4 diffusion, CH4 ebullition, and CH4 degassing models, respectively. To correct
for bias associated with developing models using log-transformed data (Newman,1993), we calculated re-
gression standard error of residuals (
s
) as

2
Σ
DF
ˆ
ii
yy
, where
i
y
is model-predicted flux,
ˆi
y
is measured
flux, and DF is the number of degrees of freedom (number of available comparisons between measurements
and model predictions minus the number of calibrated parameters in each G-res sub-model). We then ran a
Monte Carlo analysis in which all model estimates were amended with a randomly determined error (with
a Gaussian distribution and a mean equal to the standard error of residuals (
s
, defined above). Predicted
per-reservoir fluxes were subsequently exponentiated and summed. This calculation was repeated 1,000
times with random assignment of errors and the median and 95% confidence interval values for predict-
ed global total fluxes were determined based on distribution of results. Median values resulting from this
exercise are presented as the most probable estimate (i.e., highest probability density) of global reservoir
fluxes. This approach to propagating the uncertainty for each flux pathway model to estimate the global
GHG footprint means that greater model uncertainty results in higher estimated global flux. This effect is
not negligible. In the case of G-res models, the estimated global flux accounting for the uncertainty in the
prediction of each individual reservoir was 1.48, 1.99, 4.45, and 4.88-fold higher than uncorrected totals for
CO2 diffusion, CH4 diffusion, CH4 degassing and CH4 ebullition sub-models, respectively.
Fluxes reported by latitudinal region were binned into boreal, temperate, subtropical, and tropical regions,
defined as >62°N and S, 35–62°N and S, 23.5–35°N and S, and <23.5°N and S, respectively. Fluxes were also
binned and summed by climate zone, as defined by the IPCC (Rubel & Kottek,2010). Climate zones used
for this exercise were: boreal, cool temperate, temperate warm/dry, temperate warm/moist, tropical dry/
montane, and tropical moist/wet. Unless specified as climate-related (e.g., in Table 5), the terms boreal,
temperate, and tropical refer to latitudinal regions.
HARRISON ET AL.
10.1029/2020GB006888
5 of 14

Global Biogeochemical Cycles
2.6. Model Sensitivity and Efficiency Analyses
To evaluate G-res sensitivity to input data and model parameters, we
sequentially increased individual model input parameters by 10% and
evaluated the resulting change in model predictions. To evaluate the im-
portance of G-res model components, we performed an efficiency analy-
sis (as described in Nash & Sutcliffe,1970), wherein model components
were removed one at a time and the resulting change in Nash Sutcliffe
Efficiency (NSE) was evaluated. NSE is a measure of model skill where
(similar to r2) a value of 1 connotes perfect model predictions but where a
value of zero indicates that a mean of measurements is as good a predic-
tor of measurements as the model being evaluated, and negative values
indicate that a model's predictions do worse than simply using the mean
of available measurements (Nash & Sutcliffe,1970).
3. Results and Discussion
3.1. Model Performance
Although error associated with CO2 and CH4 flux predictions for indi-
vidual reservoirs was, in some cases, substantial (e.g., see Table 2 and
Figure2), each of the individual G-res submodels was reasonably proficient at predicting spatial variation
in per-area fluxes, and, more importantly for this global analysis, performed well in predicting per-reservoir
fluxes (NSE>0.8 for CO2 and CH4 diffusion, and NSE > 0.5 for CH4 ebullition and degassing, Table2,
Figure2). Further, the G-res submodels were bias-free in that slopes of least squares linear regressions be-
tween measured and model-estimated fluxes were not significantly different from unity for any of the G-res
submodels (Figure1). Submodels also performed well when outputs from more than one submodel were
summed and compared with summed measurements from several systems (NSE: 0.71 and 0.83 for total CH4
and Total C-based greenhouse gases, respectively). These comparisons were also free of apparent bias (Fig-
ure2). G-res submodels were somewhat less skilled at predicting per-area fluxes than per-reservoir fluxes,
but still performed better than an average of measurements in predicting per-area CO2 and CH4 fluxes, as
indicated by positive NSE values (≥0.29 in all submodels, Table2).
3.2. Global CO2 and CH4 Fluxes
Using G-res, we estimate global flux of GHGs from reservoirs as 1,076Tg
CO2 eq. yr−1 (range: 730–2,412Tg CO2 eq. yr−1) mainly as CH4 (328 and
748 Tg CO2 eq. yr−1 for CO2 and CH4, respectively). For methane, our
estimate of diffusive plus ebullitive CH4 emission (337Tg CO2 eq. yr−1)
is similar to that of Hertwich etal.(2013), and more than double that of
Bastviken etal. (2011), but somewhat smaller than other recent global
estimates, which are based on averages of reported fluxes and range from
606–2,380Tg CO2 eq. yr−1 (Table3). Degassing fluxes of CH4, which have
not been accounted for in other global analyses, were the largest and also
the most uncertain flux in our estimate, accounting for 411Tg CO2 eq.
yr−1 (95% confidence range: 227–1,261Tg CO2 eq. yr−1). When degassing
is included, our median global estimate of reservoir greenhouse gas flux-
es (Tg CO2 eq. for CO2 plus CH4) is among the largest that have been re-
ported, and is 45% higher than a recent estimate by Deemer etal.(2016).
G-res-estimated average global per-area GHG emissions (Tg CO2 eq. km−2
of reservoir surface area) are the highest that have been reported, exceed-
ing those of Deemer etal.(2016) and St. Louis etal.(2000), by 29% and
36%, respectively.
Globally, we estimate that diffusive CH4 fluxes from reservoirs, which,
of all CH4 flux pathways, have received the most attention to-date, are
HARRISON ET AL.
10.1029/2020GB006888
6 of 14
Flux pathway
NSE NSE Median %
n Outliers
Per-
area
Per-
reservoir Error (IQR)
CO2 Diffusion 0.37 0.89 11 (−44–52) 169 3
CH4 Diffusion 0.52 0.84 −21 (−54–121) 160 15
CH4 Degassing 0.57 0.74 21 (−76–187) 38 2
CH4 Ebullition 0.29 0.58 14 (−84–318) 46 3
Notes. The number of outliers excluded from NSE and IQR calculations
are also reported. See methods (Section2.8) for description of the method
used to exclude outliers.
Table 2
Performance Statistics for G-res Submodels, Including Nash Sutcliffe
Efficiency (NSE) Values for Both Per-Area and Per-Reservoir Predictions,
Median and Interquartile Range Values for Percent Error for the Models,
and the Number of Flux Estimates Used to Develop and Evaluate the
Models
Figure 2. G-res predicted CH4 and CO2 fluxes (Mg CO2 eq reservoir−1yr−1)
by pathway versus measurements. Nash Sutcliffe Efficiency values for
G-res submodels (each flux pathway) are shown in Table2.

Global Biogeochemical Cycles
small relative to other reservoir GHG flux pathways, accounting for just 7% of the total CH4 flux and 5% of
the greenhouse effect due to combined CH4 plus CO2 flux. In contrast, ebullition and, especially, degassing
fluxes are comparatively large, accounting for 38% and 55% of the global total CH4 flux, respectively. Overall,
diffusive fluxes of CO2 are substantial, accounting for almost one third (30.5%) of the greenhouse liability
from reservoirs, but exert a smaller impact on greenhouse warming than reservoir-sourced CH4, which
collectively accounts for 69.5% of the greenhouse warming effect due to reservoir emissions globally. This
fraction would be substantially greater if a less conservative greenhouse warming potential were used for
CH4. If, for example, we used the 20-year greenhouse warming potential (85, IPCC 2013) for CH4 instead
of a 100-year greenhouse warming potential, CH4 would constitute 85% of total annual warming potential
emitted from reservoirs. In addition, whereas virtually all of the CH4 emitted from reservoirs can be con-
sidered anthropogenic, a substantial portion of the global CO2 flux can be considered non-anthropogenic
in origin as ∼69% of these emissions would have occurred even in the absence of reservoir construction
(Prairie etal.,2017).
Although this analysis revises the estimate of global total diffusive CH4 flux and ebullitive CH4 flux down-
ward relative to other recent studies (Table 3), the total estimated global flux of CH4 (22.0 Tg y−1; range:
13.4–58.8Tg CH4 y−1) we report is higher than that estimated by other recent efforts (Table3). This is due,
in large part, to our novel inclusion of CH4 degassing in this global estimate of reservoir GHG fluxes. G-res
per-area estimates of CH4 diffusion are lower than past estimates because the G-res models attempt to ac-
count for potential sampling bias associated with reported flux measurements. Reasons that past estimates
of global diffusive CH4 fluxes, and global ebullitive CH4 fluxes may have been overestimated include: dis-
proportionate sampling of high-emission reservoirs, seasonal sampling that focuses on summer conditions
(which tend to foster large GHG fluxes), inappropriate temporal averaging that neglects winter ice cover, use
of littoral only emissions to represent emissions from the total surface area of reservoirs, and oversampling
of highly active regions within reservoirs. In the formulation of G-res models, every effort was taken to ac-
count for these potential sources of bias in the measurement data. For example, measurements of CH4 and
CO2 diffusion used in model calibration were temperature-corrected to account for potential seasonal bias
in sampling, and ebullitive emissions during ice-covered months (i.e., months with mean air temperatures
<0°C) were assumed to be zero while diffusive emissions were assumed to be emitted at ice-off after pro-
ceeding under 4°C conditions during the ice-covered period. Although sensible, these assumptions remain
largely unverified by field observations and might still overestimate CH4 diffusive flux, as a portion of the
CH4 produced under ice can be oxidized before ice-off. However, we argue that these modeling assumptions
represent an improvement over simply extrapolating ice-free period fluxes to the whole year. These efforts
resulted in comparatively low annual estimates of global diffusive CH4 fluxes and CH4 ebullition fluxes.
However, the addition of CH4 degassing, arguably a less certain flux (due to modeling assumptions and
fewer data points), more than compensated for lower estimates of other CO2 and CH4 fluxes, leading to an
estimate of global reservoir CH4 fluxes that is similar to, or greater than, other recent estimates (Table3) and
HARRISON ET AL.
10.1029/2020GB006888
7 of 14
Global reservoir area used
(103km2)CO2 (Tg CO2 eq yr−1) CH4 (Tg CO2 eq yr−1)
CO2+CH4 (Tg CO2 eq
yr−1)
CO2+CH4 (Mg
CO2 eq km−2yr−1)
This study 350 328 (276–414) 748 (454–1,998) 1,076 (730–2,412) 3,074 (2,086–6,891)
This study (no degassing) 350 328 (276–414) 337 (227–737) 664 (503–1,151) 1,897 (1,437–3,289)
Deemer etal.(2016) 311 135 606 741 2,383
Hertwich(2013) 330 279 331 610 1,815
Bastviken etal.(2011) 340 136 400
Barros etal.(2011) 500 176 680 856 1,712
St-Louis etal.(2000) 1,500 1,000 2,380 3,380 2,253
Notes. All values have been converted to mass CO2 eq. yr−1 using a 100-year greenhouse warming potential factor of 34 for CH4. Ranges in parentheses represent
95% confidence intervals for G-res v3 estimates. Differences between studies in total estimated global reservoir surface area generally result from differences
in statistical methods for estimating the cumulative surface area of small reservoirs. In some cases total CH4 fluxes differ slightly from sums of individual flux
pathways due to rounding.
Table 3
Global Reservoir, Year-2020 Fluxes of CO2 and CH4 From This Study and Other Recent Analyses

Global Biogeochemical Cycles
comparable to other major global anthropogenic CH4 sources such as landfills, biomass burning, and rice
paddies (68, 29, and 30Tg CH4 y−1, respectively; Saunois etal.,2020). We estimate that collectively reservoirs
account for ∼6% (range: 3.7%–17.4%) of total global anthropogenic CH4 emissions (340–381Tg CH4 y−1) and
∼14% (range: 7.1%–50.4%) of total global freshwater CH4 emissions (Saunois etal.,2020).
The likely global importance of CH4 degassing is also interesting from a GHG management perspective as
it suggests some potential to limit or reduce reservoir CH4 emissions. G-res output suggests that if all res-
ervoir CH4 degassing were eliminated, global GHG fluxes from reservoirs would be reduced by 31%–52%.
Because high degassing fluxes are most likely to occur when water is drawn through a dam from a reser-
voir's hypolimnion, where low-O2 conditions allow CH4 to accumulate, constructing dams that draw water
from well-oxygenated near surface portions of a reservoir's water column or managing water withdrawals
to minimize release of hypolimnetic waters could substantially reduce downstream degassing CH4 fluxes.
For example, a simulated increase in water withdrawal depth by as little as 3m (from the hypolimnion to
the metalimnion) yielded a 92% reduction in CH4 degassing emissions from a Malaysian reservoir (Batang
Ai) (Soued & Prairie,2020). It is also possible that retrofitting existing dams with epilimnetic water with-
drawal structures or hypolimnetic aeration systems could reduce downstream degassing, while also likely
mitigating other environmental impacts of dams on downstream ecosystems (Michie etal., 2020). While
the effectiveness, costs, and potential tradeoffs associated with these solutions are not yet documented, they
certainly deserve further attention and study.
3.3. Spatial Distribution of Reservoir GHG Fluxes
G-res allows a first-ever analysis of the global distribution of reservoir greenhouse gas fluxes that takes into
account characteristics of reservoirs beyond reservoir surface area. G-res estimates of reservoir per-area
greenhouse gas fluxes were quite variable, spanning more than three orders of magnitude (Range: 115–
145,472g CO2 eq. m−2y−1). Highest per-area fluxes generally occurred near the equator and decreased at
higher latitudes (Figure3a). Further, the very highest per-area fluxes were driven primarily by CH4 degas-
sing, although ebullition fluxes were also quite substantial. For example, CH4 degassing was the largest sin-
gle flux pathway for GHG emissions in all of the top 10 GHG emitting reservoirs globally. The pattern where
high fluxes occurred near the equator was not particularly surprising given G-res model structure (i.e., that
effective temperature is an input parameter to G-res diffusive CO2 and CH4 sub-models), and is consistent
with studies showing increasing CH4 production rates with increasing temperatures (Barros etal.,2011;
Thottathil etal.,2019; Yvon-Durocher etal., 2014), although see also Deemer et al. (2016) and Deemer
and Holgerson(2021). In addition, annual fluxes tended to be lower at high latitudes due to the influence
of freezing and the associated reduction of gas production. Although the pattern of decreasing fluxes with
increasing latitude was unsurprising, an understanding of the magnitude of the latitude effect on per-area
GHG fluxes is new. Further, the dominance of degassing and ebullition CH4 fluxes throughout the tropics is
a novel insight, deserving further investigation.
Similar to per-area fluxes, the highest GHG mass fluxes also occurred at low latitudes. 60.4% of total CH4
emissions were estimated to occur between the Tropic of Cancer and the Tropic of Capricorn, and 75.2% of
CH4 emissions were estimated to occur within “tropical” climate zones, as defined by the IPCC (Rubel &
Kottek,2010). The tropical latitude band contained 14.9% of the reservoirs in our global database, while the
subtropical band contained 22.7%. Despite some similarities in distribution between per-area GHG fluxes
and GHG total mass fluxes, there were also some important differences, due to the uneven distribution of
reservoir surface area by latitude (Compare Figures3a and 3b). Mass fluxes of different gases exhibited
different patterns with latitude, as did fluxes due to different CH4 flux pathways (TableS3). CO2 dominated
fluxes at high northern latitudes (>63°N), accounting for 83% of total boreal reservoir GHG fluxes, whereas
CH4 accounted for almost half (48%) of the total greenhouse warming potential from reservoirs in temper-
ate latitudes and the majority of total reservoir-sourced greenhouse warming potential in subtropical, and
tropical latitudes (66.5% and 77.9%, respectively; Figure3b). This latitudinal pattern results from the fact
that CH4 emissions (for CH4 diffusion and, consequently, CH4 degassing) tended to be depressed by low
temperatures at high latitudes or (for ebullition) by low estimates of cumulative horizontal irradiance or
low numbers of ice-free months. The relative contribution of bubbling is also variable among regions. In
the boreal zone, CH4 bubbling and degassing represented comparatively small fractions of total CH4 fluxes
HARRISON ET AL.
10.1029/2020GB006888
8 of 14

Global Biogeochemical Cycles
(33.6% and 13.2%, respectively) whereas these flux pathways comprised a greater fraction of total CH4 fluxes
in temperate and tropical zones. Degassing was particularly important at tropical latitudes, accounting for
49% of all GHG fluxes (in CO2 eq.) in that region. However, degassing accounted for a much smaller fraction
of total GHG in the temperate zone (5.2%), where ebullition was a more important flux pathway, accounting
for 38.5% of the GHG flux (in CO2 eq.).
CH4 degassing, CH4 ebullition, and CO2 diffusion were each the single largest GHG flux pathway for 287
(6.2%), 1,513 (33.2%), and 2,763 (60.6%) of the reservoirs in the G-res database, respectively, whereas CH4
diffusion was never the dominant flux pathway in any reservoir (Figure4). Hence, although CH4 degassing
constitutes a large portion of the global flux, this large flux is due to large per-area fluxes from a small frac-
tion of reservoirs globally. In fact, according to G-res, the top 100 degassing reservoirs account for >90% of
the global CH4 degassing flux, and almost half (47%) of the global CH4 degassing flux can be attributed to
just 10 reservoirs. Although degassing fluxes have not been measured in most of these systems, CH4 degas-
sing fluxes have been measured in the reservoirs with the second and third highest predicted CH4 degassing
yields globally (Balbina and Tucurui reservoirs, respectively). In each case G-res-estimated CH4 degassing
flux was within a factor of two of direct measurement-based estimates. High degassing fluxes from a rela-
tively small number of reservoirs, generally occurring at tropical latitudes, suggests that G-res predictions
are sensitive to assumptions about which reservoirs contribute CH4 to the atmosphere via this pathway.
This highlights the need for both a broader empirical assessment of degassing emissions across a diversity
HARRISON ET AL.
10.1029/2020GB006888
9 of 14
Figure 3. Global distribution and magnitudes of reservoir total (CH4+CO2) GHG fluxes. Each one-degree grid-cell is color-coded according to G-res-predicted
total CO2 plus CH4 flux from reservoirs in that cell. Panel A shows average per-reservoir-area rates of emissions (g CO2 eq. m−2yr−1), with average per-reservoir-
area fluxes for each 1-degree latitude band shown in the stacked line plot to the right (also g CO2 eq. m−2yr−1). Panel B shows total mass fluxes of CH4+CO2
GHG fluxes from reservoirs in each 1°×1° grid cell (Gg CO2 eq. reservoir−1yr−1). Distribution of total mass flux by emission pathway and 1° latitude band is
shown to the right of the map. In both panels, fluxes in the stacked line plots are additive such that in panel A the height of peaks represents average per-
reservoir-area total flux per degree of latitude, while in panel B the height of the peaks represents the total reservoir-sourced GHG flux to the atmosphere for
each 1° latitude band. White areas are regions lacking large reservoirs. Gridded model output available for download (Harrison etal.,2021).

Global Biogeochemical Cycles
of reservoirs, and more data on reservoir intake depth. Nevertheless, from a GHG mitigation perspective,
the skewed distribution of reservoir degassing is intriguing in that it suggests that substantial reductions in
CH4 emissions via degassing may be possible by focusing mitigation efforts on a relatively small number of
high-flux systems. In contrast to CH4 degassing fluxes, which were limited to a relatively small subset of the
global total number of reservoirs, CO2 diffusion was the single largest flux from more than half of reservoirs
in the G-res database, but these fluxes were on-average much smaller than degassing and ebullitive fluxes
on a per-area CO2 eq. basis. Focusing solely on CH4 fluxes, a somewhat different picture emerges. In this
case, CH4 degassing CH4 ebullition, and CH4 diffusion, were the single largest CH4 flux in 504 (11.0%), 3,338
(73.2%), 721 (15.8%) of G-res database reservoirs, respectively, highlighting the importance of ebullition as
the largest CH4 flux pathway in a majority of reservoirs. Dominant fluxes varied regionally in surprisingly
consistent ways. Regions where ebullition was the single greatest GHG flux occurred in perennially ice-free
regions with high rates of cumulative annual solar irradiance, consistent with observations in the few res-
ervoirs where solar irradiance has been considered as a potential driver of CH4 ebullition (Wik etal.,2014;
FigureS3). In contrast, reservoir GHG fluxes in temperate regions with low cumulative annual solar irradi-
ance tended to be dominated by CO2 diffusion (Figure4 and FigureS3). CO2 diffusion also dominated GHG
fluxes in north temperate and, especially, boreal regions. CO2 diffusion was also the single largest GHG
flux pathway throughout much of China, Japan, Korea, and the Eastern United States. In contrast, CH4
ebullition tended to dominate GHG fluxes in the western United States, parts of Brazil, throughout much
of southern Europe, in much of Africa, India, New Zealand, and western Australia. Degassing was only
infrequently the dominant flux pathway regionally, but the regions where it was important included: the SE
US, parts of Brazil, Eastern Europe, Central and Northern Africa (in the Nile River valley) as well as parts
of both western and eastern China (Figure4).
Importantly, the highest G-res-predicted per-area rates of emission occurred in exactly the regions where
the majority of ongoing and planned new reservoir construction is anticipated to occur in coming dec-
ades: the developing tropics and subtropics (Zarfl etal.,2015). This suggests that dam construction could
significantly increase global GHG fluxes from reservoir systems global-
ly. Furthermore, the two largest estimated pathways for CH4 emissions
(ebullition and degassing) are also the most uncertain due to limited
measurements (Tables2 and4) and comparatively unstable models (i.e.,
models that are sensitive to small changes in input parameters [see Sec-
tion3.4 below]). In the case of the degassing submodel, G-res estimates
may be too low if hypolimnetic water release is more common than we
estimate. Conversely G-res degassing estimates could be too high if hy-
polimnetic water release is less common or less constant (e.g., due to
seasonal destratification) than estimated here. More work is required to
better constrain these fluxes, both to improve understanding of aquatic C
cycling and provide GHG management-relevant information. Similarly,
HARRISON ET AL.
10.1029/2020GB006888
10 of 14
Figure 4. Dominant flux pathways for C-based greenhouse gases for each 1°×1° grid cell. White areas are regions lacking large reservoirs.
Flux pathway (Tg CO2 eq yr−1)
CH4 diffusion 54 (42–77)
CH4 ebullition 283 (184–660)
CH4 degassing 411 (227–1,261)
CO2 diffusion 328 (276–414)
Total (all flux pathways) 1,076 (730–2,412)
Table 4
G-res-Predicted Global, Year-2020 Fluxes of CH4 and CO2 by Flux
Pathway

Global Biogeochemical Cycles
G-res does not currently provide an estimate of CO2 degassing. Although
this component is likely small at the global scale, it may be significant in
individual cases and should be further investigated.
3.4. Model Sensitivity, Efficiency, and Future Directions
3.4.1. Model Efficiency Analysis
An analysis of model efficiency, wherein model components are re-
moved sequentially to evaluate the contribution of each to model pre-
dictive capacity, suggests that climate and geomorphological parameters
are particularly important in determining G-res model skill (Table6). In
particular, the removal of the temperature-associated parameters (cu-
mulative radiation and ice cover duration parameters) from the CH4
ebullition model each substantially decreased G-res model skill, with
the removal of each of these parameters from the model decreasing the
NSE from 0.58 to to −6.13 (Tables2 and6). The removal of the temper-
ature parameter from the CH4 degassing model (not directly, but rather
through its influence on the diffusive CH4 model, which in-turn is an input to the CH4 degassing model)
substantially decreased that model's skill (NSE decreased from 0.74 to −3.24; Tables2 and5), and the re-
moval of the temperature parameter from the CH4 diffusion model decreased that G-res submodel's NSE
from 0.84 to 0.46 (Tables2 and6). The temperature parameter in the CO2 diffusion model had very little
effect (decreased NSE value by only 0.17; Table6), but the strong impact of the removal of temperature
parameters on three out of the four G-res-estimated flux pathways suggests that temperature should be
an important focus for attention and improvement in future iterations of the G-res model. The predic-
tive skill of the G-res CH4 ebullition and CH4 diffusion submodels was also strongly dependent on the
estimate of littoral area. Excluding the littoral area parameter from the CH4 ebullition model decreased
NSE from 0.58 to 0.30, and excluding the littoral area parameter from
the CH4 diffusion model decreased NSE from 0.84 to −0.36. In addition
to its direct impact on CH4 ebullition and CH4 diffusion model skill, the
littoral area parameter also affects the CH4 degassing model through its
impact on predicted CH4 diffusion, which is a critical input to the CH4
degassing model. Model skill for the CH4 degassing submodel (NSE)
decreased from 0.74 to −0.26 when the littoral area parameter was re-
moved from the diffusive CH4 flux model (Table6). Because the littoral
estimate strongly influences all three CH4 flux pathways, which col-
lectively account for about 70% of the global greenhouse gas liability
due to reservoir emissions (CO2 eq.), this is a very important parameter
on which to focus energy in developing future iterations of the G-res
model. The littoral area parameter is quite uncertain as it necessarily
(due to a lack of better global scale data) relies on a simple algorithm
estimating reservoir bathymetry as a function of average and maximum
depth (See equations for "Littoral Fraction" and "Bathymetric Shape" in
Table S1). Hence by improving estimates of this parameter, for example
by using better reservoir geomorphology information, future iterations
of G-res may improve substantially. In contrast to temperature-related
and geomorphometric parameters, G-res model skill was relatively ro-
bust to removal of other input parameters such as reservoir age, water
residence time, soil C content, total P loading, and reservoir surface
area. In each case, parameter removal reduced model NSE by less than
0.32 units, and often far less (Table6).
3.4.2. Model Sensitivity Analysis
A sensitivity analysis in which G-res model inputs and coefficients were
increased by 10% in order to evaluate model response (Table 7) was
HARRISON ET AL.
10.1029/2020GB006888
11 of 14
Climate zone
CH4 and CO2% of
total(Tg CO2 eq. yr−1)
Boreal 29.6 2.8
Cool temperate 107.6 10.0
Temperate warm/dry 61.4 5.7
Temperate warm/moist 67.7 6.3
Tropical dry/montane 350.1 32.5
Tropical moist/wet 456.4 42.4
Total Temperate 129.1 22.0
Total Tropical 806.5 75.0
Table 5
G-res-Estimated Greenhouse Gas Fluxes by Climate Zone for Year-2020
Conditions
Model parameter removed
Resulting
NSE
Change in
NSE from
original model
No solar radiation (CH4 ebullition) −6.13 −6.71
No ice cover correction (CH4 ebullition) −6.13 −6.71
No temperature (CH4 degassing) −3.24 −4.08
No littoral area (CH4 diffusion) −0.36 −1.20
No littoral area (CH4 degassing) −0.26 −1.00
No temperature (CH4 diffusion) 0.46 −0.38
No reservoir age (CO2 diffusion) 0.58 −0.32
No littoral area (CH4 ebullition) 0.30 −0.28
No temperature (CO2 diffusion) 0.72 −0.17
No reservoir age (CH4 diffusion) 0.69 −0.15
No water residence time (CH4 degassing) 0.64 −0.10
No total P (CO2 diffusion) 0.84 −0.05
No soil C content (CO2 diffusion) 0.85 −0.04
No reservoir area (CO2 diffusion) 0.87 −0.02
Notes. Model parameters where removal resulted in loss of little model
skill (<0.38 NSE units) are shaded whereas parameters to which model
skill was particularly sensitive are left unshaded.
Table 6
Results From a Model Efficiency Analysis Showing How Nash Sutcliffe
Efficiency (NSE), an Indicator of Model Skill, Changes as a Function of
Removal of Individual Model Components

Global Biogeochemical Cycles
consistent with the efficiency analysis in that it suggested that G-res
model predictions are sensitive to small changes in temperature- and
geomorphology-related parameters but comparatively insensitive to
changes in other parameters. The overall model was quite insensitive
to 10% increases in the C content of inundated soils, total P content,
and reservoir age, with model predictions changing far less than 10%
in each case. As expected, based on the model formulation, predicted
fluxes scale approximately linearly with reservoir surface area. In con-
trast, but consistent with the results of the efficiency analysis described
above, G-res predictions were very sensitive to small (10%) increases
in (and hence to small errors in estimates of) littoral area fraction in
reservoirs, increasing more than 11-fold. G-res predictions were also
quite sensitive to changes in temperature-related parameters, includ-
ing effective temperature, cumulative radiation, and the length of the
ice-free season. As with the littoral area parameter, there is also room
for improvement in the temperature-related parameters. At the global
scale there is not currently a global water temperature database for res-
ervoirs; nor is there a widely accepted method to reliably and accurately
link sediment and water column temperatures to air temperatures. Any
enhancements inability to model water temperatures at large scales is
likely to enhance understanding of regional and global reservoir GHG
fluxes.
3.4.3. Future Directions
Taken together, results from efficiency and sensitivity analyses suggest that littoral area and temperature
are two major sources of G-res model uncertainty, and hence opportunities for model enhancement.
Work to link air temperatures to water temperatures (esp. bottom-water temperatures) and improve
estimates of lake bathymetry and stratification dynamics at large scales would both be useful. These
two parameters are also changing with a changing climate (Kraemer etal.,2015; O'Reilly etal.,2015)
and likely to change further in coming decades in manners that are likely to accelerate reservoir GHG
emissions (i.e., higher temperatures, deeper thermoclines; Woolway etal.,2020), highlighting a need
to study, monitor, and improve methods for estimating changing water temperatures and stratification
regimes.
In addition to better constraining temperature and geomorphometric model input parameters, there
are some additional model improvements that should be considered in future iterations of regional
and global lake and reservoir GHG emissions models. One of these improvements is the inclusion of
an explicit trophic status or primary production parameter. Several recent studies have reported strong
correlations between primary production and CH4 emissions (Beaulieu etal.,2019; Deemer etal.,2016;
DelSontro etal., 2018; Harrison etal.,2017), and there is good reason to think there is a causal link
between primary production and CH4 emission. By providing organic C and creating the anoxic condi-
tions that favor CH4 production, biological production in surface waters is likely to fuel higher rates of
CH4 emission, leading to higher emissions from eutrophic systems than oligotrophic systems. To some
extent, G-res models use cumulative radiance and total phosphorus concentrations as indirect and im-
perfect proxies for primary productivity. Yet, no large-scale models, including G-res, currently include
a primary production term as a model input. This is because there are no global datasets of lake or res-
ervoir primary productivity that are sufficiently robust to be of use for this application. However, work
is ongoing in this area (Sayers etal.,2015), and this is likely to change within the next several years.
Another area meriting additional investigation is understanding how global reservoir GHG emissions
will change in the future with the ongoing and anticipated global boom in reservoir impoundment
(Zarfl etal.,2015).
HARRISON ET AL.
10.1029/2020GB006888
12 of 14
Model parameter modified
% Change in G-res-predicted global
CO2+CH4 emissions resulting
from a 10% Increase in input
parameter values (g CO2 eq. yr−1)
Littoral area fraction 191.1
Effective temperature 60.9
Solar radiation 33.7
Ice-free period 33.7
Reservoir surface area 10.1
Actual age plus 10years −3.6
Water residence time 2.4
Total P 2.5
Inundated soil C content 0.8
Notes. Values in bold are greater than 10%, indicating the G-res model is
comparatively sensitive to changes in the associated input parameters.
Table 7
Results of a Model Sensitivity Analysis Showing G-res Predictions of Total
Global Reservoir GHG Flux Change as a Function of a 10% Increase in
Various Model Drivers

Global Biogeochemical Cycles
4. Conclusion
Despite remaining uncertainties and opportunities for further improvement, the work presented here rep-
resents a significant step forward in understanding and representing global and regional reservoir GHG
fluxes. Here we present the first-ever spatially and temporally explicit, global estimates of reservoir CO2 and
CH4 fluxes modeled for individual reservoirs and not simply based on the product of mean flux rates and
reservoir surface area. In addition, we present a first-ever spatially explicit estimate of the global reservoir
CH4 degassing flux. Analysis of these novel results grants several important new insights including the
following: (a) diffusive CH4 fluxes are probably lower than has previously been estimated; (b) CH4 fluxes
via ebullition and degassing are larger than previously recognized, but also quite poorly constrained; (c)
these fluxes are highest in the tropics and subtropics, which, together, are expected to account for 65%–75%
of new hydropower dam construction and reservoir impoundment in coming decades (Zarfl etal.,2015);
(d) global distribution of emissions shows that the contribution of CO2 flux is the most important in boreal
climate while CH4 degassing and ebullition contribution is dominant in tropical and subtropical climate;
and (e) G-res estimated reservoir GHG fluxes are quite sensitive to input parameters that are both poorly
constrained and likely to be strongly influenced by climate change in coming decades. Together these re-
sults highlight a critical need both to better understand climate-related drivers of GHG emission and the
relationship between these drivers and the highly uncertain CH4 ebullition and degassing fluxes.
Data Availability Statement
Datasets and model output for this research are available in these in-text data citation references: Harrison
etal.(2021), and Prairie etal.(2021).
References
Abril, G., Guérin, F., Richard, S., Delmas, R., Galy-Lacaux, C., Gosse, P., etal. (2005). Carbon dioxide and methane emissions and
the carbon budget of a 10-year old tropical reservoir (Petit Saut, French Guiana). Global Biogeochemical Cycles, 19(4). https://doi.
org/10.1029/2005GB002457
Barros, N., Cole, J. J., Tranvik, L. J., Prairie, Y. T., Bastviken, D., Huszar, V. L. M., etal. (2011). Carbon emission from hydroelectric reservoirs
linked to reservoir age and latitude. Nature Geoscience, 4(9), 593–596. https://doi.org/10.1038/ngeo1211
Bastviken, D., Tranvik, L. J., Downing, J. A., Crill, P. M., & Enrich-Prast, A. (2011). Freshwater methane emissions offset the continental
carbon sink. Science, 331, 50. https://doi.org/10.1126/science.1196808
Beaulieu, J. J., DelSontro, T., & Downing, J. A. (2019). Eutrophication will increase methane emissions from lakes and impoundments
during the 21st century. Nature Communications, 10. https://doi.org/10.1038/s41467-019-09100-5
Butman, D., & Raymond, P. A. (2011). Significant efflux of carbon dioxide from streams and rivers in the United States. Nature Geoscience.
Retrieved from http://www.nature.com/ngeo/journal/vaop/ncurrent/abs/ngeo1294.html
Cole, J. J., Prairie, Y. T., Caraco, N. F., McDowell, W. H., Tranvik, L. J., Striegl, R. G., etal. (2007). Plumbing the global carbon cycle: Integrat-
ing inland waters into the terrestrial carbon budget. Ecosystems, 10, 172–185. https://doi.org/10.1007/s10021-006-9013-8
Deemer, B. R., Harrison, J. A., Li, S., Beaulieu, J. J., Delsontro, T., Barros, N., etal. (2016). Greenhouse gas emissions from reservoir water
surfaces: A new global synthesis. BioScience, 66, 949–964. https://doi.org/10.1093/biosci/biw117
Deemer, B. R., & Holgerson, M. A. (2021). Drivers of methane flux differ between lakes and reservoirs, complicating global upscaling
efforts. Journal of Geophysical Research: Biogeosciences, 126. https://doi.org/10.1029/2019JG005600
DelSontro, T., Beaulieu, J. J., & Downing, J. A. (2018). Greenhouse gas emissions from lakes and impoundments: Upscaling in the face of
global change. Limnology and Oceanography Letters, 3(3), 64–75. https://doi.org/10.1002/lol2.10073
Downing, J. A., Prairie, Y. T., Cole, J. J., Duarte, C. M., Tranvik, L. J., Striegl, R. G., etal. (2006). The global abundance and size distribution
of lakes, ponds, and impoundments. Limnology & Oceanography, 51(5), 2388–2397. https://doi.org/10.4319/lo.2006.51.5.2388
Harrison, J. A., Deemer, B. R., Birchfield, M. K., & O'Malley, M. T. (2017). Reservoir water-level drawdowns accelerate and amplify meth-
ane emission. Environmental Science and Technology, 51(3), 1267–1277. https://doi.org/10.1021/acs.est.6b03185
Harrison, J. A., Prairie, Y. T., Mercier-Blais, S., & Soued, C. (2021). Year 2020 reservoir methane and CO2 emissions as predicted by the g-res
model. Zenodo. https://doi.org/10.5281/zenodo.4632428
Hertwich, E. G. (2013). Addressing biogenic greenhouse gas emissions from hydropower in LCA. Environmental Science and Technology,
47(17), 9604–9611. https://doi.org/10.1021/es401820p
Inglett, K. S., Inglett, P. W., Reddy, K. R., & Osborne, T. Z. (2012). Temperature sensitivity of greenhouse gas production in wetland soils of
different vegetation. Biogeochemistry, 108(1–3), 77–90. https://doi.org/10.1007/s10533-011-9573-3
Kraemer, B. M., Anneville, O., Chandra, S., Dix, M., Kuusisto, E., Livingstone, D. M., etal. (2015). Morphometry and average temperature af-
fect lake stratification responses to climate change. Geophysical Research Letters, 42, 4981–4988. https://doi.org/10.1002/2015GL064097
Lehner, B., Reidy Liermann, C., Revenga, C., Vorosmarty, C., Fekete, B., Crouzet, P., etal. (2011). Global reservoir and dam database, version
1 (GRanDv1): Dams, revision 01. NASA Socioeconomic Data and Applications Center (SEDAC). https://doi.org/10.7927/H4N877QK
Maavara, T., Chen, Q., Van Meter, K., Brown, L. E., Zhang, J., Ni, J., & Zarfl, C. (2020). River dam impacts on biogeochemical cycling.
Nature Reviews Earth & Environment, 1, 103–116. https://doi.org/10.1038/s43017-019-0019-0
Maavara, T., Lauerwald, R., Regnier, P., & Van Cappellen, P. (2017). Global perturbation of organic carbon cycling by river damming. Na-
ture Communications, 8. https://doi.org/10.1038/ncomms15347
HARRISON ET AL.
10.1029/2020GB006888
13 of 14
Acknowledgments
We thank Atle Harby, Jukka Alm, Sofia
D'Ambrosio, and Stephen Henderson
for input on early drafts of this manu-
script. Funding to JAH was provided
by an NSF INFEWS grant (NSF
EAR1639458), a GRIL Fellowship grant,
the Cox visiting professorship fund at
Stanford University, a U.S. Army Corps
of Engineers Climate Preparedness and
Resilience Programs grant, and a NSF
DEB Grant #1355211. Financial support
to YTP and SMB was provided by the
International Hydropower Associa-
tion for the development of the G-res
models.

Global Biogeochemical Cycles
Michie, L. E., Hitchcock, J. N., Thiem, J. D., Boys, C. A., & Mitrovic, S. M. (2020). The effect of varied dam release mechanisms and storage
volume on downstream river thermal regimes. Limnologica, 81, 125760. https://doi.org/10.1016/j.limno.2020.125760
Nash, E., & Sutcliffe, V. (1970). Part I-A discussion of principles. Journal of Hydrology, 10, 282–290. https://doi.
org/10.1016/0022-1694(70)90255-6
Newman, M. C. (1993). Regression analysis of log-transformed data: Statistical bias and its correction. Environmental Toxicology & Chem-
istry, 12, 1129–1133. https://doi.org/10.1002/etc.5620120618
O'Reilly, C. M., Sharma, S., Gray, D. K., Hampton, S. E., Read, J. S., Rowley, R. J., etal. (2015). Rapid and highly variable warming of lake
surface waters around the globe. Geophysical Research Letters. https://doi.org/10.1002/2015GL066235
Prairie, Y. T., Alm, J., Beaulieu, J., Barros, N., Battin, T., Cole, J., etal. (2018). Greenhouse gas emissions from freshwater reservoirs: What
does the atmosphere see? Ecosystems, 21(5). https://doi.org/10.1007/s10021-017-0198-9
Prairie, Y. T., Alm, J., Harby, A., Mercier-Blais, S., & Nahas, R. (2017). Technical documentation, UNESCO/IHA research project on the GHG sta-
tus of freshwater reservoirs. Version 1.1. https://www.hydropower.org/publications/the-ghg-reservoir-tool-g-res-technical-documentation
Prairie, Y. T., Mercier-Blais, S., Harrison, J. A., Soued, C., del Giorgio, P., Harby, A., etal. (2021). G-res tool modelling database. Zenodo.
https://doi.org/10.5281/zenodo.4711132
Raymond, P. A., Hartmann, J., Lauerwald, R., Sobek, S., McDonald, C., Hoover, M., etal. (2013). Global carbon dioxide emissions from
inland waters. Nature, 503(7476), 355–359. https://doi.org/10.1038/nature12760
Rosentreter, J. A., Borges, A. V., Deemer, B. R., Holgerson, M. A., Liu, S., Song, C., etal. (2021). Aquatic ecosystems are highly variable
sources contributing half of the global methane emissions. Nature Geoscience. https://doi.org/10.1038/s41561-021-00715-2
Rubel, F., & Kottek, M. (2010). Observed and projected climate shifts 1901—2100 depicted by world maps of the Koppen-Geiger climate
classification. Meteorologische Zeitschrift, 19(2), 135–141. https://doi.org/10.1127/0941-2948/2010/0430
Saunois, M., Stavert, A. R., Poulter, B., Bousquet, P., Canadell, J. G., Jackson, R. B., etal. (2020). The global methane budget 2000-2017.
Earth System Science Data, 12, 1561–1623. https://doi.org/10.5194/essd-12-1561-2020
Sayers, M. J., Grimm, A. G., Shuchman, R. A., Deines, A. M., Bunnell, D. B., Raymer, Z. B., etal. (2015). A new method to generate a
high-resolution global distribution map of lake chlorophyll. International Journal of Remote Sensing, 36(7), 1942–1964. https://doi.org
/10.1080/01431161.2015.1029099
Soued, C., & Prairie, Y. T. (2020). The carbon footprint of a Malaysian tropical reservoir: Measured versus modeled estimates highlight the
underestimated key role of downstream processes. Biogeosciences, 30(October), 1–22. https://doi.org/10.5194/bg-2019-385
St Louis, V. L., Kelly, C. A., Duchemin, É., Rudd, J. W. M., & Rosenberg, D. M. (2000). Reservoir surfaces as sources of greenhouse gases
to the atmosphere: A global estimate. BioScience, 50(9), 766–775. https://doi.org/10.1641/0006-3568(2000)050[0766:RSASOG]2.0.CO;2
Teodoru, C. R., Bastien, J., Bonneville, M.-C. C., del Giorgio, P. A., Demarty, M., Garneau, M., etal. (2012). The net carbon footprint of a
newly created boreal hydroelectric reservoir. Global Biogeochemical Cycles, 26(2). https://doi.org/10.1029/2011GB004187
Thottathil, S. D., Reis, P. C. J., & Prairie, Y. T. (2019). Methane oxidation kinetics in northern freshwater lakes. Biogeochemistry, 143(1),
105–116. https://doi.org/10.1007/s10533-019-00552-x
Tranvik, L., Downing, J., Cotner, J., Loiselle, S., Striegl, R., Ballatore, T., etal. (2009). Lakes and reservoirs as regulators of carbon cycling
and climate. Limnology & Oceanography, 54(6), 2298–2314. https://doi.org/10.4319/lo.2009.54.6_part_2.2298
Wik, M., Crill, P. M., Varner, R. K., & Bastviken, D. (2013). Multiyear measurements of ebullitive methane flux from three subarctic lakes.
Journal of Geophysical Research: Biogeosciences, 118(3), 1307–1321. https://doi.org/10.1002/Jgrg.20103
Wik, M., Thornton, B. F., Bastviken, D., MacIntyre, S., Varner, R. K., & Crill, P. M. (2014). Energy input is primary controller of methane
bubbling in subarctic lakes. Geophysical Research Letters, 41(2), 555–560. https://doi.org/10.1002/2013gl058510
Wik, M., Thornton, B. F., Bastviken, D., Uhlback, J., Crill, P. M., Uhlbäck, J., etal. (2016). Biased sampling of methane release from north-
ern lakes: A problem for extrapolation. Geophysical Research Letters, 43(3), 1256–1262. https://doi.org/10.1002/2015gl066501
Woolway, R. I., Kraemer, B. M., Lenters, J. D., Merchant, C. J., O'Reilly, C. M., & Sharma, S. (2020). Global lake responses to climate change.
Nature Reviews Earth & Environment, 1, 388–403. https://doi.org/10.1038/s43017-020-0067-5
Yvon-Durocher, G., Allen, A. P., Bastviken, D., Conrad, R., Gudasz, C., St-Pierre, A., etal. (2014). Methane fluxes show consistent tempera-
ture dependence across microbial to ecosystem scales. Nature, 507(7493), 488–491. https://doi.org/10.1038/nature13164
Zarfl, C., Lumsdon, A. E., Berlekamp, J., Tydecks, L., & Tockner, K. (2015). A global boom in hydropower dam construction. Aquatic
Sciences, 77(1), 161–170. https://doi.org/10.1007/s00027-014-0377-0
HARRISON ET AL.
10.1029/2020GB006888
14 of 14