ACPAtmospheric Chemistry and PhysicsACPAtmos. Chem. Phys.1680-7324Copernicus PublicationsGöttingen, Germany10.5194/acp-18-3779-2018How a European network may help with estimating methane emissions on the French national scalePisonIsabelleisabelle.pison@lsce.ipsl.frhttps://orcid.org/0000-0001-5471-7785BerchetAntoinehttps://orcid.org/0000-0001-6709-0125SaunoisMarielleBousquetPhilippeBroquetGrégoireConilSébastienDelmotteMarcGanesanAnitahttps://orcid.org/0000-0001-5715-8923LaurentOlivierMartinDamienO'DohertySimonRamonetMichelSpainT. GerardVermeulenAlexhttps://orcid.org/0000-0002-8158-8787Yver KwokCamillehttps://orcid.org/0000-0002-2181-2863Laboratoire des Sciences du Climat et de l'Environnement, LSCE-IPSL (CEA-CNRS-UVSQ), Université Paris-Saclay, 91191 Gif-sur-Yvette, FranceLaboratory for Air Pollution/Environmental Technology, Empa - Swiss Federal Laboratories for Materials Science and Technology, Dübendorf, SwitzerlandAgence Nationale pour la gestion des Déchets RadioActifs, Châtenay-Malabry, FranceCentre for Climate and Air Pollution Studies, School of Physics, National University of Ireland Galway, Galway, IrelandAtmospheric Chemistry Research Group, School of Chemistry, University of Bristol, Cantocks Close, Bristol, UKNational University of Ireland Galway, Galway, IrelandEnergy Research Centre of the Netherlands, Petten, the Netherlandsnow at: ICOS ERIC – Carbon Portal, Lund, SwedenIsabelle Pison (isabelle.pison@lsce.ipsl.fr)15March20181853779379817July20176October20171February20182February2018This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/4.0/This article is available from https://acp.copernicus.org/articles/18/3779/2018/acp-18-3779-2018.htmlThe full text article is available as a PDF file from https://acp.copernicus.org/articles/18/3779/2018/acp-18-3779-2018.pdf
Methane emissions on the national scale in France in 2012 are
inferred by assimilating continuous atmospheric mixing ratio measurements
from nine stations of the European network ICOS located in France and
surrounding countries. To assess the robustness of the fluxes deduced by our
inversion system based on an objectified quantification of uncertainties, two
complementary inversion set-ups are computed and analysed: (i) a regional run
correcting for the spatial distribution of fluxes in France and (ii) a
sectorial run correcting fluxes for activity sectors on the national scale.
In addition, our results for the two set-ups are compared with fluxes
produced in the framework of the inversion inter-comparison exercise of the
InGOS project. The seasonal variability in fluxes is consistent between
different set-ups, with maximum emissions in summer, likely due to
agricultural activity. However, very high monthly posterior uncertainties (up
to ≈ 65 to 74 % in the sectorial run in May and June) make it
difficult to attribute maximum emissions to a specific sector. On the yearly
and national scales,
the two inversions range from 3835 to 4050 Gg CH4 and
from 3570 to 4190 Gg CH4 for the regional and sectorial runs, respectively,
consistently with the InGOS products. These estimates are 25 to 55 % higher
than the total national emissions from bottom-up approaches (biogeochemical
models from natural emissions, plus inventories for anthropogenic ones),
consistently pointing at missing or underestimated sources in the
inventories and/or in natural sources. More specifically, in the sectorial
set-up, agricultural emissions are inferred as 66% larger than estimates
reported to the UNFCCC. Uncertainties in the total annual national budget are
108 and 312 Gg CH4, i.e, 3 to 8 %, for the regional and sectorial
runs respectively, smaller than uncertainties in available bottom-up products,
proving the added value of top-down atmospheric inversions. Therefore, even
though the surface network used in 2012 does not allow us to fully constrain all
regions in France accurately, a regional inversion set-up makes it possible
to provide estimates of French methane fluxes with an uncertainty in the
total budget of less than 10 % on the yearly timescale. Additional sites deployed
since 2012 would help to constrain French emissions on finer spatial and
temporal scales and attributing missing emissions to specific sectors.
Introduction
Methane (CH4) is the second most important anthropogenic greenhouse gas
in terms of impact on climate change (after CO2), due to its global
warming potential 28 times larger than that of CO2 over a 100 year
period , and possibly even larger .
Consequently, it is a very good candidate for climate change mitigation policies.
CH4 is emitted by a variety of sources. Most CH4 sources
(≈ 60 % in mass) are linked to microbial activity in anaerobic
environments: mainly natural wetlands, anthropogenically managed wetlands
(such as rice paddies), landfills, waste-water facilities and the intestines
of wild and domesticated animals. CH4 is also emitted from fossil-fuel-related
processes through natural geologic gas seeps or during the
exploitation and distribution of gas, oil and coal. Finally, CH4 is
emitted by biomass burning through incomplete combustion, mainly in wildfires,
biomass burning due to agricultural activities and the use of
biofuels. This variety of sources and the strong spatial and temporal
heterogeneity of emissions lead to uncertainties in CH4 global and
regional budgets, which remain large enough to impair our understanding of
atmospheric variations in CH4 concentrations, and particularly the
attribution of CH4 mixing ratio variations to specific sources and/or
zones .
CH4 emissions are reported yearly to the UNFCCC (United Nations
Framework Convention on Climate Change) by the countries that are parties to
the convention, both in the framework of the convention and of the Kyoto
protocol. Reporting CH4 emissions on the national scale to the UNFCCC is
currently done by bottom-up approaches, which include inventories (mainly for
anthropogenic emissions) and biogeochemical models (mainly for anthropogenic
emissions due to biogenic processes and natural emissions). For instance,
French methane emissions represent about 13 % of the EU-28 ones (according to
UNFCCC 2012 data) and are reported by the CITEPA (Centre Interprofessionnel
Technique d'Études de la Pollution Atmosphérique), an institute that
compiles inventories. Inventories are based on collecting and aggregating
huge amounts of data and information (e.g. activity statistics, emission
factors). The provides guidelines to build inventories for
reporting to the UNFCCC, classifying the methodologies in three tiers, from the
simplest to implement (Tier 1, which uses default activities and emission
factors provided by IPCC) to the most complex (Tier 3, which may include
models and is supposed to lead to smaller uncertainties). The Tier 1
uncertainty is the most straightforward to obtain since it combines the
uncertainties in the activity and the emission factor. From these guidelines,
the CITEPA provides annual emissions of CH4 in mainland France for
anthropogenic activity sectors together with Tier 1 uncertainties for the
major contributing sectors, ranging from 16 % (≈±212 Gg CH4
in 2012) for enteric fermentation to 104 % (≈±90 Gg CH4 in 2012)
for waste-water treatment and discharge . In October 2016,
another French inventory was released. This inventory, called
Inventaire National Spatialisé , provides emissions at a
kilometric resolution; for the year 2012, the kilometric maps are based on
the CITEPA's national totals. CH4 anthropogenic emissions for France are
also provided by larger-scale inventories: IER on the European
scale and four inventories covering the whole world: EDGAR ,
ECLIPSE , EPA and FAO . For
natural CH4 emissions in France, we use the emissions provided by
biogeochemical models on the global scale for wetlands and termites in the
context of . The difficulties of bottom-up approaches are
mainly due to missing information. For example, inventories may miss either
statistics on activity sectors or even sources. Moreover, inventory
uncertainties remain high, for instance, on the national scale due to errors
in the aggregation of statistical information or due to uncertainties in the
emission factors. Also, inventories do not often associate uncertainties to
their estimates. For UNFCCC reporting, the CITEPA provides uncertainties for
the main emitting sectors in France: the uncertainty on French anthropogenic
CH4 emissions in 2012 is then at least ±26 %.
In this context, top-down approaches may help with bringing more information to
emissions estimated by inventories. Top-down approaches are based on the
assimilation of atmospheric data (in our case, measurements of atmospheric
mixing ratios) into a chemistry-transport model using prior information on
the emissions. Within an inverse-modelling framework, the data, model and
prior emissions, together with their respective error statistics, are
optimally combined to provide posterior emissions (with their own
uncertainties, depending on the method used). The atmospheric signal
integrates all emissions so that sources which are not explicitly described
in bottom-up approaches are taken into account in top-down approaches.
Top-down approaches are widely used on the global scale for a review,
see. Recent studies have also used top-down approaches on
regional scales for large regions such as the Arctic ,
Eurasia , East Asia or the USA
. These regional studies are either global with a zoom or
focus over a specific region of interest or domain-limited on fine scales;
almost all of the studies use surface data, sometimes with the addition of
satellite data. Studies on national scales for countries about the size of
France are not numerous. In European studies, the atmospheric measurement
data are mostly provided from national and European surface networks:
estimated the Swiss national total of CH4 emissions;
examined the CH4 (and nitrous oxide) emissions in
Ireland and the United Kingdom; and
analysed methane emissions in Europe on the regional or country scales,
including France.
Although top-down studies are promising, their robustness is limited by
(i) the availability of observations, which must be numerous enough in time
and well-distributed in space over the relatively small (compared to the
global scale) area of interest; (ii) for most of them, the lack of
expert-knowledge for defining the set-up of the inverse system
(i.e. prescribing the error statistics, including the spatial and temporal
correlations in prior emissions, which may be assumed to be highly
country-dependent); and (iii) the issue of representing at best the
atmospheric transport on this scale. It is indeed important to assess which
spatio-temporal scales are actually constrained by the assimilated data in
order to exploit as much information as possible while avoiding
over-interpretation of the results (e.g. on too fine scales). This issue
arises particularly when estimating emission budgets on the national scales in
rather small countries, like France and most countries in western Europe.
Studies aiming at estimating European greenhouse gas (GHG) emissions can take
advantage of measurements from the ICOS (Integrated Carbon Observatory
System) network. ICOS is a European research infrastructure, of which one of
the main objectives is to quantify European GHG fluxes. To do so, a number of
European national measurement networks cooperate to ensure the monitoring of
GHG atmospheric concentrations and fluxes in terrestrial and marine
ecosystems, as well as the distribution of the data with a common high
quality standardization. The ICOS network of atmospheric stations performs
continuous in situ measurements, made both from ground stations and tall towers.
This study aims at estimating CH4 emissions in mainland France. We use
an inversion framework that allows us to overcome the issue of prescribed
error statistics. The data to assimilate are atmospheric measurements
available from the ICOS network in 2012. In particular, we aim at determining
whether the current status and deployment of the ICOS network is sufficient to
infer French methane emissions by answering the following questions. What
constraints may such a network bring on French emissions on the national
scale? What spatio-temporal scales are constrained in France, which is a
country with large regional variations in emissions? Which characteristics of
the French national budgets can be inferred: uncertainties, seasonal
variations, types of processes? For example, is it possible to infer seasonal
variations? Is the uncertainty in the total annual budget for France smaller
than the uncertainty in bottom-up inventories?
The methodological framework of our study is presented in
Sect. , with a focus on the tools provided for the
interpretation of the results (Sect. ). The inversion
set-ups used for inferring methane emissions in France are described in
Sect. . Results are discussed in
Sect. , first in terms of the relevancy of the features
informed by the inversion (Sects.
and ) then in terms of French methane emissions
(Sect. ).
Inverse methodGeneral inverse framework
In the framework of atmospheric inversion, the most common notations are the
following: x for the state vector, including the emission fluxes to
be optimized on the chosen spatial and temporal scales; xb for
the prior estimate of the state vector; yo for the
observation vector, consisting here of CH4 atmospheric concentration
data. The observations and the prior state are associated with their
covariance error matrices R and Pb,
respectively. R includes the errors on the measurements (e.g.
instrumental errors) plus the errors on the transport in the model and on the
representativity of the grid cell compared to the measurement. The link from
the state vector to the observation space is made by the observation operator H.
Here, H represents the atmospheric transport and mixing on the model's
grid and the space and time filtering of the simulated concentrations to
obtain the equivalent of the observation data. Since the lifetime of CH4
in the atmosphere is very long (≈ 9 years) compared to the residence
time of air masses in the domain of interest in this study (≈ 3–5 days),
chemistry is not taken into account so that H is assumed to be
linear and its Jacobian H is used, with
H(x)=Hx.
As mentioned previously, the inversion optimally combines the prior
knowledge, the knowledge on which the model is based and the knowledge
brought by the data to be assimilated: it consists of finding the probability
density function (pdf) of the state x knowing both the prior xb
and the differences between the observations yo and their equivalents
computed by the model H(xb). For any possible state x, this probability
is p(x|yo, xb). To characterize
p(x|yo, xb), it is common to use
the Bayesian framework and to assume that uncertainties in the system follow
Gaussian functions. As a result, the posterior state vector xa
and its associated covariance matrix of posterior errors Pa are given by
xa=xb+Kyo-Hxb,Pa=Pb-KHPb,
with K the Kalman gain matrix, given by
K=PbHTR+HPbHT-1.
If R and Pb are given, the inversion is a
direct computation from these formulae (providing the sizes of the matrices
are adapted to the computing resources). As stated before, R and
Pb are generally derived from expert knowledge based on
studies on the atmospheric transport, the performances of the models, etc.
Such knowledge is quite established for the global scale and large region
scales, but is not readily available yet for GHGs at the smaller national
scales. Therefore, defining R and Pb is not
an easy task on the country scale (scale of interest here), while
incorrectly specifying R and Pb and more especially
their relative weights, has a very strong impact on the results of the inversion.
Principle and main steps of the marginalized Bayesian inversion method
In order to avoid multiple tests on the structures and values of R
and Pb, we use the marginalized Bayesian inversion
method, which is an extension of the classical Bayesian inversion framework,
developed and implemented by . Instead of classically
inferring the posterior state xa and its covariance
matrix Pa directly from prescribed prior uncertainties in the
covariance matrices R and Pb, the method uses
a sample of the continuous distribution of all the possible couples of prior
uncertainties (R, Pb)i to produce an
ensemble of the posterior counterparts (xa,
Pa)i. The distribution of prior uncertainties
p(R, Pb) is computed by analysing the likelihood
of the innovation vector p(yo-xb|R, Pb, xb). The
final product of the marginalized inversion is the node of the aggregated
pdf (xa)i and its associated covariance matrix Pa.
The implementation of the method is divided into three main steps to derive the
optimal posterior state of emissions and the associated uncertainties.
First, the node of p(R, Pb) is obtained from the
maximum likelihood computed with a pseudo-Newtonian algorithm. This couple
(R, Pb)opt would actually give the
xa corresponding to the node of the posterior pdf
p(x|yo, xb) but with too small
posterior uncertainties. Therefore, in a second step, a Monte Carlo ensemble
on p(R, Pb) is used to get a sample of the whole
distribution of p(x|yo, xb), as
illustrated in Fig. . In the last step, the final
Pa is deduced from the shape of the distribution. As the
method is based on Monte Carlo estimates of the posterior distribution, the
computational costs should be tightly controlled. This is done by limiting
the detectable spatial and temporal resolutions of posterior fluxes in space
and time. The expert-knowledge required on the covariance matrices in the
classical method is then partially transferred to the definition of the
resolutions of the components of the state vector (described in
Sect. ). The relevancy of these choices may be checked a
posteriori by examining the posterior error covariances (see Sect. ).
When computing the maximum likelihood, emissions which are not constrained
enough are filtered out to avoid generating numerical artefacts on top of
aggregation errors. These under-constrained fluxes are detected with the
influence matrix, KHdefined by,
available at each step of the computation. The diagonal terms of this matrix
are between 0 and 1 and represent the sensitivity of each component of x
to the inversion. When the algorithm reaches a local minimum, the fluxes for
which the sensitivity is less than 0.5 are filtered out .
Statistic uncertainty in Bayesian inversion. The inversion infers
the posterior state xa from yo and
xb. In the classical Bayesian framework,
xa is inferred together with its uncertainty
Pa from the covariance matrices (R,
Pb) (a). To account for uncertainties in the
error statistics, an ensemble of (R,Pb) couples
can be tested to infer an ensemble of (xa,
Pa) (b), which are part of
p(x|yo, xb).
Large gradients in the concentrations, which are due to emission hotspots
are an issue. Peaks in the emissions generate fine plumes (in space and time)
that the transport model may not be able to simulate accurately. The
detection of such plumes is based on the diagonal terms in
(R, Pb) following a highly skewed pdf at the end
of the maximum likelihood . All observations for which the
uncertainty is in the largest 5 % of R are filtered out; all
regions of emissions for which the uncertainty is more than 500 % are also
filtered out. With this filtering, observations influenced by “hotspots” of
emissions are not assimilated and regions seen only through plumes are not
inverted.
Tools for the interpretation of the results
This analytical method with Monte Carlo ensemble gives access to quantifying
tools, which help to better understand the influence of the various
information sources within the inversion.
Prior uncertainty
The prior fluxes are provided by yearly inventories
(Sect. ) and their uncertainties are computed from our
marginalization (Sect. ). For unconstrained
components (for example, emission regions that never influence concentrations
at any measurement sites), prior uncertainties cannot be obtained. Therefore,
the uncertainty for these components is computed based on the mean of the
covariances of constrained components. The final prior uncertainty then
includes prior uncertainties for both constrained and unconstrained
components. This uncertainty represents the atmospheric point of view, i.e.
it estimates how well the prior fluxes enable the model to reproduce the
signal in the atmospheric concentrations. It is therefore higher when the
difference between simulated concentrations and the data is larger. In the
following it is called σprior.
Posterior fluxes and uncertainties
The posterior fluxes xa and their uncertainty
matrix Pa are determined from the Monte Carlo ensemble of
(xa, Pa)i
(Sect. ). As the distribution of (xa)i
is symmetric relative to its node, we compute
xa as the median of the Monte Carlo samples:
xa= median(xia). The posterior
uncertainties and correlations of errors are defined by the covariance matrix
of the ensemble (xa)i. Correlations are used to analyse
the temporal and spatial structure of the constraints on the fluxes provided
by the observation network. The posterior uncertainty is obtained from the
tolerance interval covering 68.27 % of the Monte Carlo ensemble of posterior
state vectors (xa)i. This uncertainty is then equivalent
to the 1σ interval in a Gaussian case and hereafter written σpost.
Temporal and spatial scales informed by the inverse system
The method provides the full posterior error covariance
matrix Pa (i.e. not only its diagonal terms). It is possible
to use the correlations in Pa to determine which
components of the state vector can be considered independent (in time and/or
space) from one another by the inversion. Due to atmospheric mixing and the
limited number of observations, the inversion may meet difficulties in
separating some regions. This is generally indicated by low uncertainty
reduction for these regions and high positive or negative correlations
between them. Here we use the correlations of errors to group blocks of
emissions (see Sect. ) as a conservative proxy for the
temporal and spatial scales constrained by the inversion.
Constrained fluxes and influence of the observation sites
The influence matrix KH gives the constraints on the fluxes. By
de-aggregating the influence according to the prior fluxes, and taking into
account the correlations, the distributed constraints on the fluxes are
obtained. They may be expected to be linked to the intensity of emissions and
to the distance to the stations.
The sensitivity matrix HK gives the sensitivity of the inversion
to a change in one component of the observation vector. An observation with a
high sensitivity brings strong constraints on the inversion. The weight of
each station in the inversion can be computed by summing up the corresponding
diagonal elements of HK.
Building inferred fluxes
As stated before, not all fluxes are constrained by the inversion because
some fluxes do not have any significant impact on the observations. Also,
some inverted fluxes may not be robust enough (see
Sect. ). To build total fluxes, we then use the
posterior emissions when available and robust, and the prior emissions
otherwise (see Sects. and ).
The obtained fluxes are hereafter called inferred fluxes (they are not the same as the posterior fluxes which result
directly from the inversion). The uncertainty on inferred fluxes is computed
from the prior and posterior uncertainties by assuming that the posterior and
prior parts are independent from each other and calculated as follows:
σinferred=σpost2+σprior2.
Error reduction
The final error reduction, after post-processing of the Monte Carlo outputs,
brought by assimilating the atmospheric data may be estimated with the following:
R=1-σinferredσprior×100.
Inversion set-ups
For this study, we use the domain-limited chemistry-transport model CHIMERE
at 10 × 10 km2 over France (Sect. ) and focus on the
year 2012 for which four stations provided CH4 measurements in France
and five in the neighbouring countries (Sect. ). Two
inversions are performed: one called “regional run” and the other “sectorial
run”. The regional run aims at estimating the total CH4 fluxes by
region. It consists of using geographical areas, defined so that the size of
the problem is reasonable but each area is physically consistent and
aggregation errors are assumed to be small. The sectorial run focuses on the
national CH4 emissions by sectors. It consists of using the various
sectors for methane sources available in the prior
(Sect. ) and assuming that each type of source is
consistent enough over the whole country to be inverted as a whole. As a
result the state vector is defined differently for the two runs (Sect. ).
The inferred fluxes of CH4 for 2012 are obtained from a series of
12 monthly inversions. In the following, the inversion set-up is given for 1 month,
the 12 monthly inversions having been run independently both for the
regional run and the sectorial run.
Horizontal grid used by CHIMERE (see Sect. ).
Resolution in the centre (mainland France): 10 km × 10 km for
98 × 98 grid cells. The sizes of grid cells increase in areas not
covering mainland France: 30, 50 and 80 km over 3, 3 and 2 rows of grid
cells.
Observation operator: CHIMERE model
The chemistry-transport model CHIMERE is an area-limited 3D Eulerian
chemistry-transport model www.lmd.polytechnique.fr/chimere/;,
embedded in the inversion system PYMAI developed at LSCE
. The full description of CHIMERE and
references are available in . The area of interest in our
study is mainland France, at a horizontal resolution of 10 × 10 km2.
Boundary conditions are interpolated from global simulations (see
Sect. for details). To limit the aggregation errors due
to the coarse resolution of boundary conditions, a buffer region around
mainland France is defined with intermediate horizontal resolutions
(Fig. ). With this grid, the global coarse information on
concentrations is only used on the scale of the hemispheric background, while
neighbouring regions are explicitly included in our simulations focussing on
mainland France. In the vertical, 29 levels are defined from the surface to
300 hPa, with a finer resolution close to the surface (first levels at
≈ 5, 40, 85 and 135 m a.g.l. then geometrical increase).
The model is forced by the European Centre for Medium-range Weather
Forecast (ECMWF) data, forecast at 12 h, available every 3 h and interpolated at
0.15∘× 0.15∘. The relevant fields (horizontal wind,
temperature, humidity, etc.) are then interpolated hourly on the horizontal
and vertical grid of CHIMERE. The transport schemes are of order 1 on the
vertical and 2 on the horizontal; deep convection is taken into account with Tiedke's scheme.
For each component of the state vector xb (see
Sect. ), response functions (i.e. the contributions of this
component to the simulated concentrations equivalent
Hxb to the observation data yo)
are computed. The 200 (for the regional run) or 136 (for the sectorial run)
simulations are then summed up.
The colours identify regions for emissions, 26 regions in France (numbers),
4 “outside” regions (letters A to D) and 1 sea region (E). Stars and names
are sites at which measurements were available in 2012 for CH4, see
characteristics in Table .
Observation vector
In 2012, measurements of atmospheric CH4 mixing ratios were available at
four stations in France and five stations in the neighbouring countries, mainly
north from France (Fig. ). Their coordinates are given
in Table . Hourly means of continuous data are all
reported on the same scale (NOAA2004). The measurements are made mostly by
optical instruments, such as Picarro or Caribou instruments and by gas
chromatographs at GIF and PUY . Taking
into account failures and maintenance of the instruments, data are not
available during the whole year, as indicated in Table
and on the time series in the Supplement (Sects. S1 and S3).
Since our problem is to be explicitly solved, the size of the error
covariance matrix for observations, R, must be small enough.
Moreover, the observation data used must be consistent with the space and
time resolutions chosen for the problem. Therefore, we used hourly means
(provided with the associated variance) computed from the continuous
measurements. When several levels are available at a site, only the highest
one is retained since the transport model is not always able to optimally
represent vertical mixing close to the surface.
Among the available data (Table ), we used hourly means
in the afternoon (defined as the period from 14:00 UTC – included – to 19:00 UTC – not included)
only when the boundary layer height is higher than 500 m
in the model (selected data displayed in Sects. S1 and S3). This choice is
made to avoid periods when the representation of vertical mixing in the model
is not adapted for atmospheric inversion .
Characteristics of the available stations at the time of the study
(see map in Fig. ). The altitude is above sea level; the
height is above ground level. The total number of available data is the number
of hourly means available for the whole year (i.e. maximum 8784). The number of
selected data is the number of hourly means available from 14:00 UTC (included)
to 19:00 UTC (not included) when the boundary layer height is higher than 500 m
in the model. The time coverage is computed over the afternoon hours
(14:00–18:00 UTC), i.e. 100 % for 1830 h. At PUY, two different instruments measure CH4.
The spatial distribution of the stations is not homogeneous throughout
France: stations are sparse in the most western part of the country and in
the south-east. The time coverage is also heterogeneous and sometimes sparse
(e.g. BIS, Table and Fig. S1 in the Supplement). Heterogeneous sampling
of atmospheric concentrations may influence the performance of the inversion,
which is further discussed in Sect. .
State vectors
For each monthly run, the fluxes are optimized on the weekly timescale: 3 weeks
of 8 days and a last “week” of 5 to 7 days depending on the month, leading to
a number of components of 4 times the number of regions or sectors. The
lateral boundary conditions are adjusted every 2 days (or 3 days at the
end of 31-day months) for each of the 4 lateral borders and the top of the
domain, leading to 75 (70 in February) components. The initial methane
concentrations are adjusted by one coefficient for the whole 3-D concentration
field at the first time step.
For the regional run, the French regions were delimited based on the land-use
and vegetation type, according to GlobCover v2.3 and
ECOCLIMAP . Limiting the size of the problem and according
to the two aforementioned maps, we chose to define 26 regions in France. Four
other regions were added to represent the neighbouring continental areas and
a last one for the sea. The 31 regions are represented in Fig. .
As a result, for the regional run, the state vector for one month has 200 components:
1 component for initial conditions;
75 components for boundary conditions (only 70 components in February); and
124 components for emissions (i.e. 31 regions during 4 “weeks”).
For the sectorial run, we use the SNAP (Selected Nomenclature for Air
Pollution) sectors from 1 to 10 for anthropogenic CH4 emissions (see
Table for the definition of the sectors). Other sources are
neglected (including natural emissions such as from wetlands and termites).
CH4 emissions are split into SNAP sectors over France only. For the
neighbouring continental regions and the sea, total emissions are used.
As a result, for the sectorial run, the state vector for one month has 136 components:
1 component for initial conditions;
75 components for boundary conditions (only 70 components in February);
40 components for emissions in France (i.e. the 10 SNAP sectors during
4 “weeks”); and
20 components for emissions in the 5 outlying areas (continental
areas A–D and sea E in Fig. ) during 4 “weeks”.
For each component, the propagation to compute the response function is 6 days
(the domain is supposed to be ventilated after this delay).
Prior yearly total methane emissions (in Gg CH4) in France from
IER interpolated on the model's grid; the crosses (x) indicate sectors which are
constrained by the atmospheric inversion (in the sectorial run).
SNAPDescriptionGg CH4% ofConstrainedthe total1combustion in energy and transformation industries20.12non-industrial combustion plants1073.4x3combustion in manufacturing industry20.14production processes20.15distribution of fossil fuel and geothermal energy943.0x6solvents and other product use007road transport210.78other mobile sources and machinery20.19waste treatment and disposal52216.8x10agriculture235675.8xTotal3108Prior informationInitial and boundary conditions
For the initial and boundary conditions in our domain, we use CH4
concentration fields optimized on the global scale for 2010, using the
inversion set-up of . The initial spatial resolution of
the 3-D fields was 3.75∘× 2.5∘, longitude and latitude
respectively, with 19 vertical levels from the surface to the stratosphere. A
time resolution of 48 h was used. These concentration fields were
spatially and temporally interpolated to our model resolution
(Sect. ). Even though 2012 was not available at the time of
our study, using 2010 values ensures that the large-scale variations at the
boundaries are realistic in terms of seasonal cycle. The impact on the final
results of using 2010 values instead of 2012 is small since boundary
conditions are optimized in the inversion (see Sect. ).
Methane emissions
Emission estimates used as prior knowledge of CH4 fluxes are taken from
the European annual anthropogenic emission inventory produced by IER
(Institut für Energiewirtschaft und Rationelle Energieanwendung,
Universität Stuttgart) for 2005 . This inventory estimates
French mainland annual CH4 emissions at 3108 Gg CH4. Emissions are
provided for 10 SNAP sectors, the main emitting sectors in France being
agriculture (SNAP10, about two-thirds of the total anthropogenic emissions)
and waste treatment and disposal (SNAP9, about 17 % of the total
anthropogenic emissions, Table ). SNAP5 (non-industrial
combustion plants) contributes ≈ 3.5 % of the total anthropogenic
emissions, and SNAP2 (distribution of fossil fuel) about 3 %. These four SNAP
sectors represent a total of 99 % of the prior emissions. SNAP6 (solvents and
other products) does not emit CH4. Sources other than those included in
these 10 sectors are neglected, including natural emissions such as from
wetlands since their total area (and contribution to atmospheric
concentrations) were assumed to be small in France. This assumption will be
further discussed in Sect. 4 when discussing the French methane yearly
budget. The choice of a larger-scale anthropogenic inventory has been made
because the CITEPA does not provide gridded emissions and the INS was not
available at the time of this study. Forward sensitivity tests have shown
that IER was the inventory ensuring the best performances over France in
simulating CH4 concentrations at stations compared to the global-scale
inventory EDGAR. The EDGAR v4.2 FT2012 inventory estimates
larger CH4 emissions over France (3866 Gg CH4 in 2012) and leads to
larger discrepancies between observations and forward simulations.
The IER CH4 inventory is available at a 10 min horizontal resolution
(about 15 km) for each SNAP sector. The emission maps were interpolated on
the grid of the model with an hourly time resolution. The total emission map
used as the prior is shown in Fig. , the emission
maps for each sector are presented in the Supplement (Sect. S2).
Results and discussion
One of the main objectives of this study is to assess CH4 emissions
using atmospheric data on the yearly timescale and to compare with bottom-up
estimates. With our method we can determine the components of the state
vector that are actually constrained in the inversion. This allows us to
define the spatial and temporal scales that are resolved by our system
(Sect. 4.1) and to determine how each station constrains the system and which
regions or sectors are constrained (Sect. 4.2). The inferred fluxes are
reconstructed from the posterior estimates of the constrained components and
the prior estimates for the un-constrained ones: this is first done on the
monthly timescale (Sect. 4.3.1) to discuss seasonal variations (Sect. 4.3.2), and
finally on the yearly timescale (Sect. 4.3.3) to compare our top-down estimate
with bottom-up ones.
Annual median CH4 emission fluxes in g CH4 m-2 in
France (a, prior from IER, inferred fluxes from the regional and
sectorial runs); differences inferred minus prior (b) for the
regional and sectorial runs; details for the sectorial run (c):
differences inferred minus prior for the four sectors which are actually seen,
SNAPs 2, 5, 9 and 10.
Spatio-temporal scales resolved by the inversion
Assessing the spatial and temporal scales resolved by an inversion system is
critical for establishing future network design strategies and correctly
analysing the outputs of the inversion. As detailed in
Sect. , the posterior error covariance matrix Pa
is used to assess which spatial and temporal scales
are solved by the inversion. Components of the state vector are considered to
be actually separated by the inversion when the associated correlations in
the posterior error covariance matrix Pa are lower
than a given threshold (see Sect. ). In the regional
run, the threshold must be set so as to avoid over-interpreting spatial
information; in the sectorial run, the threshold must be set so as to avoid
unduly separating sectors. In the following, a “block” is a set of components
that are considered correlated together (i.e. a group of components among
which the correlations are all higher than the chosen threshold). A given
block may include emissions for various weeks and various regions and/or sectors
together with initial conditions and boundary conditions. A high correlation
between fluxes and boundary conditions may be due to over-corrections of
emissions to create a background concentration signal: a few parts per billion of error in
boundary conditions can be compensated by non-realistic increments in fluxes
inside the domain; conversely, an error in emissions in the buffer regions
can be compensated by non-realistic increments in boundary conditions. This
is why, in the following, we discard such increments by taking into account
only blocks exclusively including emissions (neither initial nor boundary conditions).
The correlation threshold must be set at a value that avoids two issues
as explained by: too high a threshold
leads to always separating all the components (≥ 0.7
Fig. b), which implies a high risk of
over-interpreting small-scale results since patterns of corrections forming
dipoles in neighbouring regions are not grouped; whereas a lower value leads
to large blocks of regions covering half of France (≤ 0.3,
Fig. c). For this study, the correlation threshold
is set at a balanced value of 0.5, which gives the largest number of blocks
of more than one component (Fig. a) as well as a
small residual correlation between the blocks (Fig. d)
and the second smallest mean area covered by one block
(Fig. c). In the regional run, this mean block area
corresponds almost to the finest available spatial resolution for emissions
in the state vector (≈ 68 000 km2). The same threshold is set for
the sectorial run, which gives the second largest number of blocks of more
than one component (Fig. a).
(a) Annual number of blocks of at least two components
independent from both initial conditions (IC) and boundary conditions (BC)
for various correlation thresholds for the regional (black) and the sectorial
(blue) runs. (b) Annual total number of blocks (i.e. including
blocks of only one region also, compared to (a) independent of IC
and BC. The larger the correlation threshold is, the larger the total number
of blocks is and the smaller the number of blocks of at least two regions,
since less regions are considered correlated together. (c) Annual
mean area covered by a block for the regional run. (d) Annual mean
covariance between blocks for the regional run.
The components of interest correspond to the 26 French regions (numbered 1
to 26 in Fig. ) in the regional run or to the 10 SNAP
sectors in the sectorial run. With a correlation threshold of 0.5, in the
regional run, 260 components of interest are seen over the year, among
1248 weekly components (26 French regions × 4 “weeks” × 12 months),
and about 55 % of these 260 are correlated at least to another one. In the
sectorial run, 92 components of interest are seen over the year, among
480 components (10 sectors × 4 “weeks” × 12 months), and about 35 %
of these are in a block with at least another one.
The components that are seen and grouped indicate that the regional spatial
resolution with 26 regions is neither too coarse (individual regions are
seen) nor too fine (some regions are grouped together) from the atmospheric
point of view compared to the information that can be retrieved from the
atmospheric data into the emission space. More measurement sites would allow
the inversion to constrain emissions at a finer spatial resolution.
The weekly time resolution seems to be close to the finest resolution at
which the inversion is actually informative. The components corresponding to
a given region through the 4 weeks of a month are almost never grouped
(2 cases of 2 weeks in the same group among the 64 groups). Running inversions
with a coarser time resolution (e.g. bi-weekly or monthly), in the state
vector would therefore be equivalent to assuming perfect correlations between
weeks, which are not suggested by the information in the atmospheric signal.
Nevertheless, using results on the weekly timescale would lead to a risk of
over-interpreting the time windows when a posterior is available compared to
the weeks not seen by the inversion. Finally, the best compromise to
interpret the results of the inversion is to aggregate them at a coarser time
resolution (monthly and yearly), as described in
Sect. . Moreover, the yearly timescale is,
in fine, the one that top-down approaches have to target to be
integrated as control methods that check the national emission reports and
their trends, in order to meet societal and political needs.
Constrained areas in the regional run (described in
Sects. and ). The influence matrix
(a, influence for each grid cell given in % over the whole year) is
de-aggregated according to prior fluxes (b) to obtain the
constraints (c): the annual sum of constraints on CH4 emissions
by the atmospheric data is shown on a logarithmic scale (c,
non-dimensional).
Red is for a strong constraint. The spatial resolution is the
grid of the model (see Fig. ). Only fluxes independent from
initial and boundary conditions are used (see Sect. ).
Black bold lines show the borders of the regions; grey regions are never
constrained. The relative contributions of the stations in the inversion,
averaged over the year are shown on an arbitrary scale (d), white being for a
small contribution.
Constrained areas and sectors
By de-aggregating the influence matrix according to the prior fluxes, the
constraints on the fluxes are obtained (see Sect. ).
The constraint at a given time and location then depends both on how well a
source is detected in the atmospheric signal and on the intensity of the
flux. It is a good indicator of the efficiency of the inversion since there
is not much interest in having information on an area where the emissions are
known to be small or null. In Fig. , the total
annual constraints on regions independent from initial and boundary
conditions (see Sect. ) are displayed for the year 2012
together with the average weights of the stations (computed from the
sensitivity matrix, see Sect. ). These weights are
displayed on a scale with arbitrary units traceable to degrees of freedom of
the signal. For instance, BIS contributes more to the constraints than CBW in
the regional run (Fig. , see Sect. S3 for details
of the whole year at each site).
As the domain covers neighbouring areas as well as France itself, stations
outside France (CBW, CRP, MHD, RGL, TAC) can help with constraining fluxes outside
of France and the boundary conditions. To quantify the impact of these
stations on the constraints on the fluxes in France, a regional run was
carried out without them. The total annual sum of constraints on French
CH4 fluxes in the regional run without these outside-France stations is
more than 1.7 times smaller than the constraints provided in the reference
regional run assimilating data from all available stations.
When using the stations outside France, the influence of the components for
fluxes outside France and the boundary conditions is partly taken into
account by the information provided by the stations outside France. The
information provided by stations located in France is then more efficiently
used for constraining French fluxes, the influence of outside fluxes and
boundary conditions being otherwise taken out from the atmospheric signal.
As expected, the regional run shows that most areas where stations are sparse
are not well constrained. Thus, the south-east of France is not very well
constrained in 2012, more specifically regions in the Alps (18, 19 in
Fig. ) and close to the Mediterranean coast (9, 26,
13 in Fig. ). The Pyrenees (14 in
Fig. ) are not constrained at all, as well as regions
in the east (e.g. 1, 11 in Fig. ). Newly operated
stations in Germany or the south-east of France can thus be expected to
improve our spatial coverage of French CH4 emissions. Nevertheless, the
best constrained regions are not necessarily those where measurement sites
are located. Indeed, regions 3–5 (numbers in
Fig. ) in the west, are better constrained than
regions 22 and 23 (between OPE and GIF) and region 15 (close to PUY). The
spatial distribution of constraints actually depends on the intensity of
fluxes and of the distance to stations. The best constrained fluxes are not
necessarily the closest to the stations because plume situations are filtered
out in the inversion (see end of Sect. ). The
best constrained fluxes are then in areas upwind of the stations at distances
between 100 and 300 km, when plumes are spread out and the atmospheric signal
is smoothed enough to be compared with the transport model. As a result,
Brittany (regions 3 and 5 in Fig. ) is well
constrained (Fig. c) although it is far from the
stations, because the prior fluxes are among the most intense
(10–20 g m-2, Fig. b) and the western
circulation brings well-mixed air masses from this region to GIF. PUY does not always constrain the regions closest to the station
(15 and 20 in Fig. ) very well: the local transport
brings filtered-out plumes from local emissions when the station is in the
boundary layer and clean air masses (containing almost no information on
French surface fluxes) when the station is in the free troposphere. The
region close to BIS is not well constrained as the wind comes either from the
Atlantic ocean, with no influence from the French emissions, or from the
east, with either relatively small fluxes (0.5–2 g m-2,
Fig. b) or local plumes from nearby towns directly
impacting the station and then being filtered out.
Constraints obtained for the sectorial run (described in
Sects. and ). The influence matrix
(a, influence given in % for the whole domain over the whole year)
is de-aggregated according to prior fluxes (b) to obtain the
constraints (c): the annual sum of constraints on CH4 emissions
in the whole domain by the atmospheric data is shown on a logarithmic scale
(adimensional). Only fluxes independent from initial and boundary conditions
are used (see Sect. ). Only the sectors which are actually
seen are displayed. The relative contributions of the stations in the
inversion, averaged over the year, are shown on an arbitrary scale (map),
white being for a small contribution.
In the sectorial run, the four major contributors to methane emissions,
SNAP10, SNAP9, SNAP2 and SNAP5 are constrained (Table ,
Fig. ). The other sectors are never inverted as
the observations do not provide any constraints on them (null constraints in Fig. c).
National emissionsReconstruction of inferred emissions and error reduction
In the following sections, inferred estimates of the French emissions are
based on the posterior fluxes, where and when fluxes are constrained. Where
and when fluxes are not constrained, the values of prior fluxes are used (see
Sect. ) to reconstruct the inferred estimates of French emissions.
Regional run: French monthly total CH4 emissions (in Gg CH4)
in 2012: prior confidence range (provided by our method, see Sect. ),
fraction of prior constrained by the inversion in %, confidence range for the
inferred emissions and error reduction in % (see Sect.
for definition). No. of data used: number of data used as constraints by the
inversion for each month.
Sectorial run: French monthly total CH4 emissions (in Gg CH4)
in 2012: prior confidence range (provided by our method, see Sect. ),
fraction of prior constrained by the inversion in %, confidence range for the
inferred emissions and error reduction in % (see Sect.
for definition).
In the regional run, from February to December, between 28 and 65 % of the
national monthly prior fluxes are constrained
(Table ); fluxes in January are less constrained
(14 %), which may be linked to the smaller number of individual observations
available after selection (348 against more than 430 for the other months).
In the sectorial run, between 45 and 94 % of the national monthly total prior
fluxes are constrained, apart from February (14 %) and May (38 %), as
detailed in Table . As explained in
Sect. , these constrained fluxes belong to the four
most emitting sectors, SNAPs 2, 5, 9 and 10, representing 99 % of the total
prior emissions (Table ). The differences in the constrained
fraction of emissions between the two runs are due to the different
resolutions. A sector covering the whole of mainland France may be
constrained by any one of the available stations; conversely, if no data are
available (e.g. all are filtered out because of plume situations), the whole
sector is not constrained.
Since the inferred emissions are built from a patchwork of posterior and
prior fluxes, the differences between the prior and the inferred emissions
are larger where constraints are stronger, as displayed in
Fig. . Both runs agree on the main patterns of
correction applied to the prior emissions, with smaller fluxes around Paris
and larger fluxes in Normandy and Brittany (regions 3–5 in the
regional run) as well as in the centre (regions 15 and 8). Not surprisingly,
the regional run infers more contrasted fluxes than the sectorial run.
Indeed, the regional run can optimize regions separately and eventually
create contrasts, while the sectorial run keeps the (smoother) prior
distribution of each sector, which is scaled for the whole of France. Such a
difference is clearly visible in the centre of France
(Fig. , middle panel). The positive corrections are
due to SNAP10 (agriculture), which is also the sector with the largest
emissions. The negative corrections around Paris are due to SNAP9 (waste
treatment and disposal) and, for a smaller part, SNAP2 (non-industrial
combustion plants) and SNAP5 (distribution of fossil fuels).
On the monthly timescale, the uncertainty on inferred fluxes is smaller than on
the prior (Fig. a) for both runs. In the regional
run, the monthly error reductions (computed as explained in
Sect. , Table ) in national
budgets are larger than 25 % (up to 72 %, median at 39 %) with the exception
of January (≈ 17 %), when only 14 % of the fluxes are constrained (see
above). In the sectorial run, the error reductions are larger than 25 % for
8 months (from 37 to 90 %); for the 4 remaining months (February, April, May
and June), for which less than 50 % of the fluxes are constrained, the error
reductions are smaller than 16 % (Table ).
CH4 monthly emissions (in Gg CH4) in France in 2012 by
the regional and sectorial runs. (a) Prior fluxes (provided as
detailed in Sect. ) with the uncertainty computed by our
method (Sect. ) and confidence range of inferred fluxes
with the median shown as a solid line (Sects.
and ). (b) For the sectorial run: details of
inferred monthly emissions for the four SNAPs which are actually seen by the
inversion. c) Comparison of both runs to the inversions “S4” in the InGOS
project for which monthly emissions are available
.
Seasonal variations
In both runs, from a constant prior, the inferred fluxes vary over the year
with larger emissions during the summer (June to August for the regional run,
July and August for the sectorial run, Fig. a). The
amplitude of the monthly variations in the inferred median fluxes are
≈ 260 Gg CH4 in the regional run and ≈ 265 Gg CH4 in
the sectorial run (Fig. a). Generally, both runs are
statistically compatible, i.e. the inferred confidence ranges overlap, with
the exceptions of September and December. A similar seasonal variability was
found by the inversions in the InGOS project : among the
four systems providing monthly variations, three have a maximum in August, with
amplitudes of ≈ 130 to 170 Gg CH4 over the year
(Fig. c). The variations introduced by the inversion
may be an artefact of the variations in the number of assimilated data
(number of data used per month in Table ). Moreover,
in December, the inferred peak in emissions found in both runs may be due to
the limited spin-down period: data acquired till 31 December are
used so that emissions of the last week of the month are not well constrained
through having only a small impact at most stations. Nevertheless, the
consistency between the two runs, which use the same data but for
constraining different state vectors, and with the inversions in the InGOS
project, which do not use the same set-up and data, strongly suggests that
the inferred variations are due to actual characteristics of the fluxes. In
this case, the variations introduced by the inversion may be due to natural
sources (which are not included in our prior) and/or to seasonal variations
in anthropogenic sources, which are not taken into account in the yearly inventories.
Natural sources of CH4 in France are assumed to originate mainly from
natural wetlands or termites. Other natural emissions involve lakes and the
natural out-gassing of the Earth and are hardly quantified at the moment on
this scale, but are expected neither to be large nor to bring significant
contribution to the seasonal cycle of methane emissions. Natural wetland
emissions in France have been estimated from several vegetation models in the
framework of an international inter-comparison project 11 models;
at 200 ± 150 Gg CH4 yr-1 with a peak-to-peak
amplitude of 15–35 Gg CH4. The peak season is in September–October
(which may correspond to accelerated methanogenesis under warmer temperatures
and larger amounts of labile substrates) and the smallest emissions occur in
February–March. This contribution of wetlands therefore cannot explain by
itself the inferred seasonal variations in our total emissions. Emissions by
termites are not expected to vary much over the year, though information is
missing to document their variations.
Therefore, these results strongly suggest that anthropogenic sources largely
contribute to the seasonal variability. The sectorial run indicates that the
month-to-month variations are mainly due to agriculture (SNAP10 in
Fig. b). Indeed, since most of French CH4
emissions are due to agriculture (75 % according to our prior,
Table SNAP10), whose intensity varies during the year
(generation of agricultural waste, sensitivity of microbial decomposition to
temperature and humidity), seasonal variations in this sector may actually be
large. Nevertheless, the actual period of maximum or minimum emissions is not
easy to assess in the inventories. For example, CH4 emissions from
cattle are linked to several parameters, including the age and activity of
the animal e.g. in France,; similarly,
in Switzerland, indicate that the transhumance of cows is
not taken into account in the inventories. Emissions from waste treatment and
disposal (SNAP9), particularly water waste treatment, also display seasonal
variability .
Overall, the inferred seasonal variations are likely to be due to
agricultural (and for a smaller part, waste) emissions superimposed with
contributions of the natural sources, which the inversion has had to
attribute to one of the available sectors since natural sources were not
included in the prior emissions and no new sector could be created by the inversion.
Yearly budget
Our study estimates total yearly CH4 emissions in France to be
3835–4051 Gg CH4 based on the regional run and 3570–4193 Gg CH4
based on the sectorial run (Table ). As mentioned
previously, these two runs are consistent on the yearly timescale.
CH4 yearly emissions (in Gg CH4) in France for this study
and the other studies or inventories listed in Table .
“Total”: the inventories including only anthropogenic emissions (in red)
are summed-up with the natural emissions by wetlands and termites. For these
totals (black bars), the error bars (in black) are obtained from the range of
wetland emissions (Table ) combined with the uncertainty in
anthropogenic emissions (in red), when available i.e. only for CITEPA
(Table ). The CITEPA provides uncertainties only for the
main emitting sectors so that the error bar on the total emissions is
underestimated. “SNAP10”: the inventories including only anthropogenic
emissions (in red) are summed-up with the natural emissions by wetlands and
termites. Only the CITEPA provides an uncertainty (in red), which is combined
with the range on wetland emissions to obtain the error bar on the whole sector
(in black). Other sectors: only the CITEPA provides an uncertainty for these
sectors. N/A = not-available or the definition of sectors or activities does
not match those of SNAPs.
Estimates of yearly total CH4 emissions (in Gg CH4) in
France: top-down for our study and the European project InGOS (result from
6 different models), bottom-up for anthropogenic inventories, 11 biogeochemical
models for natural fluxes from wetlands and 1 model for emissions by termites.
Some methods do not provide uncertainties.
Type of fluxArea ofSourceEstimateYearfocus(Gg CH4)net totalFrancethis study, regional run3835–40512012net totalFrancethis study, sectorial run3570–41932012net totalEuropeInGOSa3200–47002012anthropogenicFranceINSc24692012FranceCITEPAd2430 ± 6372012EuropeIERb31072005worldEdgar4.3.2e26512012worldECLIPSE5af25632010worldEPAg26502010agriculture onlyworldFAOh17602012naturalworldwetlandsi200 [50–350]2000–2014worldtermitesi2092012
a; b, also our
prior; c Inventaire National Spatialisé ;
d, which is the reporting to UNFCCC – the
values given for uncertainties are minimum since uncertainties are provided
only for the main sources; e; f;
g; h; i GCP-CH4.
Our results are also statistically consistent (i.e. the inferred confidence
ranges overlap) with those derived from the set of atmospheric inversion
systems participating in InGOS Table or
Fig. “Total”. The range provided by InGOS
is computed from the differences between average values from the various
systems and not, as in our study, from an analysis of the errors. If the
uncertainty in each system was taken into account, the range for InGOS would
be larger still. A comprehensive inter-comparison of inversion methods and
systems with a common data set should be considered on the national scale as
it is done on the continental scale in the framework of InGOS.
The atmospheric inversions of French emissions (our study and InGOS)
consistently suggest that CH4 emissions may be up to 2 times larger
than the estimates provided by anthropogenic inventories
(Table ). As stated in Sect. , the
natural emissions were not included in the prior emissions. These natural
emissions are estimated at 200 ± 150 Gg CH4 yr-1 for wetlands and
209 Gg CH4 yr-1 for termites, i.e. 10–15 % of anthropogenic French
emissions. In the future, when finer spatial resolution maps of wetland
emissions will be available, these natural emissions should be included to
better represent the prior knowledge of the emissions on the French national
scale. Taking into account these known estimates of natural emissions, the
median values of the inferred emissions by top-down approaches (our study and
InGOS) are still systematically larger than the total estimates provided by
bottom-up approaches (any anthropogenic inventory added to wetland and
termite emissions; Fig. and Table ).
Our inferred CH4 emissions are about 25 to 55 % larger than bottom-up
estimates (median values in Table ). For example, our
atmospheric inversions lead to CH4 emissions about 35 % larger than the
most recent anthropogenic inventory dedicated to France, INS, summed-up with
the median estimate of natural emissions; the CITEPA median estimate
(reported to UNFCCC), added to the median natural source estimates, is about
35 % smaller than our estimates.
The partitioning between emission sectors is available for the sectorial run
and most of the inventories (Fig. ). Since the natural
emissions have to be attributed to an already defined sector, we chose to
assume that most of them were attributed to SNAP10. This assumption is mainly
based on the fact that the spatial distribution of agriculture makes it the
most consistent with the spatial distribution of natural emissions. Indeed,
the other sectors seen by the inversion (SNAPs 2, 5 and 9) are not diffuse
enough to match the patterns of natural emissions by wetlands or termites
(Sect. S2). Also, the atmospheric inversion attributes about 84 % of the
total emissions to agriculture (Fig. “SNAP10”), while
agriculture emissions from inventories added to natural emissions from
wetlands and termites represent 68–79 % of the total bottom-up estimates
(Fig. ). Assuming the natural emissions are included
in SNAP10 in the sectorial run, the posterior estimate for these sources is
2970–3580 Gg CH4, i.e. about 66 and 18 % larger than the agriculture
emissions by INS and IER, respectively, plus natural emissions.
Emissions due to waste treatment and disposal (SNAP9) are reduced by the
inversions and estimated at only 380–460 Gg CH4 in the sectorial run
compared to 657 Gg CH4 in the INS. SNAP9 inferred emissions are lower
than any bottom-up median estimates, except ECLIPSE.
Emissions by the distribution of fossil fuels (SNAP5) are estimated at
81–155 Gg CH4, on the higher range of the bottom-up estimates
(23–155 Gg CH4). From the atmospheric inversions, the relative
uncertainty in the SNAP5 emissions (about 30 %) is expected to be large since
these emissions are very localized in areas where natural gas distribution
systems are built and operated, and, as such, might not always be seen by the
inversion, especially after our filtering of hotspots (see Fig. S13).
Finally, emissions by the residential sector (SNAP2, non-industrial
combustion plants) stay very close to the prior by IER, mainly because it is
not strongly constrained (see Sect. and Fig. c).
Top-down estimates, from our study and the InGOS project, are in agreement.
They both find larger CH4 emissions in France than the bottom-up methods
(inventories and biogeochemical models). Moreover, in our study, the
filtering out of hotspots limits the risk of over-estimating the posterior
emissions due to the assimilation of a few high concentration peaks.
Therefore, the atmospheric inversions hint at an underestimation of French
CH4 emissions in the inventories. The possible underestimation of
CH4 emissions in the bottom-up methods could be due to an
underestimation of the emission factors or activity data, or due to
underestimations resulting from extrapolation or interpolation procedures in the
anthropogenic inventories, or to an underestimation of the natural sources
(including other natural sources than wetlands and termites).
Conclusions
In this study, we have inferred CH4 emissions in mainland France in 2012
by assimilating continuous atmospheric mixing ratios measurements from the
European network ICOS into a Bayesian inversion system. Two runs were
performed in order to use the atmospheric information in different ways: one
case is based on regions of emissions to adjust the spatial distribution of
inventory-based fluxes, and the other is based on emission sectors to adjust
source activities prescribed in inventories.
The analytical method we used allows us to compute several diagnostics and to
derive insights into the strengths and limitations of our set-up in a
consistent statistical approach. The first issue is to assess which
spatio-temporal scales are actually constrained by a relatively sparse
network in a country with large regional variations in emissions. Our results
show that, with a network of four continuous stations inside France and five
in the neighbouring countries, regions of about 50 000 km2 and a time
resolution of about 1 week are close to the finest resolutions at which
information can be retrieved from the available atmospheric data into the emission space.
The network providing continuous atmospheric mixing ratio data was set up as
a European infrastructure. Therefore, the question arises of the constraints
it can bring on emissions on the national scale. As expected, given the
relatively small number of measurement sites and their heterogeneous spatial
distribution, regions where stations were sparse in 2012 were not well
constrained, i.e. particularly in the south-east of France. This limitation
could now be overcome as two stations have been set up in the Observatoire de
Haute-Provence and at the Cap Corse in 2013 and 2014, respectively. Further work is needed
to quantitatively estimate their impact but they will certainly contribute to
better constrain the fluxes in the south-east and Corsica. Other stations
outside France are also now available in Spain, Italy, Switzerland and Germany.
From the quantitative diagnostics derived from the analytical method, we
decided to exploit the results of our inversions on the monthly and yearly
timescales for the regional and sectorial inversions. These results are ranges of
emissions, equivalent to a 1σ interval in a Gaussian framework.
The monthly totals reveal seasonal variations in French methane emissions in 2012.
Both of our inversions are statistically consistent (i.e. the inferred
confidence ranges overlap) with each other for most of the year (10 months out
of 12). The uncertainties are large (±166 to 173 Gg CH4) in May and
June for activity sectors, because of agriculture and, possibly, natural
emissions. We assume that natural emissions have mostly been attributed by
the inversion to the agriculture sector because its spatial distribution is
the closest to the diffuse pattern of natural fluxes. The seasonal variations
we find are consistent with other inversions from the InGOS project, with a
maximum in summer (July–August) and a peak magnitude of about
260 Gg CH4. We assumed that the consistency with various inversion
set-ups makes it likely that this seasonal signal is not an artefact of the
varying number of assimilated data. These seasonal variations may indeed be
due to actual variations in the agricultural (and for a smaller part, waste)
emissions superimposed with variations in the natural sources, but cannot be
explained by natural sources alone, considering the biogeochemical model
estimates for wetland emissions used in this study.
Our estimated CH4 emissions for France in 2012 range from 3835 to
4050 Gg CH4 and from 3570 to 4190 Gg CH4 for the regional run and
the sectorial run, respectively. Our two runs are statistically consistent
with each other and also with the InGOS results of a set of top-down studies
based on different chemistry-transport models and inverse systems. To compare
our estimates with bottom-up estimates, we added the emissions reported by
inventories dedicated to anthropogenic emissions with natural emissions from
wetlands and termites computed from biogeochemical models. Our atmospheric
inversions inferred total CH4 emissions about 25 to 55 % higher than
bottom-up estimates. In the sectorial run, for instance, inferred agriculture
emissions are increased by 18 % compared to the prior, leading to agriculture
emissions up to 66 % larger than the lowest bottom-up estimates (by the CITEPA).
In our study, the filtering out of high concentration peaks (in plume
situations) limits the risk of over-estimating the posterior emissions.
Therefore, the possible underestimation of CH4 emissions in the
bottom-up approaches need to be further investigated. First, it would be
useful to assess the potential origin of such an underestimation in the
anthropogenic inventories (in terms of emission factors, activity data or
extrapolation and/or interpolation procedures); second, it would be needed to better
assess natural sources of CH4 on the national scale.
The main differences between the prior bottom-up emissions and the inferred
emissions are (i) smaller fluxes around Paris, mainly due to waste
treatment and disposal and to a lesser extent to non-industrial combustion
plants; and (ii) larger fluxes in Normandy and Brittany as well as in
the centre of France, because of agriculture and, possibly, natural fluxes (wetlands and termites).
The uncertainties in our total annual budgets are ±108 and
±312 Gg CH4, for the regional and sectorial runs, respectively, which is smaller than the range of
variation of the available inventories (from 2689 to 3666,
i.e. ±488 Gg CH4, anthropogenic and natural values added). The
uncertainties in the fluxes by activity sectors could probably be decreased
with information from isotopic data or other source-specific tracers (such as
ethane for the gas and oil sector).
Further steps of this work include runs with additional observations, method
improvement and extension to other species. The building up of the ICOS
network should allow us to better constrain the different regions and refine
the results in the upcoming years. The main methodological improvement would
be to assimilate more data each day so as to make better use of the available
continuous mixing ratio measurements. In this study, night-time data and data
acquired when the boundary layer height is small are filtered out, whereas
they contain the strongest signals due to regional emissions. Cautious
integration of such data should increase our confidence in inferred local
emissions. Finally, the PYMAI-CHIMERE inversion system will have to be
adapted for the inversions of CO2 and N2O fluxes on the national scale.
Datasets relative to the results discussed here are available
online at 10.5281/zenodo.1195930. Other underlying data are available upon request.
The Supplement related to this article is available online at https://doi.org/10.5194/acp-18-3779-2018-supplement.
The authors declare that they have no conflict of interest.
Acknowledgements
This work has been supported by the GMES-MDD programme (Global Monitoring for
Environment and Security-Ministère du Développement Durable) by the
French ministry of sustainable development. The study extensively relies on
the meteorological data provided by the ECMWF. Calculations were performed
using the resources of LSCE, maintained by François Marabelle and the LSCE IT team.
We also wish to thank Simona Castaldi and Monia Santini for providing methane
emissions from termites to the Global Methane Budget project. We are grateful
to the modellers who provided estimates of methane emissions from wetlands
under the umbrella of the Global Methane Budget project: Charles Koven, Xiyan Xu
and William Riley for CLM4.5, Joe Melton and Vivek Arora for CTEM,
Hanquin Tian for DLEM, Thomas Kleinen for LPJ-MPI, Ben Poulter and Zhen Zhang
for LPJ-wsl, Renato Spahni and Fortunat Joos for LPX-Bern, Sushi Peng
for ORCHIDEE, David Beerling, Peter O. Hopcroft, Lila Taylor and David J. Wilson
for SDGVM, Zhu Qiuan for TRIPLEX and Akihiko Ito and Makoto Saito for
VISIT. We acknowledge Peter Bergamaschi for sharing InGOS results, and the
inverse modellers who participated in the InGOS project for estimating
European methane emissions: Peter Bergamaschi for TM5, Ute Karsten for
TM3-STILT, Aki Tsuruta for TM5-CTE and Alistair J. Manning for NAME. The
funding of Irish data is from the Irish Environmental Protection Agency; TAC
and RGL are funded by the UK Department of Business, Energy and Industrial
Strategy (formerly the Department of Energy and Climate Change).
Edited by: Ilse Aben
Reviewed by: two anonymous referees
References
Berchet, A.: Quantification des sources de méthane en Sibérie par
inversion atmosphérique à la méso-échelle, PhD thesis,
Université de Versailles Saint Quentin en Yvelines, Versailles, 2014.Berchet, A., Pison, I., Chevallier, F., Bousquet, P., Conil, S., Geever, M.,
Laurila, T., Lavrič, J., Lopez, M., Moncrieff, J., Necki, J., Ramonet, M.,
Schmidt, M., Steinbacher, M., and Tarniewicz, J.: Towards better error statistics
for atmospheric inversions of methane surface fluxes, Atmos. Chem. Phys., 13,
7115–7132, 10.5194/acp-13-7115-2013, 2013.Berchet, A., Pison, I., Chevallier, F., Paris, J.-D., Bousquet, P., Bonne, J.-L.,
Arshinov, M. Y., Belan, B. D., Cressot, C., Davydov, D. K., Dlugokencky, E. J.,
Fofonov, A. V., Galanin, A., Lavrič, J., Machida, T., Parker, R., Sasakawa,
M., Spahni, R., Stocker, B. D., and Winderlich, J.: Natural and anthropogenic
methane fluxes in Eurasia: a mesoscale quantification by generalized atmospheric
inversion, Biogeosciences, 12, 5393–5414, 10.5194/bg-12-5393-2015, 2015.Berchet, A., Pison, I., Chevallier, F., Bousquet, P., Bonne, J.-L., and Paris,
J.-D.: Objectified quantification of uncertainties in Bayesian atmospheric
inversions, Geosci. Model Dev., 8, 1525–1546, 10.5194/gmd-8-1525-2015, 2015a.Berchet, A., Pison, I., Chevallier, F., Paris, J. D., Bousquet, P., Bonne, J.
L., Arshinov, M. Y., Belan, B. D., Cressot, C., Davydov, D. K., Dlugokencky, E.
J., Fofonov, A. V., Galanin, A., Lavric, J., Machida, T., Parker, R., Sasakawa,
M., Spahni, R., Stocker, B. D., and Winderlich, J.: Natural and anthropogenic
methane fluxes in Eurasia: a meso-scale quantification by generalized atmospheric
inversion, Biogeosciences, 12, 5393–5414, 10.5194/bg-12-5393-2015, 2015b.Bergamaschi, P., Corazza, M., Karstens, U., Athanassiadou, M., Thompson, R. L.,
Pison, I., Manning, A. J., Bousquet, P., Segers, A., Vermeulen, A. T.,
Janssens-Maenhout, G., Schmidt, M., Ramonet, M., Meinhardt, F., Aalto, T.,
Haszpra, L., Moncrieff, J., Popa, M. E., Lowry, D., Steinbacher, M., Jordan, A.,
O'Doherty, S., Piacentino, S., and Dlugokencky, E.: Top-down estimates of
European CH4 and N2O emissions based on four different inverse models,
Atmos. Chem. Phys., 15, 715–736, 10.5194/acp-15-715-2015, 2015.Bergamaschi, P., Karstens, U., Manning, A., Saunois, M., Tsuruta, A., Berchet,
A., Vermeulen, A., Arnold, T., Janssens-Maenhout, G., Hammer, S., Levin, I.,
Schmidt, M., Ramonet, M., Lopez, M., Lavric, J., Aalto, T., Chen, H., Feist,
D., Gerbig, C., Haszpra, L., Hermansen, O., Manca, G., Moncrieff, J., Meinhardt,
F., Necki, J., Galkowski, M., O'Doherty, S., Paramonova, N., Scheeren, H.,
Steinbacher, M., and Dlugokencky, E.: Inverse modelling of European CH4
emissions during 2006–2012 using different inverse models and reassessed
atmospheric observations, Atmos. Chem. Phys., 18, 901–920, 10.5194/acp-18-901-2018, 2018.
Bousquet, P., Ciais, P., Miller, J. B., Dlugokencky, E. J., Hauglustaine, D. A.,
Prigent, C., d. Werf, G. R. V., Peylin, P., Brunke, E.-G., Carouge, C.,
Langenfelds, R. L., Lathière, J., Papa, F., Ramonet, M., Schmidt, M., Steele,
L. P., Tyler, S. C., and White, J.: Contribution of anthropogenic and natural
sources to atmospheric methane variability, Nature, 443, 439–443, 2006.Cardinali, C., Pezzulli, S., and Andersson, E.: Influence-matrix diagnostic of
a data assimilation system, Q. J. Roy. Meteorol. Soc., 130, 2767–2786, 10.1256/qj.03.205, 2004.
Champeaux, J., Masson, V., and Chauvin, F.: ECOCLIMAP: a global database of
land surface parameters at 1 km resolution, Meteorol. Appl., 12, 29–32, 2005.
CITEPA: Rapport National d'Inventaire pour la France au titre de la Convention
cadre des Nations Unies sur les Changements Climatiques et du Protocole de
Kyoto, Tech. rep., Centre Interprofessionnel Technique d'Études de la
Pollution Atmosphérique, Paris, France, 2016.
Defourny, P., Bontemps, S., van Bogaert, E., Arino, O., and Kalogirou, V.:
GLOBCOVER 2009 products description and validation report, Tech. Rep., European
Space Agency, Paris, France, 2011.EDGAR 4: Emission Database for Global Atmospheric Research (EDGAR), release
version 4.0, European Commission, Joint Research Centre (JRC)/Netherlands
Environmental Assessment Agency (PBL), http://edgar.jrc.ec.europa.eu
(last access: 4 April 2017), 2009.FAOSTAT: Food and Agriculture Organization of the United Nations:
http://www.fao.org/faostat/en/#data/GT, last access: 10 January 2017.Ganesan, A. L., Manning, A. J., Grant, A., Young, D., Oram, D. E., Sturges, W.
T., Moncrieff, J. B., and O'Doherty, S.: Quantifying methane and nitrous oxide
emissions from the UK and Ireland using a national-scale monitoring network,
Atmos. Chem. Phys., 15, 6393–6406, 10.5194/acp-15-6393-2015, 2015.Henne, S., Brunner, D., Oney, B., Leuenberger, M., Eugster, W., Bamberger, I.,
Meinhardt, F., Steinbacher, M., and Emmenegger, L.: Validation of the Swiss
methane emission inventory by atmospheric observations and inverse modelling,
Atmos. Chem. Phys., 16, 3683–3710, 10.5194/acp-16-3683-2016, 2016.Holmes, C. D., Prather, M. J., Søvde, O. A., and Myhre, G.: Future methane,
hydroxyl, and their uncertainties: key climate and emission parameters for
future predictions, Atmos. Chem. Phys., 13, 285–302, 10.5194/acp-13-285-2013, 2013.
IPCC: 2006 IPCC Guidelines for National Greenhouse Gas Inventories, Tech. rep.,
Intergovernmental Panel on Climate Change, Hayama, Japan, 2006.
IPCC: Fifth Assessment Report, Tech. rep., Intergovernmental Panel on Climate
Change, Cambridge, UK and New York, NY, USA, 2014.Janssens-Maenhout, G., Crippa, M., Guizzardi, D., Muntean, M., Schaaf, E.,
Dentener, F., Bergamaschi, P., Pagliari, V., Olivier, J. G. J., Peters, J. A.
H. W., van Aardenne, J. A., Monni, S., Doering, U., and Petrescu, A. M. R.:
EDGAR v4.3.2 Global Atlas of the three major Greenhouse Gas Emissions for the
period 1970–2012, Earth Syst. Sci. Data Discuss., 10.5194/essd-2017-79, in review, 2017.Jeong, S., Newman, S., Zhang, J., Andrews, A. E., Bianco, L., Bagley, J., Cui,
X., Graven, H., Kim, J., Salameh, P., LaFranchi, B. W., Priest, C., Campos-Pineda,
M., Novakovskaia, E., Sloop, C. D., Michelsen, H. A., Bambha, R. P., Weiss, R.
F., Keeling, R., and Fischer, M. L.: Estimating methane emissions in California's
urban and rural regions using multitower observations, J. Geophys. Res.-Atmos.,
121, 13031–13049, 10.1002/2016JD025404, 2016.Lopez, M., Schmidt, M., Ramonet, M., Bonne, J.-L., Colomb, A., Kazan, V., Laj,
P., and Pichon, J.-M.: Three years of semicontinuous greenhouse gas measurements
at the Puy de Dôme station (central France), Atmos. Meas. Tech., 8, 3941–3958,
10.5194/amt-8-3941-2015, 2015.Menut, L., Bessagnet, B., Khvorostyanov, D., Beekmann, M., Blond, N., Colette,
A., Coll, I., Curci, G., Foret, G., Hodzic, A., Mailler, S., Meleux, F., Monge,
J.-L., Pison, I., Siour, G., Turquety, S., Valari, M., Vautard, R., and Vivanco,
M. G.: CHIMERE 2013: a model for regional atmospheric composition modelling,
Geosci. Model Dev., 6, 981–1028, 10.5194/gmd-6-981-2013, 2013.Ministère de l'Environnement, de l'Énergie et de la Mer: Inventaire
National Spatialisé, http://emissions-air.developpement-durable.gouv.fr/index.html,
last access: 1 June 2017.PA: Global Anthropogenic Non-CO2 Greenhouse Gas Emissions 1990–2030,
EPA 430-R-12-006, US Environmental Protection Agency, Washington, D.C., 2012.Poulter, B., Bousquet, P., Canadell, J., Ciais, P., Peregon, A., Saunois, M.,
Arora, V., Beerling, D., Brovkin, V., Jones, C., Joos, F., Gedney, N., Ito, A.,
Kleinen, T., Koven, C., McDonald, K., Melton, J., Peng, C., Peng, S., Prigent,
C., Schroder, R., Riley, W., Saito, M., Spahni, R., Tian, H., Taylor, L., Viovy,
N., Wilton, D., Wiltshire, A., Xy, X., Zhang, B., Zhang, Z., and Zhu, Q.: Global
wetland contribution to 2000–2012 atmospheric methane growth rate dynamics,
Environ. Res. Lett., 12, 094013, 10.1088/1748-9326/aa8391, 2016.
Pregger, T., Scholz, Y., and Friedrich, R.: Documentation of the anthropogenic
GHG Emission Data for Europe Provided in the Frame of CarboEurope IP, Tech. rep.,
IER – Institute of Energy Economics and the Rational Use of Energy, University
of Stuttgart, Stuttgart, 2007.Saunois, M., Bousquet, P., Poulter, B., Peregon, A., Ciais, P., Canadell, J. G.,
Dlugokencky, E. J., Etiope, G., Bastviken, D., Houweling, S., Janssens-Maenhout,
G., Tubiello, F. N., Castaldi, S., Jackson, R. B., Alexe, M., Arora, V. K.,
Beerling, D. J., Bergamaschi, P., Blake, D. R., Brailsford, G., Brovkin, V.,
Bruhwiler, L., Crevoisier, C., Crill, P., Covey, K., Curry, C., Frankenberg, C.,
Gedney, N., Höglund-Isaksson, L., Ishizawa, M., Ito, A., Joos, F., Kim,
H.-S., Kleinen, T., Krummel, P., Lamarque, J.-F., Langenfelds, R., Locatelli,
R., Machida, T., Maksyutov, S., McDonald, K. C., Marshall, J., Melton, J. R.,
Morino, I., Naik, V., O'Doherty, S., Parmentier, F.-J. W., Patra, P. K., Peng,
C., Peng, S., Peters, G. P., Pison, I., Prigent, C., Prinn, R., Ramonet, M.,
Riley, W. J., Saito, M., Santini, M., Schroeder, R., Simpson, I. J., Spahni, R.,
Steele, P., Takizawa, A., Thornton, B. F., Tian, H., Tohjima, Y., Viovy, N.,
Voulgarakis, A., van Weele, M., van der Werf, G. R., Weiss, R., Wiedinmyer, C.,
Wilton, D. J., Wiltshire, A., Worthy, D., Wunch, D., Xu, X., Yoshida, Y., Zhang,
B., Zhang, Z., and Zhu, Q.: The global methane budget 2000–2012, Earth Sys.
Sci. Data, 8, 697–751, 10.5194/essd-8-697-2016, 2016.Saunois, M., Bousquet, P., Poulter, B., Peregon, A., Ciais, P., Canadell, J. G.,
Dlugokencky, E. J., Etiope, G., Bastviken, D., Houweling, S., Janssens-Maenhout,
G., Tubiello, F. N., Castaldi, S., Jackson, R. B., Alexe, M., Arora, V. K.,
Beerling, D. J., Bergamaschi, P., Blake, D. R., Brailsford, G., Bruhwiler, L.,
Crevoisier, C., Crill, P., Covey, K., Frankenberg, C., Gedney, N., Höglund-Isaksson,
L., Ishizawa, M., Ito, A., Joos, F., Kim, H.-S., Kleinen, T., Krummel, P.,
Lamarque, J.-F., Langenfelds, R., Locatelli, R., Machida, T., Maksyutov, S.,
Melton, J. R., Morino, I., Naik, V., O'Doherty, S., Parmentier, F.-J. W., Patra,
P. K., Peng, C., Peng, S., Peters, G. P., Pison, I., Prinn, R., Ramonet, M.,
Riley, W. J., Saito, M., Santini, M., Schroeder, R., Simpson, I. J., Spahni, R.,
Takizawa, A., Thornton, B. F., Tian, H., Tohjima, Y., Viovy, N., Voulgarakis,
A., Weiss, R., Wilton, D. J., Wiltshire, A., Worthy, D., Wunch, D., Xu, X.,
Yoshida, Y., Zhang, B., Zhang, Z., and Zhu, Q.: Variability and quasi-decadal
changes in the methane budget over the period 2000–2012, Atmos. Chem. Phys.,
17, 11135–11161, 10.5194/acp-17-11135-2017, 2017.Schmidt, M., Lopez, M., Yver Kwok, C., Messager, C., Ramonet, M., Wastine, B.,
Vuillemin, C., Truong, F., Gal, B., Parmentier, E., Cloué, O., and Ciais,
P.: High-precision quasi-continuous atmospheric greenhouse gas measurements at
Trainou tower (Orléans forest, France), Atmos. Meas. Tech., 7, 2283–2296,
10.5194/amt-7-2283-2014, 2014.Spokas, K., Bogner, J., and Chanton, J.: A process-based inventory model for
landfill CH4 emissions inclusive of seasonal soil microclimate and CH4
oxidation, J. Geophys. Res.-Biogeo., 116, G04017, 10.1029/2011JG001741, 2011.Stohl, A., Aamaas, B., Amann, M., Baker, L. H., Bellouin, N., Berntsen, T. K.,
Boucher, O., Cherian, R., Collins, W., Daskalakis, N., Dusinska, M., Eckhardt,
S., Fuglestvedt, J. S., Harju, M., Heyes, C., Hodnebrog, Ø., Hao, J., Im, U.,
Kanakidou, M., Klimont, Z., Kupiainen, K., Law, K. S., Lund, M. T., Maas, R.,
MacIntosh, C. R., Myhre, G., Myriokefalitakis, S., Olivié, D., Quaas, J.,
Quennehen, B., Raut, J.-C., Rumbold, S. T., Samset, B. H., Schulz, M., Seland,
Ø., Shine, K. P., Skeie, R. B., Wang, S., Yttri, K. E., and Zhu, T.:
Evaluating the climate and air quality impacts of short-lived pollutants,
Atmos. Chem. Phys., 15, 10529–10566, 10.5194/acp-15-10529-2015, 2015.
Thompson, R. L., Stohl, A., Zhou, L. X., Dlugokencky, E., Fukuyama, Y., Tohjima,
Y., Kim, S.-Y., Lee, H., Nisbet, E. G., Fisher, R. E., Lowry, D., Weiss, R. F.,
Prinn, R. G., O'Doherty, S., Young, D., and White, J. W. C.: Methane emissions
in East Asia for 2000–2011 estimated using an atmospheric Bayesian inversion,
J. Geophys. Res.-Atmos., 120, 4352–4369, 2015.Thompson, R. L., Sasakawa, M., Machida, T., Aalto, T., Worthy, D., Lavric, J.
V., Lund Myhre, C., and Stohl, A.: Methane fluxes in the high northern latitudes
for 2005–2013 estimated using a Bayesian atmospheric inversion, Atmos. Chem.
Phys., 17, 3553–3572, 10.5194/acp-17-3553-2017, 2017.
Vautard, R., Schaap, M., Bergström, R., Bessagnet, B., Brandt, J., Builtjes,
P., Christensen, J., Cuvelier, C., Foltescu, V., Graff, A., Kerschbaumer, A.,
Krol, M., Rouïl, L., Stern, R., Tarrason, L., Thunis, P., Vignati, E., and
Wind, P.: Skill and uncertainty of a regional air quality model ensemble, Atmos.
Environ., 43, 4822–4832, 2009.
Vermorel, M.: Emissions annuelles de méthane d'origine digestive par les
ovins, les caprins et les équins en France, INRA Prod. Anim., 10, 153–161, 1997.
Vermorel, M., Jouany, J. P., Eugène, M., Sauvant, D., Noblet, J., and
Dourmad, J.-Y.: Evaluation quantitative des émissions de méthane
entérique par les animaux d'élevage en 2007 en France, INRA Prod. Anim.,
21, 403–418, 2008.Yver-Kwok, C., Laurent, O., Guemri, A., Philippon, C., Wastine, B., Rella, C.,
Vuillemin, C., Truong, F., Delmotte, M., Kazan, V., Lebègue, B., Kaiser,
C., Xueref-Rémy, I., and Ramonet, M.: Comprehensive laboratory and field
testing of cavity ring-down spectroscopy analyzers measuring H2O, CO2,
CH4 and CO, Atmos. Meas. Tech., 8, 3867–3892, 10.5194/amt-8-3867-2015, 2015.