Understanding the recent evolution of methane emissions in the Arctic is necessary to interpret the global methane cycle. Emissions are affected by significant uncertainties and are sensitive to climate change, leading to potential feedbacks. A polar version of the CHIMERE chemistry-transport model is used to simulate the evolution of tropospheric methane in the Arctic during 2012, including all known regional anthropogenic and natural sources, in particular freshwater emissions which are often overlooked in methane modelling. CHIMERE simulations are compared to atmospheric continuous observations at six measurement sites in the Arctic region. In winter, the Arctic is dominated by anthropogenic emissions; emissions from continental seepages and oceans, including from the East Siberian Arctic Shelf, can contribute significantly in more limited areas. In summer, emissions from wetland and freshwater sources dominate across the whole region. The model is able to reproduce the seasonality and synoptic variations of methane measured at the different sites. We find that all methane sources significantly affect the measurements at all stations at least at the synoptic scale, except for biomass burning. In particular, freshwater systems play a decisive part in summer, representing on average between 11 and 26 % of the simulated Arctic methane signal at the sites. This indicates the relevance of continuous observations to gain a mechanistic understanding of Arctic methane sources. Sensitivity tests reveal that the choice of the land-surface model used to prescribe wetland emissions can be critical in correctly representing methane mixing ratios. The closest agreement with the observations is reached when using the two wetland models which have emissions peaking in August–September, while all others reach their maximum in June–July. Such phasing provides an interesting constraint on wetland models which still have large uncertainties at present. Also testing different freshwater emission inventories leads to large differences in modelled methane. Attempts to include methane sinks (OH oxidation and soil uptake) reduced the model bias relative to observed atmospheric methane. The study illustrates how multiple sources, having different spatiotemporal dynamics and magnitudes, jointly influence the overall Arctic methane budget, and highlights ways towards further improved assessments.
The climate impact of atmospheric methane (CH
Recent changes in methane concentrations are not uniform and vary with
latitude. The rise in methane in 2007 was, for example, particularly
important in the Arctic region due to anomalously high temperatures leading
to high wetland emissions (Dlugokencky et al., 2011; Bousquet et al., 2011).
The Arctic (> 60
This context points to the need to closely monitor Arctic sources. The
largest individual natural source from high latitudes is wetlands. An
ensemble of process-based land-surface models indicate that, between 2000 and
2012, wetland emissions have increased in boreal regions by 1.3 TgCH
Freshwater emissions are another important and uncertain terrestrial source
of methane. About 40 % of the world's lakes are located north of
45
Additional continental sources include anthropogenic emissions, mostly from Russian fossil fuel industries, and to a lesser extent, biomass burning, mostly originating from boreal forest fires. The Arctic is also under the influence of transported emissions from midlatitude methane sources, mostly of human origin (e.g. Paris et al., 2010; Law et al., 2014).
Marine emissions from the Arctic Ocean are smaller than terrestrial
emissions, but they too are climate sensitive and affected by large
uncertainties. Sources within the ocean include emissions from geological
seeps, from sediment biology, from underlying thawing permafrost or
hydrates, and from production in surface waters (Kort et al., 2012). The
East Siberian Arctic Shelf (ESAS, in the Laptev and East Siberian Seas),
which comprises more than a quarter of the Arctic shelf (Jakobsson et al.,
2002) and most of subsea permafrost (Shakhova et al., 2010), is a large
reservoir of carbon and most likely the biggest emission area (McGuire et
al., 2009). Investigations led by Shakhova et al. (2010, 2014) estimated
total ESAS emissions from diffusion, ebullition, and storm-induced degassing,
at 8–17 TgCH
The main sink of methane is its reaction with the hydroxyl radical (OH) in the troposphere, which explains about 90 % of its loss. Other tropospheric losses include reaction with atomic chlorine (Cl) in the marine boundary layer (Allan et al., 2007) and oxidation in soils (Zhuang et al., 2013). These sinks vary seasonally, especially in the Arctic atmosphere, and their intensity is at maximum in summer, when Arctic emissions are the highest. A good representation of the methane budget thus requires a proper knowledge of these sinks.
As mentioned before, a better understanding of methane sources and sinks and of their variations is critical in the context of climate change. Methane emissions can be estimated either by bottom-up studies, relying on extrapolation of flux measurements, on inventories and process-based models, or by top-down inversions which optimally combine atmospheric observations, transport modelling, and prior knowledge of emissions and sinks. The main input for top-down inversions is measurements of atmospheric methane mixing ratios, either at the surface or from space. Such observations are critical and should be made over long time periods to assess trends and variability. Surface methane monitoring started in the Arctic in the mid-1980s. Although more than 15 sites currently exist, six of them being in continuous operation (in addition to tower sites such as the JR-STATION tower network over Siberia; Sasakawa et al., 2010), the observational network remains limited considering the Arctic area and the variety of existing sources (AMAP, 2015).
Retrievals of methane concentrations have been made from space since the mid-2000s, from global and continuous observations. However, at high latitudes, passive spaceborne sounders are limited by the availability of clear-sky spots and by sunlight (for NIR/SWIR instruments), and have been affected by persistent biases (e.g. Alexe et al., 2015; Locatelli et al., 2015). This is why only surface measurements, which provide precise and accurate data, are used in this study.
One interesting feature of Arctic methane emissions is that they are generally more distinct spatially and temporally (no or low wetland emissions in winter; anthropogenic emissions all year round) compared to tropical emissions (e.g. in northern India). Also, fast horizontal winds more efficiently relate emissions to atmospheric measurements (e.g. Berchet et al., 2016).
Methane modelling studies that rely on Arctic measurements have been used,
for example, to assess the sensitivity of Arctic methane concentrations to
uncertainties in its sources, in particular concerning the seasonality of
wetland emissions and the intensity of ESAS emissions (Warwick et al., 2016;
Berchet et al., 2016). Top-down inversions have also led to methane surface
flux estimates and discussions of their variations. For instance, Thompson et
al. (2017) have found significant positive trends in emissions in northern
North America and northern Eurasia over 2005–2013, contradicting previous
global inversion studies based on a more limited observational network north
of 50
Combining atmospheric methane modelling using the CHIMERE chemistry-transport model (Menut et al., 2013) and surface observations from six continuous measurement sites, this paper aims to evaluate the information contained in methane observations concerning the type, the intensity, and the seasonality of Arctic sources. The study focuses on 2012, as this is the last year for which wetland emissions are available for a set of models in a controlled framework. Section 2 describes the data and modelling tools used in this study. Section 3 analyses the simulated methane mole fractions and investigates their agreement with the observations. It also discusses the sensitivity of the model to wetland and freshwater sources, as well as to methane sinks. Section 4 concludes this study.
Continuous methane measurements for the year 2012, from the six Arctic surface sites, have been gathered. The site characteristics are given in Table 1, and Fig. 1 represents their position in the studied domain. Two sites are considered to be remote background sites: Alert, located in northern Canada, where measurements are carried out by Environment Canada (EC), and Zeppelin (Ny-Alesund), located in Svalbard archipelago on a mountaintop, and operated by the Norwegian Institute for Air Research (NILU). NOAA Earth System Research Laboratory (NOAA-ESRL) is responsible for the measurements at Barrow observatory, which is located in northern Alaska, 8 km north-east of the city of Barrow, and at Cherskii. Cherskii and Tiksi are located close to the shores of the East Siberian Sea and the Laptev Sea, respectively. Pallas is located in northern Finland, with dominant influence from Europe. Measurements at these last two sites are carried out by the Finnish Meteorological Institute (FMI). No data were available in Barrow in 2012 after May due to a lapse in funding (Sweeney et al., 2016). Gaps in Cherskii (October–January), Pallas (August–mid-October), and Zeppelin (January–April) data are due to instrument issues.
Description of the six continuous measurement sites used in this study.
Delimitation of the studied Polar domain and location of the six continuous measurement sites used in this study. ALT: Alert. BRW: Barrow. CHS: Cherskii. PAL: Pallas. TIK: Tiksi. ZEP: Zeppelin.
Data from Alert, Barrow, and Pallas were downloaded from the World Data
Centre for Greenhouse Gases (WDCGG,
Methane emissions in the studied polar domain, for the reference
simulation, and for other scenarios. Total emissions for the reference
scenario amount to 68.5 TgCH
The CHIMERE Eulerian chemistry-transport model (Vautard et al., 2001; Menut
et al., 2013) has been used for simulations of tropospheric methane. It
solves the advection–diffusion equation on a regular grid, forced using
pre-computed meteorology. Our domain goes from 39
The model is forced by meteorological fields from European Centre for Medium
Range Weather Forecasts (ECMWF) forecasts and reanalyses
(
The model is run with seven distinct tracers: six correspond to the different Arctic emission sources (anthropogenic, biomass burning, geology & oceans, ESAS, wetlands, and freshwater systems) and one corresponds to the boundary conditions. This framework allows us to analyse the contribution of each source in the simulated total methane mixing ratio, defined as the sum of each tracer. No chemistry is included in the standard simulations, but a sensitivity test is carried out (see Sect. 3.4).
Surface emissions used here stem from a set of various inventories, models, and data-driven studies, from which are built a reference scenario, complemented by several sensitivity scenarios. The different emission sources used are described in Table 2, along with the amount of methane emitted in the studied domain.
All types of anthropogenic emissions are provided by the EDGAR (Emission
Database for Global Atmospheric Research) v4.2 Fast Track 2010 (FT 2010) data
(Olivier and Janssens-Maenhout, 2012), which have a
0.1
Biomass burning emissions come from the Global Fire Emissions Database
version 4 (GFED4.1) (van der Werf et al., 2010; Giglio et al., 2013) monthly
means product. Burned areas estimated from the MODIS spaceborne instrument
are combined with the biomass density and the combustion efficiency derived
from the CASA biogeochemical model, and with an empirically assessed
emission factor. The emissions are provided on a 0.25
Wetland emissions in the reference scenario come from the ORCHIDEE-WET model
(Ringeval et al., 2010, 2011), which is derived from the ORCHIDEE global
vegetation model (Krinner et al., 2005). The wetland methane flux density is
computed for each 0.5
Emissions from geological sources, including continental macro- and
micro-seepages, and marine seepages, are derived from the GLOCOS database
(Etiope, 2015). They represent 4.0 TgCH
ESAS emissions are prescribed following Berchet et al. (2016), and scaled to
2 TgCH
Since they have generally been represented poorly or not at all in former
atmospheric studies, freshwater emissions were built for the purpose of this
work. The inventory is based on the GLWD level 3 product (Lehner and
Döll, 2004), which provides a map of lake and wetland types at a 30 s
(
As a result, we built an inventory for freshwater emissions (Fig. 2a), (i) with a
total budget of 9.3 TgCH
The more recent GLOWABO (Global Water Bodies) database (Verpoorter et al.,
2014) has a higher resolution than the GLWD (0.002 vs. 0.1 km
A simulation of seven methane tracers is run with CHIMERE for 2012. On top of methane from initial and boundary conditions, these include methane from anthropogenic sources, biomass burning, East Siberian Arctic Shelf (ESAS), geology and oceans (counting as only one source and excluding ESAS), wetlands, and freshwater systems.
The boundary conditions are the dominant signal; they result from emissions coming from sources located outside of the domain, and from emissions coming from Arctic sources, which have once left the domain and then re-entered in it. The boundary condition tracer does not hold information on where the transported methane initially comes from. So, to focus on Arctic sources, the source contributions are defined here relatively to the sum of the six tracers which correspond to sources located in the domain, i.e. excluding methane resulting from the boundary conditions. The source contribution is only calculated when methane directly from Arctic sources is greater than 1 ppb. One should keep in mind that this signal represents a small fraction of total atmospheric methane.
The weight of each source varies both spatially and seasonally. Figures 3 and 4 represent the mean source contributions to methane concentrations near the surface, in winter (November to May) and in summer (from June to October), respectively.
In winter, anthropogenic methane is dominant (over winter, the daily average over the domain is in the range 18–59 %, with a mean of 42 %). More than 80 % of anthropogenic emissions come from oil, gas, and coal industries. In particular, it affects western Russia (mostly due to gas production), the Khanty–Mansia region (mostly due to oil production), and south-eastern Russia (mostly due to coal mining). Oil production is also the main contributor to atmospheric methane in continental Canada.
Geological and oceanic emissions represent an important part of atmospheric methane in the domain, particularly in winter (11–36 %, mean: 27 %). Emissions from ESAS are expected to be larger in summer, when most of the area is ice-free, than in winter. However, its relative contribution is higher in winter (8–23 %, mean: 15 %), when other sources, particularly from wetlands, are lower. Alaska and northern Siberia are particularly affected by geological and oceanic emissions in winter, including from ESAS.
Mean source contributions (in %) to the CH
Mean source contributions (in %) to the CH
Sources contributions (in %, left axis) to the CH
In summer, wetland emissions are the dominant contributor (33–56 %, mean: 50 %) (although anthropogenic emissions remain important in western Russia), while they are quite negligible in winter. Freshwater systems too are an important contributor in summer (9–29 %, mean: 19 %), but of lower intensity than wetlands, except in eastern Canada and Scandinavia, where methane from lakes can exceed methane from wetlands.
Biomass burning takes place in summer (0–7 %, mean: 4 %), when fuel characteristics and meteorological conditions foster combustion. Although the 2012 fire emissions are particularly high (e.g. almost twice as high as the 2013 emissions) and large-scale fires occur in boreal Russian and Canadian forests, their impact on methane remains limited to some regions in continental Russia.
The contribution of the different sources is more quantitatively discussed in the following, focusing on the six continuous measurement sites shown in Figs. 3 and 4.
The evolution of the daily averaged source contributions at the six sites is represented in Fig. 5. In December and from January to April, methane from Arctic sources is driven by anthropogenic, ESAS, and geological and oceanic emissions at all sites. It is confirmed by the figures in Tables 3 and 4, which give the mean relative and absolute contributions, respectively, for winter and summer. Over winter, anthropogenic sources account for more than 50 % only in Pallas and Zeppelin. For the other four sites, anthropogenic emissions contribute between 23 and 35 %, while methane from continental seepages and oceans, including ESAS, account for more than 54 % of methane from Arctic sources, and up to 68 % at Tiksi, corresponding to 18 ppb. ESAS emissions have the lowest impact in methane levels in Pallas and Zeppelin (< 1 ppb). Freshwater systems and wetlands combined contribute between 8 and 27 % in winter, corresponding to only a few ppb.
Mean source contributions (in %) to atmospheric CH
Same as Table 3, but for the absolute values, in ppb.
Wetland emissions start having an impact in May and dominate from June to October, fading in November (Fig. 5). Freshwater emissions present a similar seasonal cycle, except in Pallas where some contributions are seen in December–January. According to the lake inventory developed here, southernmost Scandinavian lakes have not frozen over and continue to emit until January. Elsewhere, their contribution follows the same seasonality as wetland emissions' but lagged by 1 month, and with a lower impact. In summer, wetland emissions are the major contributor from Arctic sources at all sites (from 48 to 70 %, or from 10 to 84 ppb), and methane from both wetland and freshwater sources amounts to at least 65 % of methane from Arctic sources, on average, for all sites. These two major sources overshadow anthropogenic sources, the impact of which remains below 16 %. Only Cherskii and Tiksi are substantially impacted by ESAS emissions in summer (10 and 17 %, or 8 and 11 ppb, respectively). Overall, biomass burning negligibly contributes to the methane abundance at the six surface sites.
Time series of simulated (in colour) and observed (black points)
methane mixing ratios in ppb, at Alert, Barrow, and Cherskii in 2012. The
baseline is the contribution of the boundary conditions alone. Time
resolution for simulations and observations is 1 day. Maximum for Cherskii
CH
Same as Fig. 6, for Pallas, Tiksi, and Zeppelin.
Difference between the means of CH
Figure 5 also shows the evolution of the simulated methane from Arctic sources (white line, right axis). Over the year, Alert, Pallas and Zeppelin mixing ratios have lower contributions from Arctic sources (always below 60 ppb) than Barrow, Cherskii, and Tiksi (sometimes more than 120 ppb). In winter, although the source repartition is different among the sites, methane levels are quite low for all of them, from 10 ppb in Alert to 26 ppb in Tiksi, on average (Table 4). However, there still are individual peaks related to either predominant anthropogenic or ESAS sources. In Alert, for example, on 1 March, methane from Arctic sources reaches 31 ppb, 77 % of which corresponds to anthropogenic sources. In Cherskii, on 5 April, 89 % of the 45 ppb methane signal came from ESAS emissions. Contributions from geological and oceanic sources can reach the highest proportions in winter, but repeatedly correspond to only a few ppb of methane, up to only 14 ppb in Barrow in 4 December.
In summer, all measurement sites see higher methane contributions from Arctic sources, predominantly from wetland emissions, with Barrow, Cherskii, and Tiksi being more affected by them. These last three sites experience contributions greater than 45 ppb on average, while, for the three others, contributions from Arctic sources remain below 26 ppb. The freshwater signal is almost always less than the wetland signal, but even for Alert and Zeppelin, which have the lowest levels of methane from freshwater emissions, it sometimes exceeds 25 %, with substantial corresponding contributions in ppb.
The simulated absolute values of total methane at the sites are shown in Figs. 6 and 7, along with the observed mixing ratios. There is good agreement between observed and simulated methane, both in terms of intensity and temporal evolution. In particular, the model shows its ability to reproduce short-term peaks and drops, which are either due to the intrusion of enriched or depleted air from outside of the domain or directly due to the evolution of Arctic sources.
Although Arctic emissions are greater in summer, Alert, Pallas, and Zeppelin have higher methane values in winter due to a higher influence of air from lower latitudes, with a methane seasonal cycle that is mostly driven by OH. Table 5 gives the differences between the mean methane in winter and the mean methane in summer for the observations and the reference simulation. The greatest seasonal cycle is seen in Pallas, the closest site to midlatitude Europe. Tiksi is less sensitive to boundary conditions, and the influence of summer sources produces an opposite seasonal cycle (maximum in summer), although with a weaker average amplitude than for the three sites mentioned above. Observations in Cherskii show no clear seasonal cycle in contradiction to the simulation, particularly in September when simulated methane from wetlands frequently exceeds 100 ppb. This discrepancy is mainly due to an overestimation of wetland emissions by ORCHIDEE in the region near Cherskii.
As we have seen above, these two kinds of seasonal cycle do not prevent the same kind of events from happening at the scale of a few days (synoptic variations). For instance, even if methane variability in Alert, Pallas, and Zeppelin is mostly driven by the boundary conditions in winter, measurements made at these sites do hold information on Arctic (anthropogenic, geological, and oceanic) sources during particular synoptic events. And in summer, methane peaks have important contributions at all sites from wetland and freshwater emissions. Overall, with the exception of biomass burning, all sources have a substantial impact on the six measurement sites, whether it is on the scale of synoptic events of a few days or regularly occurring over the course of several months.
The overall good agreement between simulations and measurements is quantified in Table 6, which gives the mean difference between observed and simulated methane during 2012. The mean daily bias remains below 7.5 ppb for all sites, except for Cherskii, where it reaches 34.8 ppb, mostly because of a large overestimation of methane from wetland emissions in September. For all sites, the bias stems from an overestimation of modelled methane in summer (in the range 4.8–8.6 ppb, Cherskii excluded), which is compensated in winter by either a lower overestimation (Pallas, Tiksi, Zeppelin) or an underestimation (Alert, Barrow, Cherskii). As a result, the seasonality is well captured in Pallas, Tiksi, and Zeppelin, but is not pronounced enough in Alert (Table 5).
At Alert (Fig. 6), simulated methane is higher than the measurements in June and July. The boundary conditions may be responsible for this disagreement, given that, for several days, the measurements are lower than methane resulting from the boundary conditions alone. The absence of the methane sinks in the reference simulation may also be a reason. It may also indicate that the emissions are not well represented in the reference simulation. In August, September, and October, then, the reference simulation agrees better with the measurements, although the intensity of some modelled peaks may be too low.
The results of our reference simulation depend on the hypotheses made, especially on source distribution (see Figs. S1–S6 in the Supplement) and the absence of methane sinks. The impacts of wetland and freshwater source distribution and of methane sinks on modelled atmospheric methane are investigated in the next sections as sensitivity tests.
As noted previously, wetland emissions represent the main source of methane in the Arctic, explaining at least 48 % of the methane signal from Arctic sources for all six measurement sites in summer on average. Therefore, the representation of wetland emissions in Arctic methane modelling is crucial. This is why the outputs of 10 other land-surface models than ORCHIDEE have been tested for June to October 2012 (assuming significant wetland emissions only take place at this time of year). The impact of the different land-surface models is assessed focusing on the four sites that provide data uniformly distributed over these 5 months (Alert, Cherskii, Tiksi, and Zeppelin).
Mean difference (and standard deviation) between observed and
simulated CH
Mean difference (and standard deviation) between observed and
simulated CH
The 11 land-surface models are described in Poulter at al. (2017)
and Saunois et al. (2016). Wetland emissions are mostly located in
Scandinavia, between the Ob and Yenisei rivers and between the Kolyma and
Indigirka rivers in Russia, Nunavut (NU), and Northwest Territories (NT) in
Canada, and in Alaska, with large discrepancies among the models even if
they use the same wetland emitting zones (see Sect. 2.3). Emissions from
all models and their evolution over the year are illustrated in Figs. S5 and S6.
For all models, emissions start in May and end in October. The maximum
in emission is reached in June (for the LPJ-wsl, CTEM, and DLEM models) or
in July. Only the LPX-Bern and SDGVM models have maximum emissions in August
and September, respectively. The latter has the highest emissions of all
models in September and October due to its
Given the sensitivity to the variability of methane from the boundary
conditions in Alert and Zeppelin, and its likely overestimation in June–July
(see Sect. 3.1.3), the bias alone is not a good criterion for evaluating
the different wetland models. Instead, Fig. 8 shows Taylor diagrams of the
comparisons between methane simulated with the outputs of 11 different
land-surface models and the measurements. At Alert, SDGVM is the best
performing model in terms of its correlation with the measurements
(correlation coefficient
Taylor diagram representations of the comparison between
observations (star marker) and CH
Difference between the absolute values of the biases between
simulated and observed CH
In Tiksi, the high variability and high values of methane peaks lead to low correlation coefficients, as the model is not fully able to reproduce the short-term variability whatever the wetland emission. However, SDGVM reaches a correlation coefficient of 0.60. SDGVM and ORCHIDEE have standard deviations similar to the measurements and two of the three lowest biases. However, ORCHIDEE's correlation coefficient is only 0.39.
In Cherskii, like in Tiksi, the model has troubles reproducing the
variability of the measurements, and this can lead to high biases. However,
CLM4.5 and LPX-Bern have biases below 9 ppb and correlation coefficients
above 0.62, with similar standard deviations. It is worth noting that SDGVM
and ORCHIDEE have the two worst correlation coefficients here. Again, the
simulation with ORCHIDEE has unexpectedly extreme values in September, up to
2925 ppb, certainly due to outlying high emissions in the Kolyma and
Indigirka regions in this month. Indeed, according to ORCHIDEE,
1.4 TgCH
The comparison between the measurements and the simulations performed with the outputs of 10 different land-surface models and with the reference scenario, show that no wetland emission model performs perfectly. SDGVM and LPX-Bern, which is overall the least biased model, seem to be the two most reliable models on average. These models are characterized by low emissions in early summer/late spring. ORCHIDEE, except in Cherskii, has a fairly average performance compared to the other models. On the contrary, LPJ-MPI is a clear outlier, leading to methane values that are too high.
The results obtained in Sect. 3.1 appear to be sensitive to the choice of the land-surface model. More effort is needed to better represent the location, timing, and magnitude of Arctic wetland emitting zones (Tan et al., 2016). Continuous observations clearly offer a good constraint to handle this challenge.
Freshwater emissions are the second main contributing source in the Arctic in summer, explaining between 11 and 26 % of the atmospheric signal at the six measurement sites on average. As was previously noted, there is a large uncertainty affecting the distribution and magnitude of this particular source. This is why an alternative lake emission inventory is tested here. bLake4Me is a one-dimensional, process-based, climate sensitive lake biogeochemical model (Tan et al., 2015; Tan and Zhuang, 2015a, b). Model output used here corresponds to the 2005–2009 average.
The difference between the inventory used in the reference simulation and
the one based on bLake4Me is shown in Fig. 2b. Since bLake4Me's output is
only available above 60
Figure 9 represents the difference between the absolute value of the bias
calculated with the simulation using the bLake4Me inventory and the absolute
value of the bias of the reference simulation. A positive value (black
dots), therefore, means that the freshwater inventory developed for the
reference simulation performs better than the bLake4Me inventory. For Alert,
Barrow, Pallas, and Zeppelin, differences in the bias generally remain
within
In Alert and Zeppelin, using bLake4Me inventory increases simulated methane by a few ppb in July–September, with no major changes during the rest of the year. This leads to an increase in the bias, although this can also improve agreement with the measurements for some periods, particularly in September, when the reference simulation underestimates some methane peaks. Table 5 shows that the changes brought by the new inventory worsen the seasonality simulated at these two stations.
Difference between the reference simulation and
Only in Pallas does the bLake4Me inventory lead to lower simulated methane,
particularly in winter, linked to the shortened season of freshwater
emissions in Scandinavia. As a consequence, the bias is improved from
Although bLake4me produces physical outputs of freshwater emissions, and is therefore far more advanced than the crude inventory developed here for the reference simulation, no significant improvement is found in comparisons between simulated and observed methane at the six measurement sites. Once again, as stated for wetlands (Sect. 3.2), the distribution and magnitude of lake emissions can be critical for correctly reproducing methane concentrations at sites located nearby (e.g. Cherskii). Using such observational stations combined with a chemistry-transport model offers a good constraint to improve the magnitude and location of methane emissions from lakes in the Arctic.
Regional modelling of atmospheric methane generally does not consider
methane sinks, focusing more on synoptic variations than on long-term
changes. This is justified by the rather long methane lifetime
(
Time series of simulated and observed methane mixing ratios at Alert in 2012. The cyan line represents the contribution of the boundary conditions; the red line represents the added direct contribution of the sources emitting in the domain; the black line includes the three added sinks (OH, soil, Cl). The blue points represent the observations. Time resolution for simulations and observations is 1 day.
The main atmospheric loss of methane results from OH oxidation in the
troposphere. OH concentrations are higher in summer and above continents, as
OH production is controlled by solar radiation, albedo, and the
concentrations of NO
Figure 10a shows the difference between the reference simulation and the simulation including methane oxidation by OH, thus representing the effect of the methane sink due to OH on the mixing ratios (set to a positive value). As expected, the impact is mostly visible in summer. Even if the general pattern is similar among the sites – a progressive increase in the OH sink effect from March to July, when it can be as high as 12 ppb, and a symmetric decrease until November – the daily variability in the OH sink effect is not the same for all sites. Pallas, for example, has the strongest variability. This variability stems from the disparity in the proximity/distance of the origin of the air masses observed at the sites combined with the heterogeneity in the distribution of OH concentrations.
The second potential chemical sink lies in the oxidation of methane by
chlorine (Cl) in the marine boundary layer. Theoretical prescribed Cl fields
were thus included in CHIMERE, following the recommended scenario described
in Allan et al. (2007). Cl atoms are concentrated in the marine boundary
layer, above ice-free zones. Daily sea ice data from the EUMETSAT Ocean and
Sea Ice Satellite Application Facility (OSI SAF,
Uptake of methane from methanotrophic soil bacteria is considered here a
surface sink. Here we use the monthly 1
We finally investigate whether the integration of these three methane sinks
improves the fit to observed methane mixing ratios. Figure 11 shows
simulated methane at Alert, including the cumulated effects of the three
sinks, and compares it to the reference simulation and to the measurements.
Indeed, for all sites, the reference simulation is too high in summer, but
in Alert in particular, it does not properly reproduce the sharp decrease in
methane from April to July (
On average, including the sink processes, and especially OH chemistry, appears to significantly improve the simulation of methane. However, as expected, these loss processes are not sufficient to fully explain the discrepancies in the seasonal variations between the model and the measurements.
Atmospheric methane simulations in the Arctic have been made for 2012 with a
polar version of the CHIMERE chemistry-transport model and implemented with a
regular 35
The simulations have been compared to six continuous measurement sites. Half of these sites have their seasonality mainly driven by air from outside of the Arctic domain studied here, with higher concentrations in winter than in summer, although Arctic sources are stronger in summer. The model is able to globally reproduce the seasonality and magnitude of methane concentrations measured at the sites. All sites are substantially impacted by all Arctic sources, except for biomass burning. In winter, when methane emitted by Arctic sources is lower, the sites are more sensitive to either anthropogenic or ESAS emissions on the scale of a few days; over the whole summer, they are more sensitive to wetland and freshwater emissions.
The main disagreement between the simulated and observed methane mixing ratios may stem from, in part, inaccurate boundary conditions, overestimation or mis-location of some of the sources, particularly during the May–July time period, or lack of methane sinks. We have conducted a series of sensitivity tests which vary wetland emissions and freshwater emissions, and include methane sinks.
On top of the wetland emissions computed by the land-surface model ORCHIDEE
(used in our reference simulation), the outputs of 10 other process-based
land-surface models have been tested. Among them, the SDGVM and LPX-Bern
models appear to be the most convincing at reconciling the simulations with
the measurements. These models have lower emissions than most of the models
in May–July, and reach a maximum of emission later in September and August,
respectively, while the others have their maximum in June–July. Over the
wetland emission season, they both have lower emissions than ORCHIDEE (19 and
26 vs. 30 TgCH
The influence of freshwater emissions, which account for 11–26 % of the
methane signal from Arctic sources in summer at the six sites, is also
assessed, and found to be significant. Our simple inventory, where a
prescribed total budget of 9.3 TgCH
The inclusion of the major methane sinks (reaction with OH and soil uptake) in regional methane modelling in the Arctic is shown to improve the agreement with the observations. The cumulated impact of the sinks significantly decreases bias in the simulations at the sites. Reaction with Cl in the marine boundary layer, on the contrary, has a negligible impact.
Our work shows that an appropriate modelling framework combined with continuous observations of atmospheric methane enables us to gain knowledge on regional methane sources, including those which are usually poorly represented such as freshwater emissions. Further understanding and knowledge of the Arctic sources may be obtained by combining tracers other than methane, such as methane isotopologues, within forward or inverse atmospheric studies. Such a study would gain in robustness with a wider and more representative atmospheric observational network. It is therefore of primary interest, considering the changing climate and the high climate sensitivity of the Arctic region, to maintain and further develop methane atmospheric observations at high latitudes, considering both remote and in situ observations. So far, remote sensing of atmospheric methane is mainly based on sunlight absorption, thus not appropriate during high latitude winter. After 2020, the MERLIN space mission, based on a lidar technique, should bring an interesting complement to the surface and actual remote sensing observations (Kiemle et al., 2014), but with lower time resolution than continuous surface stations.
Measurement data from Alert, Barrow and Pallas are
available from WDCGG (
The authors declare that they have no conflict of interest.
We thank the principal investigators of the observation sites which were used
in this study for maintaining methane measurements at high latitudes and
sharing their data. This work has been supported by the Franco-Swedish
IZOMET-FS “Distinguishing Arctic CH