What do we know about the global methane budget? Results from four decades of atmospheric CH4 observations and the way forward

Atmospheric CH4 is arguably the most interesting of the anthropogenically influenced, long-lived greenhouse gases. It has a diverse suite of sources, each presenting its own challenges in quantifying emissions, and while its main sink, atmospheric oxidation initiated by reaction with hydroxyl radical (OH), is well-known, determining the magnitude and trend in this and other smaller sinks remains challenging. Here, we provide an overview of the state of knowledge of the dynamic atmospheric CH4 budget of sources and sinks determined from measurements of CH4 and δ13CCH4 in air samples collected predominantly at background air sampling sites. While nearly four decades of direct measurements provide a strong foundation of understanding, large uncertainties in some aspects of the global CH4 budget still remain. More complete understanding of the global CH4 budget requires significantly more observations, not just of CH4 itself, but other parameters to better constrain key, but still uncertain, processes like wetlands and sinks. This article is part of a discussion meeting issue 'Rising methane: is warming feeding warming? (part 1)'.


Introduction
Atmospheric CH 4 absorbs terrestrial infrared radiation in a band at approximately 7.6 µm, so it affects Earth's radiation balance and is a primary driver of the human impact on climate. Based on measurements of air extracted from ice cores and firn, we know that atmospheric CH 4 increased from approximately 360 ppb (see note about units in electronic supplementary material) during glacial periods to approximately 700 ppb during interglacial periods [1] as receding high-northern latitude glacial ice and changes to monsoon circulation resulted in more wetland area. This change was initiated by small variations in Earth's orbital parameters that affected the seasonal distribution of solar radiation across Earth's surface, but most of the increase in atmospheric CH 4 likely resulted from feedbacks in the carbon cycle [2]. Natural emissions in the southern tropics may have driven a sustained atmospheric CH 4 increase starting 5000 years before present [3]. These Holocene and earlier paleo records demonstrate the strong sensitivity of atmospheric CH 4 to changing climate.

and stratospheric H 2 O.
Developing sensible mitigation strategies for methane [5] requires a quantitative understanding of individual source and sink terms. Strong climate feedbacks similar to those that drove changes in atmospheric CH 4 abundance in the Holocene have the potential to greatly enhance CH 4 emissions, both in the Arctic, where current warming might release huge stores of carbon (greater than 1 Pg C in the top 3 m of permafrost [6]) as CH 4 or CO 2 , depending on hydrology, and also in the tropics, where wetlands, including flooded land used for rice agriculture, and cattle may respond to warmth and rainfall.

Observations of atmospheric CH 4 abundance and δ 13 C CH4 as budget constraints (a) Experimental methods
Blake et al. [7] showed that CH 4 in the surface atmosphere varied spatially and seasonally across the globe. Recognizing the importance of improved understanding of the global CH 4 budget for climate, NOAA began measurements of atmospheric CH 4 from discrete air samples collected in its existing Cooperative Global Air Sampling Network in 1983, expanding available high-quality data both in frequency and spatially [8] to better constrain the global CH 4 budget. Since 1998, δ 13 C CH4 has been measured in a subset of the same air samples by the Institute of Arctic and Alpine Research (INSTAAR), University of Colorado [9]. Currently, many laboratories around the world monitor atmospheric CH 4 abundance and a few measure δ 13  Greenhouse Gases hosted by the Japan Meteorological Agency (https://gaw.kishou.go.jp/). Even with this structure in place, there are still 'ease of use' issues; not all laboratories report data in a timely fashion, they report data on different standard scales without providing conversion to the WMO GAW CH 4 mole fraction scale, uncertainties are not reported, etc. Isotopic data are further hindered by a lack of CH 4 -in-air reference materials and different methods of tracing δ 13 C CH4 to calcite reference materials [10]. To circumvent these issues, some research programs have collected data and made packages of observations available (see e.g. https://www.esrl. noaa.gov/gmd/ccgg/obspack/). Future effort is required to ensure high-quality measurements can be traced back to common standard scales, because our understanding of the global CH 4 budget relies on precise spatio-temporal differences from these measurements. The data should also be publicly accessible in a timely manner, fully quality controlled and ready for scientific analysis; data users should not have to add uncertainties nor ensure traceability to a common standard scale. We do not cover column CH 4 abundances retrieved from measurements of radiance by satellite or surface IR sensors, which are not direct, calibrated measurements of CH 4 abundance and have potential biases similar in magnitude to the large-scale signals we are trying to determine. While they are still useful, especially in the tropics, where direct measurements remain sparse, and in regional studies, they will always require calibrated direct measurements of CH 4 for evaluation.
Here, we focus on NOAA CH 4 and INSTAAR δ 13 C CH4 measurements in air samples collected from the Global Monitoring Laboratory's Cooperative Global Air Sampling Network. The network and sampling procedures were described in Dlugokencky et al. [11]. Briefly, weekly discrete whole-air samples are collected in pairs of borosilicate glass flasks with PTFE O-ring seals using a portable air sampler from a globally distributed network of air sampling sites (https://www.esrl.noaa.gov/gmd/ccgg/flask.html). Air samples were analysed for CH 4 by gas chromatography with flame ionization detection through mid-2019. Since then, samples have been analysed for CH 4 using a cavity ring-down spectrometer (CRDS). Responses of the GC/FID and CRDS were calibrated for CH 4 abundance against a CH 4 standard scale maintained at NOAA ( [12]; this scale serves as the WMO CH 4 mole fraction scale, X2004A). These data provide a spatially and temporally consistent set of observations. INSTAAR began analysing a subset of air samples for δ 13 C CH4 in 1998. CH 4 from the sample is pre-concentrated, separated from other remaining components in a GC, combusted to CO 2 and the CH 4 -derived CO 2 analysed by continuous flow isotope ratio mass spectrometry. CH 4 and δ 13 C CH4 data are available from NOAA's data server: ftp://aftp.cmdl.noaa.gov/data/trace_gases/xxx/flask/, where xxx = 'ch4' or 'ch4c13'.

(b) Documenting CH 4 's global burden
Observations of the spatial and temporal distribution of atmospheric CH 4 abundance provide key constraints on the global CH 4 budget. In 2020, preliminary globally averaged CH 4 determined by smoothing data as described below was 1879.2 ± 0.6 ppb, with an increase of 14.8 ± 0.5 ppb from January 1, 2020 to January 1, 2021 (https://gml.noaa.gov/ccgg/trends_ch4/; downloaded 8 July 2021). This implies an imbalance between emissions and sinks in 2020 of 41.0 Tg CH 4 (based on 1 globally averaged ppb = 2.77 Tg CH 4 [13]). The time series of these observations from 1983.5 through 2019 is shown in figure 1a. Global (and zonal) means are calculated by first smoothing data from a subset of sampling sites that capture well-mixed, background air, extracting values from the smoothed curve fits at synchronized time steps, then smoothing again as a function of latitude to define a matrix of CH 4 abundance as a function of time (48 values per year) and latitude (intervals of sine (latitude) = 0.05). The quality of the data ensures that the variations in growth rate are significant, as indicated by the uncertainties (dashed lines in figure 1b) determined with a combination of resampling of the network (bootstrap; [14]) and Monte Carlo (to assess measurement uncertainty) methods (see electronic supplemenatry material for details).   For 1983-2006, if we assume no trend in the atmospheric CH 4 lifetime, then the atmospheric CH 4 burden was approaching steady state [15]. Starting from a global mass balance and rearranging to solve for emissions we have: where E is emissions, [CH 4 ] is the global annual mean CH 4 burden and d [CH 4 ]/dt its annual increase, both determined from our surface global means, and τ is the atmospheric CH 4 lifetime. After calculating emissions using τ = 9.1 yr [16] and fitting a straight line to emissions from 1984 through 2006, we find no trend in emissions (electronic supplementary material, figure S1; slope = 0.0 ± 0.3 Tg CH 4 yr −1 ; uncertainty is 68% C.I.). It may seem surprising that atmospheric CH 4 continues to increase while emissions and lifetime are constant, but this is expected from a chemical system with pseudo-first-order loss approaching steady state,  4 ] 0 ), and τ to give a steady-state value of 1794 ppb and a lifetime of 9.2 yr, in agreement with IPCC's Fifth Assessment Report CH 4 budget lifetime of 9.1 yr [16]. A curve calculated using the coefficients of the fit, extrapolated to 2020, is plotted in figure 1a (green), and the difference between it and the deseasonalized trend (green, in ppb) is shown with the growth rate (red) in figure 1b.
Since 2007, when this difference starts increasing dramatically, there has been a significant change in the global CH 4 budget that will be explored later. Our conclusion of no trend in emissions from 1983 to 2006 is true only if there has been no trend in τ CH4 , which is the largest source of uncertainty in this simple global mass balance. For changes in global CH 4 emission rates to be consistent with the observed global means, τ CH4 would have had to change in a way that still coincidentally fits this steady-state model.

(d) Interannual variability
In addition to the long-term changes in atmospheric CH 4 , IAV is also evident in the growth rate and the residuals (figure 1b), and it has inherent information about budget processes. When the process, or processes, responsible for IAV can be identified, it allows us to quantitatively test our understanding. An example is a large increase in CH 4 growth rate in 1991 (red line in figure 1b) and the positive anomaly in the difference between the trend and steady-state fit in 1991 (green line in figure 1b). In this smoothed representation, which is similar to a 12-month running mean, the anomaly peaked at approximately 5 ppb; at weekly time resolution, it was approximately 9 ppb (see electronic supplementary material, figure S2). The subsequent decrease in 1992 growth rate is a result of the trend adjusting back to the long-term behaviour as indicated by the difference between the trend and steady-state fit returning to near-zero (green line in figure 1b).
Dlugokencky et al. [17] attributed this positive anomaly to the impact of the eruption of Mt. Pinatubo on the atmospheric CH 4 sink. SO 2 and ash from the eruption, and subsequent sulfate aerosol produced from the oxidation of SO 2 , decreased transmission of UV radiation, which decreased production of O( 1 D) and, ultimately, primary production of hydroxyl radical (OH). Qualitatively, this made sense, since anomalies in CO abundance were coincident and consistent in magnitude with the changes in CH 4 . Bândȃ et al. [18] investigated this change with a threedimensional global chemistry model and found SO 2 and sulfate reduced the tropospheric CH 4 sink from reaction with OH by 17.8 Tg CH 4 from June, 1991 to June, 1993, which only explained approximately 40% of the change in atmospheric CH 4 burden from June, 1991 to May, 1992. Bândȃ et al. [19] identified additional factors that contributed significantly to changes in atmospheric CH 4 growth rate including: changes in stratospheric O 3 , which also affected UV flux; atmospheric temperature, which affected CH 4 emission rates from wetlands and the rate of oxidation of CH 4 (k OH+CH4 is strongly temperature-dependent, approximately 2% K −1 ); biomass burning; and changes in rates of emission of other compounds that affect OH concentration (e.g. CO and NMHCs).

(e) Spatial patterns as budget constraints
Because atmospheric mixing is not instantaneous, signals of sources and sinks exist in the atmospheric CH 4 distribution that are useful for quantifying emissions. Fung et al. [13] compared the time-averaged latitude gradient calculated with a chemical transport model (CTM) to NOAA CH 4 observations (electronic supplementary material, figure S3) to rule out specific emission scenarios that were inconsistent with atmospheric observations. They showed, for example, that early estimates of CH 4 emissions from northern wetlands were too large to be consistent with the observed latitude gradient in surface CH 4 .
We exploited the gradient further by analysing temporal changes in spatial patterns to identify changes in the distribution of emissions, although this analysis cannot quantitatively determine the magnitude of the emission changes. Following the economic collapse in the former Soviet Union, Dlugokencky et al. [20] showed that a change in inter-polar CH 4 difference (IPD; the difference between zonal averages calculated for 53-90°N and 53-90°S) was consistent with a decrease in CH 4 emissions from the former Soviet Union of approximately 10 Tg between 1991 and 1992. Agreement between observations and a forward simulation with an atmospheric transport model (TM3) using emission estimates from Emissions Database for Global Atmospheric Research (EDGAR) was quite good; both showed a decrease of approximately 12 ppb in annual means calculated for high-northern latitude sites relative to the South Pole between the mid-1980s and the late-1990s (electronic supplementary material, figure S4). Such analyses are only qualitative, because observed trends in spatial patterns are affected by changes in atmospheric transport, not just emissions [21], so quantitative estimates of emission changes must be assessed with a CTM. However, even without the rigour of a transport model, the observations show that, as of 2019, IPD has not recovered to the levels seen in the late-1980s (see electronic supplementary material, figure S5), suggesting there have not been large increases in Arctic CH 4 emissions from clathrates and permafrost carbon in response to increasing temperature.
(f) Other tracers: δ 13 C CH4 We get additional information about processes responsible for emitting CH 4 from other tracers. For example, anomalies in CO abundance measured in the same air samples measured for CH 4 were the right magnitude to be consistent with a decreased atmospheric OH sink after the eruption of Mt. Pinatubo. CO is also used as a tracer for biomass burning, where its molar emission ratio with CH 4 is significantly larger than for the sink process. Ethane (C 2 H 6 ) has been used as a tracer for fossil fuel emissions of CH 4 . Both are also emitted from other sources, which can complicate their interpretation. For example, Lan et al. [22] analysed long-term observations of CH 4 and a suite of NMHCs from samples collected at a site significantly influenced by oil and gas production. Because emission ratios of C 2 H 6 /CH 4 from fossil fuel production vary with time, predicting trends in CH 4 enhancements from trends in C 2 H 6 (and other NMHCs) enhancements and an assumed constant C 2 H 6 /CH 4 emission ratio results in significant over-estimates of those trends.
A much better tracer than CO or C 2 H 6 to identify processes affecting atmospheric CH 4 is the isotopic composition of CH 4 , which is emitted and removed by exactly the same processes as CH 4 itself, though with isotope signatures that vary by source, and fractionation factors that are unique to various sink pathways. Here, we focus on δ 13 C CH4 , which can be used to partition the relative fraction of emissions from broad categories: microbial (mic), fossil (fe; including natural fossil emissions)) and biomass burning (bb). Globally averaged atmospheric δ 13 C CH4 is equal to the average δ 13 C CH4 of sources mass-weighted by their emissions (equation (2.4)) and corrected for fractionation by sinks [23] (equation (2.5)). δ 13 C Q · Q Total = δ 13 C mic · Q mic + δ 13 C fe · Q fe + δ 13 C bb · Q bb , (2.4) and, at steady-state, where the subscript 'Q' refers to sources. An isotopic fractionation factor, α, is defined as the ratio of rate coefficients (k 13 C/k 12 C) for the reaction of 13 CH 4 relative to that for 12 CH 4 for a specific sink (e.g. reaction with OH). It is related to ε by ε = (α − 1) In equation (2.5), ε is weighted by the relative contributions of each sink process and commonly reported in per mille (‰). Interpretation of the isotope data requires knowledge of the mass-weighted isotopic composition of the sources that emit CH 4 to the atmosphere. An isotope source signature database compiled by Sherwood et al. [24] greatly increased the sample size and reduced the uncertainty of source signatures for a global CH 4 isotope mass balance. Within each source category, a range of δ 13 C CH4 values has been observed as a result of systematic spatial differences in these source signatures. These differences help constrain the spatial distribution of CH 4 emissions. This database was complemented by other studies on the spatial patterns in δ 13 C CH4 of sources ( [25] for oil, natural gas, coal, biomass burning and ruminants; [26] for wetlands; [27] for geological seeps; and [28] for ruminants), which can help to better leverage the spatial information in observed δ 13 C CH4 . . Adjustment for fractionation by sinks, as with emissions, is weighted by the magnitude of each sink process. Since atmospheric observations of CH 4 isotopes are sensitive to CH 4 sources and sinks with their distinct fingerprints, they are an extremely powerful constraint on the CH 4 budget (see the following section for the δ 13 C CH4 constraints on global CH 4 budget). In figure 2, CH 4 abundance (2a) and δ 13 C CH4 (2b) are plotted. The start of the decline in δ 13 C CH4 is nearly coincident with the increase in atmospheric CH 4 budget. Unfortunately, INSTAAR measurements of δ 13 C CH4 started after the large El Niño of 1997/1998, when there were large biomass burning signals, which δ 13 C CH4 should be very sensitive to. But based on analysis of measurements of CH 4 abundance and δ 13 C CH4 at Ny Ålesund, Spitzbergen, Morimoto et al. [29] found significant enhancements of CH 4 emissions from both wetlands and biomass burning in 1998, highlighting that the interpretation of the observations can be complicated by multiple processes changing nearly simultaneously. From measurements of air extracted from ice cores, we know that δ 13 C CH4 was increasing for approximately 200 yr [30] before the recent decline began. This suggests a rather significant change in the global budget of CH 4 sources and sinks that will be explored below, although the timing of the change in emissions is difficult to match directly with the δ 13 C CH4 observations because of the slow response of δ 13 C CH4 to changes in CH 4 emissions or average source signatures [31].

The global CH 4 budget, past and present
Based on measurements of CH 4 abundance from ice cores and firn, and assuming its lifetime was similar to today (approx. 9 yr), pre-industrial emissions were approximately 220 Tg CH 4 yr −1 . The dominant source then was emissions from wetlands, but there is substantial uncertainty with regard to emissions from natural fossil sources, which impacts estimates of anthropogenic fossil emissions in today's budget. δ 13 C CH4 can be used in an isotope mass balance to assess the fraction of total CH 4 emissions from fossil sources [32], but it cannot distinguish between natural seeps and anthropogenic emissions from current fossil fuel exploitation.
A potential tracer for that is 14 C CH4 in pre-industrial air extracted from ancient ice, because all 14 C has decayed in fossil CH 4 (lifetime approximately 8000 yr). While the number of observations of 14 C CH4 in air extracted from ice cores remains very small (in large part because of the enormous amount of ice needed for sufficient air for a measurement), two studies now indicate that emissions from fossil sources prior to industrialization were small. Petrenko et al. [33] found emissions were less than approximately 15.4 Tg CH 4 yr −1 based on ice with air age from 11.6 kyr ago. In a study representative of the atmosphere just prior to industrialization, Hmiel et al. [34] reported a fossil component of less than 5.4 Tg CH 4 yr −1 . These conclusions were based on very few 14 C CH4 measurements, but if we assume natural fossil emissions have not changed since pre-industrial, then there is a large discrepancy between this direct quantification and aggregating assessments that suggest much greater natural geological fossil emissions with a range of 43-50 Tg CH 4 yr −1 [27]. This large, important discrepancy will only be resolved with far more measurements of 14 C CH4 from ancient pre-industrial ice complemented by systematic measurement of fluxes from seeps.
The main sources of CH 4 to the atmosphere today are known, and its budget of emissions and sinks has been summarized periodically by the Global Carbon Project, most recently by Saunois et al. [35]. While these budget assessments use in situ observations and satellite retrievals of CH 4 , they do not apply the constraints imposed by measurements of CH 4 isotopic composition.
Briefly, from comparison with a pre-industrial atmospheric CH 4 mass balance assuming the same lifetime as today, we know approximately two-thirds of current CH 4 emissions are anthropogenic and come from agriculture (ruminants and their manure, and rice agriculture), waste management (landfills and waste treatment), fossil fuel use (coal, oil and natural gas) and biomass burning (including biofuels). The remaining emissions are from natural processes, predominantly tropical and high-northern latitude wetlands. There remains a great disparity between bottom-up and top-down estimates of natural emissions [35]. If natural fossil emissions are indeed small, as Hmiel et al. [34] infer, then to balance the isotopic budget, anthropogenic fossil fuel emissions are correspondingly larger.
The largest atmospheric loss process in the global CH 4 budget is mostly initiated by reaction with OH, especially in the tropical mid-troposphere [36], but also by Cl and O( 1 D) (stratosphere only). Oxidation by microbes in soils is likely a small sink, but uncertainty in its magnitude and trend remain large [37].
Atmospheric observations of spatial gradients and their changes with time are used in atmospheric CTM to infer emissions from specific regions. In figure 3, contours of CH 4 zonally averaged growth rate determined from NOAA's observations are plotted; warm colours show where the combined effects of source/sink imbalance and transport result in greater than zero growth rate and cool colours where it is less than zero. Data assimilation systems (also known as inverse modelling) start with an initial distribution of atmospheric CH 4 and a first-guess distribution and magnitude of emissions. This first-guess is used to calculate atmospheric CH 4 at the next model time step after winds and sinks have acted. Based on a comparison of the model and measured CH 4 , the emissions are optimized to get the best fit to the data. North-south gradients in growth rate, to first order, are qualitative indicators of where emissions have likely changed, but to be quantitative, the use of a CTM is absolutely necessary, except perhaps when global totals are estimated. This method has a limitation; it depends on the a priori first-guess for emissions. Optimized emissions can only be varied within the uncertainty limits specified for the first-guess, and they are also confined to the prior's spatial distribution. If that first-guess is grossly biased, the assimilation system cannot correct it enough, so emissions from another source are adjusted, perhaps erroneously, to match the observations. Improvements to bottom-up inventories and process models that supply priors would improve top-down assessments of the CH 4 budget.
The renewed long-term increase in atmospheric CH 4 and near-coincident decrease in δ 13 C CH4 in 2007 suggests a significant change in the global CH 4 budget with potentially important implications for climate. Many scenarios have been proposed to explain the increase; Lan et al. [25] investigated some of them to determine which are consistent with observations of atmospheric CH 4 abundance and δ 13 C CH4 . Most scenarios look at a single process or combination of competing processes acting from 2007 to present, but the reality is likely more complicated than that.
Dlugokencky et al. [38] attributed the initial rise to increased CH 4 emissions from tropical wetlands responding to increased precipitation in tropical land regions during an intense La Niña phase of ENSO in 2007 and 2008. But this change was not sufficient to sustain the increase; other factors, some perhaps that had acted over the previous decade must have contributed.
An isotope mass balance for CH 4 requires that emissions from fossil sources (natural seeps and exploitation of fossil fuels) are a larger fraction of total emissions than captured in emission  inventories [25,32,39], although with no significant trend [32]. Further studies using δ 13 C CH4 as a constraint on emission changes after 2006 are not consistent with increased fossil emissions dominating the increase in atmospheric CH 4 burden [25,39]. A large increase in U.S. shale gas production contributing to the increase in global CH 4 burden is not consistent with continental observations of CH 4 abundance [22] nor with the heavy isotopic signature of U.S. shale gas [40]. Another possible explanation of rising methane is that [OH] has had a sustained decline, thereby increasing τ CH4 . But this hypothesis is also inconsistent with the decade-long decreasing trend in δ 13 C CH4 . Lan et al. [25] used a three-dimensional atmospheric transport model to investigate a scenario where the OH component of τ CH4 was adjusted to match the increase in CH 4 burden. The impact on δ 13 C CH4 through 2016 is shown in figure 2b (green). As the proportion of oxidation by non-OH sinks becomes a larger fraction of the total sink, fractionation due to loss increases and the net effect is an increase in δ 13 C CH4, opposite to what is observed. More generally, there is little evidence for a profoundly decreasing trend in [OH] (see e.g. [41]), but [OH], particularly its trend, remains a large uncertainty in quantifying the global CH 4 budget and, especially, how it is changing.
Lan et al. [25] tested competing hypotheses explaining the post-2006 global CH 4 increase in a common three-dimensional tracer model framework. The scenarios most consistent with both CH 4 and δ 13 C CH4 observations and most plausible in regard to other evidence involve increased emissions from microbial sources (agriculture and wetlands) in the tropics with a small contribution from fossil sources [39,[42][43][44], with a small decrease in biomass burning to maintain isotope mass balance.

Uncertainties that limit our understanding of the global CH 4 budget
There are many processes for which our ability to assess IAV and trends is limited, and this imposes uncertainty in assessing atmospheric CH 4 's global budget. Here, we highlight some of the larger uncertainties in our understanding.

(a) Sink processes
Oxidation of atmospheric CH 4 by OH is the largest term in the global CH 4 budget, and small trends, much less than 1% yr −1 , are important. We cannot accurately evaluate magnitudes and trends in global total CH 4  its trend have been assessed indirectly through observations of 1,1,1-trichloroethane (CH 3 CCl 3 ; common name: methyl chloroform, which we abbreviate to MC), an anthropogenic tracer predominantly removed from the atmosphere by reaction with OH. To be effective, the method required accurate estimates of MC emissions. When MC production was phased out by the Montreal Protocol on Substances that Deplete the Ozone Layer, it allowed Montzka et al. [45] to exploit the greatly decreased emissions of MC to estimate [OH] averaged over multiple years based on the rate of decay in the atmospheric burden of MC. However, as the abundance of MC in the atmosphere has decreased closer to its analytical detection limit, this approach has become more sensitive to small remaining emissions. Moreover, while this approach was useful for assessing the lifetimes of CH 4 and MC averaged over some years and their IAV, it was less suitable to assess small trends in [OH].
Recent reports of a decreasing trend in [OH] driving the renewed growth in atmospheric CH 4 since 2007 may be artefacts of the box-modelling methods used ( [46] and references therein) and are inconsistent with models of atmospheric chemistry and transport [41]. This scenario is also inconsistent with trends in δ 13 C CH4 as discussed above. While the reaction rate coefficient for OH + CH 4 , including its temperature dependence, has been greatly studied [47], uncertainty in the fractionation of the reaction on atmospheric δ 13 C CH4 is still substantial, and that uncertainty affects our ability to quantitatively partition emissions among broad source categories. Saueressig et al. [48] reported a significantly smaller KIE for fractionation of 13 C by OH (1.0039 ± 0.0004) than reported by Cantrell et al. [49] (1.0054 ± 0.0009). Uncertainty in this factor muddles interpretation of measurements of atmospheric δ 13 C CH4 imposing a 20% uncertainty on emissions from the fossil fuel sector (approx. 30 Tg CH 4 yr −1 ) [25].
Oxidation of CH 4 initiated by Cl has little impact on atmospheric CH 4 total loss, but strongly fractionates δ 13 C CH4 . Using a CTM (TOMCAT) and sources of Cl from natural and anthropogenic halocarbons and dechlorination of sea-salt aerosol, Hossaini et al. [50] estimated a sink for atmospheric CH 4 by reaction with Cl of 12-13 Tg CH 4 yr −1 . Based on analysis of δ 13 C CO measurements in the extra-tropical Southern Hemisphere, Gromov et al. [51] concluded that the loss of CH 4 by reaction with Cl is no greater than 1% of the total atmospheric CH 4 sink (less than half the estimate of Hossaini et al. [50]), and even 1% makes it difficult to balance the 13 C(CO) budget. Wang et al. [52] calculated a similar size sink for CH 4 from reaction with Cl, 5.3 Tg CH 4 yr −1 . The impact of Cl on the interpretation of δ 13 C CH4 observations is still clear; Strode et al. [53] compared multiple tropospheric Cl spatial and seasonal distributions and found the choice impacts the north-south gradient and seasonal cycle of δ 13 C CH4 . And Lan et al. [25] found that the magnitude of the Cl sink chosen (none or 13 Tg CH 4 yr −1 ) impacts the amount of fossil emissions that can be consistent with observations by approximately 20%.
Oxidation of CH 4 in soils is intermediate in magnitude, much smaller than loss initiated by reaction with OH and likely much larger than reaction with Cl. Saunois et al. [35] reported a bottom-up range for soil loss of 12 to 49 Tg CH 4 yr −1 based on a literature review. Fractionation of CH 4 carbon isotopes by soil loss is reported to be approximately −20‰, with a range of measured values of −16 to 27‰ (e.g. [54]). Since the sink mostly occurs in forested soils, which are susceptible to land use, the magnitude [55] and distribution of the sink may be changing. Given that the fluxes of soil loss are very small and occur over large areas of forested and Savannah soils, better estimates of global soil loss will remain challenging.
(b) Wetlands CH 4 emissions from wetlands are the largest natural source of atmospheric CH 4 . Emissions from wetlands vary greatly over small spatial scales and in time; this variability, combined with relatively few flux measurements, makes up-scaling uncertain. Prior estimates of emissions, therefore, rely on process models, but the range of emissions estimated among different wetland process models is quite large, approximately 40% [56]. One limitation is the accurate assessment of time-varying wetland area and inundation extent. While space-based methods seem ideal to determine wetland area, these methods are limited by cloud cover [57]. New methods based on the Cyclone Global Navigation Satellite System (CYGNSS), which operate in a frequency region that allows them to see through clouds, rain and most vegetation, may provide an important improvement [58]. Wetlands remain a limitation for top-down studies of the entire CH 4 budget, because estimates of emissions from all other sources are impacted by inaccurate estimates of the magnitude and distribution of CH 4 emissions from wetlands.
Using the isotopic constraint effectively requires that the source signatures of different wetland ecosystems are accounted for. Recently, Ganesan et al. [26] developed a map of spatially varying δ 13 C CH4 source signatures for wetlands, which vary by approximately 10‰ between highnorthern latitudes and the tropics. The δ 13 C CH4 of CH 4 emitted from a wetland will depend on factors such as the mechanism of CH 4 production (CO 2 versus acetate reduction), the source of the carbon (the proportion of plants with C3 versus C4 photosynthetic pathways), and oxidation between production and emission to the atmosphere, which depends on the transport route. Further efforts to identify the underlying processes that control wetland isotopic signature can improve the spatial mapping and thus improve the accuracy of attributing wetland emissions spatially.

Conclusion
Changes to the global CH 4 budget are recorded in the atmosphere in CH 4 abundance and δ 13 C CH4 spatial and temporal gradients. High-quality, calibrated observations of atmospheric CH 4 and its stable isotopic composition are crucial for decreasing uncertainty in quantifying the evolving global CH 4 budget, yet the world is still greatly under-sampled. If we want to track future changes in the global CH 4 budget, including potential climate feedbacks, we need to expand these observations to under-sampled regions, such as the tropics, Siberia, Africa and South America. Most studies using CH 4 isotope composition as a constraint use stable carbon isotopes, but more extensive observations of stable isotopes of hydrogen would further constrain the budget [39], especially for constraining the OH sink, which has a large fractionation factor for δ 2 H CH4 . However, the use of δ 2 H CH4 would require both an extended time series from an adequate global network of air sampling sites and detailed work to define source signatures by type of source and geography.
Explaining the renewed and accelerating increase in atmospheric CH 4 burden since 2007 remains challenging, and the exact causes are not yet clear. But, the observations we describe suggest that increased emissions from microbial sources are the strongest driver, with a relatively smaller contribution from other processes, e.g. fossil fuel exploitation. A more difficult question to answer is the one posed by this special issue: is warming feeding the warming? We cannot say for certain, but we cannot rule out the possibility that climate change is increasing CH 4 emissions. The strong signals from the tropics combined with the isotopic data are consistent with increased emissions from natural wetlands, but large IAV and inter-decadal variability in wetland drivers like precipitation make it difficult to identify small trends. Observations are needed that will help process models capture this variability. The size of the IAV illustrates the potential scope of uncontrollable near-future change and emphasizes the urgency of reducing the global methane burden by mitigating the methane emissions that we can control, from the fossil fuel and agricultural sectors.
Authors' contributions. All authors contributed to writing the manuscript. E.J.D. and S.E.M. provided data. Competing interests. The authors declare that they have no competing interests. Funding. The sampling and measurements used were supported by NOAA's Climate Program Office. Acknowledgements. This work would not have been possible without the dedicated assistance from participants in NOAA's Cooperative Global Air Sampling Network and the skilled contributions of NOAA Carbon Cycle Group members over the past four decades. This work was supported in part by the U.S. National Aeronautics