Three decades of global methane sources and sinks

4 levels 20,21 . Here we construct a global CH 4 budget for the past three decades by combining bottom-up and top-down estimates of CH 4 sources and the chemical CH 4 sink (Box 1). We use chemical transport models — constrained by atmospheric CH 4 measurements — to estimate CH 4 fluxes using top-down atmospheric inversions. We compare these and using a combination of atmospheric measurements and results from chemical transport models, ecosystem models, climate chemistry models and inventories of anthropogenic emissions. The resultant budgets suggest that data-driven approaches and ecosystem models overestimate total natural emissions. We build three contrasting emission scenarios — which differ in fossil fuel and microbial emissions — to explain the decadal variability in atmospheric methane levels detected, here and in previous studies, since 1985. Although uncertainties in emission trends do not allow definitive conclusions to be drawn, we show that the observed of methane can potentially be explained by decreasing-to-stable fossil fuel emissions, combined with stable-to-increasing microbial emissions. We show that a rise in natural wetland emissions and fossil fuel emissions probably accounts for the renewed increase in global methane levels after 2006, although the relative contribution of these two sources remains uncertain.


Sources and sinks
The global atmospheric CH 4 budget is determined by many terrestrial and aquatic surface sources, balanced primarily by one sink in the atmosphere. CH 4 emissions can be broadly grouped into three categories: biogenic, thermogenic and pyrogenic. Biogenic sources contain CH 4 -generating microbes (methanogens) 17 , and comprise anaerobic environments such as natural wetlands and rice paddies, oxygen-poor freshwater reservoirs (such as dams), digestive systems of ruminants and termites, and organic waste deposits (such as manure, sewage and landfills). Thermogenic CH 4 , formed over millions of years through geological processes, is a fossil fuel. It is vented from the subsurface into the atmosphere through natural features (such as terrestrial seeps, marine seeps and mud volcanoes), and through the exploitation of fossil fuels, that is, through the exploitation of coal, oil and natural gas. Pyrogenic CH 4 is produced by the incomplete combustion of biomass and soil carbon during wildfires, and of biofuels and fossil fuels. These three types of emissions have different isotopic δ 13 C signatures (δ 13 C = [( 13 C/ 12 C) sample /( 13 C/ 12 C) standard ] − 1) × 1000): −55 to −70‰ for biogenic emissions, −25 to −55‰ for thermogenic emissions, and −13 to −25‰ for pyrogenic emissions 20,22,23 . The isotopic composition of atmospheric CH 4 -measured at a subset of surface stations -has therefore been used to constrain its source [20][21][22][23][24] . CH 4 emissions by living plants under aerobic conditions do not seem to play a significant role in the global CH 4 budget (Supplementary Section ST8); some very large 25 estimates of this source published in 2006 have not been confirmed 26 .
The primary sink for atmospheric CH 4 is oxidation by hydroxyl radicals (OH), mostly in the troposphere, which accounts for around 90% of the global CH 4 sink. Additional oxidation sinks include methanotrophic bacteria in aerated soils 27,28 (~4%), reactions with *A full list of authors and their affiliations appears at the end of the paper. chlorine radicals and atomic oxygen radicals in the stratosphere 17 (~3%), and reactions with chlorine radicals from sea salt in the marine boundary layer 29 (~3%).

Global decadal budget
We combine state-of-the-art top-down and bottom-up approaches (Box 1) using a consistent methodology (see Methods) to assess global CH 4 sources and sinks over the past three decades. At the global scale for the 2000s, top-down inversions yield total global emissions of 548 Tg of CH 4 per year with a minimum-maximum range of 526-569 (six models in Table 1) and a global sink of 540 [514-560] Tg CH 4 yr -1 . The source-sink mismatch reflects the observed average imbalance of 6 Tg CH 4 yr -1 of the CH 4 growth rate in the 2000s, which is smaller than that of the 1980s and 1990s (34 Tg CH 4 yr -1 and 17 Tg CH 4 yr -1 , respectively; Fig. 1). In fact, stabilization of atmospheric CH 4 prevailed in the early 2000s, and the atmospheric increase resumed after 2006.
Summing up all bottom-up emission estimates, a different picture emerges for the global source for the 2000s. We obtain a value of 678 Tg CH 4 yr -1 , which is 20% larger than the inversion-based estimate (P<0.01; Table 1). The higher global source in bottom-up estimates is explained by a larger sum of natural emissions (from wetlands, freshwater, and geological sources) than in the inversions (Table 1). For the 2000s, the bottom-up estimate of the total sink is 632 Tg CH 4 yr -1 , with a large range (592-785). Most of this sink -604 Tg CH 4 yr -1 -is due to the hydroxyl radical CH 4 sink, as estimated Atmospheric CH 4 mole fraction (ppb) by the nine bottom-up chemistry climate models (CCMs) 30 . The OH sink simulated by the seven models that run time slices from the 1980s to the 2000s is found to increase with time, which contrasts with the stability of the OH sink inferred from top-down inversions for the 1990s and the 2000s ( Table 1). The positive trend in the OH sink in the CCMs can be explained by the fact that the chemical consumption of OH, for instance through reactions with CH 4 and carbon monoxide, is offset by the production of OH through photochemical reactions, involving water vapour, nitrogen oxides and stratospheric ozone. The stable OH sink inferred from top-down inversions relates to the observed atmospheric record of methyl chloroform, which is used to infer OH changes on decadal scales 30 . We group decadal estimates of emissions (top-down and bottomup) into five categories: natural wetlands; other natural emissions (termites, geological, fresh water systems, permafrost and hydrates); agriculture and waste; fossil fuels; and biomass and biofuel burning (Table 1). Freshwater systems include lakes, reservoirs, streams and rivers. In the 2000s, natural wetland emissions (top-down, 142-208 Tg CH 4 yr -1 ; and bottom-up, 177-284 Tg CH 4 yr -1 ) and agriculture and waste emissions (top-down, 180-241 Tg CH 4 yr -1 ; and bottom-up, 187-224 Tg CH 4 yr -1 ) dominate CH 4 emissions, followed by anthropogenic fossil fuel emissions, other natural emissions and emissions from biomass and biofuel burning (Table 1). Together with natural CH 4 emissions from lake and freshwater sources 31,32 , we find an imbalance of almost 50 Tg CH 4 yr -1 (in the 2000s) between the mean global emission and the mean global sink in the Categories are split into: natural wetlands, biomass burning, fossil fuels, agriculture and waste, other sources (see Table 1), soil uptake and chemical loss by OH oxidation. Error bars spread between minimum and maximum values.
bottom-up approach, which is larger than the observed growth rate of around 6 Tg CH 4 yr -1 .
This discrepancy, combined with the fact that the global mean emission is 130 Tg CH 4 yr -1 greater in the bottom-up approach than in the top-down approach ( Table 1), suggests that CH 4 emissions are overestimated in the bottom-up approach. Indeed, the bottom-up global emission estimate is obtained by adding up independently estimated flux components, and thus lacks a constraint on its global magnitude. In contrast, the global CH 4 emission derived from the top-down approach is constrained at the global scale by the atmospheric CH 4 growth rate, using atmospheric CH 4 measurements, and by the magnitude of the chemical sink, using proxy atmospheric observations, such as the concentration of methyl chloroform, to estimate OH concentrations. Such proxy methods have proven to be reliable indicators of mean OH levels in the troposphere, although their ability to capture OH changes has been widely discussed 33,34 . These proxy methods suggest that the mean global chemical sink for CH 4 derived from bottom-up estimates may also be overestimated, especially in the 2000s (Table 1).
When summing up anthropogenic fossil emissions, natural fossil CH 4 from onshore and offshore seeps 35,36 (part of geological emissions in Table 1) and hydrates, bottom-up total fossil emissions account for 28% (~156 Tg CH 4 yr -1 ) of the global CH 4 source between 1985 and 2000. This is consistent with an analysis of 14 C-CH 4 atmospheric measurements 37 in both hemispheres inferring a 30 ± 2% fossil fraction in the global CH 4 source. However, fossil emissions of this magnitude are not confirmed by a recent analysis of the global atmospheric record of ethane 15 , which is co-emitted with geological CH 4 . Top-down inversions cannot provide useful information to settle this debate, as they generally do not separate this source from other natural emissions ( Table 1). Consideration of the natural fossil CH 4 source, neglected in previous Intergovernmental Panel on Climate Change (IPCC) assessments, thus represents a significant update to the global CH 4 budget, although it is still debated.

Global budget uncertainty
Uncertainties associated with decadal CH 4 budgets are expressed by the minimum-maximum range between different decadal estimates, due to the small number of studies available for calculating reliable standard deviations (Table 1). For the 2000s, the uncertainty range for bottom-up estimates -defined as (max−min)/meanis 50% for natural wetlands and typically 100% for other natural sources, though the other individual natural sources have smaller fluxes than wetlands. Anthropogenic sources seem to be known more precisely, with an uncertainty range of 30% for agriculture/ waste-and fossil-fuel-related emissions, and 20% for biomass burning. The uncertainty range of the global sink is 40%, but drops to 20% when removing one outlier with very high total OH loss in a recent comparison of climate chemistry models 30,38 . Note that the uncertainties reported in Table 1 are correlated to some extent. Because of more recent and robust estimates for each decade, each term in the budget has a smaller error range than in the IPCC AR4 report: 50% smaller for wetlands, 60% smaller for biomass burning, and 40% smaller for agriculture and waste emissions ( Table 1).
Natural wetlands have the largest absolute uncertainty of any of the emission categories, with a min-max range of 107 Tg CH 4 yr -1 in the bottom-up approach (177−284 Tg CH 4 yr -1 ). This large range is confirmed by a recent multi-model analysis 39 showing a ±40% range of wetland emissions around an average of 190 Tg CH 4 yr -1 . In the three wetland emission models used here [40][41][42] , emissions were calculated for each grid point as the product of a flux rate and a wetland area, both having uncertainties. Uncertainties in wetland extent seem to be the dominant source of discrepancy in modelled CH 4 emissions 39,43 .
The OH sink seems to have a smaller error range using proxy methods in the top-down approach (max-min range of 30 Tg CH 4 ) than in bottom-up CCMs (max-min range of 250 Tg CH 4 , dropping to 110 Tg CH 4 when removing one outlier model from the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) 30,38 ), in which different humidity and temperature fields cause a large spread of the OH sink 38 .
Following IPCC AR5 guidelines for the treatment of uncertainties 44 , we defined a level of confidence for both top-down estimates and bottom-up estimates, based on robustness (number of published studies) and agreement (difference between maximum and minimum estimates, relative to the mean). Many studies have focused on constraining the CH 4 budget during the 1990s and 2000s, but fewer estimates are available for the 1980s. As a result, The top-down approach is based on atmospheric inversion models, which determine 'optimal' surface fluxes 92,93 that best fit atmospheric CH 4 observations given an atmospheric transport model including chemistry, prior estimates of fluxes, and their uncertainties. Global atmospheric inversions provide a timevarying distribution of CH 4 fluxes, albeit with limited insight into the underlying processes when different sources overlap in the same region. This is, for example, often the case for agricultural, waste and fossil emissions in densely populated areas of east Asia, Europe and North America. We collected results from nine inversion systems (Supplementary Table S1).
The bottom-up approach includes process-based models estimating CH 4 emissions, and CCMs estimating the OH sink. Eight bottom-up models for wetland and fire CH 4 emissions are parameterized with empirical knowledge of local processes and driven by global data sets of climate, or satellite-observed burned area, to simulate CH 4 fluxes on spatial and temporal scales relevant for regional and global budgets (Supplementary Section II). Bottom-up emission inventories 56,81,82 based on energy use, agricultural activity, and emission factors from different sectors provide yearly or decadal mean estimates of anthropogenic wasterelated, rice, livestock, biofuel, and fossil fuel emissions, usually at national scales. Three inventories for anthropogenic emissions are used, updated to 2008 (Supplementary Information).
The photochemical sink of CH 4 is large and difficult to quantify, given the very short lifetime of OH (~1 sec) and its control by a myriad of precursor species. Direct measurements of atmospheric OH radicals do not have the required accuracy and coverage to derive global OH concentrations and consequently the magnitude of the CH 4 sink. We estimated CH 4 loss due to OH from the output of nine numerical CCMs 65 , which are categorized here as an atmospheric bottom-up approach. The OH concentration as calculated by CCMs can be further adjusted, at a large scale, by inversions based on measurements of tracers with known emissions and whose dominant sink is oxidation by OH, such as methyl chloroform 34,49,85,94 or chloromethanes 33,34 .
Combining top-down and bottom-up approaches allows us to investigate the consistency of each term of the CH 4 budget 21 . In this comparison, it should be noted that bottom-up models and inventories are not independent from inversions, because they are usually used in inversions to prescribe a prior spatial, and sometimes temporal, distribution of the emissions and sinks. However, inversions use independent atmospheric observations to partially correct the prior values.

NATURE GEOSCIENCE DOI: 10.1038/NGEO1955
estimates for all source categories during the 2000s are more robust, especially for inversions (Fig. 2). Agreement among studies is high (difference is less than 33%) for agriculture and waste (top-down and bottom-up), biomass burning and fossil fuels (bottom-up) and OH loss (top-down), whereas agreement is only medium (33−66% difference) for natural wetlands (top-down and bottom-up), fossil fuel emissions (top-down) and OH sink (bottom-up) estimates. Low agreement (> 66% difference) is found for biomass burning (top-down) and other natural sources (bottom-up). Increasing the number of studies does not necessarily lead to enhanced agreement. This can be seen for the fossil fuel and other sources categories, partly because of poorly constrained models, and partly because the results from a single new study can produce a large increase in the spread of emission estimates when very few studies are available.
No source or sink category reaches the highest level of confidence (highest agreement and highest robustness), emphasizing the large uncertainties that remain in our understanding of CH 4 emissions. Overall, higher confidence in global emissions is found for agriculture and waste (top-down) than for fossil fuels, the OH sink, natural wetlands and other natural sources. Top-down and bottom-up estimates are listed separately for the different categories in Fig. 1. For top-down inversions, the 1980s decade starts in 1984. Numbers in square brackets represent minimum and maximum values. A balance with the atmospheric annual increase and the sum of the sources has been assumed for inversions not reporting their global sink. Stratospheric loss for bottom-up is the sum of the loss by radicals, a 10 Tg yr -1 loss due to O( 1 D) radicals 22 and a 20-35% contribution due to Cl radicals 29 . Ranges of total chemical loss are about half the reported ranges (for example, [509-619] for the 2000s) when removing one outlier.

Regional decadal budget
The geographical breakdown of emissions per category and per region reveals major CH 4 emission zones worldwide and the level of consistency between top-down and bottom-up approaches ( Fig. 3 and Supplementary Section ST2 and Tables S2 and S3). Anthropogenic emissions dominate in Europe, North America, China, and the fossil-fuel-producing countries of eastern Europe and central Asia, with good agreement between top-down and bottom-up approaches (Fig. 3). Emission ranges are given in Table S2. Densely populated regions usually emit fossil, agricultural and waste CH 4 , making these sources difficult to separate in top-down inversions. Noteworthy is the large range of estimates for anthropogenic fossil CH 4 emissions from China in the top-down approach, possibly due to the low density of atmospheric CH 4 measurements in this region, and to biases in inventories 45 Table S2). Tropical South America shows the largest regional discrepancy between top-down When aggregated over large regions, emissions from biomass burning are the largest in Africa (top-down, 9 Tg CH 4 yr -1 ; bottom-up, 8 Tg CH 4 yr -1 ) and in tropical South America (top-down, 5 Tg CH 4 yr -1 ; bottom-up, 4 Tg CH 4 yr -1 ), but play only a minor role in temperate and boreal regional budgets. The bottom-up estimates are likely to be conservative compared to top-down estimates, as small fires are often undetected by satellite retrieval algorithms 47 . For biomass burning, simulated emission areas are consistent between models for 38 ± 9% of global emissions over the period 1997-2000, revealing robust large emission zones around the thermal equator in Africa (for example, Central African Republic, Democratic Republic of the Congo, Republic of the Congo, Angola, Zambia and Cameroon), central South America (Brazil and Bolivia), Indonesia, and to a lesser extent in eastern Russia, Laos, and Mexico ( Supplementary Fig. S0). Emission zones in northern Australia and in boreal regions (Canada and Siberia) can also be clearly identified.
Other natural sources, including termites, lakes and other fresh waters, and onshore geological emissions show maximum values in Africa and tropical South America, due to the relatively strong contribution of emissions by termites 48 Table 1. Circle size depicts the robustness of the estimate (number of studies). Circle colour illustrates the level of agreement among studies (min-max ranges): green, high confidence; yellow, medium confidence; red (with black dot), low confidence. Circles are grey when only one study has been used. A large green circle, for example, indicates a very good level of confidence 44 . termite source, contributing 30% and 36%, respectively, of the total (Supplementary Section ST7). Finally, CH 4 loss due to OH radicals is largest in the tropical atmosphere, both over land and oceans, as the tropics are the major region of OH production 49 .

Attribution of temporal changes
Year-to-year variations of CH 4 fluxes have been intensively studied 4,14,21,47,50 . The present study confirms the findings from previous ones showing that, over the last three decades, variations in wetland emissions have dominated the year-to-year variability in surface emissions ( Supplementary Fig. S5). Interannual variability in wetland emissions surpasses that of biomass burning emissions, except during intensive fire periods 21,50 . Analyses of anomalies in CH 4  The observed decadal changes remain much more enigmatic than yearly anomalies ( Supplementary Fig. S5). We use a scenario approach, built from our synthesis and from recent publications, to investigate these changes, and the contribution of the different CH 4 sources to them (see Methods). We assume that decadal changes in global mean CH 4 emissions since 1985 are well represented by the mean of those five atmospheric inversions covering the past three decades 53 , averaged on a five-year basis ( Fig. 4

and Methods).
A global mass balance model 54

1985-2005.
The S 0 storyline clearly overestimates global emissions after 1990, which calls for corrections to the magnitude of one or several sources in the S 0 scenario (Fig. 4). Using ethane firn air and atmospheric measurements, two recent studies indicated that CH 4 emissions from the fossil fuel sector decreased between 1985 and 2000 at a rate of −0.4 to −0.8 Tg CH 4 yr -1 , and attributed such a decline to decreasing fugitive emissions (leaks during extraction, treatment and use of fossil fuels) from oil and gas industries 15,57 . One of these studies further extended the ethane record up to 2010 15 , with either a slower decline or a stabilization of fossil fuel emissions    15 , and keeping the other sources as in S 0 , leads to a first plausible scenario that is consistent with the observation-driven global emissions (S 1 in Fig. 4). An alternative scenario (Sʹ 1 ), using bottom-up ecosystem model results for wetland emissions as a storyline instead of top-down inversions, is also consistent with the observation-driven global emissions. Two different analyses of δ 13 C-CH 4 isotopic composition trends 58,59 for 1990-2005 reached contradictory conclusions. In one, constant fossil fuel emissions but decreasing microbial emissions in the Northern Hemisphere were inferred 58 , the latter mainly attributed to decreasing rice emissions. In the other 59 , fossil fuel and microbial emissions remained constant. Assuming constant fossil fuel emissions during 1985-2005 and decreasing microbial emissions 58 produces a second scenario that is mostly consistent with observation-driven global emissions when using wetland fluxes from top-down inversions (S 2 in Fig. 4), but not when using wetland fluxes from bottom-up ecosystem models (Sʹ 2 ). Assuming decreasing fossil fuel emissions before 1990 (as in S 1 ), but constant fossil fuel and microbial emissions between 1990 and 2005 59 , produces a third scenario that is consistent with observation-driven global emissions, with either top-down or bottom-up wetland emission estimates (S 3 and Sʹ 3 in Fig. 4).
Overall, the three plausible scenarios, among many other possible source compositions matching global decadal changes, suggest that a decrease in fossil fuel CH 4 emissions is a more likely explanation for the stability of global CH 4 emissions between 1990 and 2005 than a reduction in microbial CH 4 emissions. An actual decrease in rice paddy emissions may have been surpassed by an increase in other microbial emissions (natural wetlands, animals, landfills and waste) as found by ecosystem models combined with the EDGAR4.2 inventory. Considering the significant uncertainties reported in a recent isotope study 59 for the 1990-2005 period, decreasing-to-stable fossil fuel emissions, combined with stable-to-increasing total microbial emissions, would reconcile the atmospheric ethane trends with the 13 C-CH 4 trends, at least for one 13 C-CH 4 data set 59 . Finally, trends in the magnitude of the OH CH 4 sink, which remain uncertain over decadal timescales, can still modulate these incomplete conclusions 34 .  -year basis averages). This is a 30% overestimation compared with the mean increase derived from the observations (17-22 Tg CH 4 yr -1 , see above). Thus, either the increase in fossil fuel emissions is overestimated by inventories, or the sensitivity of wetland emissions to precipitation and temperature is too large in some wetland emission models 39 . The contribution of microbial versus fossil emissions to this increase remains largely uncertain; respective contributions vary from 20 to 80%, if accounting for all additional top-down inversions available for the 2000s (Supplementary Fig. S5 and Table 1).

Shortcomings and uncertainty reductions
Our analyses suggest four main shortcomings in the assessment of regional to global CH 4 budgets. First, decadal means and interannual changes in CH 4 emissions from natural wetlands and freshwater systems are too uncertain. It is critically important to improve wetland mapping, both by refining land surface models (for example, through improving estimates of tropical flood plains in hydrological models, specific model developments for peatlands, and the integration of freshwater systems) and by further developing remotely sensed inundation data sets 61 (for instance for dense tropical forests). The scarcity of wetland CH 4 flux measurements and data sets limits the ability to validate large-scale modelled CH 4 emissions for natural wetlands and fresh waters 43 . The extension of the CO 2 FLUXNET measurements and database 62 to CH 4 fluxes is probably achievable at a reasonable cost, and would provide useful constraints for land surface models. For interannual variations in wetland emissions, the sensitivity of emission rates to warming at Scenarios : high northern latitudes and to rainfall changes in the tropics needs to be more consistently quantified in wetland models. The Amazon drought in 2010 63 should have resulted in a drop in wetland CH 4 emissions, and ongoing analyses may allow researchers to test the hypothesis that tropical wetland CH 4 emissions respond strongly to rainfall anomalies and trends. Second, the partitioning of CH 4 emissions by region and process is not sufficiently constrained by atmospheric observations in topdown models. Regional partitioning of total emissions would benefit from denser and more evenly distributed CH 4 concentration data. This can be achieved by further developing synergies between high precision monitoring of the surface and the lower atmosphere, including poorly sampled key areas such as the Amazon Basin, Siberia and tropical Africa on one hand, and retrievals of global-scale CH 4 columns by satellites and by high precision remote sensing from the ground on the other. Including continuous measurements of the δ 13 C stable isotope ( 13 CH 4 ) at surface stations would help separate biogenic emissions from other sources. Measurements of the δD stable isotope (CH 3 D) would provide constraints on the uncertain OH CH 4 sink, which can also be constrained by new proxy tracers 33,34 . Radiocarbon CH 4 data ( 14 CH 4 ) would help constrain the uncertain fossil part of CH 4 emissions, if 14 CH 4 emissions from nuclear installations can be accurately estimated 37 . Estimating long-term trends of fluxes and concentrations requires equally long-term observations, which in turn require stable and coordinated networks 64 .

REVIEW ARTICLE
Third, decadal trends in natural and anthropogenic emissions are still very uncertain and limit our ability to definitively attribute changes in emissions from specific sources to observed atmospheric changes since the 1990s. In addition to the (already noted) improvements in land surface models required, inventories for anthropogenic emissions should systematically include an uncertainty assessment, and should improve their representation of emission trends (for instance by more frequently updating the time-dependent factors used in their calculations).
Fourth, uncertainties in the modelling of atmospheric transport and chemistry limit the optimal assimilation of atmospheric observations by increasing uncertainties in top-down inversions. Such uncertainties are also only partly estimated in current inversions. We therefore recommend the continuation of ongoing international model inter-comparisons, which can provide a quantification of transport and chemistry errors to be included in top-down inversions 65,66 .

From challenge to opportunity
Our decadal CH 4 budgets reveal that bottom-up models may overestimate total natural CH 4 emissions. The various emission scenarios tested -designed to explain the temporal changes in atmospheric CH 4 levels observed in this and previous studies -suggest that the stabilization of atmospheric CH 4 in the early 2000s is likely to be due to a reduction in or stabilization of fossil fuel emissions, combined with a stabilization of or increase in microbial emissions. After 2006, the renewed global increase in atmospheric CH 4 is consistent with higher emissions from wetlands and fossil fuel burning, but the relative contributions remain uncertain.
In the context of climate change mitigation, atmospheric CH 4 poses both an opportunity and a challenge. The challenge lies in more accurately quantifying the CH 4 budget and its variations. Our synthesis suggests that improvements in models of natural wetland and freshwater emissions, the integration of surface networks monitoring CH 4 concentrations and fluxes (including isotopic composition) and new satellite missions (including active space-borne observations 67 ), improvements in anthropogenic emission trends in inventories, and uncertainty reductions in models of atmospheric transport and chemistry, could all help. The opportunity lies in the possibility of developing short-term climate change mitigation policies that take advantage of the relatively short atmospheric lifetime of CH 4 of about 10 years, and the known technological and agronomical options available for reducing emissions 68 .
The potential intensive exploitation of natural gas from shale formations around the world may lead to significant additional CH 4 release into the atmosphere 69 , although the potential magnitude of these emissions is still debated 70 . Such additional emissions, and combustion of this 'new' fossil fuel source, may offset mitigation efforts and accelerate climate change. In the longer term, the thawing of permafrost or hydrates could increase CH 4 emissions significantly, and introduce large positive feedbacks to long-term climate change 71 . A better quantification of the global CH 4 budget, with regular updates as done for carbon dioxide 72 , will be key to both embracing the opportunities and meeting the challenge.

Data analysis.
Top-down and bottom-up studies addressing the evolution of the CH 4 cycle after 1980 and covering at least five years of a decade were gathered. Therefore, the number and the nature of studies used in this work vary from one decade to another. Top-down inversions include atmospheric chemistry transport models and assimilation systems 19,46,53,[73][74][75][76][77] . Bottom-up approaches comprise modelling studies for wetland [40][41][42] and biomass-burning emissions 47,78-80 , emission inventories for anthropogenic 55,56,81 and natural sources 82 , and a suite of atmospheric chemistry models within the ACCMIP intercomparison project providing CH 4 chemical loss 30,39,83 .
The monthly fluxes (emissions and sinks) provided by the different groups were post-processed similarly. They were re-gridded on a common grid (1°× 1) and converted into the same units (Tg CH 4 per grid cell); then monthly, annual and decadal means were computed for 12 regions based on the TransCom 84 intercomparison map, with subdivisions in high-emission regions. Regional and global means were used to construct Figs 1, 2 and 3, Supplementary Figs S2 and S3, Table 1 and Supplementary Tables S2 and S3. The reported ranges and error bars represent the minimum and maximum values obtained among the different studies (Figs 1, 3 and 4 and Table 1). The small number of studies for some categories makes it difficult to properly apply a standard deviation.
Interannual variability (IAV) was computed as the difference between the 12-month running mean and the long-term mean. However a consistent period for estimating the long-term mean was not compatible with all data sources ( Supplementary Fig. S5).

Observation-driven global CH 4 emissions.
For 'attribution of temporal changes' , we used the only top-down study that estimates CH 4 emissions over the past 30 years 53 with five different set-ups. The mean of these five inversions was assumed to represent average global emissions. However these five inversions only partially represented the full range of global CH 4 emissions, due to differences in prior emission scenarios and errors, observations and their errors, OH fields and atmospheric transport representation. To estimate the full range of global CH 4 emissions we complemented the mean inversion with a sensitivity analysis based on a one-box model for the whole atmosphere 54 . The change in the global burden of CH 4 is given by: where [CH 4 ] is the global CH 4 burden, E is the sum of all emissions, and τ is the total atmospheric CH 4 lifetime. Equation (1) can be rearranged to calculate the annual CH 4 source strength E as follows: In this equation, the annual increase d [CH 4 ]/dt and the burden [CH 4 ] were given by the yearly-averaged growth rates and mole fractions of Fig. 1. Global CH 4 emissions were generated by computing emissions with equation (2) for each of the four networks and for a lifetime τ varying from 8 to 10 years to include uncertainties in OH changes 34,85 . Minimum and maximum values of E were extracted for five-year periods to produce the range of emissions plotted around the mean of atmospheric inversions (blue shaded area in Fig. 4, top panel).