Introduction
The global increase of the important greenhouse gas methane in the
atmosphere since the beginning of the industrial period is very well
established (Dlugokencky et al., 2009, 1996, 1998; Etheridge et al., 1998; Khalil et al., 2007;
Loulergue et al., 2008; MacFarling Meure et al., 2006; Rasmussen and Khalil,
1981; Spahni et al., 2005). The existing CH4 mole fraction measurement
data enable accurate assessment of the source–sink imbalance through time
and, together with the estimated total sink strength, they allow for a
top-down constraint on the global source of methane to the atmosphere
(Bergamaschi et al., 2013; Houweling et al., 2014). Bottom-up estimates
of the global methane budget carry much larger uncertainties, which are
inherent to the assumptions made in the extrapolation of local-scale
measurements to larger scales (Bruhwiler et al., 2014; Kirschke et al.,
2013; Nisbet et al., 2014). The advantage of bottom-up estimates is,
however, the possibility to distinguish different sources and to link
observations to process-level understanding of the emissions.
An independent approach for distinguishing between source categories of
CH4 is the analysis of its isotopic composition, which is strongly
linked to the source–sink processes. This is particularly true for methane
from biogenic, thermogenic and pyrogenic sources (Gros et al., 2004;
Houweling et al., 2008; Quay et al., 1999; Sapart et al., 2012). A more
detailed differentiation within one source category, e.g., biogenic CH4,
for emissions from wetlands, ruminants, rice paddies or termites, however,
is complicated because of the overlap of the respective isotopic source
signatures. Further complications arise because individual source signatures
can show pronounced dependence on environmental parameters and metabolized
substrates (Kawagucci et al., 2014; Klevenhusen et al., 2010). In
addition to the source contributions, the sink processes (mainly chemical
removal by the hydroxyl radical (OH), but also soil deposition and
stratospheric loss) also affect the isotopic composition of atmospheric
methane (Brenninkmeijer et al., 1995; Röckmann et al., 2011;
Saueressig et al., 1996, 2001; Snover and Quay, 2000).
Nevertheless, over the past decades, numerous studies have shown the
potential of isotope measurements to identify individual source categories
from isotope observations (Beck et al., 2012; Lassey et al., 1993;
Tarasova et al., 2006; Umezawa et al., 2012b; Zazzeri et al., 2015) and to
constrain budgets (Ferretti et al., 2005; Fischer et al., 2008; Houweling
et al., 2008; Lassey et al., 2000; Lowe et al., 1994; Sapart et al., 2012;
Umezawa et al., 2012a).
The isotopic composition is commonly reported in δ notation, where
δ quantifies the relative deviation of an isotope ratio
(13R=13C / 12C for carbon isotopes and
2R=2H / 1H, abbreviated as D/H, for hydrogen
isotopes) in a sample from a standard ratio. The international standard for
reporting δ(13C, CH4) values is Vienna Pee Dee Belemnite
(VPDB, 13RVPDB=0.0112372; Craig, 1957) and
for δ(D, CH4) it is Vienna Standard Mean Ocean Water (VSMOW,
2RVSMOW=0.0020052; Baertschi, 1976).
δ(13C, CH4) and δ(D, CH4) are abbreviated as
δ13C and δD in the following and given in per mill
(‰). CH4 mole fractions χ(CH4) are
reported in nmol mol-1 = 10-9 and µmol mol-1 = 10-6. For
interpretation of global- or continental-scale atmospheric data the expert
group of the WMO/IAEA has set a scientifically desirable level of
compatibility of 2 nmol mol-1, 0.02 and 1 ‰ for CH4 fraction, δ13C and
δD, respectively (WMO, 2014). For regionally focused
studies with large local fluxes, extended compatibility goals of 5 nmol mol-1,
0.2 and 5 ‰ for χ(CH4),
δ13C and δD were defined.
Due to the complexity of the involved measurement techniques, CH4
isotope measurements have been limited mostly to relatively low-frequency
sampling in the field followed by isotope analysis in the laboratory
(Bock et al., 2010; Brass and Röckmann, 2010; Sapart et al., 2011;
Sperlich et al., 2013; Umezawa et al., 2009; Yamada et al., 2003). For many
decades, the dominant method for high-precision isotope analysis of
atmospheric methane was isotope ratio mass spectrometry. In particular, the
development of continuous-flow isotope ratio mass spectrometry (IRMS) in the past 2 decades
(Merritt et al., 1994, 1995) has
greatly increased the throughput of IRMS methods, making this the technique
of choice in most laboratories, also because of the small sample amounts
required.
Recently, mid-infrared laser absorption spectroscopy has proven its
potential for high-precision isotope ratio analysis. First attempts of
measuring the isotopic composition of methane (Bergamaschi et al., 1998a,
b, 1994) were restricted to enhanced CH4 mole fractions (> 50 µmol mol-1 for δ13C and > 2000 µmol mol-1
for δD) and required cryogenic cooling for both the laser source and
the detector, which impeded in situ and long-term applications. The
invention of room temperature quantum cascade laser (QCL) sources has
triggered the development of a novel generation of spectrometers suitable
for in situ analysis of the isotopic composition of greenhouse gases
(Eyer et al., 2016; Tuzson et al., 2008; Wächter et al., 2008). Their
capability for high temporal resolution led to new applications aiming for
source attribution (Mohn et al., 2012; Tuzson et al., 2011; Wolf et al.,
2015). The advantages of in situ measurements are particularly apparent in
combination with atmospheric modeling techniques, which enables the
identification of specific source regions (Rigby et al., 2012; Sturm et
al., 2013). Similarly, high-frequency, high-precision CH4 isotope data
are expected to greatly reduce uncertainties of national and global source
estimations, as demonstrated in an observing system simulation experiment
(Rigby et al., 2012).
In this paper we present the analytical setup and results of a 5-month
campaign at the Cabauw tall tower site in the Netherlands, where the
isotopic composition (δ13C and δD) of CH4 was
measured with two instruments, one IRMS system developed at Utrecht
University and one quantum cascade laser
absorption spectroscopy (QCLAS) instrument developed at Empa. The compatibility of
the two analytical techniques for CH4 mole fraction and isotopic
composition (δ13C and δD) is assessed and the obtained
high-resolution isotope dataset is exploited using a novel moving Keeling
plot (MKP) method. A comparison of measurement results with calculations from two
different models (TM5 and FLEXPART-COSMO) and two emission inventories
(EDGAR, TNO-MACC) indicates the potential of this approach to better
constrain on isotope source signatures and emissions in atmospheric models.
Methods
Site description
The 213 m tall tower is the central construction of the Cabauw Experimental
Site for Atmospheric Research (CESAR; http://www.cesar-observatory.nl/;
51∘58′ N, 4∘55′ E; 2 m a.s.l.). CESAR is
dedicated to atmospheric research and hosts a wide variety of instruments
for in situ and remote sensing measurements of meteorological parameters,
trace gases, pollutants, aerosols and clouds. The site is located in an
agricultural landscape, with CH4 emissions originating from ruminants
and other agricultural activities as well as from the peaty soil and the
drainage ditches between the surrounding fields (Peltola et al., 2014).
The small town Lopik (∼ 7500 inhabitants) is located 1 km east
of the tower. Population and road density increase steeply further away from
the tower towards the country's major cities: Utrecht (at about 20 km
distance), Rotterdam (30 km), the Hague (40 km) and Amsterdam (45 km). An
estimated 7 million people inhabit these cities and their many
neighboring settlements. The location and surroundings are described in more
detail in Peltola et al. (2014, 2015) and Vermeulen et al. (2011). The instruments were operated in a room on the ground floor of the
CESAR building. Since this room is not commonly used as laboratory, it has
air conditioning with limited cooling capacity and the temperature varied
between 25 and 30 ∘C.
IRMS system
Air was continuously drawn through 1/2 in. o.d. (outer diameter)
Dekabon tubing from 20 m height at a total flow of 16 L min-1 (STP)
provided by a Varian scroll pump (Agilent Technologies Inc., USA). The
sample gas flow was adjusted by means of a flow restriction at the inlet of
the pump in order to maintain the pressure in the sampling line above 950 hPa. The sample gas flows for the methane isotope analyzers were branched
off upstream of the scroll pump and the restriction, using 1/4 in. o.d. Dekabon lines.
Air sampling at the Cabauw tall tower
The new IRMS method for δ13C and δD analysis of
atmospheric CH4 is based on the ISAAC system as developed at the Max Planck Institute for Biogeochemistry (MPI-BGC) in Jena (Brand et al., 2016).
Importantly, the system does not require liquid nitrogen coolant for the
preconcentration and focusing steps but rather uses a massive copper block cooled
down to about -145 ∘C, to which the cold traps for
preconcentration and cryofocusing are connected via standoffs (see Sect. 2.3.1).
This cold assembly is contained in an evacuated steel dewar to prevent
condensation of moisture. During the campaign, the extraction unit and two
IRMS instruments (Thermo Delta Plus XL for hydrogen isotopes and Thermo
Delta Plus XP for carbon isotopes, both Thermo Fisher Scientific Inc.,
Germany) were operated at CESAR. The system is schematically shown
in Fig. 1.
Schematics of the preconcentration and extraction system developed
for the IRMS technique. MFC denotes mass flow controller. The eight-port valve
through which the reference air cylinder (Ref) was connected to the first selection valve
is not shown to reduce complexity. For further description see the main
text.
Cryogenic trapping
A Polycold compact cooler compressor (Brooks Automation Inc., USA), filled
with coolant PT-30, cooled a cold end on which a copper cylinder (70 mm
diameter, 85 mm height, 3 kg) was mounted. In this configuration, the copper
block reached a temperature of -145 ∘C. The preconcentration
trap (PreCon) was a 10 cm length, 1/8 in. o.d. SS tube filled with 4 cm 60/80 mesh
HayeSep D in the center and 3 cm 60/80 glass beads on each end. It was
connected with Valco fittings and the packing material was retained in the
trap using removable frits (CEF1F, Valco Instruments Company Inc., USA). The
focus trap (Focus) was a 10 cm length, 1/16 in. o.d. SS tube filled with 2 cm HayeSep D
and 4 cm glass beads at both ends, connected with Valco fittings
(ECEF211.0F, Valco Instruments Company Inc., USA). The traps could be heated
with 0.5 m Thermsys heating wire wrapped around the tubes. The PreCon and
Focus trapping units were glued together with a PT-100 temperature sensor in
heat-conducting two-component epoxy on a brass standoff. These brass
standoffs were mounted to the copper cylinder. In the “trapping”
configuration the temperatures of the traps were usually kept at -135 ∘C.
Measurement procedure
A three-port two-position Valco valve (3PV, Fig. 1) selected either ambient air
drawn from the tower through a Mg(ClO4)2 dryer or cylinder air
that was injected via one port of an eight-port multi-position Valco valve (MPV).
To check the system performance, a reference air cylinder (Ref) was measured
alternately with ambient air, and three other target gas cylinders were
measured occasionally. The inlet line was connected to a four-port two-position
Valco valve (4PV1), which directed either Helium (He, BIP quality, Air
Products and Chemicals Inc., USA) or the selected airflow to the PreCon
unit, which was connected in the loop position of a six-port two-position Valco
valve (6PV). All He and air flows were controlled by MKS mass flow
controllers (MFCs; MKS Instruments Inc., USA).
The preconcentration and cryofocusing was done similarly to Brass and
Röckmann (2010). After flushing the inlet line with
> 20 mL air, the 6PV was switched to the load position and air
was admitted to the PreCon unit. The duration of the air sampling for the
IRMS system was 10 min at a flow rate of 5 mL min-1 for δ13C and 7 mL min-1 for δD (273 K, 1 bar). The flow was
provided by a Xavitech mini pump (P200-GAS-12V, Xavitech AB, Sweden). During
this step, the temperature measured at the PreCon stayed below -132 ∘C. At this temperature CH4 and several other trace species
were retained on the HayeSep D, while the air matrix was efficiently flushed
out.
After preconcentration, the PreCon unit was heated to -30 ∘C and
a He flow of 4 mL min-1 (273 K, 1 bar) transported the CH4 in 90 s
to the Focus unit, which was held at a temperature <-137 ∘C. After transfer of the sample to the Focus, the 6PV was
switched to the load position and the PreCon was heated to -10 ∘C
to release any remaining trapped gases such as CO2.
The Focus was then heated to release the CH4, which was directed via
4PV2 and 4VP3 either to the combustion oven and the Delta plus XP IRMS for
13C analysis or to the pyrolysis oven and the Delta plus XL IRMS for D
analysis.
For δD analysis, the CH4 was injected into a pyrolysis
tube furnace (1400 ∘C), where CH4 was converted to H2
and carbon. The H2 entered the IRMS, after passing a 2 m CarboPLOT
column at room temperature and a Nafion dryer, via the GasBench
interface. No krypton interference (Schmitt et al., 2013) could be
determined in this setup. The repeatability for δD was
generally better than 2 ‰ (reported as SD), based on
consecutive analyses of reference air.
For δ13C, the CH4 was injected from the cryofocus
unit into a combustion oven containing a nickel / nickel oxide wire catalyst
at 1100 ∘C, where the CH4 was converted to CO2 and
H2O. The resulting gas mixture passed a Nafion dryer and a 10 m
PoraPLOT Q column (5 ∘C) to eliminate interference from
co-trapped krypton (Schmitt et al., 2013) before entering the IRMS via
the GasBench interface. The repeatability of δ13C was
better than 0.07 ‰ (reported as SD), based on
consecutive analyses of reference air.
The typical measurement order during the Cabauw campaign was Ref δ13C – Air δ13C – Ref δD – Air δD. A
full measurement cycle took 84 min. On a regular basis, pressurized air from
a cylinder, applied as a target gas, was analyzed as a quality-control tool
in order to monitor the long-term stability of the analytical technique. The
CH4 mole fraction and isotopic composition in ambient air and target
gas were calculated using an interpolation of the reference air analyzed
before and afterwards. A custom-made LabView software program (National
Instruments Corp., USA) was used to control and log the temperature of the
traps, the valve switching and the flow set points of the MFCs.
IRMS system isotope calibration
The isotope calibration of the IRMS system was based on a reference air
cylinder that contains ambient air collected at the IMAU in 2014, with 1888 nmol mol-1
of CH4 and isotope values of δ13C = (-47.89 ± 0.05) ‰ and δD = (-88.08 ± 1.1) ‰. We used the average of the reference air measurement
before and after the sample air measurement to calculate the mole fraction
and δ values. The linear response of the analytical system
(independence of the δ value on the amount of CH4 analyzed) was
verified by injecting various volumes of reference air up to a volume
equivalent to 2700 nmol mol-1. Occasionally, the long-term stability of the
system was checked by measuring three target cylinders with different CH4
mole fractions and isotopic compositions. The δ values are reported on the recently established scale that links the isotopic composition of methane to the international reference materials VPDB and VSMOW (Sperlich et al., 2016).
QCLAS system
The analytical procedure of the laser-based measurement system involves two
steps: preconcentration of the CH4 from 7.5 L of ambient air in a trace
gas extractor (TREX) by adsorption on HayeSep D (Eyer et al., 2014; Mohn
et al., 2010) and analysis of CH4 isotopologues with a modified
commercial QCLAS (QCL-76-D, Aerodyne Inc., USA). Details on the development,
optimization and validation of the TREX-QCLAS system are given by Eyer et
al. (2016).
The present paper comprises the first application of the TREX-QCLAS
system for in situ analysis of CH4 isotopologues at a field site for an
extended period of time. In comparison to the original setup, the heating
power of the polyimide foil on the cold trap was reduced to 60 W to increase
its lifetime. Due to the lower heating power, the duration of the desorption
step had to be extended, which led to an improved separation from residual
bulk gases (e.g., N2 and O2). Lowering the O2 enhancement in
the gas matrix is also the main reason for a lower offset in δ13C of (1.58 ± 0.1) ‰, with respect to the
MPI scale, as compared to 2.3 ‰ in previously published
results (Eyer et al., 2016). The offset was related to a higher O2 mole fraction in the gas matrix after CH4 preconcentration. One
measurement cycle consisted of four consecutive measurements of ambient air
samples and one sample of pressurized air used as a target gas, followed by
a calibration phase, and took around 4.5 h. This translates into an
analysis time of 54 minutes per sample of ambient or pressurized air.
A calibration gas (CG1, (1200 ± 50) µmol mol-1 CH4 in high-purity synthetic air (79.5 % N2 and 20.5 % O2); δ13C = -(44.24 ± 0.10) ‰, δD =
-(104.7 ± 1.1) ‰) was diluted with the same
synthetic air to 688 µmol mol-1 and analyzed between every
preconcentrated sample as an anchor to correct the measurements for
instrumental drift. A second calibration gas (CG2, (1103.8 ± 3.5) µmol mol-1
CH4; δ13C =-(36.13 ± 0.10) ‰, δD =-(180.6 ± 1.1) ‰),
diluted to a similar CH4 mole fraction of 681 µmol mol-1, was used to calculate calibration factors for δ13C and δD values. Furthermore, gas cylinders of pressurized
ambient air, referred to as target gas (TG1, TG2), were frequently measured
over the entire campaign to determine and verify the repeatability of the
measurement system, which was found to be 0.28 and 1.7 ‰ for δ13C and δD (1σ),
respectively. Additional adjustments in the preconcentration procedure and
in the analytical routine for isotope analysis improved the repeatability to
0.18 and 0.85 ‰ for δ13C and δD in the last month of the campaign. One example is
the improved temperature control of the trap during adsorption, which in
turn stabilized the O2 content in the measuring gas and thereby reduced
variations in δ13C.
The isotopic composition of the calibration gases, as well as the target gases (CH4 in pressurized air),
was determined by the Stable Isotope Laboratory at MPI-BGC. The calibrated values for the target gases are as follows: TG1: (2639.5 ± 0.6) nmol mol-1 CH4,
δ13C =-(46.48 ± 0.10) ‰, δD =-(119.0 ± 1.1) ‰; TG2: (2659.8 ± 0.6) nmol mol-1 CH4, δ13C =-(45.87 ± 0.10) ‰,
δD =-(114.1 ± 1.1) ‰. CH4 mole fraction
measurements were linked to the WMO-X2004 calibration scale (Dlugokencky
et al., 2005) through calibration of the target gases against NOAA reference
standards at Empa.
Modeling
Two complementary atmospheric transport models (TM5, FLEXPART-COSMO), both
in combination with two different emissions inventories
(TNO-MACC_2, EDGAR/LPJ-WHyMe), were applied to support
interpretation of the measurements. The Eulerian tracer model TM5 simulated
the distribution of CH4 and 13CH4 at global scale with a
zoom on Europe at 1∘ × 1∘ resolution and considered both
the isotopic signatures of different sources and the fractionation by
different removal pathways of CH4 in the atmosphere. The Lagrangian
particle dispersion model FLEXPART-COSMO, conversely, was run in backward
mode at a higher resolution of 0.06∘ × 0.06∘ but only
over Europe. This model is better able to represent the spatial variability
of CH4 sources in the near field of Cabauw but it only simulated the
contributions from the last 4 days of emissions within Europe and not the
large-scale background. Chemical loss of CH4 was not considered due to
the short transport times between the sources and the receptor point at
Cabauw.
TM5 modeling
Simulations of atmospheric CH4 and δ13C were performed
using the global tracer model TM5 (Krol et al., 2005). The Eularian
offline model was driven by meteorological fields from the European Centre
for Medium-Range Weather Forecast (ECMWF) reanalysis project ERA-Interim
(Dee et al., 2011), preprocessed for use in TM5. For
vertical transport due to moist convection we made use of ERA-Interim
archived convective mass fluxes, replacing the use of the Tiedke scheme in
Krol et al. (2005). The model was run at a horizontal resolution of
6∘ × 4∘ globally and 1∘ × 1∘ inside a zoom domain covering western Europe. The model uses 25 hybrid sigma-pressure levels from the
surface to top of atmosphere.
European CH4 emissions and isotope source signatures
(δ13C, δD) for the different source categories used in
TM5. Bold numbers refer to the total of natural, anthropogenic and all emissions, respectively.
Process
Yearly emissions
Source signature
(Europe, Tg CH4 yr-1)
δ13C/‰
Natural emissions
22.1
–59.2
Natural wetlandsa
Peatland
9.3
-68
Wet mineral soils
4.6
-65
Inundated wetlands
1.3
-60
Geological emissionsb
6.5
-42
Termitesc
0.4
-63
Anthropogenic emissions
45.3
–52.4
Biomass burningd
0.3
-23.6
Agriculturee
Domestic ruminants
11
-64
Manure
3
-54
Rice paddies
0.17
-65
Energy sectore
Coal mining
3.4
-47
Oil production
3
-42
Gas production and distribution
12
-42
Oil combustion
0.41
-32
Residential sectore
1.6
-32
Waste treatmente
Landfills
9
-54
Waste waters
3
-50
Total
67.4
–54.6
a Spahni et al. (2011); b Etiope et al. (2008); c Sanderson (1996);
d GFED3/4 (http://www.globalfiredata.org/); e EDGAR4.2FT (EDGAR,
2010).
Two parallel (forward) TM5 simulations were performed with CH4 and
13CH4 as transported tracers. In the standard configuration,
anthropogenic CH4 emissions were taken from EDGAR4.2 FT2010 (EDGAR, 2010), extrapolated to 2014 and 2015 using annual
statistics from the Food and Agriculture Organization of the United Nations
(FAO) and the British Petroleum Company (BP), as described in Houweling et
al. (2014). For natural wetland emissions, an average of the emission
estimates derived by Spahni et al. (2011) for the period 2003–2008 was
taken, using the LPJ-WHyMe model. For a complete description of the CH4
emissions (Table 1), see Monteil et al. (2013) and references therein.
13CH4 emissions were derived from the CH4 emissions using
prescribed δ13C source signatures (Table 1). The emission
inventory was built according to a double constraint: first, each source
signature must be chosen within its own uncertainty interval; second,
the resulting global average source signature must be compatible with the
global source signature that is inferred from the observations (and that is
known with a much better precision than the individual source signatures)
(Monteil et al., 2011). In a second set of simulations, anthropogenic
emissions in a regional domain centered on Cabauw were replaced by emissions
from the European TNO-MACC_2 inventory, which was used as the
standard inventory in the FLEXPART-COSMO simulations (see below). Outside
the regional domain covered by TNO-MACC_2, the EDGAR
emissions were used.
Atmospheric removal of CH4 was modeled as described in Monteil et al. (2013), using kinetic fractionation factors α
=k(12C) /k(13C) of αOH=1.0055, αCl=1.066 and αO(1D)=1.013 for the
reactions between CH4 and OH (Sander et al., 2006), Cl (Saueressig et
al., 1995) and O(1D) (Saueressig et al., 2001),
respectively. The simulations were initialized at steady state (obtained via
a spin-up run) in 2005, and simulations of the period 2005–2015 were used to
calculate a realistic state of the atmosphere at the start of the
measurement campaign, including the imbalance between emissions and
atmospheric CH4 mixing ratio/isotopic composition in 2014. Time series
were extracted from model-simulated mole fraction fields after interpolation
to the horizontal coordinate and height of the Cabauw tower air inlet.
FLEXPART-COSMO modeling
The Lagrangian particle dispersion model (LPDM) FLEXPART
(Stohl et al., 2005) was used in a modified version coupled
to the mesoscale numerical weather forecast model COSMO (Baldauf
et al., 2011) to simulate the regional contribution of different source
categories to the concentrations and isotopic signatures of CH4 at
Cabauw. FLEXPART–COSMO was driven by hourly operational analysis fields
generated by the Swiss national weather service MeteoSwiss for a domain
covering entire western and central Europe from Ireland, Denmark and Poland
in the north to Portugal and southern Italy in the south with a horizontal
resolution of approximately 7 km × 7 km and 60 vertical levels. Every 3 h, 50 000 particles (air parcels) were released from the position of the
inlet 20 m above surface and traced backward in time for 4 days to compute
the sensitivity of each 3-hourly measurement to upwind sources. The
corresponding source sensitivity maps or footprints (Seibert and
Frank, 2004) were multiplied with gridded CH4 emissions to compute the
mole fraction enhancement above background expected from different sources.
Emissions were taken from the TNO-MACC_2 inventory for Europe
representative of the year 2009 and available at 0.125∘ × 0.0625∘
resolution (Kuenen et al., 2014)
or, alternatively, from the same version of EDGAR/LPJ-WHyMe inventory
driving TM5 at a resolution of 1∘ × 1∘. Methane mole
fractions were computed separately for a number of SNAP (Standardized
Nomenclature for Air Pollutants) source categories with specific isotopic
signatures as summarized in Table 2.
SNAP (Standardized Nomenclature for Air Pollutants) source
categories and corresponding δ13C and δD source
signatures from the TNO-MACC_2 inventory as used in
FLEXPART-COSMO.
SNAP category
Description
δ13C/‰
δD/‰
1
Energy industries, oil or gas production
-42
-175
2
Residential combustion
-32
-175
3+4
Industrial combustion and non-combustion processes
-60
-175
5
Extraction and distribution of fossil fuels including distribution of natural gas
-42
-175
7
Road transport
-20
-175
9
Waste including emissions from landfills
-54
-293
10
Agriculture including emissions from ruminants and manure management
-64
-319
6+8
Other emissions (negligible)
-42
-175
For the domain covered by the FLEXPART-COSMO simulations, which includes
most of western and central Europe, total anthropogenic emissions are 20.6 Tg CH4 yr-1 in EDGAR and 18.3 Tg CH4 yr-1 in TNO-MACC, which
corresponds to a difference of 12.5 %. CH4 emissions from gas/oil
production and distribution are 89 % higher, CH4 emissions from
agriculture 19 % lower and CH4 emissions from waste 12 % higher in
EDGAR than in TNO-MACC.
Source-specific emissions were combined with isotopic signatures of the
various categories from Table 2 to derive mean δ13C and δD isotopic signatures for the CH4 that was picked up by the air parcel
along the trajectory.
Interpretation of CH4 isotope data
Data analysis by a Keeling plot technique
The isotopic composition of CH4 emissions were estimated using the
Keeling plot technique (Keeling, 1961; Pataki et al., 2003).
This method allows the isotopic signature of a single source process or the
mean isotopic signature of combined source processes that mix into a
background reservoir to be determined from the observed ambient isotopic
composition and mole fraction. An implicit assumption of the Keeling plot
approach is that the isotopic composition and mole fraction of the
background reservoir and the isotopic composition of the source or the
combined source stay constant over the time range of the analysis. This may
not always apply as the relative contribution of individual CH4 sources
or their isotopic signature may change over time
To exploit the high temporal resolution of our data, we applied a novel
approach of moving Keeling plot (MKP) method. Data within a moving window
of 12 h were used to calculate the source isotopic composition. This
window was moved in 1 h time steps over the data series. In addition,
values for background conditions within a 48 h period, centered on the
respective 12 h window, were included in the analysis. These background
values were chosen between 10:00 and 18:00 local time, because during this
period a convective boundary layer usually develops and hence local
influence is weak; pollution events with CH4 mole fractions above
2100 nmol mol-1 were filtered out additionally. For each time window, an
orthogonal least squares fit was applied to the δ values vs. the
inverse CH4 mole fractions and R2 values were calculated. A
Keeling plot analysis only returns meaningful values for the source isotopic
composition if the variations in CH4 mole fraction are
significant and if the emissions are from a source with a well-defined
isotopic composition. Therefore, two additional filters were applied: (i) the
mole fraction had to vary by more than 200 nmol mol-1 within each time window
and (ii) the R2 of the fit had to be larger than 0.8. If R2 < 0.8, the 12 h interval was reduced consecutively by 1 h to a
minimum of 6 h until either the R2 of the fit was > 0.8 or the number of data points was lower than five. On average this
technique accumulated 22 data points per 12 h time window.
Results
Overview of the field measurements at the Cabauw site
CH4 mole fraction, χ(CH4) and isotopic composition
(δ13C, δD) measured at the Cabauw tall tower from 17 October 2014 until 29 March 2015. Real-time measurements by IRMS (Utrecht
University) are indicated in yellow, and TREX-QCLAS (Empa) data in blue.
The full record of the methane mole fraction and isotopic composition
obtained with the two measurement techniques at CESAR is shown in
Fig. 2. The IRMS system started with δD measurements first and
after 3 weeks delivered both δ13C and δD data. The
TREX-QCLAS system started later and ran continuously from mid-December to
mid-January and from mid-February to the end of the campaign. Despite a
number of interruptions mainly due to various kinds of instrument
malfunction, the combined time series of both techniques shows a high
temporal coverage with more than 2500 measurements performed for both
δ13C and δD.
A qualitative inspection of the time series already conveys the obvious
features that will be discussed below in more detail: the methane mole
fraction χ(CH4) shows a large number of substantial increases
above background level, and these positive methane excursions are
accompanied by negative excursions in the δ values from the
background level. Thus the additional methane is generally depleted in both
13C and D.
Comparison of the two analytical techniques
Before presenting a detailed analysis of the CH4 isotopic composition
in ambient air, we compare the results obtained with the IRMS and QCLAS
techniques in order to evaluate their performance and to combine the results
into one final dataset. Although both systems measured air from the same
intake line, the sampling intervals could not be synchronized since both
instruments operated in different measurement cycles. A full measurement
cycle (including measurement of the reference gas) took 84 min for the
IRMS system and 54 min for the TREX-QCLAS system. The actual duration of
the air sampling was 10 min for the IRMS system and 15 min for the
QCLAS system. So even if the systems coincidentally started sampling at the
same time, they never actually analyzed exactly the same air mass.
Consequently, differences between the systems contain contributions from
natural variability, random fluctuations due to limited measurement
precision, and system offsets.
Correlation diagrams for CH4 mole fraction, δ13C
and δD analyzed with IRMS (Utrecht University) and TREX-QCLAS
(Empa). The dashed black lines are 1:1 lines; dashed red lines mark the
extended WMO compatibility goals of ± 5 nmol mol-1, ± 0.2 ‰ and ± 5 ‰ for CH4 mole
fraction, δ13C and δD, respectively. The temporal
difference between IRMS and TREX-QCLAS sampling is indicated by the point
size (large: 20 min; medium: 40 min; small: 60 min). For δ13C
and δD, the differences in the CH4 mole fraction of the
measurements are represented by the shading (black: identical mole
fractions; white: 50 nmol mol-1 difference).
Figure 3 shows a comparison of the χ(CH4), as well as δ13C and δD values that were obtained with the TREX-QCLAS and
the IRMS technique. To visualize the possible effect of time shifts, the
size of the points corresponds to the proximity of the sampling intervals. A
total of 727, 333 and 277 measurement pairs for χ(CH4), δ13C and δD, respectively, analyzed by both techniques, were
combined in this way.
The mole fraction comparison shows good agreement along the 1:1 line but
with a large scatter, which has two contributions: (i) instrumental noise, as
the isotope systems have a relatively large uncertainty for measurement of
the mole fraction compared to existing high-precision CH4 analyzers,
and (ii) natural variability associated with the sampling of different air
masses as described above. The second point is supported by the fact that
the average difference in CH4 mole fractions between the two analytical
techniques was larger for larger temporal differences in the sampling
intervals.
For the isotope intercalibration plots, the grey-black shading of the
circles indicates the difference in χ(CH4) of the respective
measurement pair analyzed by both techniques. The overall difference between
the measurements conducted with the two systems (QCLAS-IRMS) is (+0.25 ± 0.04) ‰ for δ13C and (-4.3 ± 0.4) ‰
for δD (the stated errors are
standard errors of the mean). The mean offsets are slightly outside the WMO
extended compatibility goals for δ13C (0.2 ‰) and within the WMO extended compatibility goals for
δD (5 ‰), as indicated by the red dashed lines
(WMO, 2014). Individual measurement pairs can show significantly
larger deviations for aforementioned reasons. Differences between the two
techniques are higher than expected as both laboratories refer their
measurements to MPI-BGC, who recently established a link between the
CH4 isotopic composition and the international reference materials VPDB
and VSMOW, in the framework of the INGOS project (Sperlich et al., 2016).
Therefore, remaining differences can only be rationalized by uncertainties
in propagating the scale or by instrumental issues. The enhanced
discrepancies for low δD values might originate from a
nonlinear response of one of the applied analytical techniques. The mean
offset values determined above were applied to the QCLAS data to create one
combined dataset with 2610 data points for δ13C and 2673 data
points for δD.
FLEXPART-COSMO source attribution
Absolute (top) and relative (bottom) contributions of methane
emissions that are picked up along the 4-day FLEXPART-COSMO trajectories
during the campaign. The results shown are from the FLEXPART-COSMO
simulations with the TNO-MACC inventory. They indicate major contributions
of the following source categories: “agriculture” (mainly ruminants),
“waste” (mainly landfills) and “fossil” (fugitive losses from coal, oil
and natural gas production and from gas transportation and distribution) to
the increase in CH4 mole fractions at Cabauw. The category “rest”
primarily represents residential CH4 emissions.
In FLEXPART-COSMO, the contributions of the individual source types are
simulated separately and added up to obtain the cumulative CH4 mole
fraction. Figure 4 shows these contributions in absolute (top) and relative
terms (bottom). According to the model, the relative contributions at the
Cabauw site are quite uniform, with agricultural sources accounting for more
than 60 %, waste (mostly landfills) around 20–40 % and fossil sources
between 0 and 40 %. We note that significant contributions from fossil
sources are only detected episodically, during several events that usually
last a few days. Contributions from other source categories are generally
negligible at the Cabauw site.
TM5 and FLEXPART-COSMO modeling including isotopes
The TM5 model calculates the combined influence of the global methane
sources and sinks on CH4 and δ13C at the Cabauw tower, and
therefore the TM5 results can be compared directly to the measured time
series. For FLEXPART-COSMO, a representative background mole fraction and
isotopic signature need to be added for comparison with the observations.
For simplicity we assumed a constant background similar to the observed
values for background conditions: 1930 nmol mol-1 for χ(CH4) with
δ13C =-47.1 ‰ and δD
=-86 ‰.
Comparison of the modeled and measured time series of CH4 mole
fraction and isotopic composition (δ13C and δD).
Measurements are shown as circles and model results as lines. The top graph shows two
selected model configurations for the entire campaign: FLEXPART-COSMO using
the TNO-MACC inventory (blue) and TM5 using the Edgar/WHyMe inventory
(red). The bottom graph shows time series for March 2015 with all four model–inventory combinations. For δD, only the synthetic FLEXPART-COSMO
results are available for comparison since TM5 does not simulate δD.
Figure 5 shows a comparison of these model-generated time series with the
measured data for the entire campaign. Both models capture the amplitude and
the temporal variability of χ(CH4) well. Most of the methane
pollution events observed at CESAR are also present in the modeled
time series and the increase in χ(CH4) is of a comparable size.
In addition, the results of the TM5 and the FLEXPART-COSMO model for
CH4 mole fractions agree relatively well with each other
(R2=0.69), in particular when both models are run with the same
inventory at the same coarse spatial resolution, i.e., with EDGAR/LPJ-WHyMe.
A few pronounced CH4 events in Fig. 5 show larger differences between
the models. On 2 November, FLEXPART-COSMO simulates an emission signal that
is not captured by TM5. Unfortunately no measurements are available for this
event to decide on which model performs better. On 30 November TM5 simulates
a CH4 plume, which is absent in FLEXPART-COSMO, and this event is also
not supported by the measurements. The global model has the advantage that
it includes the influence of long-range transport. As expected, however, the
observed variability is predominantly influenced by local and regional
emissions.
Regarding the time series of the δ values, both TM5 and
FLEXPART-COSMO qualitatively display the expected anti-correlations between
CH4 and δ13C. However, the amplitude of the δ13C variability is generally underestimated in the model runs,
especially when using the EDGAR inventory. In addition, the modeled
background level of δ13C in TM5 is offset by up to 1 ‰, but this offset is also present at clean background
sites in the Northern Hemisphere.
Using the TNO-MACC inventory in FLEXPART-COSMO results in better agreement
with the observed variability of δ13C. In TM5, the TNO-MACC
emissions reduce the amplitude of the CH4 variability, which is
explained by the 13 % lower emissions in TNO-MACC compared with EDGAR.
Furthermore, the results of both models are consistent with the emissions
being more depleted in δ13C in TNO-MACC than in EDGAR. The
measurements indicate emissions that are even more depleted in δ13C than TNO-MACC values. These results suggest that the fractional
contribution of isotopically heavy fossil emissions is overestimated in
EDGAR, at least in the area sampled by Cabauw, although the uncertainty in
the assumed δ13C source signatures could also contribute. For
instance, recent literature showed that landfill emissions from the UK are
more depleted in 13CH4 due to the implementation of gas extraction
systems (Zazzeri et al.,
2015).
The δD time series simulated with FLEXPART-COSMO using the TNO-MACC
inventory is in good agreement with the measurements. This further indicates
that TNO-MACC has a realistic source mixture, but the uncertainties in the
mean δD signature are too large to draw firm conclusions at this
stage. Despite these uncertainties, Fig. 5 clearly demonstrates how isotopic
measurements highlight differences between emission inventories, which would
go unnoticed looking only at CH4 mole fractions. Additional information
may be available from the combination of both isotope signatures. For
several of the CH4 elevation events shown in Fig. 5b, the relative
changes in δ13C and δD modeled with FLEXPART-COSMO vary
when using the two different inventories (TNO-MACC and EDGAR). Some of the
anomalies show differences pointing in the same direction for δ13C and δD, and some others do not. This suggests that δD
provides additional independent information, which will be discussed in more
detail in Sect. 4.3 using a double-isotope plot of the source signatures
(Fig. 7). The benefit of the high-resolution dual isotope measurements for
validating emissions used in the models will be investigated in Sect. 4.4.
Discussion
Diurnal and synoptic variability
A prominent feature of the high-resolution dataset is the pronounced diurnal
variability, with large increases in CH4 mole fraction that occur often
during the night due to the shallow planetary boundary layer. In addition,
there are also several synoptic (but much smaller) pollution events, where
CH4 mole fractions stay above the unpolluted background level for
several days. These elevations are likely caused by synoptic-scale advection
of CH4 plumes from other source regions with a different source mix.
Isotope identification of the mean CH4
source
Keeling plot of all data using an orthogonal regression method. The
dashed line indicates the regression line and the shaded area the confidence
interval taking into account the measurement uncertainties. The color code
indicates all measured data (grey points) and daily background values (red
points). Left panels show the region near the y axis intercept.
In Fig. 6, the Keeling plot technique is applied to identify the mean
isotopic signatures (δ13C, δD) of the combined CH4
emissions detected at the Cabauw site. An orthogonal regression method was
applied to determine the fit parameters. This analysis yields well-defined
mean isotopic signatures of the cumulative source (the y intercept of the
regression analysis) of δ13C =-(60.8 ± 0.2) ‰ and δD =-(298 ± 1) ‰.
The inferred mean isotopic signature agrees well
with emission from ruminants, which are expected to be the main source of
CH4 in this rural area. This is plausible because the mean isotopic
signature is largely determined by the pronounced nighttime CH4
elevations, which represent the local emissions close to the tower. Also the
source contributions modeled by FLEXPART-COSMO suggest the dominant
influence of agricultural emissions in this rural area (Fig. 4).
Interestingly, the mean isotopic signature for the much smaller synoptic
CH4 variations of the background (red points labeled “Background” in Fig. 6) is not
significantly different from the one for the complete dataset.
Short-term variability
Given the high temporal resolution of the dataset presented here, the
isotope variations can be interpreted in much more detail than the overall
analysis performed above. This allows identifying varying contributions of
CH4 sources during different periods of the campaign. To do so, we
applied a 12 h MKP method to the data, as described
in Sect. 2.6.1.
MKP intercepts of δD vs. δ13C. The colored
areas indicate typical isotope signatures for different source categories.
Circles show the 6 h-averaged source signatures. Large colored symbols
indicate data from the three events (event 1: 10–12 March;
event 2: 16–18 March; event 3: 22–24 March) that are highlighted in Fig. 9. The labels a and b refer to day 1 and
day 2 of the 2-day events, respectively. For the source signatures, the
δ13C values are taken from Table 1 and the δD values
from recent literature (Snover et al., 2000; Rigby
et al., 2012).
Keeling plots for the period between 16 and 18 March, illustrating a
rapid change in δ values over the course of hours, which is most
probably related to a change from mainly ruminant-derived CH4 to a
significant contribution of fossil and/or waste CH4. The dashed lines
indicate the regression line; the shaded areas show the uncertainty (1
standard deviation) of the regression line. Left panels show the region near
the y axis intercept. Times indicated are central European time (CET).
Figure 7 summarizes the results of the MKP method in the form of a δD
vs. δ13C plot. To combine δ13C and δD
measurements performed at different times, MKP intercepts were averaged over
6 h intervals. Mean δ13C signatures range between -68 and -55 ‰ and mean δD
signatures cover a relatively wide range between -350
and -260 ‰, indicating emissions mainly from microbial
sources as derived from the cumulative Keeling plot analysis. During some
periods, however, elevated mean δ13C and δD signatures
reveal significant additional contributions from waste and/or fossil
emissions.
The colored symbols in Fig. 7 highlight the mean isotopic signatures of
three 48 h events (10–12, 16–18 and 22–24 March) that are discussed in more
detail in the following. For the event of 16–18 March, selected results of
the 12 h MKP method are displayed in Fig. 8, demonstrating the
advantage of the high-temporal-resolution data. It is possible to clearly
distinguish variations in the mean isotopic signatures during this event by
variations in the y axis intercepts. The increase by about 6 ‰ for δ13C and
about 50 ‰ for δD, in the source isotopic signature for
this event, clearly indicates the gradually increasing contribution of
CH4 from isotopically enriched sources, e.g., fossil-fuel- or
waste-related CH4.
Detailed analysis of three 2-day periods with large CH4
elevations in March 2015. The top panel exhibits CH4 mole fraction
(grey) with background values in red (10:00–18:00, > 2100 nmol mol-1). The middle panels show the mean isotopic signatures (δ13C, δD) derived with the 12 h MKP method. The color-coding in
the middle panels (red, light blue, purple) indicates characteristic
contributions from different sources; red is microbial, light blue is fossil
and
purple is waste. For consistency, the same color-coding was chosen in Fig. 7.
The bottom graph presents CH4 source contributions as computed with the
FLEXPART-COSMO model using the TNO-MACC inventory, averaged over 24 h.
The temporal evolution of the observed source mixture is investigated in
further detail in Fig. 9, where the 16–18 March period (labeled as 2) is
compared to two other 48 h periods (10–12 March: label 1; 22–24 March: label 3), each with significant diurnal CH4 elevations. For
event 1, the mean isotopic signatures stayed rather constant at values
around δ13C =-63 ‰ and δD =-320 ‰. These values are typical for microbial
emissions from an agricultural source and agree well with the source
contributions predicted for this period by the FLEXPART-COSMO model.
Period 2 is characterized by much stronger isotopic change within the 48 h
period. The δ13C signature increases to above -60 ‰ and the δD signature
increases to -240 ‰ by the end of the period (see Fig. 9). The
double-isotope plot in Fig. 7 shows that the change in δD during
event 2b clearly points towards fossil fuel sources, which provides
independent support for the FLEXPART-COSMO simulations, where the
contributions from fossil-fuel-derived emissions are higher for the second
day.
For period 3, the mean δ13C isotopic signatures increased
during the 48 h by about 2–3 ‰, whereas the δD
signatures remained constant around -300 ‰. For this
period, the double-isotope plot of Fig. 7 indeed shows a shift towards the
waste category. Also this observation is independently confirmed (at least
qualitatively) by the FLEXPART-COSMO model-derived source attribution, which
indicates the largest fraction of waste-derived CH4 for the first day
and a small addition of fossil CH4 for the second day of event 3. These
examples show that even at a location like Cabauw, where one source category
strongly dominates, contributions from isotopically different sources can be
identified if sufficiently high-resolution dual isotope ratio data are
available. We note that the “directional” information in the double-isotope plot is only available by combining δ13C and δD
measurements. It would be much harder, if not impossible, to distinguish an
addition from fossil-fuel- or landfill-derived CH4 based on δ13C or δD data alone.
Evaluation of emission databases with high-temporal-resolution
CH4 isotope data
As described in Sect. 3.4, both the TM5 and the
FLEXPART-COSMO model-generated time series of CH4 mole fractions show
an adequate agreement with the CH4 measurements at the Cabauw site.
Therefore, the comparison between measurement data and the models can be
used to evaluate the methane budget in more detail. In this context, the
measured and modeled isotopic composition can be employed to assess the
validity of the emission inventories, EDGAR and TNO-MACC, with respect to the
magnitude and spatial distribution of source categories. To compare the
measured mean isotopic signatures to the model results, the simulated
isotope time series were linearly interpolated and evaluated in the same way
as the observations using the 12 h MKP method. This analysis was performed
for both models (TM5 and FLEXPART-COSMO), each using both the
EDGAR/LPJ-WHyMe and the TNO_MACC inventories. Additionally,
time series for the mean isotopic signatures were calculated directly from
FLEXPART-COSMO data, without using of the MKP method. This direct method
allowed an independent estimation of the mean isotopic signatures and, thus,
also provided an opportunity to evaluate the MKP method.
The statistics of the mean isotopic signatures from all four model–inventory
combinations are shown as histograms in Fig. 10, together with the
measurement-derived mean isotopic signatures and the directly derived
signatures from FLEXPART-COSMO modeling. The numerical values are given in Table 3. A clear difference can be observed
between the mean isotopic signatures derived with the two different emission
inventories. Model runs with the EDGAR/LPJ-WHyMe emission inventory (red in
Fig. 10) tend to produce mean CH4 isotopic signatures that are more
enriched in 13C and D than the model runs with TNO-MACC emissions.
These differences are very similar for the simulations using TM5 and
FLEXPART-COSMO, suggesting that differences originate from the emission
inventories rather than from differences between the models themselves. The
δ13C source signatures derived from the measurements at the
Cabauw tower are significantly more depleted than any of the model-generated
datasets. For δD, the mean isotopic signatures using TNO-MACC
emissions are relatively close to the measurements at Cabauw, whereas the
values using EDGAR emissions are much more enriched in CH3D.
The high-temporal-resolution isotope data that are described in this paper
thus provide relevant information to further constrain models and/or
emission inventories because the mean isotopic signatures can change
rapidly. The comparison of our first high-resolution isotope measurements at
Cabauw to model calculations clearly identifies differences between the
modeled inventories, where the EDGAR inventory produced too-enriched mean
isotopic signatures due to a higher contribution from fossil fuel sources.
Similar differences in terms of source contributions between EDGAR and
TNO-MACC_2 were also reported by Hiller et al. (2014) for Switzerland, and Henne et al. (2015) concluded that natural gas emissions in
Switzerland are likely overestimated in EDGAR.