Global Methane Budget 2000–2020

Abstract

Understanding and quantifying the global methane (CH₄) budget is important for assessing realistic pathways to mitigate climate change. CH₄ is the second most important human-influenced greenhouse gas in terms of climate forcing after carbon dioxide (CO₂), and both emissions and atmospheric concentrations of CH₄ have continued to increase since 2007 after a temporary pause. The relative importance of CH₄ emissions compared to those of CO₂ for temperature change is related to its shorter atmospheric lifetime, stronger radiative effect, and acceleration in atmospheric growth rate over the past decade, the causes of which are still debated. Two major challenges in quantifying the factors responsible for the observed atmospheric growth rate arise from diverse, geographically overlapping CH₄ sources and from the uncertain magnitude and temporal change in the destruction of CH₄ by short-lived and highly variable hydroxyl radicals (OH). To address these challenges, we have established a consortium of multidisciplinary scientists under the umbrella of the Global Carbon Project to improve, synthesise, and update the global CH₄ budget regularly and to stimulate new research on the methane cycle. Following Saunois et al. (2016, 2020), we present here the third version of the living review paper dedicated to the decadal CH₄ budget, integrating results of top-down CH₄ emission estimates (based on in situ and Greenhouse Gases Observing SATellite (GOSAT) atmospheric observations and an ensemble of atmospheric inverse-model results) and bottom-up estimates (based on process-based models for estimating land surface emissions and atmospheric chemistry, inventories of anthropogenic emissions, and data-driven extrapolations). We present a budget for the most recent 2010–2019 calendar decade (the latest period for which full data sets are available), for the previous decade of 2000–2009 and for the year 2020.

The revision of the bottom-up budget in this 2025 edition benefits from important progress in estimating inland freshwater emissions, with better counting of emissions from lakes and ponds, reservoirs, and streams and rivers. This budget also reduces double counting across freshwater and wetland emissions and, for the first time, includes an estimate of the potential double counting that may exist (average of 23 Tg CH₄ yr⁻¹). Bottom-up approaches show that the combined wetland and inland freshwater emissions average 248 [159–369] Tg CH₄ yr⁻¹ for the 2010–2019 decade. Natural fluxes are perturbed by human activities through climate, eutrophication, and land use. In this budget, we also estimate, for the first time, this anthropogenic component contributing to wetland and inland freshwater emissions. Newly available gridded products also allowed us to derive an almost complete latitudinal and regional budget based on bottom-up approaches.

For the 2010–2019 decade, global CH₄ emissions are estimated by atmospheric inversions (top-down) to be 575 Tg CH₄ yr⁻¹ (range 553–586, corresponding to the minimum and maximum estimates of the model ensemble). Of this amount, 369 Tg CH₄ yr⁻¹ or $\sim$ 65 % is attributed to direct anthropogenic sources in the fossil, agriculture, and waste and anthropogenic biomass burning (range 350–391 Tg CH₄ yr⁻¹ or 63 %–68 %). For the 2000–2009 period, the atmospheric inversions give a slightly lower total emission than for 2010–2019, by 32 Tg CH₄ yr⁻¹ (range 9–40). The 2020 emission rate is the highest of the period and reaches 608 Tg CH₄ yr⁻¹ (range 581–627), which is 12 % higher than the average emissions in the 2000s. Since 2012, global direct anthropogenic CH₄ emission trends have been tracking scenarios that assume no or minimal climate mitigation policies proposed by the Intergovernmental Panel on Climate Change (shared socio-economic pathways SSP5 and SSP3). Bottom-up methods suggest 16 % (94 Tg CH₄ yr⁻¹) larger global emissions (669 Tg CH₄ yr⁻¹, range 512–849) than top-down inversion methods for the 2010–2019 period. The discrepancy between the bottom-up and the top-down budgets has been greatly reduced compared to the previous differences (167 and 156 Tg CH₄ yr⁻¹ in Saunois et al. (2016, 2020) respectively), and for the first time uncertainties in bottom-up and top-down budgets overlap. Although differences have been reduced between inversions and bottom-up, the most important source of uncertainty in the global CH₄ budget is still attributable to natural emissions, especially those from wetlands and inland freshwaters.

The tropospheric loss of methane, as the main contributor to methane lifetime, has been estimated at 563 [510–663] Tg CH₄ yr⁻¹ based on chemistry–climate models. These values are slightly larger than for 2000–2009 due to the impact of the rise in atmospheric methane and remaining large uncertainty ($\sim$ 25 %). The total sink of CH₄ is estimated at 633 [507–796] Tg CH₄ yr⁻¹ by the bottom-up approaches and at 554 [550–567] Tg CH₄ yr⁻¹ by top-down approaches. However, most of the top-down models use the same OH distribution, which introduces less uncertainty to the global budget than is likely justified.

For 2010–2019, agriculture and waste contributed an estimated 228 [213–242] Tg CH₄ yr⁻¹ in the top-down budget and 211 [195–231] Tg CH₄ yr⁻¹ in the bottom-up budget. Fossil fuel emissions contributed 115 [100–124] Tg CH₄ yr⁻¹ in the top-down budget and 120 [117–125] Tg CH₄ yr⁻¹ in the bottom-up budget. Biomass and biofuel burning contributed 27 [26–27] Tg CH₄ yr⁻¹ in the top-down budget and 28 [21–39] Tg CH₄ yr⁻¹ in the bottom-up budget.

We identify five major priorities for improving the CH₄ budget: (i) producing a global, high-resolution map of water-saturated soils and inundated areas emitting CH₄ based on a robust classification of different types of emitting ecosystems; (ii) further development of process-based models for inland-water emissions; (iii) intensification of CH₄ observations at local (e.g. FLUXNET-CH₄ measurements, urban-scale monitoring, satellite imagery with pointing capabilities) to regional scales (surface networks and global remote sensing measurements from satellites) to constrain both bottom-up models and atmospheric inversions; (iv) improvements of transport models and the representation of photochemical sinks in top-down inversions; and (v) integration of 3D variational inversion systems using isotopic and/or co-emitted species such as ethane as well as information in the bottom-up inventories on anthropogenic super-emitters detected by remote sensing (mainly oil and gas sector but also coal, agriculture, and landfills) to improve source partitioning.

The data presented here can be downloaded from 10.18160/GKQ9-2RHT (Martinez et al., 2024).

1 Introduction

The average surface dry air mole fraction of atmospheric methane (CH₄) reached 1912 ppb in 2022 (Fig. 1; Lan et al., 2024), 2.6 times greater than its estimated pre-industrial value in 1750. This increase is attributable in large part to increased anthropogenic emissions arising primarily from agriculture (e.g. livestock production, rice cultivation, biomass burning), fossil fuel production and use, waste disposal, and alterations to natural CH₄ fluxes due to increased atmospheric CO₂ concentrations, land use (Woodward et al., 2010, Fluet-Chouinard et al., 2023), and climate change (Ciais et al., 2013; Canadell et al., 2021). An equal mass of CH₄ emissions has a stronger impact on climate than carbon dioxide (CO₂), which is reflected by its global warming potential (GWP) relative to CO₂ on a given time horizon. For a 100-year time horizon the GWP of CH₄ emitted by fossil sources is 29.8 (GWP of CH₄ emitted by microbial sources is 27), whereas the values reach 82.5 over a 20-year horizon for CH₄ emitted by fossil sources and 79.7 for CH₄ emitted by microbial sources (Forster et al., 2021). Although global anthropogenic emissions of CH₄ are estimated at around 359 Tg CH₄ yr⁻¹ (Saunois et al., 2020), representing around 2.5 % of the global CO₂ anthropogenic emissions when converted to units of carbon mass flux for the recent decade, the emissions-based effective radiative forcing of CH₄ concentrations contributed $\sim$ 31 % (1.19 W m⁻²) to the additional radiative forcing from anthropogenic emissions of greenhouse gases and their precursors (3.84 W m⁻²) over the industrial era (1750–2019) (Forster et al., 2021). Changes in other chemical compounds such as nitrogen oxides (NO_x) or carbon monoxide (CO) also influence atmospheric CH₄ through changes to its atmospheric lifetime. Emissions of CH₄ contribute to the production of ozone, stratospheric water vapour, and CO₂, and most importantly they affect its own lifetime (Myhre et al., 2013; Shindell et al., 2012). CH₄ has a short lifetime in the atmosphere (about 9 years for the year 2010; Prather et al., 2012; Szopa et al., 2021). Hence a stabilisation or reduction of CH₄ emissions leads to the stabilisation or reduction of its atmospheric concentration (assuming no change in the chemical oxidants), and therefore its radiative forcing, in only a few decades. While reducing CO₂ emissions is necessary to stabilise long-term warming, reducing CH₄ emissions is recognised as an effective option to limit climate warming in the near-term future (Shindell et al., 2012; Jackson et al., 2020; Ocko et al., 2021; UNEP, 2021) because of its shorter lifetime compared to CO₂.

Figure 1

Globally averaged atmospheric CH₄ concentrations (ppb) (a) and annual growth rates $G_{\mathrm{ATM}}$ (ppb yr⁻¹) (b) between 1983 and 2022, from four measurement programmes: National Oceanic and Atmospheric Administration (NOAA), Advanced Global Atmospheric Gases Experiment (AGAGE), Commonwealth Scientific and Industrial Research Organisation (CSIRO), and University of California, Irvine (UCI). Detailed descriptions of methods are given in the supplementary material of Kirschke et al. (2013).

[Figure omitted. See PDF]

The momentum around the potential of CH₄ to limit near-term warming led to the launch of the Global Methane Pledge at the November 2021 Conference of the Parties (COP 26). Signed by 158 countries (update on October 2024), this collective effort aims at reducing global CH₄ anthropogenic emissions by at least 30 % from 2020 levels by 2030 (Global Methane Pledge, 2023). Given that global baseline CH₄ emissions are expected to grow through 2030 (by an additional 20–50 Mt of CH₄, UNEP, 2022), the CH₄ emission reductions currently needed to reach the Global Methane Pledge objective (UNEP, 2022) correspond to 36 % of the projected baseline emissions in 2030 (i.e. if no further emission reductions were implemented). This implies that large reductions of CH₄ emissions are needed to meet the Global Methane Pledge, which is also consistent with the 1.5–2 °C target of the Paris Agreement (UNEP, 2022). Moreover, because CH₄ is a precursor of important air pollutants such as ozone, CH₄ emissions reductions are required by two international conventions – the United Nations Framework Convention on Climate Change (UNFCCC) and the Convention on Long-Range Transport of Air Pollution (CLRTAP) – making this global CH₄ budget assessment all the more critical.

Changes in the magnitude and temporal variation (annual to interannual) in CH₄ sources and sinks over the past decades are characterised by large uncertainties (e.g. Kirschke et al., 2013; Saunois et al., 2017; Turner et al., 2019). Also, the decadal budget suggests relative uncertainties (hereafter reported as min–max ranges) of 20 %–35 % for inventories of anthropogenic emissions in specific sectors (e.g. agriculture, waste, fossil fuels; Tibrewal et al., 2024), 50 % for biomass burning and natural wetland emissions, and up to 100 % for other natural sources (e.g. inland waters, geological sources). The uncertainty in the chemical loss of CH₄ by OH, the predominant sink of atmospheric CH₄, has been estimated using Prather et al. (2012) and Rigby et al. (2017). The former study estimated this uncertainty at $\sim$ 10 % from the uncertainty in the reaction rate between CH₄ and OH, and the latter study was based on methyl-chloroform measurements. Bottom-up approaches (chemistry transport models) estimate the uncertainty of the chemical loss by OH at around 15 %–20 % (Saunois et al., 2016, 2020). This uncertainty in the OH-induced loss translates, in the top-down methods, into the minimum relative uncertainty associated with global CH₄ emissions, as other CH₄ sinks (atomic oxygen and chlorine oxidations, soil uptake) are much smaller and the atmospheric growth rate is well defined (Dlugokencky et al., 2009). Globally, the contribution of natural CH₄ emissions to total emissions can be quantified by combining lifetime estimates with reconstructed pre-industrial atmospheric CH₄ concentrations from ice cores (assuming natural emissions have not been perturbed during the Anthropocene) (e.g. Ehhalt et al., 2001). Regionally or nationally, uncertainties in emissions may reach 40 %–60 % (e.g. for South America, Africa, China, and India; see Saunois et al., 2016). Another difficulty of the CH₄ budget lies in the necessity to also match the isotopic signal and in particular reflect the decreasing methane isotopic signal ¹³C (Nisbet et al., 2016, 2019). The previous budgets were tested against the isotopic observations (Saunois et al., 2017) and follow an exhaustive assessment (Zhang et al., 2021b). To date only a couple of atmospheric inverse systems are able to assimilate both CH₄ mixing ratios and stable isotopic signal to retrieve fluxes at the global scale (Thanwerdas et al., 2024; Basu et al., 2022), but these systems still need improvements in terms of configuration set-up and computing time resources, in addition to characterisation of source signatures and chemical kinetic effect (Chandra et al., 2024). We hope to be able to report isotopic constrained budgets in the coming years or at least test the budget against the isotopic balance.

To monitor emission reductions, for example to help conduct the Paris Agreement's stocktake, sustained and long-term monitoring of anthropogenic emissions per sector is needed in particular for hotspots of emissions that may be missed in inventories (Bergamaschi et al., 2018; Pacala, 2010; Lauvaux et al., 2022). At the same time, reducing uncertainties in all individual CH₄ sources, and thus in the overall CH₄ budget, remains challenging for at least four reasons. First, CH₄ is emitted by multiple processes, including natural and anthropogenic sources, point and diffuse sources, and sources associated with at least three different production origins (i.e. microbial, thermogenic, and pyrogenic). These multiple sources and processes require the integration of data from diverse scientific communities and across multiple temporal and spatial scales. The production of accurate bottom-up estimates is complicated by the fact that anthropogenic emissions result from leakage from fossil fuel production with large differences between countries depending on technologies and practices, the fact that many large leak events are sporadic, and the location of many emissions hotspots is not well known, and from uncertain emission factors used to summarise complex microbial processes in the agriculture and waste sectors. For the latter, examples include difficulties in upscaling methane emissions from livestock without considering the variety of animal weight, diet, and environment and difficulties in assessing emissions from landfills depending on waste type and waste management technology. Second, atmospheric CH₄ is removed mainly by chemical reactions in the atmosphere involving OH and other radicals that have very short lifetimes (typically $\sim$ 1 s). Due to the short lifetime of OH, the spatial and temporal distributions of OH are highly variable. While OH can be measured locally, calculating global CH₄ loss through OH measurements requires high-resolution global OH measurements (typically half an hour to integrate cloud cover and 1 km spatially to consider OH high reactivity and heterogeneity), which is impossible from direct OH observations. As a result, OH can only be calculated through large-scale atmospheric chemistry modelling. Those simulated OH concentrations from transport–chemistry models prescribed with emissions of precursor species affecting OH still show uncertain spatiotemporal distribution from regional to global scales (Zhao et al., 2019). Third, only the net CH₄ budget (sources minus sinks) is well constrained by precise observations of atmospheric growth rates (Dlugokencky et al., 2009), leaving the sum of sources and the sum of sinks uncertain. One distinctive feature of CH₄ sources compared to CO₂ fluxes is that the oceanic contribution to the global CH₄ budget is small ($\sim$ 1 %–3 %), making CH₄ source estimation predominantly a terrestrial endeavour (USEPA, 2010b). Finally, we lack comprehensive observations to constrain (1) the areal extent of different types of wetlands and inland freshwater (Kleinen et al., 2012, 2020, 2021, 2023; Stocker et al., 2014; Zhang et al., 2021a), (2) models of wetland and inland freshwater emission rates (Melton et al., 2013; Poulter et al., 2017; Wania et al., 2013; Bastviken et al., 2011; Wik et al., 2016a; Rosentreter et al., 2021; Bansal et al., 2023; Lauerwald et al., 2023a; Stanley et al., 2023), (3) inventories of anthropogenic emissions (Höglund-Isaksson et al., 2020; Crippa et al., 2023; USEPA, 2019), and (4) atmospheric inversions, which aim to estimate CH₄ emissions from global to regional scales (Houweling et al., 2017; Jacob et al., 2022).

The global CH₄ budget inferred from atmospheric observations by atmospheric inversions relies on regional constraints from atmospheric sampling networks, which are relatively dense for northern mid-latitudes, with various high-precision and high-accuracy surface stations, but are sparser at tropical latitudes and in the Southern Hemisphere (Dlugokencky et al., 2011). Recently, the density of atmospheric observations has increased in the tropics due to satellite-based platforms that provide column-averaged CH₄ mixing ratios. Despite continuous improvements in the precision and accuracy of space-based measurements (e.g. Buchwitz et al., 2016), systematic errors greater than several parts per billion on total column observations can still limit the usage of such data to constrain surface emissions (e.g. Jacob et al., 2022). The development of robust bias corrections on existing data can help overcome this issue (e.g. Inoue et al., 2016, Lorente et al., 2023; Balasus et al., 2023), and satellite data are now widely used in atmospheric inversions where they provide more global information on the distribution of fluxes and highly complement the surface networks (e.g. Lu et al., 2021).

In this context, the Global Carbon Project (GCP) seeks to develop a complete picture of the carbon cycle by establishing common, consistent scientific knowledge to support policy development and actions to mitigate greenhouse gas emissions to the atmosphere (https://www.globalcarbonproject.org, last access: 1 April 2025). The objectives of this paper are (1) to analyse and synthesise the current knowledge of the global CH₄ budget, (2) to better understand and quantify the main robust features of this budget and its remaining uncertainties, and (3) to make recommendations for improvement. We combine results from a large ensemble of bottom-up approaches (e.g. process-based models for natural wetlands, data-driven approaches for other natural sources, inventories of anthropogenic emissions and biomass burning, and atmospheric chemistry models) and top-down approaches (including CH₄ atmospheric observing networks, atmospheric inversions inferring emissions and sinks from the assimilation of atmospheric observations into models of atmospheric transport and chemistry). The focus of this work is to update the previous assessment made for the period 2000–2017 (Saunois et al., 2020) to the more recent 2000–2020 period. More in-depth analyses of trends and year-to-year changes are left to future publications. Our current paper is a living review, published at about 4-year intervals, to provide an update and new synthesis of available observational, statistical, and model data for the overall CH₄ budget and its individual components.

Kirschke et al. (2013) carried out the first CH₄ budget synthesis, followed by Saunois et al. (2016) and Saunois et al. (2020), with companion papers by Stavert et al. (2021) on regional CH₄ budgets and Jackson et al. (2020) focusing on the last year of the budget (2017). Saunois et al. (2020) covered 2000–2017 and reported CH₄ emissions and sinks for three time periods: (1) the latest calendar decade at that time (2000–2009), (2) data for the latest available decade (2008–2017), and (3) the latest available year (2017) at the time. Here, the Global Methane Budget (GMB) covers 2000–2020 split into the 2000–2009 decade, the 2010–2019 decade (where data are available), the year 2020 affected by COVID-induced changes in human activity, and briefly for 2021–2023 as per data availability (Sect. 6). The CH₄ budget is presented at global, latitudinal, and regional scales, and data can be downloaded from 10.18160/GKQ9-2RHT (Martinez et al., 2024). A global, regional, and sectoral assessment of methane emission changes over the last 2 decades is discussed in Jackson et al. (2024) based on the data of Martinez et al. (2024).

Six sections follow this introduction. Section 2 presents the methodology used in the budget – units, definitions of source categories, regions, and data analysis – and discusses the delay between the period of study of the budget and the release date. Section 3 presents the current knowledge about CH₄ sources and sinks based on the ensemble of bottom-up approaches reported here (models, inventories, data-driven approaches). Section 4 reports atmospheric observations and top-down atmospheric inversions gathered for this paper. Section 5, based on Sects. 3 and 4, provides the updated analysis of the global CH₄ budget by comparing bottom-up and top-down estimates and highlighting differences. Section 6 discusses the recent changes in atmospheric CH₄ in relation to changes in CH₄ sources and sinks. Finally, Sect. 7 discusses future developments, missing components, and the most critical remaining uncertainties based on our update to the global CH₄ budget. For easier reading, the list of content of this article is presented in the first section (Sect. S1) of the Supplement.

2 Methodology

2.1 Units used

Unless specified, fluxes are expressed in teragrams of CH₄ per year (1 Tg CH₄ yr${}^{-\mathrm{1}}={\mathrm{10}}^{\mathrm{12}}$ g CH₄ yr⁻¹), while atmospheric mixing ratios are expressed as dry air mole fractions, in parts per billion (ppb), with atmospheric CH₄ annual increases, $G_{\mathrm{ATM}}$ (expressed in ppb yr⁻¹). In the tables, we present mean values and ranges for the two decades 2000–2009 and 2010–2019, together with results for the most recent available year (2020). Results obtained from previous syntheses (i.e. Saunois et al., 2020, and Saunois et al., 2016) are also given for the decade 2000–2009. Following Saunois et al. (2016) and considering that the number of studies is often relatively small for many individual source and sink estimates, uncertainties are reported as minimum and maximum values of the available studies, given in brackets. In doing so, we acknowledge that we do not consider the uncertainty of the individual estimates, and we express uncertainty as the range of available mean estimates, i.e. differences across measurements and methodologies considered. These minimum and maximum values are those presented in Sect. 2.5 and exclude identified outliers.

The CH₄ emission estimates are provided with up to three significant digits, for consistency across all budget flux components and to ensure the accuracy of aggregated fluxes. Nonetheless, given the values of the uncertainties in the CH₄ budget, we encourage the reader to consider no more than two digits as significant for the global total budget.

2.2 Period of the budget and availability of data

The bottom-up estimates rely on global anthropogenic emission inventories, an ensemble of process-based models for wetland emissions, and published estimates in the literature for other natural sources. The global gridded anthropogenic inventories (see Sect. 3.1.1) are updated irregularly, generally every 3 to 5 years. The last reported years of available inventories were 2018 or 2019 when we started the top-down modelling activity. In order to cover the period 2000–2020, it was necessary to extrapolate the anthropogenic inventory EDGARv6 (Crippa et al., 2021) to 2020 to use it as prior information for the anthropogenic emissions in the atmospheric inversion systems as explained in the Supplement (Sect. S4). However, EDGARv7 (EDGAR, 2022; Crippa et al., 2023) spanning until 2021 was then released and was used for the bottom-up budget. EDGARv8 (EDGAR, 2023; Crippa et al., 2023), spanning until 2022 and released in 2024, is used in Sect. 6 to discuss the post-2020 methane budget. The land surface (wetland) models were run over the full period 2000–2020 using dynamical wetland areas, derived by remote sensing data or other models of flooded area variability (Sect. 3.2.1).

The atmospheric inversions run until mid-2021, but the last year of reported inversion results is 2020, which represents a 3-year lag with the present. This is due to the long time period it takes to acquire atmospheric in situ data and integrate models. Even though satellite observations are processed operationally and are generally available with a latency of days to weeks, by contrast surface observations can lag from months to years because of the time for flask analyses and data quality checks in (mostly) non-operational chains. In addition, the final 6 months of inversions must be generally ignored because the estimated fluxes are not constrained by as many observations as the previous periods. Lastly, this budget presents an extended synthesis of the most recent development regarding inland water emissions (Sect. 3.2.2) and corrections associated with double counting with wetlands.

2.3 Definition of regions

Geographically, emissions are reported globally and for three latitudinal bands (90° S–30° N, 30–60° N, 60–90° N, only for gridded products). When extrapolating emission estimates forward in time (see Sect. 3.1.1), and for the regional budget presented by Stavert et al. (2021), a set of 19 regions (oceans and 18 continental regions; see Fig. S3 in the Supplement) were used. As anthropogenic emissions are often reported by country, we define these regions based on a country list (Table S1 in the Supplement). This approach was compatible with all top-down and bottom-up approaches considered. The number of regions was chosen to be close to the widely used Transcom intercomparison map (Gurney et al., 2004) but with subdivisions to separate the contributions from important countries or regions for the CH₄ cycle (China, South Asia, tropical America, tropical Africa, the USA, and Russia). The resulting region definition is the same as that used for the Global Carbon Project (GCP) N₂O budget (Tian et al., 2020). Compared to Saunois et al. (2020), the Oceania region has been replaced by Australasia, including only Australia and New Zealand. Other territories formerly in Oceania were included in Southeast Asia.

2.4 Definition of source and sink categories

CH₄ is emitted by different processes (i.e. biogenic, thermogenic, or pyrogenic) and can be of anthropogenic or natural origin. Biogenic CH₄ is the final product of the decomposition of organic matter by methanogenic Archaea in anaerobic environments, such as water-saturated soils, swamps, rice paddies, marine and freshwater sediments, landfills, sewage and wastewater treatment facilities, or inside animal digestive systems. Thermogenic methane is formed on geological timescales by the breakdown of buried organic matter due to heat and pressure deep in the Earth's crust. Thermogenic CH₄ reaches the atmosphere through marine and land geological gas seeps. These CH₄ emissions are increased by human activities, for instance, the exploitation and distribution of fossil fuels. Pyrogenic CH₄ is produced by the incomplete combustion of biomass and other organic materials. Peat fires, biomass burning in deforested or degraded areas, wildfires, and biofuel burning are the largest sources of pyrogenic CH₄. CH₄ hydrates, ice-like cages of frozen CH₄ found in continental shelves and slopes and below subsea and land permafrost, can be of either biogenic or thermogenic origin. Each of these three process categories has both anthropogenic and natural components.

In the following, we present the different CH₄ sources depending on their anthropogenic or natural origin, which is relevant to climate policy. Compared to the previous budgets, marginal changes have been made regarding source categories (naming and grouping), to reflect the improved estimates for inland water sources and their indirect anthropogenic component. In the previous Global Methane Budget articles (Saunois et al., 2016, 2020), natural and anthropogenic emissions were split in a way that did not correspond exactly to the definition used by the UNFCCC following the IPCC guidelines (IPCC, 2006), where, for pragmatic reasons, all emissions from managed land are typically reported as anthropogenic. For instance, we considered all wetlands to be natural emissions, despite some wetlands being on managed land and their emissions being partly reported as anthropogenic in UNFCCC national communications. Separating natural from anthropogenic sources could be quite challenging, especially over regions where sources overlap, such as over heavily human-dominated floodplain deltas for example. The human-induced perturbation of climate, atmospheric CO₂, and nitrogen and sulfur deposition may also cause changes in wetland sources we classified as natural. Following our previous definition, emissions from wetlands, inland freshwaters, thawing permafrost, or geological leaks are accountable for “natural” emissions, even though we acknowledge that climate change and other human perturbations (e.g. eutrophication) may cause changes in those emissions. CH₄ emissions from reservoirs were also considered natural even though reservoirs are human-made. Indeed, since the 2019 refinement to the IPCC guidelines (IPCC, 2019), emissions from reservoirs and other flooded lands have been considered to be anthropogenic by the UNFCCC and should be reported as such. However, these estimates are not provided by inventories and not systematically reported by all countries (especially non-Annex-I countries). In this budget we rename “natural sources” to “natural and indirect anthropogenic sources” to acknowledge that CH₄ emissions from reservoirs, as well as from water bodies that were perturbed by agricultural activities (drainage, eutrophication, land use change), are indirect anthropogenic emissions. As a result, here, “natural and indirect anthropogenic sources” refer to “emissions that do not directly originate from fossil, agricultural, waste, and biomass burning sources” even if they are perturbed by anthropogenic activities and climate change. Natural and indirect anthropogenic emissions are split between “wetlands and inland freshwaters” and “other natural” emissions (e.g. wild animals, termites, land geological sources, oceanic geological and biogenic sources, and terrestrial permafrost). “Direct anthropogenic sources” are caused by direct human activities since pre-industrial/pre-agricultural times (3000–2000 BCE; Nakazawa et al., 1993) including agriculture, waste management, fossil-fuel-related activities, and biofuel and biomass burning (yet we acknowledge that a small fraction of wildfires are naturally ignited). Direct anthropogenic emissions are split between “agriculture and waste emissions”, “fossil fuel emissions”, and “biomass and biofuel burning emissions”, assuming that all types of fires are caused by anthropogenic activities. To conclude, this budget reports “direct anthropogenic” and “natural and indirect anthropogenic” methane emissions for the five main source categories explained above for both bottom-up and top-down approaches.

The sinks of methane are split into the soil uptake that can be derived from land surface models in the bottom-up budget and the chemical sinks. The chemical sinks are estimated by either chemistry climate or chemistry transport models in the bottom-up budget and are further detailed in terms of vertical distribution (troposphere and stratosphere) and oxidants.

Bottom-up estimates of CH₄ emissions for some processes are derived from process-oriented models (e.g. biogeochemical models for wetlands, models for termites), inventory models (agriculture and waste emissions, fossil fuel emissions, biomass and biofuel burning emissions), satellite-based models (large-scale biomass burning), or observation-based upscaling models for other sources (e.g. inland water, geological sources). From these bottom-up approaches, it is possible to provide estimates for more detailed source subcategories inside each main category described above (see budget in Table 3). However, the total CH₄ emission derived from the sum of independent bottom-up estimates remains unconstrained.

For atmospheric inversions (top-down approach), atmospheric methane concentration observations provide a constraint on the global methane total source if we assume the global sink is known (OH and other oxidant prescribed), or inversions also optimise the chemical sink. OH estimates are constrained by methyl chloroform inversion (Montzka et al., 2011; Rigby et al., 2017; Patra et al., 2021). The inversions reported in this work solve for the total net CH₄ flux at the surface (sum of sources minus soil uptake) (e.g. Pison et al., 2013) or a limited number of source categories (e.g. Bergamaschi et al., 2013). In most of the inverse systems the atmospheric oxidant concentrations were prescribed with pre-optimised or scaled OH fields, and thus the atmospheric sink is not optimised. The assimilation of CH₄ observations alone, as reported in this synthesis, can help to separate sources with different locations or temporal variations but cannot fully separate individual sources where they overlap in space and time in some regions. Top-down global and regional CH₄ emissions per source category were nevertheless obtained from gridded optimised fluxes, for the inversions that separated emissions into the five main GCP categories. Alternatively, for the inversion that only solved for total emissions (or for categories other than the five described above), the prior contribution of each source category at the spatial resolution of the inversion was scaled by the ratio of the total (or embedding category) optimised flux divided by the total (or embedding category) prior flux (Kirschke et al., 2013). In other words, the prior relative mix of sources at model resolution is kept in each grid cell while total emissions are given by the atmospheric inversions. The soil uptake was provided separately to report total gross surface emissions instead of net fluxes (sources minus soil uptake).

In summary, bottom-up models and inventory emissions are presented for all relevant source processes and grouped if needed into the five main categories defined above. Top-down inversion emissions are reported globally and for the five main emission categories.

2.5 Processing of emission maps and boxplot representation of emission budgets

Common data analysis procedures have been applied to the different bottom-up models, inventories, and atmospheric inversions whenever gridded products exist. Gridded emissions from atmospheric inversions and land surface models for wetland or biomass burning were provided at the monthly scale. Emissions from anthropogenic inventories are usually available as yearly estimates. These monthly or yearly fluxes were provided on a 1° $\times$ 1° grid or regridded to 1° $\times$ 1° and then converted into units of Tg CH₄ per grid cell. Inversions with a resolution coarser than 1° were downscaled to 1° by each modelling group. Land fluxes in coastal pixels were reallocated to the neighbouring land pixel according to our 1° land–sea mask and vice versa for ocean fluxes. Annual and decadal means used for this study were computed from the monthly or yearly gridded 1° $\times$ 1° maps.

Budgets are presented as boxplots with quartiles (25 %, median, 75 %), outliers, and minimum and maximum values without outliers. Outliers were determined as values below the first quartile minus 3 times the interquartile range or values above the third quartile plus 3 times the interquartile range. Mean values reported in the tables are represented as “$+$” symbols in the corresponding figures.

3 Methane sources and sinks: bottom-up estimates

For each source category, a short description of the relevant processes, original data sets (measurements, models), and related methodology is given. More detailed information can be found in original publication references, in Annex A2 where the sources of data used to estimate the different sources and sinks are summarised and compared with those used in Saunois et al. (2020), and in the Supplement of this study when specified in the text. The emission estimates for each source category are compared with Saunois et al. (2020) in Table 3 and with Saunois et al. (2016) in Table S12 for the decade 2000–2009.

3.1 Direct anthropogenic sources

3.1.1 Global inventories

The main bottom-up global inventory data sets covering direct anthropogenic emissions from all sectors (Table 1) are from the United States Environmental Protection Agency (USEPA, 2019), the Greenhouse gas and Air pollutant Interactions and Synergies (GAINS) model developed by the International Institute for Applied Systems Analysis (IIASA) (Höglund-Isaksson et al., 2020), and the Emissions Database for Global Atmospheric Research (EDGARv6 and v7; Crippa et al., 2021, 2023) compiled by the European Commission Joint Research Centre (EC-JRC) and Netherlands Environmental Assessment Agency (PBL). We also used the Community Emissions Data System for historical emissions (CEDS) (Hoesly et al., 2018) developed for climate modelling and the Food and Agriculture Organization (FAO) FAOSTAT emission database (Tubiello et al., 2022), which covers emissions from agriculture and land use (including peatland fires and biomass fires). These inventories are not independent as they may use the same activity data or emission factors, as discussed below.

Table 1

Bottom-up (BU) models and inventories for anthropogenic and biomass burning used in this study.

BU models and inventories	Contribution	Time period (resolution)	Gridded	References
CEDS (country-based)	fossil fuels, agriculture and waste, biofuel	1970–2019 (yearly)	no	Hoesly et al. (2018)
CEDS (gridded)a	fossil fuels, agriculture and waste, biofuel	1970–2020 (monthly)	$\mathrm{0.5}\times \mathrm{0.5}$°	Hoesly et al. (2018), O'Rourke et al. (2021)
EDGARv6	fossil fuels, agriculture and waste, biofuel	1990–2018b (yearly, monthly for some sectors)	$\mathrm{0.1}\times \mathrm{0.1}$°	Oreggioni et al. (2021), Crippa et al. (2021)
EDGARv7	fossil fuels, agriculture and waste, biofuel	1990–2021 (yearly)	$\mathrm{0.1}\times \mathrm{0.1}$°	Crippa et al. (2023)
IIASA GAINS v4.0	fossil fuels, agriculture and waste, biofuel	1990–2020 (yearly)	$\mathrm{0.5}\times \mathrm{0.5}$°	Höglund-Isaksson et al. (2020)
USEPA	fossil fuels, agriculture and waste, biofuel, biomass burning	1990–2030 (10-year interval, interpolated to yearly)	no	USEPA (2019)
FAO-CH4	agriculture, biomass burning	1961–2020 1990–2020 (yearly)	no	Federici et al. (2015), Tubiello et al. (2013), Tubiello (2019)
FINNv2.5	biomass burning	2002–2020 (daily)	1 km resolution	Wiedinmyer et al. (2023)
GFASv1.3	biomass burning	2003–2020 (daily)	$\mathrm{0.1}\times \mathrm{0.1}$°	Kaiser et al. (2012)
GFEDv4.1s	biomass burning	1997–2020 (monthly)	$\mathrm{0.25}\times \mathrm{0.25}$°	Giglio et al. (2013), van der Werf et al. (2017)
QFEDv2.5	biomass burning	2000–2020 (daily)	$\mathrm{0.1}\times \mathrm{0.1}$°	Darmenov and da Silva (2015)

^a Due to its limited sectoral breakdown, this data set is not used in Table 3. ^b Data set extends up to 2020, as stated in Sect. 2.2. and detailed in the Supplement (Sect. S4).

These inventory data sets report emissions from fossil fuel production, transmission, and distribution; livestock enteric fermentation; manure management and application; rice cultivation; and solid waste and wastewater. Since the level of detail provided by country and by sector varies among inventories, the data were reconciled into common categories according to Table S2. For example, agricultural waste-burning emissions treated as a separate category in EDGAR, GAINS, and FAO are included in the biofuel sector in the USEPA inventory and in the agricultural sector in CEDS. The GAINS, EDGAR, and FAO estimates of agricultural waste burning were excluded from this analysis (these amounted to 1–3 Tg CH₄ yr⁻¹ in recent decades) to prevent any potential overlap with separate estimates of biomass burning emissions (e.g. GFEDv4.1s; Giglio et al., 2013; van der Werf et al., 2017). In the inventories used here, emissions for a given region/country and a given sector are usually calculated following IPCC methodology (IPCC, 2006), as the product of an activity factor and its associated emission factor. An abatement coefficient may also be used, to account for any regulations implemented to control emissions (see e.g. Höglund-Isaksson et al., 2015). These data sets differ in their assumptions and data used for the calculation; however, they are not completely independent because they often use the same activity data, and some of them follow the same IPCC guidelines (IPCC, 2006). While the USEPA inventory adopts emissions reported by the countries to the UNFCCC, other inventories (FAOSTAT, EDGAR, and the GAINS model) produce their own estimates using a consistent approach for all countries, typically IPCC Tier 1 methods or deriving IPCC Tier 2 emission factors from country-specific information using a consistent methodology. These other inventories compile country-specific activity data and emission factor information or, if not available, adopt IPCC default factors (Tibrewal et al., 2024; Oreggioni et al., 2021; Höglund-Isaksson et al., 2020; Tubiello, 2019). CEDS takes a different approach (Hoesly et al., 2018) and combines data from GAINS, EDGAR, and FAO depending on the sector. Then their first estimates are scaled to match other individual or region-specific inventory values when available. This process maintains the spatial information in the default emission inventories while preserving consistency with country level data. The FAOSTAT data set (hereafter FAO-CH₄) provides estimates at the country level and is limited to agriculture (CH₄ emissions from enteric fermentation, manure management, rice cultivation, energy usage, burning of crop residues, and prescribed burning of savannahs) and land use (peatland fires and biomass burning). FAO-CH₄ uses activity data mainly from the FAOSTAT crop and livestock production database, as reported by countries to FAO (Tubiello et al., 2013), and applies mostly the Tier 1 IPCC methodology for emissions factors (IPCC, 2006), which depend on geographic location and development status of the country. For manure, the country-scale temperature was obtained from the FAO global agro-ecological zone database (GAEZv3.0, 2012). Although country emissions are reported annually to the UNFCCC by Annex-I countries, and episodically by non-Annex-I countries, data gaps of those national inventories do not allow the inclusion of these estimates in this analysis.

In this budget, we use the following versions of these inventories that were available at the start and during the analysis (see Table 1):

• EDGARv6 provides yearly gridded emissions by sector from 1970 to 2018 (Crippa et al., 2021; Oreggioni et al., 2021; EDGARv6 website https://edgar.jrc.ec.europa.eu/dataset_ghg60 (last access: 1 April 2025); Monforti Ferrario et al., 2021).

• EDGARv7 provides yearly gridded emissions by sector from 1970 to 2020 (monthly for some sectors), but emissions from fossil fuel energy are not separated (oil–gas and coal are lumped together – see Table S2) (EDGARv7 website: https://edgar.jrc.ec.europa.eu/dataset_ghg70 (last access: 1 April 2025); Crippa et al., 2023).

• GAINS model scenario version 4.0 (Höglund-Isaksson et al., 2020) provides an annual sectorial gridded product from 1990 to 2020 both by country and gridded. The USEPA (USEPA, 2019) provides 5-year sectorial totals by country from 1990 to 2020 (estimates from 2015 onward are a projection), with no gridded distribution available. The USEPA data set was linearly interpolated to provide yearly values from 1990–2020.

• CEDS version v_2021_04_21 provides gridded monthly and annual country-based emissions by sector from 1970 to 2019 (Hoesly et al., 2018; O'Rourke et al., 2021). Fossil fuel emissions for 2020 have been updated using the methodology described for CO in Zheng et al. (2023).

• FAO-CH₄ (database accessed in December 2022, FAO, 2022) contains annual country level data for the period 1961–2020, for rice, manure, and enteric fermentation and for 1990–2020 for burning savannah, crop residue, and non-agricultural biomass burning.

3.1.2 Total direct anthropogenic emissions

We calculated separately the total anthropogenic emissions for each inventory by adding its values for “agriculture and waste”, “fossil fuels”, and “biofuels” with additional large-scale biomass burning emissions data (Sect. 3.1.5). This method avoids double counting and ensures consistency within each inventory. This approach was used for the EDGARv6 and v7, CEDS, and GAINS inventories, but we kept the USEPA inventory as originally reported because it includes its own estimates of biomass burning emissions. FAO-CH₄ was only included in the range reported for the “agriculture and waste” category. For the latter, we calculated the range and mean value as the sum of the mean and range of the three anthropogenic subcategory estimates “enteric fermentation and manure”, “rice”, and “landfills and waste”. The values reported for the upper-level anthropogenic categories (“agriculture and waste”, “fossil fuels”, and “biomass burning and biofuels”) are therefore consistent with the sum of their subcategories, although there might be small percentage differences between the reported total anthropogenic emissions and the sum of the three upper-level categories. This approach provides a more accurate representation of the range of emission estimates, avoiding an artificial expansion of the uncertainty attributable to subtle differences in the definition of subsector categorisations between inventories.

Table 2

Biogeochemical models that computed wetland emissions used in this study. Model runs were performed with two climate inputs: CRU and GSWP3-W5E5. Models were run with prognostic (using their own calculation of wetland areas) and/or diagnostic (using WAD2M (Zhang et al., 2021b)) wetland surface areas (see Sect. 3.2.1).

Model	Institution	Prognostic		Diagnostic		References
		CRU	GSWP3-W5E5	CRU	GSWP3-W5E5
CH4MODwetland	Institute of Atmospheric Physics, CAS	n	n	y	y	Li et al. (2010)
CLASSIC	Environment and Climate Change Canada	y	y*	y	y*	Arora et al. (2018); Melton and Arora (2016)
DLEM	Boston College	y	y	y	y	Tian et al. (2015, 2023)
ELM-ECA	Lawrence Berkeley National Laboratory	y	y	y	y	Riley et al. (2011)
ISAM	University of Illinois, Urbana-Champaign	y	y	y	y	Shu et al. (2020) Xu et al. (2021)
JSBACH	MPI	y	y	y	y	Kleinen et al. (2020, 2021, 2023)
JULES	UKMO	y	y	y	y	Gedney et al. (2019)
LPJ-GUESS	Lund University	n	n	y	y	McGuire et al. (2012)
LPJ-MPI	MPI	y	y	y	y	Kleinen et al. (2012)
LPJ-WSL	NASA GSFC	y	y	y	y	Zhang et al. (2016b)
LPX-Bern	University of Bern	y	y	y	y	Spahni et al. (2011), Stocker et al. (2014)
ORCHIDEE	LSCE	y	y	y	y	Ringeval et al. (2011)
SDGVM	University of Birmingham/University of Sheffield	y	y	y	y	Beerling and Woodward (2001), Hopcroft et al. (2011, 2020)
TEM-MDM	Purdue University	n	n	y	y	Zhuang et al. (2004)
TRIPLEX-GHG	UQAM	n	n	y	y	Zhu et al. (2014, 2015)
VISIT	NIES	y	y	y	y	Ito and Inatomi (2012)

^* CLASSIC uses GSWP3-W5E version 2 that covers the time period till 2016. All other models use GSWP-W5E5 version 3.

Table 3

Global methane emissions by source type (in Tg CH₄ yr⁻¹) from Saunois et al. (2020) (left column pair) and from this work using bottom-up and top-down approaches. Because top-down models cannot fully separate individual processes, only five categories of emissions are provided (see text). Uncertainties are reported as [min–max] range of reported studies. The mean, minimum, and maximum values are calculated while discarding outliers, for each category of source and sink. As a result, discrepancies may occur when comparing the sum of categories and their corresponding total due to differences in outlier detections. Differences of 1 Tg CH₄ yr⁻¹ in the totals can also occur due to rounding errors. Compared to Saunois et al. (2020), emissions are split between “direct anthropogenic” emissions and “natural and indirect anthropogenic” sources. We also propose an estimate of the double counting between bottom-up wetland and inland freshwater ecosystem emissions. NA denotes “not available”.

	Saunois et al. (2020)		This work
Period of time	2000–2009		2000–2009		2010–2019		2020
Approaches	bottom-up	top-down	bottom-up	top-down	bottom-up	top-down	bottom-up	top-down
Natural and indirect anthropogenic sources
Combined wetlands and inland freshwaters	306 [229–391]	180 [153–196]	242 [156–355]	158 [145–172]	248 [159–369]	165 [145–214]	251 [171–364]	175 [151–229]
Wetlands    Inland freshwatersa    Double countingb	147 [102–179]  159 [117–212] NA	180 [153–196]	153 [116–189]g 112 [49–202] –23 [–9 to –36]	158 [145–172]	159 [119–203]g 112 [49–202] –23 [–9 to –36]	165 [145–214]	161 [131–198]g 112 [49–202] –23 [–9 to –36]	175 [151–229]
Other natural sources	63 [26–94]	35 [21–47]	63 [24–93]	44 [40–46]	63 [24–93]	43 [40–46]	63 [24–93]	44 [40–47]
Land sources   Geological (onshore)   Wild animals   Termites   Wildfires   Permafrost soils (direct)   Vegetation	50 [17–72] 38 [13–53] 2 [1–3] 9 [3–15] ** 1 [0–1] *		51 [18–73] 38 [13–53] 2 [1–3] 10 [4–16] ** 1 [0–1] *
Coastal and oceanic sourcesc   Biogenic   Geological (offshore)	13 [9–22] 6 [4–10] 7 [5–12]		12 [6–20] 5 [3–10] 7 [5–12]
Total natural and indirect sources	369 [245–485]	215 [176–243]	305 [180–448]	204 [189–223]	311 [183–462]	206 [188–225]	314 [195–457]	216 [193–241]
Direct anthropogenic sources
Agriculture and waste	192 [178–206]	202 [198–219]	194 [181–208]	210 [197–223]	211 [195–231]	228 [213–242]	211 [204–216]	245 [232–259]
Agriculture	132 [NA]		134 [125–142]		143 [132–155]		147 [143–149]
Livestock    and manure   Rice cultivation	104 [93–109]  28 [23–34]		104 [100–110]  30 [24–34]		112 [107–118]  32 [25–37]		117 [114–124]  32 [29–37]
Landfills and waste	60 [55–63]		61 [52–71]		69 [56–80]		71 [60–84]
Fossil fuels   Coal mining   Oil and gas   Industry   Transport	110 [94–129] 32 [24–42] 73 [60–85] 2 [0–6] 4 [1–11]	101 [71–151]	105 [97–123]h 30 [26–32] 65 [63–71] 4 [1–8] 3 [1–8]	105 [88–115]	120 [117–125]h 40 [37–44] 67 [57–74] 5 [1–9] 2 [1–3]	115 [100–124]	128 [120–133]h 41 [38–43] 74 [67–80] 5 [1–8] 2 [1–3]	122 [101–133]
Biomass and biofuel burning   Biomass burning   Biofuel burning	31 [26–46] 19 [15–32] 12 [9–14]	29 [23–35]	30 [22–44] 19 [14–29] 11 [8–14]	26 [22–29]	28 [21–39] 17 [12–24] 11 [8–14]	27 [26–27]	27 [20–41] 17 [13–27] 10 [7–14]	26 [22–27]
Total direct anthropogenicsources	334d [321–358]	332 [312–347]	333d [305–365]	341 [319–355]	358d [329–387]	369 [350–391]	372d [345–409]	392 [368–409]
Sinks
Total chemical loss   Tropospheric OH   Stratospheric loss   Tropospheric Cl	595 [489–749] 553 [476–677] 31 [12–37] 11 [1–35]	505 [459–516]	585 [481–716] 546 [446–663] 37 [ 27–51] 6 [1–13]	504e [496–511]	602 [496–747] 563 [462–663] 37 [28–43] 6 [1–13]	521e [485–532]	602 [496–747] 563 [462–663] 37 [28–43] 6 [1–13]	538e [503–554]
Soil uptake	30 [11–49]	34 [27–41]	30 [11–49]	34 [34–34]	31 [11–49]	35 [35–35]	31 [11–49]	36 [35–36]
Total sinks	625 [500–798]	540 [486–556]	615 [492–765]	538 [530–545]e	633 [507–796]	554 [520–567]e	633 [507–796]	575 [566–589]e
Source–sink imbalance
Total sources	703 [566–842]	547 [524–560]	638 [485–813]	543 [526–558]	669 [512–849]	575 [553–586]	685 [540–865]	608 [581–627]
Total sinks	625 [500–798]	540 [486–556]	615 [492–765]	538 [530–545]e	633 [507–796]	554 [550–567]e	633 [507–796]	575 [566–589]e
Imbalance	78	3 [$-$10 to 38]	23	5 [$-$4 to 13]e	36	21 [19–33]e	52	32 [15–38]e
Atmospheric growthf		5.8 [4.9–6.6]f		6.1 [5.2–6.9]f		20.9 [20.1–21.7]f		41.8 [40.7–42.9]f

^* Uncertain but likely small for upland forest and aerobic emissions, potentially large for forested wetland, but likely included elsewhere. ^** We stop reporting this value to avoid potential double counting with satellite-based products of biomass burning (see Sect. 3.1.5). ^a Freshwater includes lakes, ponds, reservoirs, streams, and rivers; part of it is due to anthropogenic disturbances estimated in Sect. 3.2.2. ^b The double counting estimate is discussed in Sect. 3.2.2. ^c Flux from hydrates considered at 0 for this study is included, including estuaries. ^d Total anthropogenic emissions are based on estimates of full anthropogenic inventory and not on the sum of “agriculture and waste”, “fossil fuels“ and “biofuel and biomass burning” categories (see Sect. 3.1.2). ^e Some inversions did not provide the chemical sink. These values are derived from a subset of the inversion ensemble. ^f Atmospheric growth rates are given in the same unit (Tg CH₄ yr⁻¹), based on the conversion factor of 2.75 Tg CH₄ ppb⁻¹ given by Prather et al. (2012) and the atmospheric growth rates provided in the text (in ppb yr⁻¹). ^g Here the numbers are from prognostic runs. To ensure a fair comparison with previous budgets (Saunois et al., 2020), the numbers are 163 [117–195] for 2000–2009 from diagnostic runs with CRU/CRU-JRA-55 climate inputs (see Sect. 3.2.1). ^h  Up to 8 Tg of additional emissions could account for ultra-emitters (Lauvaux et al., 2022), as in Tibrewal et al. (2024), that are fully or partly missed in regular anthropogenic inventories.

Based on the ensemble of databases detailed above, total direct anthropogenic emissions were 358 [329–387] Tg CH₄ yr⁻¹ for the decade 2010–2019 (Table 3, including biomass and biofuel burning) and 331 [305–365] Tg CH₄ yr⁻¹ for the decade 2000–2009. Our estimate for the 2000–2009 decade is within the range of Saunois et al. (2020) (334 [321–358]), Saunois et al. (2016) (338 Tg CH₄ yr⁻¹ [329–342]), and Kirschke et al. (2013) (331 Tg CH₄ yr⁻¹ [304–368]) for the same period. The slightly larger range reported herein with respect to previous estimates is due to the USEPA lower estimate for agriculture, waste, and fossil emissions associated with the lowest estimate of biomass burning.

Figure 2 (left) summarises or projects global CH₄ emissions of anthropogenic sources (including biomass and biofuel burning) by different data sets between 2000 and 2050. The data sets consistently estimate total anthropogenic emissions of $\sim$ 300 Tg CH₄ yr⁻¹ in 2000. For the Sixth Assessment Report of the IPCC, seven main Shared Socioeconomic Pathways (SSPs) were defined for future climate projections in the Coupled Model Intercomparison Project 6 (CMIP6) (Gidden et al., 2019; O'Neill et al., 2016) ranging from 1.9 to 8.5 W m⁻² radiative forcing by the year 2100 (as shown by the number in the SSP names). For the 1970–2015 period, historical emissions used in CMIP6 (Feng et al., 2019) combine anthropogenic emissions from CEDS (Hoesly et al., 2018) and a climatological value from the GFEDv4.1s biomass burning inventory (van Marle et al., 2017). The harmonised scenarios used for CMIP6 activities start in 2015 at 388 Tg CH₄ yr⁻¹, which corresponds to the higher range of our estimates. Since CH₄ emissions continue to track scenarios that assume no or minimal climate policies (SSP5 and SSP3), it may indicate that climate policies, when present, have not yet produced sufficient results to change the emissions trajectory substantially (Nisbet et al., 2019). After 2015, the SSPs span a range of possible outcomes, but current emissions appear likely to follow the higher-emission trajectories, given that over the past decade their trend has followed such trajectories and because the peak emission year has not yet been reached. High or medium emission reduction rates as suggested by scenarios SSP1 and SSP2 have not yet happened. This illustrates the challenge of methane mitigation that lies ahead to help reach the goals of the Paris Agreement (Nisbet et al., 2020; Shindell et al., 2024). In addition, estimates of methane atmospheric concentrations (Meinshausen et al., 2017, 2020) from the harmonised scenarios (Riahi et al., 2017) indicate that observations of global CH₄ concentrations fall well within the range of scenarios in absolute values, but their trend over the past few years is closest to those of scenario SSP5-8.5 (Fig. 2 right). The CH₄ concentrations are estimated using a simple exponential decay with inferred natural emissions (Meinshausen et al., 2011), and the emergence of any trend between observations and scenarios needs to be confirmed in the following years. However, the current observed concentrations and emissions estimates lie in the upper range of the former RCP scenarios starting in 2005 (Fig. S1). In the future, it will be important to monitor the trends from 2015 (the Paris Agreement) and from 2020 (Global Methane Pledge) estimated in inventories and from atmospheric observations and compare them to various scenarios.

Figure 2

(a) Global anthropogenic methane emissions (including biomass burning) over 2000–2050 from historical inventories (black line and shaded grey area) and future projections (coloured lines) (in Tg CH₄ yr⁻¹) from selected scenarios harmonised with historical emissions (CEDS) for CMIP6 activities (Gidden et al., 2019). Historical mean emissions correspond to the average of anthropogenic inventories listed in Table 1 added to the GFEDv4.1s (van der Werf et al., 2017) biomass burning historical emissions. (b) Global atmospheric methane concentrations for NOAA surface site observations (black) and projections based on SSPs (Riahi et al., 2017) with concentrations estimated using MAGICC (Meinshausen et al., 2017, 2020). Red dots show the last year available (2022 for observations).

[Figure omitted. See PDF]

3.1.3 Fossil fuel production and use

Most anthropogenic CH₄ emissions related to fossil fuels come from the exploitation, transportation, and usage of coal, oil, and natural gas. Additional emissions reported in this category include small industrial contributions such as the production of chemicals and metals, fossil fuel fires (e.g. underground coal mine fires and the Kuwait oil and gas fires), and transport (road and non-road transport). CH₄ emissions from the oil processing industry (e.g. refining) and production of charcoal are estimated to be a few teragrams of CH₄ per year only and are included in the transformation industry sector in the inventory. Fossil fuel fires are included in the subcategory “oil and gas”. Emissions from industry, road transport, and non-road transport are reported apart from the two main subcategories “oil and gas” and “coal”, as in Saunois et al. (2020) and contrary to Saunois et al. (2016); each of these amounts to about 2 to 5 Tg CH₄ yr⁻¹ (Table 3). The large range (1–9 Tg CH₄ yr⁻¹) is attributable to difficulties in allocating some sectors to these subsectors consistently among the different inventories (see Table S2). The spatial distribution of CH₄ emissions from fossil fuels is presented in Fig. 3 based on the mean gridded maps provided by CEDS, EDGARv6, and GAINS for the 2010–2019 decade; the USEPA lacks a gridded product.

Figure 3

Methane emissions from four source categories: natural wetlands (excluding lakes, ponds, and rivers), biomass and biofuel burning, agriculture and waste, and fossil fuels for the 2010–2019 decade (in mg CH₄ m⁻² d⁻¹). The wetland emission map represents the mean daily emission average of the 16 biogeochemical models listed in Table 2 and over the 2010–2019 decade. Fossil fuel and agriculture and waste emission maps are derived from the mean estimates of gridded CEDS, EGDARv6, EDGARv7, and GAINS models. The biomass and biofuel burning map results from the mean of the biomass burning inventories listed in Table 1 added to the mean of the biofuel estimate from CEDS (O'Rourke et al., 2021), EDGARv6 (Crippa et al., 2021), EDGARv7 (Crippa et al., 2023), and GAINS (Höglund-Isaksson et al., 2020) models.

[Figure omitted. See PDF]

Global mean emissions from fossil-fuel-related activities, other industries, and transport are estimated from the four global inventories (Table 1) to be 120 [117–125] Tg CH₄ yr⁻¹ for the 2010–2019 decade (Table 3), but with large differences in the rate of change during this period across inventories. The sector accounts on average for 34 % (range 31 %–42 %) of total global anthropogenic emissions in 2010–2019. This contribution has slightly increased from 32 % on average in 2000–2009.

Coal mining

During mining, CH₄ is emitted primarily from ventilation shafts, where large volumes of air are pumped in and out of the mine to keep the CH₄ mixing ratio below 0.5 % to avoid accidental ignition and from dewatering operations. In countries of the Organization for Economic Co-operation and Development (OECD), coalbed CH₄ is often extracted as fuel up to 10 years before the coal mine starts operation, thereby reducing the CH₄ channelled through ventilation shafts during mining. In many countries, large quantities of ventilation air CH₄ are still released to the atmosphere or flared, despite efforts to extend coal mine gas recovery under the UNFCCC Clean Development Mechanism (http://cdm.unfccc.int, last access: 1 April 2025). CH₄ leaks also occur during post-mining handling, processing, and transportation. Some CH₄ is released from coal waste piles and abandoned mines; while emissions from these sources were believed to be low (IPCC, 2000), recent work has estimated these at $\mathrm{22}\times {\mathrm{10}}^{\mathrm{9}}$ m³ (compared to $\mathrm{103}\times {\mathrm{10}}^{\mathrm{9}}$ m³ from functioning coal mines) in 2010 with emissions projected to increase into the future (Kholod et al., 2020).

In 2020, more than 35 % (IEA, 2023a) of the world's electricity is still produced from coal. This contribution grew in the 2000s at the rate of several percent per year, driven by Asian economic growth where large reserves exist, but global coal consumption declined between 2014 and 2020. In 2020, the top 10 largest coal-producing nations accounted for $\sim$ 90 % of total world CH₄ emissions from coal mining; among them, the top three producers (China, the USA, and India) produced almost two-thirds (66 %) of the world's coal (IEA, 2021).

Global estimates of CH₄ emissions from coal mining show a reduced range of 37–44 Tg CH₄ yr⁻¹ for 2010–2019 (Table 3), compared to the previous estimate for 2008–2017 in Saunois et al. (2020) reporting a range of 29–61 Tg CH₄ yr⁻¹ for 2008–2017. This reduced range probably results from using similar activity data (mostly from IEA statistics) in the different inventories. The highest value of the range in Saunois et al. (2020) came from the CEDS inventory, while the lowest came from the USEPA. CEDS seems to have revised downward their estimate compared to the previous version used in Saunois et al. (2020). There were previously large discrepancies in Chinese coal emissions, with a large overestimation from EDGARv4.2, on which CEDS was based. As highlighted by Liu et al. (2021a), a county-based inventory of Chinese methane emissions also confirms the overestimation of previous EDGAR inventories and estimated total anthropogenic Chinese emissions at 38.2 $\pm$ 5.5 Tg CH₄ yr⁻¹ for 2000–2008 (Liu et al., 2021a). Coal mining emission factors depend strongly on the type of coal extraction (underground mining emits up to 10 times more than surface mining), the geological underground structure (region-specific), history (basin uplift), and the quality of the coal (brown coal (lignite) emits less than hard coal (anthracite)). Finally, the different emission factors derived for coal mining are the main reason for the differences between inventories globally (Fig. 2).

For the 2010–2019 decade, methane emissions from coal mining represent 33 % of total fossil-fuel-related emissions of CH₄ (40 [37–44] Tg CH₄ yr⁻¹; Table 3). An additional assumed very small source corresponds to fossil fuel fires, which are mostly underground coal fires. This source is estimated at around 0.15 Tg yr⁻¹ in EDGARv7, though this value remains the same across EDGAR versions and for all years despite the changes in coal production, which could influence this estimate. However, to date, insufficient data are available to better estimate this largely unknown source.

Oil and natural gas systems

This subcategory includes emissions from both conventional and shale oil and gas exploitation. Natural gas is composed primarily of CH₄, so both fugitive and planned emissions during the drilling of wells in gas fields, extraction, transportation, storage, gas distribution, end use, and incomplete combustion in gas flares emit CH₄ (Lamb et al., 2015; Shorter et al., 1996). Persistent fugitive emissions (e.g. due to leaky valves and compressors) should be distinguished from intermittent emissions due to maintenance (e.g. purging and draining of pipes) or incidents. During transportation, fugitive emissions can occur in oil tankers, fuel trucks, and gas transmission pipelines, attributable to corrosion, manufacturing, and welding faults. According to Lelieveld et al. (2005), CH₄ fugitive emissions from gas pipelines should be relatively low; however, old distribution networks in some cities may have higher rates, especially those with cast-iron and unprotected steel pipelines (Phillips et al., 2013). Measurement campaigns in cities within the USA (e.g. McKain et al., 2015) and Europe (e.g. Defratyka et al., 2021) revealed that significant emissions occur in specific locations (e.g. storage facilities, city natural gas fuelling stations, well and pipeline pressurisation/depressurisation points, sewage systems, and furnaces of buildings) along the distribution networks (e.g. Jackson et al., 2014; McKain et al., 2015; Wunch et al., 2016). However, CH₄ emissions vary significantly from one city to another depending, in part, on the age of city infrastructure and the quality of its maintenance, making urban emissions difficult to scale up from measurement campaigns, although attempts have been made (e.g. Defratyka et al., 2021). In many facilities, such as gas and oil fields, refineries, and offshore platforms, most of the associated and other waste gas generated will be flared for security reasons with almost complete conversion to CO₂; however, due to the large quantities of waste gas generated, small fractions of gas still being vented make up relatively large quantities of methane. These two processes are usually considered together in inventories of oil and gas industries. In addition, single-point failure of natural gas infrastructure can leak CH₄ at high rate for months, such as at the Aliso Canyon blowout in Los Angeles, CA (Conley et al., 2016), or the shale gas well blowout in Ohio (Pandey et al., 2019), thus hampering emission control strategies. Production of natural gas from the exploitation of hitherto unproductive rock formations, especially shale, began in the 1970s in the USA on an experimental or small-scale basis, and then, from the early 2000s, exploitation started at a large commercial scale. The shale gas contribution to total dry natural gas production in the USA reached 82 % in 2023, growing rapidly from 48 % in 2013 (IEA, 2023b). The possibly larger emission factors from shale gas compared to conventional gas have been widely debated (e.g. Cathles et al., 2012; Howarth, 2019; Lewan, 2020). The latest studies tend to infer emission factors from the oil–gas production chain of about 1 % to 6 % (e.g. Schneising et al., 2020; Varon et al., 2023; Zhang et al., 2020), but the loss rate could be as high as more than 10 % in low-producing well sites (e.g. Omara et al., 2022, Williams et al., 2025).

CH₄ emissions from oil and natural gas systems vary greatly in different global inventories (67 to 80 Tg yr⁻¹ in 2020; Table 3). The inventories generally rely on the same sources and magnitudes for activity data, with the derived differences therefore resulting primarily from different methodologies and parameters used, including emission factors. Those factors are country- or even site-specific, and the few field measurements available often combine oil and gas activities (Brandt et al., 2014), resulting in high uncertainty in emission estimates for many major oil- and gas-producing countries. Depending on the region, the IPCC 2006 default emission factors may vary by 2 orders of magnitude for oil production and 1 order for gas production. For instance, the GAINSv4.0 estimate of CH₄ emissions from US oil and gas systems in 2015 is 16 Tg, which is almost twice as high as EDGARv8.0 (EDGAR, 2023) at 8.4 Tg and the USEPA (USEPA, 2019) at 9.5 Tg. The difference can partly be explained by GAINS using a bottom-up methodology to derive country- and year-specific flows of associated petroleum gas and attributing these to recovery/reinjection, to flaring or venting (Höglund-Isaksson, 2017), and partly to GAINS using a higher emission factor for unconventional gas production (Höglund-Isaksson et al., 2020). Recent quantifications using satellite observations and inversion estimate a relatively stable trend for US oil and gas system emissions since 2010, with Lu et al. (2023) estimating 14.6 Tg for 2010, 15.9 Tg for 2014, and 15.6 Tg for 2019; Shen et al. (2022) estimating a mean of 12.6 Tg for 2018–2020; and Maasakkers et al. (2021) a mean of 11.1 Tg for 2010 to 2015. The stable top-down trend for the USA appears not well captured in the bottom-up inventories from GAINS and EDGAR, which tend to show an increasing trend driven by an increase in production volumes.

Most recent studies (e.g. Zhang et al., 2020; Shen et al., 2023; Li et al., 2024, Tibrewal et al., 2024; Sherwin et al., 2024) still suggest that the methane emissions from oil and gas industry are underestimated by inventories, industries, and agencies, including the USEPA and UNFCCC reporting. Lauvaux et al. (2022) showed that emissions from a few high-emitting facilities, i.e. super-emitters ($>$ 20 t h⁻¹), which are usually sporadic in nature and not accounted for in the inventories, could represent 8 %–12 % of global oil and gas emissions, or around 8 Tg CH₄ yr⁻¹. These high-emitting points, located on the conventional part of the facilities, could be avoided through better operating conditions and repair of malfunctions. Over the last decade, absolute CH₄ emissions have almost certainly increased, since US crude oil production doubled and natural gas production rose by about 50 % (IEA, 2023a). However, global implications of the rapidly growing shale gas activity in the USA remain to be determined precisely.

For the 2010–2019 decade, CH₄ emissions from upstream and downstream oil and natural gas sectors are estimated to represent about 56 % of total fossil CH₄ emissions (67 [57–74] Tg CH₄ yr⁻¹; Table 3) based on global inventories, with a lower uncertainty range than for coal emissions for most countries. However, it is worth noting that 8 Tg CH₄ yr⁻¹ should be added on top of this estimate to acknowledge the ultra-emitter contribution, as done in Tibrewal et al. (2024).

3.1.4 Agriculture and waste sectors

This main category includes CH₄ emissions related to livestock production (i.e. enteric fermentation in ruminant animals and manure management), rice cultivation, landfills, and wastewater handling. Of these activities, globally and in most countries, livestock is by far the largest source of CH₄, followed by waste handling and rice cultivation. Conversely, field burning of agricultural residues is a minor source of CH₄ reported in emission inventories (a few teragrams at the global scale). The spatial distribution of CH₄ emissions from agriculture and waste handling is presented in Fig. 3 based on the mean gridded maps provided by CEDS, EDGARv6, and GAINS over the 2010–2019 decade.

Global emissions from agriculture and waste for the period 2010–2019 are estimated to be 211 [195–231] Tg CH₄ yr⁻¹ (Table 3), representing 60 % of total direct anthropogenic emissions. Agriculture emissions amount to 144 Tg CH₄ yr⁻¹, 40 % of the direct anthropogenic emissions, with the rest coming from the fossil fuel sector (34 %), waste (19 %), and biomass (5 %) and biofuel (3 %) burning.

Livestock: fermentation and manure management

Enteric domestic ruminants such as cattle, buffalo, sheep, goats, and camels emit CH₄ as a by-product of the anaerobic microbial activity in their digestive systems (Johnson et al., 2002). The very stable temperatures (about 39 °C) and pH (6.5–6.8) within the rumen of domestic ruminants, along with a constant plant matter flow from grazing (cattle graze many hours per day), allow methanogenic Archaea residing within the rumen to produce CH₄. CH₄ is released from the rumen mainly through the mouth of multi-stomached ruminants (eructation, $\sim$ 90 % of emissions) or absorbed in the blood system. The CH₄ produced in the intestines and partially transmitted through the rectum is only $\sim$ 10 % (Hill et al., 2016).

The total number of livestock continues to grow steadily. There are currently (2020) about 1.5 billion cattle globally, almost 1.3 billion sheep, and nearly as many goats (http://www.fao.org/faostat/en/#data/GE, last access: 1 April 2025). Livestock numbers are linearly related to CH₄ emissions in inventories using the Tier 1 IPCC approach such as FAOSTAT. In practice, some non-linearity may arise due to dependencies of emissions on the total weight of the animals and their diet, which are better captured by Tier 2 and higher approaches. Cattle, due to their large population, large individual size, and particular digestive characteristics, account for the majority of enteric fermentation CH₄ emissions from livestock worldwide (Tubiello, 2019; FAO, 2022), particularly in intensive agricultural systems in wealthier and emerging economies, including the USA (USEPA, 2016). CH₄ emissions from enteric fermentation also vary from one country to another as cattle may experience diverse living conditions that vary spatially and temporally, especially in the tropics (Chang et al., 2019).

Anaerobic conditions often characterise manure decomposition in a variety of manure management systems globally (e.g. liquid/slurry treated in lagoons, ponds, tanks, or pits), with the volatile solids in manure producing CH₄. In contrast, when manure is handled as a solid (e.g. in stacks or dry lots) or deposited on pasture, range, or paddock lands, it tends to decompose aerobically and to produce little or no CH₄. However aerobic decomposition of manure tends to produce nitrous oxide (N₂O), which has a larger global warming impact than CH₄. Ambient temperature, moisture, energy contents of the feed, manure composition, and manure storage or residency time affect the amount of CH₄ produced. Despite these complexities, most global data sets used herein apply a simplified IPCC Tier 1 approach, where amounts of manure treated depend on animal numbers and simplified climatic conditions by country.

Global CH₄ emissions from enteric fermentation and manure management are estimated in the range of 114–124 Tg CH₄ yr⁻¹, for the year 2020, in the GAINS model and CEDS, USEPA, FAO-CH₄, and EDGARv7 inventories (Table 3). Using the Tier 2 method adopted from the 2019 refinement to 2006 IPCC guidelines, a recent study (Zhang et al., 2022) estimated that global CH₄ emissions from livestock increased from 31.8 [26.5–37.1] (mean [minimum$-$maximum] of 95 % confidence interval) Tg CH₄ yr⁻¹ in 1890 to 131.7 [109.6–153.7] Tg CH₄ yr⁻¹ in 2019, a 4-fold increase in the past 130 years. Chang et al. (2021) estimate enteric fermentation and manure management emissions based on mixed Tier 1–2 and Tier 1 approaches and calculate livestock emissions of 120 $\pm$ 13 and 136 $\pm$ 15 Tg CH₄ yr⁻¹ respectively for 2018. Chang et al. (2021) and Zhang et al. (2022) estimates for 2018 and 2019 are on average a bit higher than the inventories estimates but in agreement considering the uncertainties. It is worth recalling here that the ranges provided in this study correspond to the minimum–maximum of the existing estimates and do not include the uncertainty of the individual estimate; these uncertainties could be larger than the range proposed here.

For the period 2010–2019, we estimated total emissions of 112 [107–118] Tg CH₄ yr⁻¹ for enteric fermentation and manure management, about one-third of total global anthropogenic emissions (Table 3).

Rice cultivation

Most of the world's rice is grown in flooded paddy fields (Elphick, 2010). The water management systems, particularly flooding, used to cultivate rice are one of the most important factors influencing CH₄ emissions and one of the most promising approaches for CH₄ emission mitigation: periodic drainage and aeration not only cause existing soil CH₄ to oxidise, but also inhibit further CH₄ production in soils (Simpson et al., 1995; USEPA, 2016; Zhang et al., 2016a). Upland rice fields are not typically flooded and therefore are not a significant source of CH₄. Other factors that influence CH₄ emissions from flooded rice fields include fertilisation practices (i.e. the use of urea and organic fertilisers), soil temperature, soil type (texture and aggregated size), and rice variety and cultivation practices (e.g. tillage, seeding, and weeding practices) (Conrad et al., 2000; Kai et al., 2011; USEPA, 2012; Yan et al., 2009). For instance, CH₄ emissions from rice paddies increase with organic amendments (Cai et al., 1997) but can be mitigated by applying other types of fertilisers (mineral, composts, biogas residues) or using wet seeding (Wassmann et al., 2000).

The geographical distribution of rice emissions has been assessed by global (e.g. Janssens-Maenhout et al., 2019; Tubiello, 2019; USEPA, 2012) and regional (e.g. Castelán-Ortega et al., 2014; Chen et al., 2013; Chen and Prinn, 2006; Peng et al., 2016; Yan et al., 2009; Zhang and Chen, 2014) inventories and land surface models (Li et al., 2005; Pathak et al., 2005; Ren et al., 2011; Spahni et al., 2011; Tian et al., 2010, 2011; Zhang et al., 2016a). The emissions show a seasonal cycle, peaking in the summer months in the extratropics, associated with monsoons and land management. Emissions from rice paddies are influenced not only by the extent of rice field area, but also by changes in the productivity of plants (Jiang et al., 2017) as these alter the CH₄ emission factor used in inventories. However, the inventories considered herein are largely based on IPCC Tier 1 methods, which mainly scale with cultivated areas and include region-specific emission factors but do not account for changes in plant productivity and detailed cultivation practices.

The largest emissions from rice cultivation are found in Asia, accounting for 30 % to 50 % of global emissions (Fig. 3). The decrease in CH₄ emissions from rice cultivation over recent decades is confirmed in most inventories because of the decrease in rice cultivation area, changes in agricultural practices, and a northward shift of rice cultivation since the 1970s, as in China (e.g. Chen et al., 2013).

Based on the global inventories considered in this study, global CH₄ emissions from rice paddies are estimated to be 32 [25–37] Tg CH₄ yr⁻¹ for the 2010–2019 decade (Table 3), or about 9 % of total global anthropogenic emissions of CH₄. These estimates are consistent with the 29 Tg CH₄ yr⁻¹ estimated for the year 2000 by Carlson et al. (2017).

Waste management

This sector includes emissions from managed and non-managed landfills (solid waste disposal on land) and wastewater handling, where all kinds of waste are deposited. CH₄ production from waste depends on the pH, moisture, and temperature of the material. The optimum pH for CH₄ emission is between 6.8 and 7.4 (Thorneloe et al., 2000). The development of carboxylic acids leads to low pH, which limits methane emissions. Food or organic waste, such as leaves and grass clippings, ferment quite easily, while wood and wood products generally ferment slowly and cellulose and lignin even more slowly (USEPA, 2010a).

Waste management was responsible for about 11 % of total global direct anthropogenic CH₄ emissions in 2000 (Kirschke et al., 2013). A recent assessment of CH₄ emissions in the USA found landfills to account for almost 26 % of total US anthropogenic CH₄ emissions in 2014, the largest contribution of any single CH₄ source in the USA (USEPA, 2016). In Europe, gas control has been mandatory on all landfills since 2009, and more importantly for CH₄ emissions, the EU Landfill Directive (1999) with subsequent amendments, has diverted most biodegradable waste away from landfills towards source separation, recycling, composting, and energy recovery, with a legally binding target not to landfill more than 10 % of municipal solid waste by 2035.

Wastewater from domestic and industrial sources is treated in municipal sewage treatment facilities and private effluent treatment plants. The principal factor in determining the CH₄ generation potential of wastewater is the amount of degradable organic material in the wastewater. Wastewater with high organic content is treated anaerobically, which leads to increased emissions (André et al., 2014). Excessive and rapid urban development worldwide, especially in Asia and Africa, could enhance methane emissions from waste unless adequate mitigation policies are designed and implemented rapidly.

The GAINS model and CEDS and EDGAR inventories give robust emission estimates from solid waste in the range of 37–42 Tg CH₄ yr⁻¹ for the year 2019 and more uncertain wastewater emissions in the range 20–45 Tg CH₄ yr⁻¹1.

In our study, the global emission of CH₄ from waste management is estimated in the range of 56–80 Tg CH₄ yr⁻¹ for the 2010–2019 period with a mean value of 69 Tg CH₄ yr⁻¹, about 19 % of total global anthropogenic emissions (Table 3).

3.1.5 Biomass and biofuel burning

This category includes CH₄ emissions from biomass burning in forests, savannahs, grasslands, peats, and agricultural residues, as well as from the burning of biofuels in the residential sector (stoves, boilers, fireplaces). Biomass and biofuel burning emit CH₄ under incomplete combustion conditions (i.e. when oxygen availability is insufficient for complete combustion), for example in charcoal manufacturing and smouldering fires. The amount of CH₄ emitted during the burning of biomass depends primarily on the amount of biomass, burning conditions, fuel moisture, and the specific material burned.

In this study, we use large-scale biomass burning (forest, savannah, grassland, and peat fires) from five biomass burning inventories (described below) and the biofuel burning contribution from anthropogenic emission inventories (EDGARv6 and v7, CEDS, GAINS, and USEPA). The spatial distribution of emissions from the burning of biomass and biofuel over the 2010–2019 decade is presented in Fig. 3 based on data listed in Table 1.

At the global scale, during the period of 2010–2019, biomass and biofuel burning generated CH₄ emissions of 28 [21–39] Tg CH₄ yr⁻¹ (Table 3), of which 30 %–50 % is from biofuel burning.

Biomass burning

Fire is an important disturbance event in terrestrial ecosystems globally (van der Werf et al., 2010) and can be of either natural (typically $\sim$ 10 % of fires, ignited by lightning strikes or started accidentally) or anthropogenic origin ($\sim$ 90 %, human-initiated fires) (USEPA, 2010b, Chap. 9.1). As previously noted all fires are counted as anthropogenic in Table 3. Anthropogenic fires are concentrated in the tropics and subtropics, where forests, savannahs, and grasslands may be burned to clear land for agricultural purposes or to maintain pastures and rangelands. Small fires associated with agricultural activity, such as field burning and agricultural waste burning, are often not detected by moderate-resolution remote sensing methods and are instead estimated based on cultivated area or through in situ measurements such as dedicated airborne campaigns (e.g. Barker et al., 2020).

Emission rates of biomass burning vary with biomass loading (depending on the biomes) at the location of the fire, the efficiency of the fire (depending on the vegetation type), the fire type (smouldering or flaming), and emission factor (mass of the considered species/mass of biomass burned). Depending on the approach, these parameters can be derived using satellite data and/or biogeochemical model or through simpler IPCC default approaches.

In this study, we use five products to estimate biomass burning emissions. The Global Fire Emission Database (GFED) is the most widely used global biomass burning emission data set and provides estimates from 1997 onwards. Here, we use GFEDv4.1s (van der Werf et al., 2017), based on the Carnegie–Ames–Stanford approach (CASA) biogeochemical model (van der Werf et al., 2010) driven by satellite-derived vegetation characteristics and burned area mostly from the MODerate resolution Imaging Sensor, MODIS (Giglio et al., 2013). GFEDv4.1s (with small fires) is available at a 0.25° resolution and on a daily basis from 1997 to 2020. One characteristic of the GFEDv4.1s burned area is that small fires are better accounted for compared to GFEDv4.1 (Randerson et al., 2012), increasing carbon emissions by approximately 35 % at the global scale. The latest version GFEDv5 (Chen et al., 2023) suggests a 61 % higher burned area than GFEDv4.1s, in closer agreement with burned area products from higher-resolution satellite sensors. The next budget would benefit from GFEDv5 to revisit the estimates of biomass burning emissions (which would likely go up) based on more specific comparison studies. The Quick Fire Emissions Dataset (QFED) is calculated using the fire radiative power (FRP) approach, in which the thermal energy emitted by active fires (detected by MODIS) is converted to an estimate of CH₄ flux using biome-specific emissions factors and a unique method of accounting for cloud cover. Further information related to this method and the derivation of the biome-specific emission factors can be found in Darmenov and da Silva (2015). Here we use the historical QFEDv2.5 product available daily on a $\mathrm{0.1}\times \mathrm{0.1}$ grid for 2000 to 2020.

The Fire INventory from the National Center for Atmospheric Research (FINNv2.5; Wiedinmyer et al., 2023) provides daily, 1 km resolution estimates of gas and particle emissions from open burning of biomass (including wildfire, agricultural fires, and prescribed burning) over the globe for the period 2002–2020. FINNv2.5 uses MODIS and VIIRS satellite observations for active fires, land cover, and vegetation density.

We use v1.3 of the Global Fire Assimilation System (GFAS; Kaiser et al., 2012), which calculates emissions of biomass burning by assimilating fire radiative power (FRP) observations from MODIS at a daily frequency and 0.5° resolution and is available for 2000–2020.

The FAO-CH₄ yearly biomass burning emissions are based on the most recent MODIS 6 burned area products (Prosperi et al., 2020), coupled with a pixel level (500 m) implementation of the IPCC Tier 1 approach, and are available from 1990 to 2020 (Table 1).

The differences in emission estimates for biomass burning arise from specific geographical and meteorological conditions and fuel composition, which strongly impact combustion completeness and emission factors. The latter vary greatly according to fire type, ranging from 2.2 g CH₄ kg⁻¹ dry matter burned for savannah and grassland fires up to 21 g CH₄ kg⁻¹ dry matter burned for peat fires (van der Werf et al., 2010). Biomass burning emissions encountered large interannual variability related to meteorological conditions, with generally higher emissions during El Niño periods as in 2019 (20 [14–28] Tg CH₄ yr⁻¹), 2015 (22 [15–28] Tg CH₄ yr⁻¹), and 2010 to a lesser extent (18 [15–29] Tg CH₄ yr⁻¹).

In this study, based on the five aforementioned products, biomass burning emissions are estimated at 17 Tg CH₄ yr⁻¹ [12–24] for 2010–2019, representing about 5 % of total global anthropogenic CH₄ emissions (Table 3).

Biofuel burning

Burning of biomass to produce energy for domestic, industrial, commercial, or transportation purposes is hereafter called biofuel burning. A largely dominant fraction of CH₄ emissions from biofuel burning comes from domestic cooking or heating in stoves, boilers, and fireplaces, mostly in open cooking fires where wood, charcoal, agricultural residues, or animal dung are burned. It is estimated that more than 2 billion people, mostly in developing countries, use solid biofuels to cook and heat their homes daily (André et al., 2014), yet CH₄ emissions from biofuel combustion have received relatively little attention. Biofuel burning estimates are gathered from the CEDS, USEPA, GAINS, and EDGAR inventories. Due to the sectoral breakdown of the EDGAR and CEDS inventories, the biofuel component of the budget has been estimated as equivalent to the “RCO – Energy for buildings” sector as defined in Worden et al. (2017) and Hoesly et al. (2018) (Table S2). This is equivalent to the sum of the IPCC 1.A.4.a - Commercial/Institutional, 1.A.4.b - Residential, 1.A.4.c - Agriculture/Forestry/Fishing/Fish farms, and 1.A.5 - Non-Specified reporting categories. This definition is consistent with that used in Saunois et al. (2016) and Kirschke et al. (2013). While this sector incorporates biofuel use, it also includes the use of other combustible materials (e.g. coal or gas) for small-scale heat and electricity generation within residential and commercial premises. Data provided by the GAINS inventory suggest that this approach may overestimate biofuel emissions by between 5 % and 50 %. Further study into this category would be needed to better disentangle biofuels from fossil combustibles.

In our study, biofuel burning is estimated to contribute 11 [8–14] Tg CH₄ yr⁻¹ to the global CH₄ budget, about 3 % of total global anthropogenic CH₄ emissions for 2010–2019 (Table 3).

3.1.6 Other anthropogenic sources (not explicitly included in this study)

Other anthropogenic sources not included in this study are related to agriculture and land use management. In particular, increases in agricultural areas (such as global palm oil production) have led to the clearing of natural peat forests, reducing natural peatland area and associated natural CH₄ emissions. Peatlands planted to forests (like in northern Europe) also lead to reduced CH₄ emissions. While studies have long suggested that CH₄ emissions from peatland drainage ditches are likely to be significant (e.g. Minkkinen and Laine, 2006; Peacock et al., 2021), CH₄ emissions related to palm oil plantations have yet to be properly quantified (e.g. Manning et al., 2019). Taylor et al. (2014) have quantified global palm oil wastewater treatment fluxes to be 4 $\pm$ 32 Tg CH₄ yr⁻¹ for 2010–2013. This currently represents a small and highly uncertain source of methane but one potentially growing in the future.

3.2 Natural and indirect anthropogenic sources

As introduced in Sect. 2.4, natural and indirect anthropogenic sources refer to pre-agricultural CH₄ emissions even if they are perturbed by anthropogenic climate change or other global change factors (e.g. eutrophication) and indirect emissions resulting from anthropogenic perturbation of the landscape (reservoirs) and the biogeochemical characteristics of soil. They include vegetated wetland emissions and inland freshwater systems (lakes, small ponds, reservoirs, and rivers), land geological sources (gas–oil seeps, mud volcanoes, microseepage, geothermal manifestations, and volcanoes), wild animals, wildfires, termites, thawing terrestrial and marine permafrost, and coastal and oceanic sources (biogenic, geological, and hydrate). In water-saturated or flooded ecosystems, the decomposition of organic matter gradually depletes most of the oxygen in the soil or the sediment zone, resulting in anaerobic conditions and CH₄ production. Once produced, CH₄ can reach the atmosphere through a combination of three processes: (1) diffusive loss of dissolved CH₄ across the air–water boundary, (2) ebullition flux from sediments, and (3) flux mediated by emergent aquatic macrophytes and terrestrial plants (plant transport). On its way to the atmosphere, in the soil or water columns, CH₄ can be partly or completely oxidised by microorganisms, which use CH₄ as a source of energy and carbon (USEPA, 2010b). Concurrently, methane from the atmosphere can diffuse into the soil column and be oxidised (see Sect. 3.3.4 on soil uptake).

3.2.1 Wetlands

Wetlands are generally defined as ecosystems in which mineral or peat soils are water-saturated at some depth or where surface inundation (permanent or not) has a dominating influence on the soil biogeochemistry and determines the ecosystem species composition (USEPA, 2010b). To refine such an overly broad definition for CH₄ emissions, we define wetlands as ecosystems with inundated or saturated soils or peats where anaerobic conditions below the water table lead to CH₄ production (Matthews and Fung, 1987; USEPA, 2010b). Brackish water emissions are discussed separately in Sect. 3.2.6. Our definition of wetlands includes ombrotrophic and minerotrophic peatlands (i.e. bogs and fens), mineral soil wetlands (swamps and marshes), and seasonal or permanent floodplains. It excludes exposed water surfaces without emergent macrophytes, such as lakes, rivers, estuaries, ponds, and reservoirs (addressed in the next section), as well as rice agriculture (see Sect. 3.1.4, rice cultivation paragraph) and wastewater ponds. It also excludes coastal vegetated ecosystems (mangroves, seagrasses, salt marshes) with salinities usually $>$ 0.5 (see Sect. 3.2.6). Even with this definition, some wetlands could be considered anthropogenic systems, being affected by human land use changes such as impoundment, drainage, or restoration (Woodward et al., 2012). In the following, we retain the generic denomination “wetlands” for natural and human-influenced wetlands, as discussed in Sect. 2.2.

The three most important factors influencing CH₄ production in wetlands are the spatial and temporal extent of anoxia (linked to water saturation), temperature, and substrate availability (Valentine et al., 1994; Wania et al., 2010; Whalen, 2005; Delwiche et al., 2021; Knox et al., 2021).

Land surface models estimate CH₄ emissions through a series of processes, including CH₄ production, oxidation, and transport. The models are then forced with inputs accounting for changing environmental factors (Melton et al., 2013; Poulter et al., 2017; Tian et al., 2010; Wania et al., 2013; Xu et al., 2010). CH₄ emissions from wetlands are computed as the product of an emission flux density and a CH₄-producing area or surface extent (Bohn et al., 2015; Melton et al., 2013). The areal extent of different wetland types (having large differences in areal CH₄ emission rates) appears to be a primary contributor to uncertainties in the absolute flux of CH₄ emissions from wetlands, with meteorological response being the main source of uncertainty for seasonal and interannual variability (Poulter et al., 2017; Kuhn et al., 2021; Parker et al., 2022; McNicol et al., 2023; Karlson and Bastviken, 2023). However, large uncertainty remains in both spatial and temporal emission distributions, especially over tropical wetlands where data are lacking to evaluate the models but are nevertheless a key region for climate feedbacks (Nisbet, 2023; Zhang et al., 2023). Direct measurement campaigns and remote sensing are providing key insights into where to improve the land surface models (e.g. France et al., 2022; Shaw et al., 2022).

In this work, 16 land surface models computing net CH₄ emissions (Table 2) were run under a common protocol with a spin-up using repeated climate data from 1901–1920 to pre-industrial conditions followed by a transient simulation through the end of 2020. Of the 16 models, 13 previously contributed to Saunois et al. (2020), and 3 models were new to this release (CH₄MOD_wetland (Li et al., 2010), ISAM (Shu et al., 2020; Xu et al., 2021), and SDGVM (Beerling and Woodward, 2001; Hopcroft et al., 2011; Hopcroft et al., 2020)) (Table 2; see also in the Supplement Table S3 for a history of the contributing models). Climatic forcing uncertainties are considered in the ensemble estimate by using two climate data sets: CRU/CRU-JRA55 (Harris et al., 2014) and GSWP3-W5E5 (Dirmeyer et al., 2006; Kim, 2017; Lange, 2019; Cucchi et al., 2020). Atmospheric CO₂ was also prescribed in the models. For all models, two wetland area dynamic schemes were applied: a diagnostic scheme using a remote-sensing-based wetland area and dynamics data set called WAD2M (Wetland Area Dynamics for Methane Modeling; Zhang et al., 2021a, b) available at 0.25° of horizontal resolution, as in Saunois et al. (2020), and a prognostic scheme using internal model-specific hydrologic models.

The diagnostic wetland extent product WAD2Mv1.0 (Zhang et al., 2021a) has been updated since the work of Saunois et al. (2020) to WAD2Mv2.0 (Zhang et al., 2021b) and extended to 2020. It uses the same Surface Water Microwave Product Series (SWAMPSv3.2) for capturing inundation dynamics (Jensen and McDonald, 2019), which was extended to 2020. To reduce potential double counting with the freshwater budget, the surface areas of rivers/streams and lakes/ponds are excluded by using the products Global River Widths from Landsat (GRWL) database v01.01 (Allen and Pavelsky, 2018) and HydroLAKES v1.0 (Messenger et al., 2006), instead of the Global Surface Water (GSW) product (Pekel et al., 2016) used in WAD2Mv1.0. The GRWL and HydroLAKES are also the data sets used separately in the upscaling of the freshwater budget, allowing for a more consistent approach between the wetland and freshwater CH₄ budgets (Sect. 3.2.2). This update in WAD2M leads to a downward revised annual average wetland extent by 0.5 Mkm² for the mid-high latitudes (mainly due to larger lake extent in HydroLAKES than in the GSW data set) with small impacts in other regions. However, since HydroLAKES includes only vectorised lakes larger than 0.1 km², smaller lakes/ponds under 0.1 km² are implicitly still included as wetlands in WAD2Mv2.0. For the high-latitude region, the recent peatland extent product from Hugelius et al. (2020) is applied, which indicates a slightly higher peatland area by 0.2 Mkm², primarily in regions above 60° N, compared to the Northern Circumpolar Soil Carbon Database (NCSCD) product (Hugelius et al., 2013) used in WAD2Mv1.0. Rice agriculture was removed using the Monthly Irrigated and Rainfed Crop Areas (MIRCA2000; Portmann et al., 2010) data set from circa 2000, as a fixed distribution.

The combined remote-sensing and inventory WAD2Mv2.0 product leads to a maximum wetland area of 13.6 Mkm² during the peak season (7.9 Mkm² on annual average, with a range of 7.5 to 8.4 Mkm² from 2000–2020, about 5.2 % of the global land surface). The largest wetland areas in WAD2Mv2.0 are in Amazonia, the Congo Basin, and the Western Siberian lowlands, which in previous studies were underestimated by inventories (Bohn et al., 2015). However, the SWAMPS v3.2 data set, which serves as a proxy of temporal variations in wetland extent, has had discontinuity issues over a few tropical hotspots since 2015 and hence affects the temporal variations in WAD2M. Consequently, this affects CH₄ emissions estimates for a subset of land surface models that are particularly sensitive to inundation in these hotspots. Meanwhile, prognostic estimates show moderate consistency in capturing the spatial distribution of wetland areas with WAD2M, with an annual average wetland area of 8.0 $\pm$ 2.0 Mkm² during the peak season for 2000–2020. The ensemble mean of annual wetland area anomaly by the prognostic models shows reasonable agreement with satellite-based estimates in capturing the response of wetland area to climate variations (Zhang et al., 2025), with higher agreement over temperate and boreal regions than in the tropics.

For the wetland methane emissions estimate, we use the decadal mean from the prognostic runs and adjust these flux estimates for double counting from inland waters (described in next section) given the reliance of the prognostic models on satellite flooded area data like WAD2Mv2 to parameterise maximum wetland extent (Zhang et al., 2025). The average emission from wetlands for 2010–2019 for the 16 models is plotted in Fig. 3. The zones with the largest emissions are the Amazon basin, equatorial Africa and Asia, Canada, western Siberia, eastern India, and Bangladesh. Regions where CH₄ emissions have high intermodel agreement (defined as regions where mean flux is larger than the standard deviation of the models, on a decadal mean) represent 72 % of the total CH₄ flux due to natural wetlands. The different sensitivities of the models to temperature, vapour pressure, precipitation, and radiation can generate substantially different patterns, such as in India. Emission estimates over regions with lower emissions (in total) are also consistently inferred between models (e.g. Scandinavia, continental Europe, eastern Siberia, central USA, and southern Africa).

The resulting global flux range for vegetated wetland emissions from the prognostic runs is 117–195 Tg CH₄ yr⁻¹ for the 2000–2020 period, with an average of 157 Tg CH₄ yr⁻¹ and a 1$\mathit{\sigma }$ standard deviation of 24 Tg CH₄ yr⁻¹. Using the prognostic set of simulations, the average ensemble emissions were 159 [119–203] Tg CH₄ yr⁻¹ for the 2010–2019 period (Table 3). The estimated average ensemble annual total from the two sets of simulations by CRU/CRU-JRA55 and GSWP3-W5E5 are 158 [126–193] and 158 [118–203] for 2010–2019 respectively. Generally, the magnitude and interannual variability agree between these two sets of simulations (Zhang et al., 2025). Wetland emissions represent about 25 % of the total (natural plus anthropogenic) CH₄ sources estimated by bottom-up approaches. The large range in the estimates of wetland CH₄ emissions results from difficulties in defining wetland CH₄-producing areas as well as in parameterising terrestrial anaerobic conditions that drive sources and the oxidative conditions leading to sinks (Melton et al., 2013; Poulter et al., 2017; Wania et al., 2013). The ensemble mean emission using the same simulation set-up (i.e. diagnostic wetland extent and CRU/CRU-JRA55) in the models is 163 [117–195] Tg CH₄ yr⁻¹, higher by $\sim$ 22 Tg CH₄ yr⁻¹ than the one previously reported (see Table 3, for 2000–2009 with comparison to Saunois et al., 2020). This difference is mainly due to the updated model structure and parameterisations in the wetland CH₄ models compared to the versions in the previous budget and the inclusion of three new land surface models.

For the last decade 2010–2019, we report in this budget an average ensemble estimate of 159 Tg CH₄ yr⁻¹ with a range of 119–203 (based on prognostic wetland extent runs; Table 3).

3.2.2 Inland freshwater ecosystems (lakes, ponds, reservoirs, streams, rivers)

This category includes CH₄ emissions from freshwater systems (lakes, ponds, reservoirs, streams, and rivers). Numerous advances have been made in the freshwater greenhouse gas knowledge base in the last few years (Lauerwald et al., 2023a). These advances include improvements in the underlying databases used to estimate inland water surface areas and model their dynamics, a rapidly growing number of direct measurements of methane fluxes, and improvements in our process-based understanding of methane biogeochemistry. Despite this, aspects of global freshwater methane estimates remain rather crude and continue to have large uncertainties. This includes the overall temperature dependency of methane emissions, the relative role of ebullition (i.e. bubble flux, which may represent the most important but most difficult-to-capture emission path in many standing water bodies), fluxes from the smallest standing water bodies (sometimes referred to as ponds) having large emissions per square metre but uncertain area extent, and the magnitude of anthropogenic influence on emissions, all which are discussed below.

Streams and rivers

The last global CH₄ budget used an estimate of 27 Tg CH₄ yr⁻¹ for global streams and rivers based largely on a data compilation by Stanley et al. (2016). This estimate was scaled from a simple data compilation without a spatial component or an estimate of ebullition. More recently, Rosentreter et al. (2021) performed a new data compilation of 652 flux estimates, including diffusive and ebullition fluxes, coupled to an ice-corrected surface area estimate of $\sim$ 625 000 km² that was aggregated to five latitudinal bands to come up with a global estimate of 6 and 31 $\pm$ 17 Tg CH₄ yr⁻¹ (respectively for the median and mean $\pm$ CI 95 %). We believe, due to better data representation in underlying data sets, that the mean estimate of Rosentreter et al. (2021) is more representative statistically because the median does not capture hotspots and hot moments of intense ebullition fluxes. Finally, Rocher-Ros et al. (2023) used a new Global River Methane (GRiMeDB) database (Stanley et al., 2023) with $>$ 24 000 observations of CH₄ concentrations to predict $\sim$ 28 $\pm$ 17 Tg CH₄ yr⁻¹ ($\pm$ CI 95 %) river emissions globally. This approach used machine learning methods coupled to the latest spatially and temporally explicit mapping of monthly stream surface area (the smallest streams are still extrapolated), which incorporates drying and freezing effects (yearly average 672 000 km²; Liu et al., 2022) and includes an ebullition flux estimated from a correlation between measured diffusive and ebullition emissions in the GRiMeDB database (Stanley et al., 2023). Thus, for this study we use an estimate of 29 $\pm$ 17 ($\pm$ CI 95 %) Tg CH₄ yr⁻¹ for streams and rivers (Fig. 4), which averages the mean estimate of Rosentreter et al. (2021) and Rocher-Ros et al. (2023). Currently, ebullition fluxes remain a major unknown quantity in streams and rivers but appear to be coarsely linearly correlated in a log space to diffusive fluxes and of similar magnitude (Rocher-Ros et al., 2023). Methodologically, the high-water velocity of many streams and rivers makes measurement of ebullition fluxes challenging (Robison et al., 2021). Effluxes are also linked to hydrology (Aho et al., 2021), although very few studies have sampled over a representative hydrograph. Plant-mediated effluxes of CH₄ in running waters also remain difficult to constrain, with a recent compilation highlighting very few measurements (Bodmer et al., 2024). Connected adjacent wetlands are a common source of CH₄ to streams and rivers (Borges et al., 2019), which may be important for the regulation of running water emissions but is currently difficult to assess at the global scale. Overall, the poor representation of sites and deficient mechanistic understanding make it difficult to model and predict methane evasion from streams and rivers using process-based models.

Figure 4

Estimation of wetland and inland freshwater emissions over the 2010–2019 decade (in Tg CH₄ yr⁻¹). The fluxes related to voluntary (such as through reservoirs or farm ponds) or involuntary (land use or eutrophication-related) perturbations of the methane cycle are shown here in pink. They are accounted for in the “natural and indirect anthropogenic” sources in the Table 3 budget and depicted as “natural and indirect anthropogenic” sources (darker green and pink hatches) in Fig. 7. Infographic designed by WeDoData (https://wedodata.fr, last access: 1 April 2025).

[Figure omitted. See PDF]

Lakes and ponds

The previous global CH₄ budget used an estimate of 71 Tg CH₄ yr⁻¹ for lakes and 18 Tg CH₄ yr⁻¹ for reservoirs. These estimates were based on an early study by Bastviken et al. (2011) coupled with a newer estimate for lakes north of 50° N (Wik et al., 2016b). There have been three new lake studies that have published their data with global estimates of 56 and 151 $\pm$ 73 (Rosentreter et al., 2021; respectively for the median and mean $\pm$ CI 95 %, 22 $\pm$ 8 (Zhuang et al., 2023; $\pm$ lake-area-weighted normalised RMSE for all parameterised lake types), process-based model) and 41 $\pm$ 36 Tg CH₄ yr⁻¹ (Johnson et al., 2022a, mean $\pm$ CI 95 %). This large range in estimated emissions can be attributed to the differences in the data sets and methods used to calculate the surface area of small waterbodies, as well as the differences between how the flux data were analysed and extrapolated between studies. For instance, total surface areas of all lakes and ponds of 3712–5688 $\times$ 10³ km² (Rosentreter et al., 2021) and 2806 $\times$ 10³ km² (Johnson et al., 2022a) were used along with measurement data from 198 and 575 individual lake systems respectively. In contrast, Zhuang et al. (2023) generated estimates using higher-temporal-resolution data from just 54 lakes to build a process-based model, which generated much lower flux estimates from tropical lakes than previously implemented statistical approaches, but in line with the most recent assessments by Borges et al. (2022). For this study, we explicitly excluded lakes $<$ 0.1 km², which are treated separately (see below). If we re-assess these three studies for only lakes greater than 0.1 km², we obtain global effluxes of 17 and 42.9 $\pm$ 20.8 Tg CH₄ yr⁻¹ (Rosentreter et al., 2021; median and mean $\pm$ CI 95 % of global flux), 21.9 $\pm$ 8.0 (Zhuang et al., 2023, $\pm$ lake-area-weighted normalised RMSE for all parameterised lake types), and 35.3 $\pm$ 31.0 Tg CH₄ yr⁻¹ (Johnson et al., 2022a, b, $\pm$95 % CI) (with areas of 2556–3468 $\times$ 10³, $\mathrm{2640}\times {\mathrm{10}}^{\mathrm{3}}$, and $\mathrm{2676}\times {\mathrm{10}}^{\mathrm{3}}$ km² respectively). Thus, for lakes $>$ 0.1 km², we propose an efflux of 33 $\pm$ 26 Tg CH₄ yr⁻¹ (an average of the mean from Rosentreter et al., 2021, Zhuang et al., 2023, and Johnson et al., 2022a, b, with the average CI 95 % from Rosentreter et al., 2021, and Johnson et al., 2022a, b) as represented in Fig. 4. Small waterbody emissions, hereafter small lakes and ponds $<$ 0.1 km², remain difficult to assess. Evidence is emerging that there is a lower limit to the power scaling laws that early studies used to extrapolate the surface area of these small systems (Bastviken et al., 2023; Kyzivat et al., 2022). Thus, for small lakes and ponds $<$ 0.1 km² (and $>$ 0.001 km²), we disregard the higher end surface area used in Rosentreter et al. (2021), which relied on these earlier estimates and scale their numbers to the evasion estimates to the lower-end surface area of $\mathrm{1002}\times {\mathrm{10}}^{\mathrm{3}}$ to obtain a mean flux of 33 Tg CH₄ yr⁻¹ (Rosentreter et al., 2021). Johnson et al. (2022a) estimated a surface area of only $\mathrm{166}\times \mathrm{103}$ km² for this size class to obtain an efflux of 6.3 Tg CH₄ yr⁻¹, which we acknowledge as a lower limit. Averaging these two values provides a conservative estimate of 20 [6–33] Tg CH₄ yr⁻¹, which is close to the number proposed by Holgerson and Raymond (2016) for diffusion effluxes only for this size class. The experts involved in this assessment have low confidence in this estimate. This estimate also does not include artificial ponds, which we discuss below. As a result, combined CH₄ emissions from large lakes ($>$ 0.1 km²) and small lakes and ponds ($<$ 0.1 km²) are estimated at 53 [19–86] Tg CH₄ yr⁻¹ (Fig. 4), which is lower than the 71 Tg estimated in the previous budget.

Reservoirs

New mean estimates of diffusive $+$ ebullition CH₄ emissions from reservoirs include 15 and 24 $\pm$ 8 (the median and mean $\pm$ CI 95 % from Rosentreter et al., 2021), 10 $\pm$ 4 (Johnson et al., 2021, mean $\pm$ 95 % CI), 10 (Harrison et al., 2021, low and high 95 % CI of 7 and 22 respectively), and 2.1 Tg CH₄ yr⁻¹ (Zhuang et al., 2023). We compile the first three estimates to a direct efflux of $\sim$ 14 Tg CH₄ yr⁻¹ (with $\pm$ CI 95 % of 9 and 23). We note the fourth estimate as a lower bound but exclude it from this budget given that it was generated via a model that only included data from six reservoir systems (Zhuang et al., 2023). We also add in an additional 12 Tg CH₄ yr⁻¹ (low and high 95 % CI of 7 and 37 respectively) that is estimated to degas in dam turbines (Harrison et al., 2021), which was not addressed in the studies by Rosentreter et al. (2021), Zhuang et al. (2023), or Johnson et al. (2021). Rocher-Ros et al. (2023) also excluded river observations below dams when executing their statistical model and so did not capture downstream dam emissions. Thus, we use a direct reservoir emission here of $\sim$ 13 [6–28] Tg CH₄ yr⁻¹ and estimate an additional $\sim$ 12 [7–37] Tg CH₄ yr⁻¹ from dam turbine degassing fluxes, giving a total of 25 [13–65] Tg CH₄ yr⁻¹ from reservoirs (Fig. 4).

Uncertainties and confidence levels

The emission estimates of lakes, reservoirs, and ponds described above are limited by several uncertainties. First, a major unknown for lakes remains the size cut-off and the representation of small lakes and ponds (Deemer and Holgerson, 2021), which are also more variable than larger water bodies in their CH₄ concentrations and fluxes (Rosentreter et al., 2021; Ray et al., 2023). Interestingly, there is also a lack of methane data representation from large lakes that are a large component of global lake surface area (Deemer and Holgerson, 2021; Messager et al., 2016). There is also a growing knowledge base on the importance of high CH₄ fluxes from lake littoral zones that is not yet well incorporated into global scaling efforts (e.g. Grinham et al., 2011; Natchimuthu et al., 2016) and emergent vegetation (Bastviken et al., 2023; Kyzivat et al., 2022). Ebullition is more constrained in lakes/reservoirs compared to streams/rivers but is still difficult to measure and model accurately. Finally, for all inland water systems a greater scrutiny for the limiting factors (including the impact of ice cover and seasonality, stratification of the water column) of different CH₄ production, consumption, and transport pathways is needed. In addition, a better understanding of the climatic, environmental, and geomorphological controls on key CH₄ processes (e.g. sedimentary diffusive and ebullition production, bubble dissolution, CH₄ oxidation) on the large-scale remains critically needed. For instance, the consistently lower global emissions determined by the process-based model of Zhuang et al. (2023) compared to observations suggest that current data sets are too limited to fully capture the spatiotemporal variability in CH₄ dynamics and their key control factors, possibly leading to biased-high estimates.

The majority of the inland water CH₄ estimates are from a limited number of studies, some without spatial representation or reported statistical uncertainties. Furthermore, as mentioned above the knowledge base of the surface area of these ecosystems is new and rapidly expanding but not standardised between studies leading to uncertainty (but see Lauerwald et al., 2023b), particularly for ponds. For this study, we are able to provide confidence intervals from the original studies for all fluxes except the smallest lake/pond size class.

The surface area of inland freshwaters

For all of these ecosystems, determining their surface area remains a central challenge. Since the last GMB, several methodological advances have reduced the uncertainty associated with the surface area estimates of rivers, streams, lakes, and reservoirs. Using a single geospatial data set that includes both lakes and reservoirs (Messager et al., 2016) has decreased double counting of lakes and reservoirs (Johnson et al., 2022a; Rosentreter et al., 2021). For rivers and streams, high-resolution global streamflow simulations, informed by satellite observations, enabled a much finer-scale estimate of surface areas for rivers with a new temporal component (Allen and Pavelsky, 2018; Lin et al., 2019; Liu et al., 2022), although the surface for the smaller streams are still estimated indirectly, and mapping of human-created drainage ditches and canals is lacking. Seasonal ice cover and melt turnover corrections also have been newly incorporated into rivers, streams, lakes, and reservoirs (Harrison et al., 2021; Johnson et al., 2022a; Lauerwald et al., 2023b; Rocher-Ros et al., 2023; Rosentreter et al., 2021; Zhuang et al., 2023). Finally, removing open-water-body surface areas from wetland surface areas based on geographic location has reduced double counting between these two land cover types, as described in the wetlands section of the GMB. However, the surface area of small lakes and ponds ($<$ 0.1 km²) is still highly uncertain, and new techniques for counting these systems and determining the overlap with wetland data bases are paramount.

Anthropogenic contributions to inland freshwater emissions

We argue that all reservoirs should be categorised as a direct anthropogenic source of emissions. Most of the surface area of reservoirs are human-made, and reservoir construction leads to anoxic sediments and/or bottom waters with labile organic matter sourced from the watershed and to in situ nutrient augmented phytoplankton production (Deemer et al., 2016; Maavara et al., 2017; Prairie et al., 2018). It is also clear that the cultural eutrophication of natural lakes driven by run-off of agricultural nitrogen fertiliser and manure augments CH₄ emissions (DelSontro et al., 2018; Li et al., 2021), with shallow lakes particularly likely to experience eutrophication (Qin et al., 2020). For instance, Beaulieu et al. (2019) modelled a 15 % reduction in lake CH₄ with a 25 % reduction in lake phosphorus concentrations. Several recent studies have estimated that anywhere between 30 % and 50 % of lakes are eutrophic (Cael et al., 2022; Qin et al., 2020; Sayers et al., 2015; Wu et al., 2022). These studies estimate numerical percentages (one by depth class: Qin et al., 2020), but none have estimated the percent of lake surface area that is eutrophic nor have any determined the extent of anthropogenic vs. natural eutrophication. Still, numerous studies have noted widespread increases in eutrophication indicators across lakes due to nutrient loading and warming (Griffiths et al., 2022; Ho et al., 2019; Taranu et al., 2015); thus we estimate that one-third or 11 Tg CH₄ yr⁻¹ of CH₄ emissions from lakes $>$ 0.1 km² could be anthropogenic (Fig. 4). Similarly, CH₄ emissions from small lakes and ponds are influenced by human factors, with emissions increasing with eutrophication (Deemer and Holgerson, 2021), erosion and runoff in agricultural landscapes (Heathcote et al., 2013), and warming, the latter likely to have a disproportionately greater effect in small, shallow systems (Woolway et al., 2016). Thus, we adopt the same $\mathrm{1}/\mathrm{3}$ number as for lakes for the proportion of anthropogenic emissions in small lakes and ponds ($<$ 0.1 km²), which amounts to 6 Tg CH₄ yr⁻¹ (Fig. 4).

There are also human-made small lakes and ponds, notably for agriculture, aquaculture, and recreation, that generally have conditions favourable for high CH₄ emissions (Downing, 2010; Holgerson and Raymond, 2016; Malerba et al., 2022; Ollivier et al., 2019; Zhao et al., 2021; Dong et al., 2023). Downing (2010) estimated that farm ponds comprise a global surface area of $\sim$ 77 000 km²; using a conservative emission rate of 265 mg CH₄ m⁻² d⁻¹ and an ice correction factor of 0.6 leads to an emission of 4.5 Tg yr⁻¹ that is anthropogenically sourced from farm ponds. Here the value is rounded to 5 Tg yr⁻¹ (Fig. 4). Clearly, more work is required to assess the anthropogenic component of CH₄ emissions from lakes and ponds.

It remains difficult to parse out an anthropogenic component to stream and river CH₄ fluxes. Although some studies have noticed a temperature dependence with stream sediments (Comer-Warner et al., 2018; Zhu et al., 2020), Rocher-Ros et al. (2023) noted a small temperature dependence of CH₄ emissions in streams and rivers compared to other freshwater ecosystems, potentially due to the many other external processes affecting fluxes in these dynamic flowing ecosystems. Urbanisation can lead to elevated river CH₄ emissions, particularly in regions with elevated organic matter and nutrient loading due to limited wastewater treatment (Begum et al., 2021; Nirmal Rajkumar et al., 2008; Wang et al., 2021a). Some studies have found agricultural streams and ditches can have higher effluxes due to inputs of fine sediments (Comer-Warner et al., 2018; Crawford and Stanley, 2016), organic carbon, and nutrients (Borges et al., 2018) that lead to in situ methane production. Furthermore, the creation of drainage ditches in organic soils taps CH₄-rich waters from water-logged horizons and increases the emissions from ex situ sources (Peacock et al., 2021). However, limitations in both the geographic coverage of existing ditch emission estimates and our ability to estimate the global surface area of ditches preclude their inclusion in this budget. Finally, extremely high rates of CH₄ emission have been linked to ongoing permafrost thaw in Asia's Qinghai–Tibet Plateau (Zhang et al., 2020). However, the loss and disconnection of wetlands to rivers may have resulted in a decrease in the input of dissolved CH₄ from this source. A recent expert elicitation (Rosentreter, et al., 2024) reported that 35 % of all inland freshwater sources were anthropogenic; given that some of the river flux is from upstream reservoirs, we assign a 30 % anthropogenic contribution to the stream and river flux (9 Tg CH₄ yr⁻¹, Fig. 4), which approximates the expert elicitation via the impact of eutrophication and urban influences.

Combination (lakes, ponds, reservoirs, streams and rivers, farm ponds)

Combining the aforementioned emissions from lakes $>$ 0.1 km² (33 [13–53] Tg CH₄ yr⁻¹), small lakes and ponds $<$ 0.1 km² (20 [6–33] Tg CH₄ yr⁻¹), reservoirs (25 [13–65] Tg CH₄ yr⁻¹), streams and rivers (29 [12–46] Tg CH₄ yr⁻¹) and farm ponds (5 Tg CH₄ yr⁻¹) leads to a total of $\sim$ 112 Tg CH₄ yr⁻¹ from freshwater systems, with a range of [49–202] Tg CH₄ yr⁻¹. This estimate is about 50 Tg lower than in Saunois et al. (2020) and is broadly consistent with the recent regionalised estimate by Lauerwald et al. (2023b) compiled for the Regional Carbon Cycle Assessment and Processes (RECCAP2, https://www.globalcarbonproject.org/reccap/, last access: 1 April 2025; 103 Tg CH₄ yr⁻¹, IQR $=$ 82.1–134.8). The updated budget from these ecosystems and their anthropogenic components are represented in Fig. 4. The gridded products for emissions from lakes and ponds by Johnson et al. (2022b), from reservoirs by Johnson (2021), and from streams and rivers by Rocher-Ros (2023) have been combined into a single map presented in Fig. 5.

Figure 5

Methane emissions (mg CH₄ m⁻² d⁻¹) from four natural and indirect anthropogenic sources: inland freshwaters (including lakes, ponds (Johnson et al., 2022b), reservoirs (Johnson et al., 2021; Johnson, 2021), and stream and rivers (Rocher-Ros et al., 2023; Rocher-Ros, 2023) with a global total scaled to 89 Tg yr⁻¹), geological sources (Etiope et al., 2019), termites (this study), and oceans (Weber et al., 2019).

[Figure omitted. See PDF]

Double counting inland freshwater ecosystems in the bottom-up estimates

To address the differences found between bottom-up and top-down CH₄ budgets, and to acknowledge advances in addressing the central issue of double counting CH₄ emissions for inland freshwater ecosystems, we introduce here a new correction term. Historically, the bottom-up estimate of global CH₄ emissions has been higher than the top-down estimate, first recognised in Kirschke et al. (2013) and confirmed in Saunois et al. (2016, 2020). The larger bottom-up emissions estimate has been partly attributed to double counting vegetated wetland emissions with inland freshwater emissions (including lakes, ponds, rivers, streams, and reservoirs) and the emissions of CH₄ produced in vegetated wetlands and then transported via aquatic processes and emitted from inland freshwaters (Pangala et al., 2017; Kirk and Cohen, 2023). The Saunois et al. (2020) CH₄ budget addressed the issue of double counting through the use of a revised vegetated wetland area data set, WAD2M v1.0 (Zhang et al., 2021a), which removed inland waters from the SWAMPS (Jensen and McDonald, 2019) surface–inundation data set, allowing for independent vegetated wetlands and inland freshwater CH₄ emissions to be compiled. However, the Saunois et al. (2020) CH₄ budget still had a $\sim$ 150 Tg CH₄ yr⁻¹ difference between bottom-up and top-down estimates. In this budget, we refined the vegetated wetland area data set with WAD2M v2.0 (see Sect. 3.2.1, where HydroLAKES is used to remove lakes and ponds $>$ 0.1 km²). Additionally, we applied numbers from peer-reviewed publications and expert elicitation to account for lateral CH₄ flux emissions. This most recent bottom-up budget estimates 159 [119–203] Tg CH₄ yr⁻¹ from vegetated wetlands for 2010–2019 and 112 Tg CH₄ yr⁻¹ from inland freshwaters, which includes 83 Tg CH₄ yr⁻¹ from lakes, ponds, and reservoirs and 29 Tg CH₄ yr⁻¹ from rivers and streams, leading to a combined wetland and inland freshwater flux of 271 Tg CH₄ yr⁻¹. Here, we propose a correction of 20 Tg CH₄ yr⁻¹ to account for double counting of small lakes and ponds ($<$ 0.1 km²) that are likely included in our vegetated wetlands estimate and removing 1–3 Tg CH₄ yr⁻¹ from river emissions due to lateral transport of CH₄ originating in adjacent vegetated wetlands. The river flux correction arises from assuming that for catchments with $>$ 10 % wetlands, rivers provide 5 %–10 % of vegetated CH₄ emissions. The total double counting correction term of 23 Tg CH₄ reduces the bottom-up budget for combined wetlands and inland waters from 271 to 248 Tg CH₄ yr⁻¹ (see Fig. 4 and Table 3). Comparing the 2000–2009 decadal emissions from wetlands and inland freshwater ecosystems across the last three previous assessments of the budget shows a significant downward revision with 305 (183 $+$ 122) Tg CH₄ yr⁻¹, 356 (147 $+$ 209) Tg CH₄ yr⁻¹ and 248 (159 $+$ 112 $-$ 23) Tg CH₄ yr⁻¹ (respectively from Saunois et al. 2016, 2020, and this work).

Finally, it is worth noting that inland freshwater ecosystems can overlap with geological seepage systems in some areas; i.e. they may occur in correspondence with geological structures that emit fossil (microbial, thermogenic, or abiotic) CH₄ generated in the Earth's crust. Examples have been documented in the Fisherman Lake in Canada (Smith et al., 2005), in Lake Baikal (Schmid et al., 2007), and in rice paddies in Japan (Etiope et al., 2011). Thus, some gas emissions in freshwater environments, particularly as bubble plumes, can be incorrectly attributed to modern biological (ecosystem) activities if appropriate isotopic and molecular analyses are not performed.

3.2.3 Onshore and offshore geological sources

Significant amounts of CH₄, produced within the Earth's crust, naturally migrate to the atmosphere through tectonic faults and fractured rocks. Major emissions are related to hydrocarbon formation in sedimentary basins (microbial and thermogenic methane), through continuous or episodic exhalations from onshore and shallow marine hydrocarbon seeps and through diffuse soil microseepage (Etiope, 2015). Specifically, five source categories have been considered. Four are onshore sources: gas–oil seeps, mud volcanoes, diffuse microseepage, and geothermal manifestations including volcanoes. One source is offshore: submarine seepage, which may include the same types of gas manifestations occurring on land. Etiope et al. (2019) have produced the first gridded maps of geological CH₄ emissions and their isotopic signature for these five categories, with a global total of 37.4 Tg CH₄ yr⁻¹ (reproduced in Fig. 5). However, these maps are based on incomplete data on geological sites due to missing information and difficulties in defining all current geological emitting sites. Combining the best estimates for the five categories of geological sources (from grid maps or from previous statistical and process-based models), the breakdown by category reveals that onshore microseepage dominates (24 Tg CH₄ yr⁻¹), with the other categories having similar smaller contributions: as mean values, 4.7 Tg CH₄ yr⁻¹ for geothermal manifestations, about 7 Tg CH₄ yr⁻¹ for submarine seepage, and 9.6 Tg CH₄ yr⁻¹ for onshore seeps and mud volcanoes. These values lead to a global bottom-up geological emission mean of 45 [27–63] Tg CH₄ yr⁻¹ (Etiope and Schwietzke, 2019).

While all bottom-up and some top-down estimates, following different and independent techniques from different authors, consistently suggest a global geo-CH₄ emission on the order of 40–50 Tg yr⁻¹, the radiocarbon (¹⁴C–CH₄) data in ice cores reported by Hmiel et al. (2020) appear to give a much lower estimate, with a minimum of about 1.6 Tg CH₄ yr⁻¹ and a maximum value of 5.4 Tg CH₄ yr⁻¹ (95 % confidence) for the pre-industrial period. Dyonisius et al. (2020) also suggest a low range of geological emissions over the last deglaciation period and for the Late Holocene (0–10 Tg CH₄ yr⁻¹). The discrepancy between Hmiel et al. (2020) and all other estimates has been discussed in Thornton et al. (2021), which demonstrated that the global near-zero geologic CH₄ emission estimate in Hmiel et al. (2020) is incompatible with the sum of multiple independent bottom-up estimates, based on a wide variety of methodologies, from individual natural geological seepage areas: for example, from the Black Sea (up to 1 Tg CH₄ yr⁻¹), the Eastern Siberian Arctic Shelf (ESAS; up to 4.6 Tg CH₄ yr⁻¹, referring mostly to thermogenic gas), onshore Alaska (up to 1.4 Tg CH₄ yr⁻¹), and a single seepage site in Indonesia (releasing 0.1 Tg CH₄ yr⁻¹ as estimated by satellite measurement) (see Thornton et al., 2021, and references therein). Jackson et al. (2020) expressed doubt about the low Hmiel et al. (2020) estimates, noting that they are difficult to reconcile with the results of many other researchers and with bottom-up approaches in general. This discrepancy highlights another main unresolved uncertainty in the methane budget and calls for further investigations to reconcile the different estimates and reduce the uncertainty on geological emissions. Waiting for further investigation to better understand discrepancies between radiocarbon approaches and other studies, we decided to keep the estimates from Etiope and Schwietzke (2019) for the mean values and associate them with the lowest estimates reported in Etiope et al. (2019), as in Saunois et al. (2020). Thus, we report a total global geological emission of 45 [18–63] Tg CH₄ yr⁻¹, with a breakdown between offshore emissions of 7 [5–10] Tg CH₄ yr⁻¹ and onshore emissions of 38 [13–53] Tg CH₄ yr⁻¹ (Table 3), similar to Saunois et al. (2020). This bottom-up estimate is slightly lower than in the Saunois et al. (2016) budget mostly due to a reduction of estimated emissions of onshore and offshore seeps (see Sect. 3.2.6 for more offshore contribution explanations).

Table 4

Top-down studies used here with their contribution to the decadal and yearly estimates noted. For decadal means, top down studies must provide at least 8 years of data over the decade to contribute to the estimate. Details on each inverse system and inversions are provided in Tables S8 to S11 in the Supplement.

Model	Institution	Observation	Time	Number of	2000–2009	2010–2019	2020	References
		used	period	inversions
Carbon Tracker – Europe CH4	FMI	Surface stations	2000–2020	4	y	y	y	Tsuruta et al. (2017)
LMDz-CIF	LSCE/CEA	Surface stations	2000-2020	4	y	y	y	Thanwerdas et al. (2022a)
LMDz-PYVAR	LSCE/CEA/THU	GOSAT Leicester v9.0	2010–2020	4	n	y	y	Zheng et al. (2018a, b, 2019)
MIROC4-ACTM	JAMSTEC	Surface stations	2000–2020	5	y	y	y	Patra et al. (2018); Chandra et al. (2021)
NISMON-CH4	NIES/MRI	Surface stations	2000–2020	2	y	y	y	Niwa et al. (2022, 2024)
NIES-TM-FLEXPART (NTFVAR)	NIES	Surface stations	2000–2020	2	y	y	y	Maksyutov et al. (2021), Wang et al. (2019a)
NIES-TM-FLEXPART (NTFVAR)	NIES	GOSAT NIES L2 v02.95	2010–2020	1	n	y	y	Maksyutov et al. (2021), Wang et al. (2019a)
TM5-CAMS	TNO/VU	Surface stations	2000–2020	1	y	y	y	Segers et al. (2022)
TM5-CAMS	TNO/VU	GOSAT ESA/CCI v2.3.8 (combined with surface observations)	2010–2020	1	n	y	y	Segers et al. (2022)
Total number of runs				24	18	24	24

Table 5

Global and latitudinal total methane emissions (in Tg CH₄ yr⁻¹), as decadal means (2000–2009 and 2010–2019) and for the year 2020 from this work using bottom-up and top-down approaches. Global and latitudinal emissions for 2000–2009 are also compared with Saunois et al. (2016, 2020) for top-down and bottom-up approaches when available. Uncertainties are reported as [min–max] range. The mean, minimum, and maximum values are calculated while discarding outliers, for each category of source and sink. As a result, discrepancies may occur when comparing the sum of categories and their corresponding total due to differences in outlier detections. Differences of 1 Tg CH₄ yr⁻¹ in the totals can also occur due to rounding errors. For the latitudinal breakdown, bottom-up anthropogenic estimates are based only on the gridded products (see Table 1). As a result, the total from the latitudinal breakdown (line called “This work (gridded BU products only”)) is slightly different from the values provided in Table 3 and recalled in the line “This work (all BU products)”. BU stands for bottom-up. S2020 and S2016 refer to Saunois et al. (2020) and Saunois et al. (2016) respectively.

Period	2000–2009		2010–2019		2020
Approach	Bottom-up	Top-down	Bottom-up	Top-down	Bottom-up	Top-down
Global
This work (all BU products)	638 [485–813]	543 [526–558]	669 [512–849]	575 [553–586]	685 [540–865]	608 [581–627]
This work (gridded BU products only)	642 [501–809]		676 [526–845]		691 [565–862]
S2020	703 [566–842]	547 [524–560]	–	–	–	–
S2016	719 [583–861]	552 [535–566]	–	–	–	–
90° S–30° N
This work	367 [254–487]	337 [311–361]	388 [275–503]	364 [337–390]	395 [292–521]	386 [353–425]
S2020	408 [322–532]	346 [320–379]	–	–	–	–
S2016	–	356 [334–381]	–	–	–	–
30–60° N
This work	234 [169–335]	182 [162–197]	250 [184–345]	187 [160–204]	256 [186–356]	197 [170–215]
S2020	252 [202–342]	178 [159–199]	–	–	–	–
S2016	$-$	176 [159–195]	–	–	–	–
60–90° N
This work	42 [22–79]	26 [22–33]	38 [17–73]	24 [18–29]	39 [17–74]	25 [20–32]
S2020	42 [28–70]	23 [17–32]	–	–	–	–
S2016	$\mathbf{-}$	20 [15–25]	–	–	–	–

3.2.4 Termites

Termites are decomposers playing a central role in ecosystem nutrient fluxes at tropical and subtropical latitudes, in particular (Abe et al., 2000). Termites represent a natural CH₄ source due to methanogenesis occurring in their hindgut during the symbiotic metabolic breakdown of lignocellulose (Sanderson, 1996; Brune, 2014). The upscaling of CH₄ emissions from termites from site to global level is characterised by high uncertainty (Sanderson, 1996; Kirschke et al., 2013; Saunois et al., 2016) due to the combination of factors that need to be considered and the scarcity of information for each of these factors for global upscaling. Needed data include termite biomass density (Sanderson, 1996), species distribution within and among ecosystems (Sugimoto et al., 1998), variation in termite CH₄ emission rates per species and dietary group (Sanderson, 1996), and the role played by the termite mound structure in affecting the fraction of produced CH₄ that is effectively released into the atmosphere (Sugimoto et al., 1998; Nauer et al., 2018). In Kirschke et al. (2013) and Saunois et al. (2016), a global upscaling of termite CH₄ emissions was proposed, where CH₄ emissions, $E_{{\mathrm{CH}}_{\mathrm{4}}}$ (kg CH₄ ha⁻¹ yr⁻¹), were estimated as the product of three terms: termite biomass (Bio_TERM g fresh weight m⁻²), a scalar correction factor (LU) expressing the effect of land use/cover change on termite biomass density, and a termite CH₄ emission factor (EF_TERM, $\mathrm{\mu }\mathrm{g}$ CH₄ g⁻¹ Bio_TERM h⁻¹). The approach between the two re-analyses of CH₄ emissions varied only for the data sources of gross primary productivity (GPP) and land use, which were used to attribute biomass values of termite per ecosystem surface unit, in order to cover different time spans: 1980s, 1990s, and 2000s in Kirschke et al. (2013) and 2000–2007 and 2010–2016 in Saunois et al. (2016). For the present update, additional changes have been introduced compared with the previous versions. Here we summarise the key data used for the new upscaling. CH₄ fluxes were modelled between 45° S and 45° N and within 35° S and 35° N. The termite biomass density, Bio_TERM, for tropical ecosystems was estimated as a function (Kirschke et al., 2013; Bio${}_{\mathrm{TERM}}=\mathrm{1.21}\times \exp (\mathrm{0.0008}\cdot \mathrm{GPP}$)) of the gross primary production (GPP; g C m⁻² yr⁻¹) using the 0.25° native resolution VODCA2GPP data set covering the period 2001–2020 (Wild et al., 2022). Wetlands, barren areas, water bodies, and artificial surfaces were excluded from this estimation and set as no data (no emissions). The scalar correction factor LU of 0.4 (60 %) for agricultural areas (i.e. croplands) (Kirschke et al., 2013) was applied to the GPP value of the nearest natural areas to account for anthropic disturbance. The annual (2001–2020) land cover information was obtained from the MODIS Terra$+$Aqua Combined Land Cover product (MCD12C1v006; https://lpdaac.usgs.gov/products/mcd12c1v006/, last access: 1 April 2025), using the International Geosphere-Biosphere Programme (IGP) classification with a 0.05° spatial resolution. For desert and arid lands, within 35° S and 35° N, a fixed Bio_TERM value of 1.56 g m⁻² was instead used (Sanderson, 1996; Heděnec et al., 2022). Similarly, fix values from the few available studies reported in literature were used to estimate Bio_TERM between 35–45° N and 35–45° S as follows: 1.83 g m⁻² for temperate forests and grasslands (Wood and Sands, 1978; Petersen and Luxton, 1982; Sanderson, 1996; Bignell and Eggleton, 2000; King et al., 2013; conversion factor from dry to fresh biomass is 0.27 from Petersen and Luxton, 1982), 5.3 g m⁻² for shrublands and Mediterranean areas of Australia (Sanderson, 1996), and 1.09 g m⁻² for the other Mediterranean shrubland ecosystems (Heděnec et al., 2022). Other climates and land covers were set as no data. Climate zoning was defined using the Köppen–Geiger climate zone data set (Beck et al., 2018); this product is representative of the 1980–2016 time period and has a 0.0083° native resolution. The EF_TERM was revised compared with previous estimates (Kirschke et al., 2013; Saunois et al., 2016), in order to consider the different distribution of termite families and subfamilies in the different continents and ecosystems, characterised by different feeding habits and nest typologies, as reported by Sugimoto et al. (1998), which might influence the EF. The species of each family and subfamily of the two major groups of lower and higher termites, listed by Sugimoto et al. (1998), were associated with EF values based on emissions from in vitro experiments as reported by Sanderson (1996) and Eggleton et al. (1999), to which a correction factor (cf_MOUND) of 0.5 (Nauer et al., 2018; Chiri et al., 2020, 2021) was applied in order to take into account the mound effect on the CH₄ produced by termites, once inside the nest. The average EF_TERM for tropical and temperate areas was hence estimated as the weighted EF_TERM derived from the product of the percentage weight of each family or subfamily of termites in the “community composition” in each geographical area and ecosystem (Sugimoto et al. (1998; Table 6), the respectively calculated EF of each family or subfamily, and a scalar or correction factor which considers the nest type (as in Table 5 from Sugimoto et al., 1998). For desert and arid lands and temperate areas, which were not reported in Sugimoto et al. (1998), EF rates were calculated directly from data reported in literature for the most representative species, which were the genus Amitermes for the former (EF from data by Sanderson, 1996; Eggleton et al., 1999; Jamali et al., 2011) and the genus Reticulitermes (family Rhinotermitidae) for the latter (EF from data by Odelson and Breznak, 1983; Sanderson, 1996; Eggleton et al., 1999; Myer et al., 2021). The following EF_TERMs were hence obtained to scale up emissions: 3.26 $\pm$ 1.79 $\mathrm{\mu }$g CH₄ g⁻¹ termite h⁻¹ (28.56 mg CH₄ g⁻¹ termite yr⁻¹) for tropical ecosystems; 1.82 $\pm$ 1.54 $\mathrm{\mu }$g CH₄ g⁻¹ termite h⁻¹ for temperate forests, grasslands, and Mediterranean areas; and 1.24 $\pm$ 1.22 $\mathrm{\mu }$g CH₄ g⁻¹ termite h⁻¹ for deserts and arid lands (warm climate). Annual CH₄ fluxes were computed for all the years from 2001 to 2020, producing 20 global maps at 0.05° resolution of yearly total emissions. A further map of the estimated error representative of the entire time period was elaborated at the same resolution as the emissions data set.

Table 6

Latitudinal methane emissions (in Tg CH₄ yr⁻¹) for the last decade 2010–2019, based on top-down and bottom-up approaches. Uncertainties are reported as [min–max] range of reported studies. The mean, minimum, and maximum values are calculated while discarding outliers, for each category of source and sink. As a result, discrepancies may occur when comparing the sum of categories and their corresponding total due to differences in outlier detections. Differences of 1 Tg CH₄ yr⁻¹ in the totals can also occur due to rounding errors. For bottom-up approaches, natural and indirect anthropogenic sources are estimated based on available gridded data sets (see text Sect. 5.2). As some emissions are missing gridded products (wild animals, permafrost, and hydrates), discrepancies may occur in terms of totals proposed in Table 3. Bottom-up direct anthropogenic estimates are based only on the gridded products (see Table 1).

Latitudinal band	90° S–30° N		30–60° N		60–90° N
Approach	Bottom-up	Top-down	Bottom-up	Top-down	Bottom-up	Top-down
Natural and indirect anthropogenic sources	178 [95–276]	148 [133–164]	100 [43–188]	42 [36–50]	28 [9–53]	14 [10–21]
Combined wetland   and inland freshwaters	151 [85–234]	128 [112–155]	73 [32–147]	27 [20–42]	24 [9–53]	9 [7–17]
Other natural	27 [11–42]	22 [20–29]	27 [10–41]	19 [16–22]	4 [2–6]	3 [1–5]
Direct anthropogenic sources	210 [180–227]	215 [191–238]	151 [142–157]	144 [121–162]	10 [6–14]	10 [6–16]
Agriculture &   waste	140 [121–150]	150 [135–168]	81 [77–84]	77 [56–88]	1 [1–2]	2 [2–2]
Fossil fuels	52 [44–65]	46 [36–62]	65 [61–71]	61 [50–69]	7 [4–10]	7 [3–13]
Biomass & biofuel   burning	22 [18–30]	19 [16–21]	7 [4–10]	6 [2–7]	1 [0–1]	1 [1–2]
Sum of sources	388 [275–503]	364 [337–390]	250 [184–345]	187 [160–204]	38 [7–73]	24 [18–29]

Termite CH₄ emissions over the period 2001–2020 varied between 9.7–10.8 Tg CH₄ yr⁻¹, with an average of 10.2 $\pm$ 6.2 Tg CH₄ yr⁻¹. Considering a 20-year average, tropical and subtropical moist broadleaf forests contributed to 46 % of the total average flux, while tropical and subtropical grasslands, savannas, and shrublands to another 36 %. In terms of regional contribution, 37.2 % of fluxes were attributed to South America, 31.5 % to Africa, 18.1 % to Asia, 5.5 % to Australia, 7.4 % to North America, and less than 1 % to Europe. The present estimate value is within the range of previous upscaling studies, spanning from 2 to 22 Tg CH₄ yr⁻¹ (Ciais et al., 2013). In this study, we report a decadal value of 10 Tg CH₄ yr⁻¹ with a range of [4–16] (Table 3).

3.2.5 Wild animals

Wild ruminants emit CH₄ through microbial fermentation that occurs in their rumen, similarly to domesticated livestock species (USEPA, 2010b). Using a total animal population of 100–500 million, Crutzen et al. (1986) estimated the global emissions of CH₄ from wild ruminants to be in the range of 2–6 Tg CH₄ yr⁻¹. More recently, Pérez-Barbería (2017) lowered this estimate to 1.1–2.7 Tg CH₄ yr⁻¹ using a total animal population estimate of 214 million (range of 210–219), arguing that the maximum number of animals (500 million) used in Crutzen et al. (1986) was poorly justified. Moreover Pérez-Barbería (2017) also stated that the value of 15 Tg CH₄ yr⁻¹ found in the last IPCC reports is much higher than their estimate because this value comes from an extrapolation of Crutzen's work for the last glacial maximum when the population of wild animals was much larger, as originally proposed by Chappellaz et al. (1993). Recently, based on the modelling of grassland extent, Kleinen et al. (2023) also suggest that the populations of wild animals during the last glacial maximum proposed by Crutzen et al. (1986) and further used by Chappellaz et al. (1993) were overestimated. However, the estimate of 1–3 Tg CH₄ yr⁻¹ seems underestimated when considering that Hempson et al. (2017) found actual CH₄ emissions from African wildlife alone to be around 9 Tg CH₄ yr⁻¹ but without discussing the uncertainty of this value. As a result, high uncertainty remains and recalls the need for further investigation of this natural source of CH₄.

Based on these findings and waiting for further global estimates, the range adopted in this updated CH₄ budget is 2 [1–3] Tg CH₄ yr⁻¹ (Table 3).

3.2.6 Coastal and oceanic sources

Coastal and oceanic sources comprise CH₄ release from estuaries, coastal vegetated habitats, and marine waters, including seas and oceans. Possible sources of coastal and oceanic CH₄ include (1) in situ biogenic production through various pathways in oxygenated sea surface waters (Oremland, 1979; Karl et al., 2008; Lenhart et al., 2016; Repeta et al., 2016), a flux that can be enhanced in the coastal ocean because of submarine groundwater discharge (USEPA, 2010b); (2) production from shallow and marine (bare and vegetated) sediments including free gas or destabilised hydrates and thawing subsea permafrost containing modern (¹⁴C-bearing) microbial gas; and (3) geological marine seepage (see also Sect. 3.2.3), including hydrates, containing fossil (¹⁴C-free) microbial or thermogenic CH₄. CH₄ produced in marine sediments and seabed CH₄ seepage can be transported across the water column to the sea surface by upwelling waters (once at the surface methane can cross the sea–air interface via diffusion) and gas bubble plumes (for instance from geological marine seeps; e.g. Judd, 2004; Etiope et al., 2019). Gas bubble plumes generally reach the atmosphere in relatively shallow waters ($<$ 400 m) of continental shelves depending on the intensity of the events (e.g. Westbrook et al., 2009); however massive deep-water seepage events could contribute a significant amount of CH₄ to the atmosphere, even from depths $>$ 1000 m (e.g. Schmale et al., 2005.; Greinert et al., 2006; Solomon et al., 2009). In coastal vegetated habitats, CH₄ can also be transported to the atmosphere through the aerenchyma of emergent aquatic plants (Purvaja et al., 2004).

We distinguish between coastal and oceanic “geological” and “modern biogenic” CH₄ sources. Coastal and oceanic “geological” emissions refer to CH₄ seepage from the Earth's crust (mostly in hydrocarbon-rich sedimentary basins), which is typically evaluated by combining geochemical analyses (isotopic and molecular, including radiocarbon, ¹⁴C, analyses) and geological observations (degassing along faults, seeps, mud volcanoes). Geological emissions do not contain modern biogenic gas that is fossil (¹⁴C-free). Coastal and oceanic “biogenic” CH₄ refers to CH₄ formed in situ in coastal and marine sediments and in the water column by recent or modern microbial activity (therefore with measurable amounts of radiocarbon (¹⁴C)). To avoid double counting, we assume that all diffusive CH₄ emissions outside of geological seepage regions (identified in global grid maps; Etiope et al., 2019) are fuelled by biogenic CH₄. Finally, we briefly discuss the case of CH₄ hydrates, which can be considered either a “geological” source when they host fossil CH₄ or a “biogenic” source when they host modern CH₄.

Coastal and oceanic modern biogenic methane emissions

Area-integrated diffusive modern biogenic CH₄ emissions from coastal ecosystems are 1–2 magnitudes lower than from inland freshwaters but significantly higher than biogenic emissions from the open ocean (Rosentreter et al., 2021; Rosentreter et al., 2023; Weber et al., 2019). Particularly the shallow vegetated coastline fringed by mangroves, salt marshes, and seagrasses is a CH₄ hotspot in the coastal ocean, characterised by significantly higher flux densities than other coastal settings such as estuaries or the continental shelves (Rosentreter et al., 2021; Rosentreter et al., 2023). Coastal ecosystems are thus being increasingly recognised as weak global sources to the atmosphere (Weber et al., 2019; Saunois et al., 2020; Rosentreter et al., 2021). Hydrogenotrophic and acetoclastic methanogenesis are largely outcompeted by sulfate reduction in coastal/marine sediments, which is often shown by a decreasing trend of CH₄ concentrations with increasing salinity from upper tidal (low salinity) to marine (high salinity) regions. Much of the CH₄ produced below the sulfate-reduction zone is indeed re-oxidised by sedimentary anaerobic methane oxidation or re-oxidised in the water column, leading to small emissions despite much larger production (Knittel and Boetius, 2009; Regnier et al., 2011). Methylated compounds such as methylamines and methyl sulfides are non-competitive substrates that are exclusively used by methanogens; therefore, methylated methanogenesis can occur in coastal regions with high sulfate concentrations, for example, in organic-rich (Maltby et al., 2018), vegetated (Schorn et al., 2022), and hypersaline coastal sediments (Xiao et al., 2018). Coastal CH₄ can be driven by the exchange of pore water or groundwater (high in CH₄) with coastal surface waters in tidal systems, referred to as tidal pumping (Ovalle et al., 1990; Call et al., 2015). Anthropogenic impacts such as wastewater pollution and land use change can increase CH₄ fluxes in estuaries (Wells et al., 2020). A large increase of CH₄ emissions follows the conversion of natural coastal habitats to aquaculture farms (Yuan et al., 2019; Yang et al., 2022).

Currently available global modern biogenic CH₄ flux data show high spatiotemporal variability within and between coastal systems because of the overall global paucity of data. Therefore, global estimates have high uncertainties and show large ranges in both empirical (Rosentreter et al., 2021) and machine-learning-based approaches (Weber et al., 2019). According to a recent data-driven metadata analysis, global estuaries, including tidal systems and deltas, lagoons, and fjords, are estimated to emit a median (Q1–Q3) of 0.25 (0.07–0.46) Tg CH₄ yr⁻¹ (Rosentreter et al., 2023). Coastal vegetation, including mangrove forests, salt marshes, and seagrasses, is estimated to emit 0.77 (0.47–1.41) Tg CH₄ yr⁻¹, which is 3 times more than global estuaries (Rosentreter et al., 2023). The combined median (Q1–Q3) emission of 1.01 (0.54–1.87) Tg CH₄ yr⁻¹ for coastal vegetation and estuaries by Rosentreter et al. (2023) is lower than the recent observation-based global synthesis including tidal flats and aquaculture ponds (median 1.49 (0.22–6.48) Tg CH₄ yr⁻¹) by Rosentreter et al. (2021). Total shallow coastal modern biogenic CH₄ emissions based on existing data including emissions from estuaries, coastal vegetation (Rosentreter et al., 2023), tidal flats, and human-made coastal aquaculture ponds (Rosentreter et al., 2021) amount to a median (Q1–Q3) of 1.8 (0.59–5.57) Tg CH₄ yr⁻¹. This range is about 3–4 times lower than the earlier global assessment by Borges and Abril (2011) and also lower than the value of 4–5 Tg CH₄ yr⁻¹ reported in the previous CH₄ budget for inner and outer estuaries including marshes and mangroves (Saunois et al., 2020), which was based on a significantly smaller data set ($n=\mathrm{80}$) and larger estuarine surface areas (Laruelle et al., 2013) than used here (Laruelle et al., 2025).

The nearshore (0–50 m), inner-shelf diffusive modern biogenic CH₄ flux of a median (Q1–Q3) of 1.33 (0.93–2.10) Tg CH₄ yr⁻¹ by Weber et al. (2019) based on machine-learning is similar to the combined shallow coastal (estuaries and coastal vegetation) median by Rosentreter et al. (2021, 2023). Adding the diffusive modern biogenic CH₄ flux for the outer shelf (50–200 m) (median (Q1–Q3) of 0.54 (0.40–0.73) Tg CH₄ yr⁻¹) and for the slope (200–2000 m) (median (Q1–Q3) of 0.28 (0.22–0.37) Tg CH₄ yr⁻¹) (Weber et al., 2019) and excluding geological seepage regions (Etiope et al., 2019; see below) gives a total median (Q1–Q3) of 3.95 (2.14–8.77) Tg CH₄ yr⁻¹ for combined coastal shallow, nearshore, outer shelf, and slope diffusive modern biogenic CH₄ emissions. The previous budget by Saunois et al. (2020) also included poorly constrained emissions (upper bound value: 1–2 Tg CH₄ yr⁻¹) from large river plumes protruding onto the shelves. However, here we assume that emissions from large river plumes are accounted for in the nearshore and outer shelf estimates by Weber et al. (2019). Area-integrated diffusive CH₄ emissions from the open ocean and deep seas ($>$ 2000 m) are much lower than from other coastal systems but amount to a median (Q1–Q3) of 0.91 (0.75–1.12) Tg CH₄ yr⁻¹ because of the large surface area of the open ocean ($>$ 300 $\times$ 10⁶ km²) (Weber et al., 2019). Overall, these marine biogenic emissions are sustained by a mixture of sedimentary production and in situ production in the sea surface layers (including the methylphosphonate pathway) (e.g. Karl et al., 2008; Repeta et al., 2016; Resplandy et al., 2024). The total coastal and ocean diffusive modern biogenic emissions retained here amount to 5 (3–10) Tg CH₄ yr⁻¹ (Table 3).

Coastal and oceanic geological methane emissions

Submarine geological CH₄ emission is the offshore component of the general geological emissions of natural gas from the Earth's crust (Judd, 2004; Etiope, 2009; Etiope et al., 2019). The onshore components include terrestrial seeps, mud volcanoes, microseepage, and geothermal manifestations, addressed in Sect. 3.2.3. Natural gas seeping at the seabed as bubble plumes can reach the surface in relatively shallow waters ($<$ 400 m) but CH₄-rich bubble plumes reaching the atmosphere from depths $>$ 500 m have been observed in some cases (e.g. Solomon et al., 2009), and upwelling of bottom marine waters can, in theory, transport geological CH₄ (dissolved) to the surface from any depth. This represents, however, a small and poorly known fraction of geological CH₄ emission. Geological CH₄ can be either microbial or thermogenic, produced throughout diverse geological periods in hydrocarbon source rocks in sedimentary basins (therefore it is always fossil, ¹⁴C-free). The seepage at the seafloor is typically related to tectonic faults, sometimes forming mud diapirs and mud volcanoes (Mazzini and Etiope, 2017). Published estimates of geological CH₄ submarine emissions range from 3 to 20 Tg yr⁻¹, with a best guess of 7 Tg yr⁻¹ (Etiope and Schwietzke, 2019; Etiope et al., 2019, and references therein).

Here, the diffusive geological CH₄ emissions are estimated at 0.16 (0.11–0.24) Tg CH₄ yr⁻¹ for nearshore (0–50 m), 0.03 (0.02–0.05) Tg CH₄ yr⁻¹ for outer shelf (50–200 m), and 0.02 (0.01–0.03) Tg CH₄ yr⁻¹ for slope (200–2000 m) by calculating the fraction of the Weber et al. (2019) diffusive fluxes that occur within the identified geological seepage regions from Etiope et al. (2019). No geological seepage regions were identified in the open ocean and deep seas ($>$ 2000 m).

In this study, we consider the ebullition flux as geologically sourced CH₄. While modern biogenic CH₄ gas production appears ubiquitous in shallow sediments (Fleischer et al., 2001; Best et al., 2006), no global data set is currently available to estimate the biogenic ebullition CH₄ flux to the atmosphere. Omission of this flux thus constitutes a significant knowledge gap in the coastal and oceanic CH₄ budget. Global geological CH₄ ebullition from continental shelf and slope, referring only to depths $<$ 200 m, were estimated at 5.06 (1.99–8.16) Tg CH₄ yr⁻¹ (Weber et al., 2019). This estimate is based on prior estimates of the geological flux from the seafloor (Hovland et al., 1993) and bubble transfer efficiency to the ocean surface (McGinnis et al., 2006). Etiope et al. (2019) estimated a partial fraction of geological emissions in the form of gas bubbles of 3.9 (1.8–6) Tg CH₄ yr⁻¹, only referring to the sum of published estimates from 15 geological seepage regions, which are also deeper than 200 m. Global extrapolation including other 16 identified seepage zones (where flux data are not available) was suggested to be at least 7 (3–10) Tg CH₄ yr⁻¹ (Etiope et al., 2019), and this value coincides with the mean emission value (best guess) derived by combining literature data; see Etiope and Schwietzke (2019) for further details. It is worth noting that the Weber et al. (2019) estimate of 5.06 (1.99–8.16) Tg CH₄ yr⁻¹, which considers only the continental shelf at depths $<$ 200 m, is compatible with the overall submarine emission of 7 (3–10) Tg CH₄ yr⁻¹ (including seeps $>$ 200 m deep) indicated in Etiope and Schwietzke (2019) and Etiope et al. (2019). Although 300–400 m is considered a general depth limit for efficient transport (with limited oxidation and dissolution) of CH₄ bubbles to the atmosphere (e.g. Judd, 2004; Schmale et al., 2005; Etiope et al., 2019), in some cases oil coatings on bubbles inhibit gas dissolution so that CH₄-rich bubbles can reach the atmosphere from depths $>$ 500 m (e.g. Solomon et al., 2009). As mentioned above, a fraction of geological CH₄ released in deep seas (such as in the areas with gas-charged sediments inventoried in Fleischer et al., 2001) can also be transported to the surface by upwelling bottom waters. Further research is needed to better evaluate the atmospheric impact of such deep seeps.

Geological submarine emissions, thus, would amount to 0.21 (0.14–0.32) Tg CH₄ yr⁻¹ in the form of a diffusive flux, while the ebullition flux would be 5.06 (3.01–7.88) Tg CH₄ yr⁻¹, considering only $<$ 200 m deep seepage and 7 (3–10) Tg CH₄ yr⁻¹ considering all data available (Etiope and Schwietzke, 2019). Here, we select the Etiope and Schwietzke (2019) assessment in order to account for all potential seepage areas, including those located at water depths $>$ 200 m. While we use the Etiope and Schwietzke (2019) estimate, we acknowledge that high uncertainty remains, and other studies suggest lower ranges of emissions based on radiocarbon (¹⁴C–CH₄) data in ice cores (e.g. Hmiel et al., 2020). The suggested estimate may overestimate this source and be part of the top-down bottom-up discrepancy as discussed in Sect. 5.1.2.

As a result, here we report a (rounded) median of 12 Tg CH₄ yr⁻¹ with a range of 6–20 Tg CH₄ yr⁻¹ for all coastal and oceanic sources (Table 3).

Methane emissions from gas hydrates

Among the different origins of coastal and oceanic CH₄, hydrates have attracted a lot of attention. CH₄ hydrates (or clathrates) are ice-like crystals formed under specific temperature and pressure conditions (Milkov, 2005). Hydrates may host either modern microbial CH₄, containing ¹⁴C and formed in situ in shallow sediments (this type of hydrates is also called “autochthonous”) or fossil, microbial, or thermogenic CH₄, migrated from deeper sediments, generally from reservoirs in hydrocarbon-rich sedimentary basins (this type of hydrates is also called “allochthonous”; Milkov, 2005; Foschi et al., 2023). The total stock of marine CH₄ hydrates is large but uncertain, with global estimates ranging from hundreds to thousands of Pg CH₄ (Klauda and Sandler, 2005; Wallmann et al., 2012). Note that the highly climate-sensitive subsea permafrost reservoir beneath Arctic Ocean shelves also contributes to the hydrate inventory (Ruppel and Kessler, 2017).

Concerning more specifically atmospheric emissions from marine hydrates, Etiope (2015) points out that current estimates of CH₄ air–sea flux from hydrates (2–10 Tg CH₄ yr⁻¹ in Ciais et al., 2013, or Kirschke et al., 2013) originate from the hypothetical values of Cicerone and Oremland (1988). No experimental data or estimation procedures have been explicitly described along the chain of references since then (Denman et al., 2007; IPCC, 2001; Kirschke et al., 2013; Lelieveld et al., 1998). It was estimated that $\sim$ 473 Tg CH₄ has been released into the water column over 100 years (Kretschmer et al., 2015). Those few teragrams per year become negligible once consumption within the water column has been accounted for. While events such as submarine slumps may trigger local releases of considerable amounts of CH₄ from hydrates that may reach the atmosphere (Etiope, 2015; Paull et al., 2002), on a global scale, present-day atmospheric CH₄ emissions from hydrates do not appear to be a significant source to the atmosphere, and at least formally, we should consider 0 ($<$ 0.1) Tg CH₄ yr⁻¹ emissions.

3.2.7 Terrestrial permafrost

Permafrost is defined as frozen soil, sediment, or rock having temperatures at or below 0 °C for at least 2 consecutive years (Harris et al., 1988). The total extent of permafrost in the Northern Hemisphere is about $\mathrm{14}\times {\mathrm{10}}^{\mathrm{6}}$ km² or 15 % of the exposed land surface (Obu et al., 2019). As the climate warms, a rise in soil temperatures has been observed across the permafrost region, and permafrost thaw occurs when temperatures pass 0 °C, often associated with melting of ice in the ground (Biskaborn et al., 2019). Permafrost thaw is most pronounced in southern and spatially isolated permafrost zones, but it also occurs in northern continuous permafrost (Obu et al., 2019). Thaw occurs either as a gradual, often widespread, deepening of the active layer (surface soils that thaw every summer) or as more rapid localised thaw associated with loss of massive ground ice (thermokarst) (Turetsky et al., 2020). A total of 1000 $\pm$ 200 Pg of carbon can be found in the upper 3 m of permafrost region soils, or 1400–2000 Pg C for all permafrost (Hugelius et al., 2014; Strauss et al., 2021).

The thawing permafrost can generate direct and indirect CH₄ emissions. Direct CH₄ emissions are from the release of CH₄ contained within the thawing permafrost. This flux to the atmosphere is small and estimated to be a maximum of 1 Tg CH₄ yr⁻¹ at present (USEPA, 2010b). Increased release of CH₄ from deep geogenic sources that occurs as seepage along permafrost boundaries and lake beds may also be considered direct, and this is estimated to be 2 $\pm$ 0.4 Tg CH₄ yr⁻¹ (Walter Anthony et al., 2012). Indirect CH₄ emissions are probably more important. They are caused by (1) methanogenesis induced when the organic matter contained in thawing permafrost becomes available for microbial decomposition; (2) thaw-induced soil wetting and changes in land surface hydrology possibly enhancing CH₄ production (McCalley et al., 2014; Schuur et al., 2022); and (3) the landscape topography changes driven by abrupt thaw processes and loss of ground ice, including the formation of thermokarst lakes, hillslope thermokarst, and wetland thermokarst (Turetsky et al., 2020). Such CH₄ production is probably already significant today and is likely to become more important in the future associated with climate change and strong positive feedback from thawing permafrost (Schuur et al., 2022). However, indirect CH₄ emissions from permafrost thawing are difficult to estimate at present, with very few data to refer to, and in any case largely overlap with wetland and freshwater emissions occurring above or around thawing areas. In a recent synthesis of full permafrost region CH₄ budgets for the period 2000–2017, Hugelius et al. (2024) compared CH₄ budgets from bottom-up and top-down (atmospheric inversion models) approaches. They estimate an integrated bottom-up budget of 50 (23, 53; mean upper and lower 95 % CI) Tg CH₄ yr⁻¹, while the top-down estimate is 19 (15, 24) Tg CH₄ yr⁻¹. The bottom-up estimate is based on a combination of data-driven upscaling reported by Ramage et al. (2024) and process-based model estimates for wetland CH₄ flux calculated from model ensembles used in Saunois et al. (2020). The top-down estimate is calculated from ensembles of atmospheric inversion models used in Saunois et al. (2020). Although it is difficult with direct process-attribution, fluxes of ca. 20–30 Tg CH₄ yr⁻¹ in the bottom-up budget are caused by land cover types affected by previous permafrost thaw (thermokarst lakes, wetlands, hillslope). Because pre-thaw land cover types often have near neutral CH₄ balances (Ramage et al., 2024), these fluxes can largely be seen as driven by permafrost thaw; however the thaw may have occurred decades, or even centuries, before today.

Here, we choose to report only the direct emission range of 0–1 Tg CH₄ yr⁻¹ (Table 3), keeping in mind that current wetland, thermokarst lakes, and other freshwater methane emissions already likely include a significant indirect contribution originating from thawing permafrost.

3.2.8 Vegetation

Three distinct pathways for the production and emission of CH₄ by living vegetation are considered here (see Covey and Megonigal (2019) and Bastviken et al. (2023) for extensive reviews). Firstly, plants produce CH₄ through an abiotic photochemical process induced by stress (Keppler et al., 2006). This pathway was initially questioned (e.g. Dueck et al., 2007; Nisbet et al., 2009), and although numerous studies have since confirmed aerobic emissions from plants and better resolved its physical drivers (Fraser et al., 2015), global estimates still vary by 2 orders of magnitude (Liu et al., 2015). This plant source has not been confirmed in the field however, and although the potential implication for the global CH₄ budget remains unclear, emissions from this source are certainly much smaller than originally estimated in Keppler et al. (2006) (Bloom et al., 2010; Fraser et al., 2015). Second, and of clearer significance, plant stems act as “straws”, drawing up and releasing microbially produced CH₄ from anoxic soils (Cicerone and Shetter, 1981; Rice et al., 2010; Nisbet et al., 2009). For instance, in the forested wetlands of Amazonia, tree stems are the dominant ecosystem flux pathway for soil-produced CH₄; therefore, including stem emissions in ecosystem budgets can reconcile regional bottom-up and top-down estimates (Pangala et al., 2017; Gauci et al., 2022). Third, the stems of both living trees (Covey et al., 2012) and dead wood (Covey et al., 2016) provide an environment suitable for microbial methanogenesis. Static chambers demonstrate locally significant through-bark flux from both soil-based (Pangala et al., 2013, 2015) and tree-stem-based methanogens (Pitz and Megonigal, 2017; Wang et al., 2016). A synthesis indicates stem CH₄ emissions significantly increase the source strength of forested wetlands and modestly decrease the sink strength of upland forests (Covey and Megonigal, 2019). Recently, fieldwork suggested that trees may also act as a CH₄ sink (Machacova et al., 2021; Gorgolewski et al., 2023; Gauci et al., 2024). The scientific activity covering CH₄ emissions in forested ecosystems reveals a far more complex story than previously thought, with an interplay of productive/consumptive, aerobic/anaerobic, and biotic/abiotic processes occurring between upland/wetland soils, trees, and atmosphere. Understanding the complex processes that regulate CH₄ source–sink dynamics in forests and estimating their contribution to the global CH₄ budget requires cross-disciplinary research, more observations, and new models that can overcome the classical binary classifications of wetland versus upland forest and of emitting versus uptaking soils (Barba et al., 2019; Covey and Megonigal, 2019). Although we recognise these emissions are potentially large (particularly tree transport from inundated soil), global estimates for each of these pathways remain highly uncertain and/or are currently included here within other flux category sources (e.g. inland waters, wetlands, upland soils).

3.3 Methane sinks and lifetime

CH₄ is the most abundant reactive trace gas in the troposphere, and its reactivity is important to both tropospheric and stratospheric chemistry. The main atmospheric sink of CH₄ ($\sim$ 90 % of the total sink mechanism) is oxidation by the hydroxyl radical (OH), mostly in the troposphere (Ehhalt, 1974). Other losses are by photochemistry in the stratosphere (reactions with chlorine atoms (Cl) and excited atomic oxygen (O(¹D)), oxidation in soils (Curry, 2007; Dutaur and Verchot, 2007), and by photochemistry in the marine boundary layer (reaction with Cl; Allan et al. (2007), Thornton et al. (2010)). Uncertainties in the total sink of CH₄ as estimated by atmospheric chemistry models are of the order of 20 %–40 % (Saunois et al., 2016). It is much less (10 %–20 %) when using atmospheric proxy methods (e.g. methyl chloroform; see below) as in atmospheric inversions (Saunois et al., 2016). In the present release of the global CH₄ budget, we estimate bottom-up CH₄ chemical sinks and lifetime mainly based on global model results from the Chemistry Climate Model Initiative (CCMI) 2022 activity (Plummer et al., 2021) and CMIP6 simulations (Collins et al., 2017).

3.3.1 Tropospheric OH oxidation

OH radicals are produced following the photolysis of ozone (O₃) in the presence of water vapour. OH is destroyed by reactions with carbon monoxide (CO), CH₄, and non-methane volatile organic compounds.

Following the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP), which studied the long-term changes in atmospheric composition between 1850 and 2100 (Lamarque et al., 2013), a new series of experiments was conducted by several chemistry–climate models and chemistry–transport models participating in the Chemistry-Climate Model Initiative (CCMI) (Plummer et al., 2021). Mass-weighted OH tropospheric concentrations do not directly represent CH₄ loss, as the spatial and vertical distributions of OH affect this loss through, in particular, the temperature dependency and the distribution of CH₄ (e.g. Zhao et al., 2019). However, estimating OH concentrations and spatial and vertical distributions is a key step in estimating methane loss through OH. Over the period 2000–2010, the global mass-weighted OH tropospheric concentration is estimated at 13.3 [11.7–18.2] $\times$ 10⁵ molecules cm⁻³ by eight CCMI-2022 models and at 11. 8 [9.4–13.5] $\times$ 10⁵ molecules cm⁻³ by nine models contributing CMIP6 historical run (Collins et al., 2017) (see Table S4). The ranges calculated here are similar to the ones proposed previously in Saunois et al. (2020), where the multi-model mean (11 models) global mass-weighted OH tropospheric concentration was 11.7 $\pm$ 1.0 $\times$ 10⁵ molecules cm⁻³ (range 9.9–14.4 $\times$ 10⁵ molecules cm⁻³; Zhao et al. (2019)) consistent with the previous estimates from ACCMIP (11.7 $\pm$ 1.0 $\times$ 10⁵ molecules cm⁻³, with a range of 10.3–13.4 $\times$ 10⁵ molecules cm⁻³; Voulgarakis et al. (2013) for year 2000) and the estimates of Prather et al. (2012) of 11.2 $\pm$ 1.3 $\times$ 10⁵ molecules cm⁻³. Nicely et al. (2017) attribute the differences in OH simulated by different chemistry–transport models to, in decreasing order of importance, different chemical mechanisms, various treatments of the photolysis rate of O₃, and modelled O₃ and CO. Besides the uncertainty on global OH concentrations, there is an uncertainty in the spatial and temporal distribution of OH. Models often simulate higher OH in the Northern Hemisphere (NH) than in the Southern Hemisphere (SH), leading to a NH $/$ SH OH ratio greater than 1 (e.g. Zhao et al., 2019). However, there is evidence for parity in interhemispheric OH concentrations (Patra et al., 2014), which needs to be confirmed by other observational and model-derived estimates. The analyses of the latest CCMI (Plummer et al., 2021) and CMIP6 (Collins et al., 2017) model outputs show that structural uncertainties in the atmospheric chemistry models remain large, probably due to inherent biases in OH precursors. Such biases have been highlighted in the OH 3D fields simulated by two atmospheric chemistry models (Zhao et al., 2023) and were corrected using OH precursor observations. Such corrections resulted in tropospheric OH mean concentrations lowered by $\mathrm{2}\times {\mathrm{10}}^{\mathrm{5}}$ molecules cm⁻³, leading to around $\mathrm{10}\times {\mathrm{10}}^{\mathrm{5}}$ molecules cm⁻³, and a NH $/$ SH OH ratio closer to 1, in better agreement with methyl chloroform (MCF)-based approaches. This study highlights the need for further improvement of the atmospheric chemistry model.

OH concentrations and their changes can be sensitive to climate variability (e.g. Nicely et al., 2018; Anderson et al., 2021), biomass burning (e.g. Anderson et al., 2024), and anthropogenic emissions of precursors (Peng et al., 2022; Stevenson et al., 2020). OH distributions calculated by chemistry–climate models show large regional differences and various vertical profiles (Zhao et al., 2019). OH changes present also regional differences in the long term (Stevenson et al., 2020). Despite large regional changes, the global mean OH concentration was suggested to have changed only slightly from 1850 to 1980 but followed by strong (9 %) increases up to the present day (Stevenson et al., 2020). This increase simulated by models over 2000–2015 is however not in agreement with observation-based approaches (Thompson et al., 2024; Patra et al., 2021; Nicely et al., 2018; Rigby et al., 2017; Turner et al., 2018) where OH decreases or remains constant over the period. CCMI and CMIP6 models show OH interannual variability ranging from 0.9 % to 1.8 % over 2000–2010 (Table S4), in agreement with the values of interannual variability (IAV) derived from some observationally constrained studies (e.g. Thompson et al., 2024; Montzka et al., 2011) but lower than values deduced from methyl chloroform measurements (Patra et al., 2021; Naus et al., 2021). However, chemistry–climate simulations consider meteorology variability but not fully emission interannual variability (e.g. from biomass burning) and thus are expected to simulate lower OH interannual variability than in reality. Using an empirical model constrained by global observations of O₃, water vapour, CH₄, and temperature as well as the simulated effects of changing NO_x emissions and tropical expansion, Nicely et al. (2017) found an interannual variability in OH of about 1.3 %–1.6 % between 1980 and 2015, in agreement with methyl-chloroform-based estimates (Montzka et al., 2011).

Over 2000–2009, the tropospheric loss (tropopause height at 200 hPa) of CH₄ by OH oxidation derived from the 10 simulations contributing to CMIP6 modelling activity (see Table S5) is estimated at 546 [446–663] Tg CH₄ yr⁻¹ (Table 3), which is similar to the one reported previously in Saunois et al. (2020) from the CCMI model (553 [476–677] Tg CH₄ yr⁻¹) and still slightly higher than the one from the ACCMIP models (528 [454–617] Tg CH₄ yr⁻¹) reported in Kirschke et al. (2013) and Saunois et al. (2016).

For the recent 2010–2019 decade, we report a climatological value based on only five models that contributed to CMIP6 runs (historical run followed by SSP3-7.0 projections starting in 2015, Collins et al., 20217) to acknowledge the impact of the rise in atmospheric methane on the methane chemical sink. Hence, for 2010–2019, we report the climatological value of 563 [462–663] Tg CH₄ yr⁻¹ (Table 3).

3.3.2 Stratospheric loss

In the stratosphere, CH₄ is lost through reactions with excited atomic oxygen O(¹D), atomic chlorine (Cl), atomic fluorine (F), and OH (Brasseur and Solomon, 2005; le Texier et al., 1988). Uncertainties in the chemical loss of stratospheric CH₄ are large, due to uncertain interannual variability in stratospheric transport (Zhang et al., 2023) as well as its chemical interactions and feedbacks with stratospheric O₃ (Morgenstern et al., 2018). Particularly, the fraction of stratospheric loss due to the different oxidants is still uncertain, with possibly 20 %–35 % due to halons, about 25 % due to O(¹D) mostly in the high stratosphere, and the rest due to stratospheric OH (McCarthy et al., 2003).

In this study, six chemistry–climate models that contributed to CMIP6 modelling activities (Table S5) provided estimates of CH₄ chemical loss, including reactions with OH, O(¹D), and Cl; CH₄ photolysis is also included but occurs only above the stratosphere. Considering a 200 hPa tropopause height, these six CMIP6 simulations suggest an estimate of 34 [10–51] Tg CH₄ yr⁻¹ for the CH₄ stratospheric sink for the 2000–2009 decade (Table S5), similar to the value derived from the previous CCMI activity reported in Saunois et al. (2020) (31 [12–41] Tg CH₄ yr⁻¹). The lowest estimate provided by a model (10 Tg CH₄ yr⁻¹) is quite unrealistic and would yield a methane stratospheric lifetime of several hundreds of years. As a result, this outlier is excluded, and we prefer to report a mean of 39 Tg CH₄ yr⁻¹ associated with a range of [27–51] for 2000–2009.

For 2010–2019, we report here a climatological range of 28–43 Tg CH₄ yr⁻¹ associated with a mean value of 37 Tg CH₄ yr⁻¹ (Table 3) based on five models that contributed to CMIP6 runs (historic followed by SSP3-7.0 projections starting in 2015; Table S5).

3.3.3 Tropospheric reaction with Cl

Halogen atoms can also contribute to the oxidation of CH₄ in the troposphere. Allan et al. (2005) measured mixing ratios of methane and ${\mathit{\delta }}^{\mathrm{13}}$C–CH₄ at two stations in the Southern Hemisphere from 1991 to 2003 and found that the apparent kinetic isotope effect (KIE) of the atmospheric CH₄ sink was significantly larger than that explained by OH alone. A seasonally varying sink due to Cl in the marine boundary layer of between 13 and 37 Tg CH₄ yr⁻¹ was proposed as the explanatory mechanism (Allan et al., 2007; Platt et al., 2004). This sink was estimated to occur mainly over coastal and marine regions, where sodium chloride (NaCl) from evaporated droplets of seawater react with NO₂ to eventually form Cl₂, which then UV-dissociates to Cl. However significant production of nitryl chloride (ClNO₂) at continental sites has been recently reported (Riedel et al., 2014) and suggests the broader presence of Cl, which in turn would expand the significance of the Cl sink in the troposphere. Recently, Hossaini et al. (2016), Sherwen et al. (2016), and Wang et al. (2019b, 2021b) have made significant improvements in tropospheric chemistry modelling, and they suggest a contribution to the total oxidation of 2.6 %, 2 %, 1 %, and 0.8 % respectively. These values correspond to a tropospheric CH₄ loss of around 12–13 Tg CH₄ yr⁻¹, 9 Tg CH₄ yr⁻¹, 5 Tg yr⁻¹, and 3 Tg CH₄ yr⁻¹ respectively, much lower than the first estimates by Allan et al. (2007). The recent work of Wang et al. (2021b) is the most comprehensive modelling study and based upon Sherwen et al. (2016) and Wang et al. (2019b). Both the KIE approach and chemistry–transport model simulations carry uncertainties (extrapolations based on only a few sites and use of indirect measurements, for the former and missing sources, coarse resolution, underestimation of some anthropogenic sources for the latter). However, Gromov et al. (2018) found that Cl can contribute only 0.23 % the tropospheric sink of CH₄ (about 1 Tg CH₄ yr⁻¹) in order to balance the global ¹³C(CO) budget (see their Table S1). While tropospheric Cl has a marginal impact on the total CH₄ sink (few percents), it influences more significantly the atmospheric isotopic ${\mathit{\delta }}^{\mathrm{13}}$C–CH₄ signal, and improved estimates of the tropospheric Cl amount should be used for isotopic CH₄ modelling studies (Strode et al., 2020; Thanwerdas et al., 2022b).

Each recent Cl estimate suggests a reduced contribution to the methane loss than previously reported by Allan et al. (2007). As a result, we suggest here to use the mean, minimum, and maximum of the last five estimates published since 2016, leading to a climatological value of 6 [1–13] Tg CH₄ yr⁻¹ (Table 3), thus reducing both the magnitude and the uncertainty range compared to Saunois et al. (2020).

3.3.4 Soil uptake

Unsaturated oxic soils are sinks of atmospheric CH₄ due to the presence of methanotrophic bacteria, which consume CH₄ as a source of energy. Dutaur and Verchot (2007) conducted a comprehensive meta-analysis of field measurements of CH₄ uptake spanning a variety of ecosystems. Extrapolating to the global scale, they reported a range of 36 $\pm$ 23 Tg CH₄ yr⁻¹ but also showed that stratifying the results by climatic zone, ecosystem, and soil type led to a narrower range (and lower mean estimate) of 22 $\pm$ 12 Tg CH₄ yr⁻¹. Modelling studies, employing meteorological data as external forcing, have also produced a considerable range of estimates. Using a soil-depth-averaged formulation based on Fick's law with parameterisations for diffusion and biological oxidation of CH₄, Ridgwell et al. (1999) estimated the global sink strength at 38 Tg CH₄ yr⁻¹, with a range 20–51 Tg CH₄ yr⁻¹ reflecting the model structural uncertainty in the base oxidation parameter. Curry (2007) improved on the latter by employing an exact solution of the one-dimensional diffusion–reaction equation in the near-surface soil layer (i.e. exponential decrease in CH₄ concentration below the surface), a land surface hydrology model, and calibration of the oxidation rate to field measurements. This resulted in a global estimate of 28 Tg CH₄ yr⁻¹ (9-47 Tg CH₄ yr⁻¹), the result reported by Zhuang et al. (2013), Kirschke et al. (2013), and Saunois et al. (2016). Ito and Inatomi (2012) used an ensemble methodology to explore the variation in estimates produced by these parameterisations and others, which spanned the range 25–35 Tg CH₄ yr⁻¹. For the period 2000–2020, as part of the wetland emissions modelling activity, JSBACH (Kleinen et al., 2020) and VISIT (Ito and Inatomi, 2012) models compute a global CH₄ soil uptake to 18 and 35 Tg CH₄ yr⁻¹ respectively. Murguia-Flores et al. (2018) further refined the Curry (2007) model's structural and parametric representations of key drivers of soil methanotrophy, demonstrating good agreement with the observed latitudinal distribution of soil uptake (Dutaur and Verchot, 2007). Their model (MeMo) simulates a CH₄ soil sink of 37.5 Tg CH₄ yr⁻¹ for the period 2010–2019 (Fig. S4), compared to 39.5 and 31.3 Tg CH₄ yr⁻¹ using the Ridgwell et al. (1999) and Curry (2007) parameterisations respectively under the same meteorological forcing, run specifically for this study. For the 2000s period, the simulations estimate the soil uptake at 30.4, 36.7, and 38.3 Tg CH₄ yr⁻¹ based on the parameterisation of Curry, MeMo, and Ridgwell respectively. As part of a more comprehensive model accounting for a range of CH₄ sources and sinks, Tian et al. (2010, 2015, 2016) computed vertically averaged CH₄ soil uptake including the additional mechanisms of aqueous diffusion and plant-mediated (aerenchyma) transport, arriving at the estimate 30 $\pm$ 19 Tg CH₄ yr⁻¹ (Tian et al., 2016) for the 2000s. The still more comprehensive biogeochemical model of Riley et al. (2011) included vertically resolved representations of the same processes considered by Tian et al. (2016), in addition to grid cell fractional inundation and, importantly, the joint limitation of uptake by both CH₄ and O₂ availability in the soil column. Riley et al. (2011) estimated a global CH₄ soil sink of 31 Tg CH₄ yr⁻¹ with a structural uncertainty of 15–38 Tg CH₄ yr⁻¹ (a higher upper limit resulted from an elevated gas diffusivity to mimic convective transport; as this is not usually considered, we adopt the lower upper bound associated with no limitation of uptake at low soil moisture). A model of this degree of complexity is required to explicitly simulate situations where the soil water content increases enough to inhibit the diffusion of oxygen, and the soil becomes a methane source (Lohila et al., 2016). This transition can be rapid, thus creating areas (for example, seasonal wetlands) that can be either a source or a sink of methane depending on the season.

The previous Curry (2007) estimate can be revised upward slightly based on subsequent work and the increase in CH₄ concentration since that time. Indeed, Murguia-Flores et al. (2021) estimated that the global soil uptake doubled between 1900 and 2015 and could further increase due to enhanced diffusion of CH₄ into soil as a result of increases in atmospheric CH₄ mole fraction. Further investigation of the soil uptake is required to better constrain this process at the global scale, while it is highly dependent on local-scale microbial activity and environmental conditions (e.g. D'Imperio et al., 2023; Fest et al., 2017).

Considering the latest estimates (based on VISIT, JSBACH, and MeMo models; Table S6 in the Supplement), we report here a mean estimate of 31 [17–39] Tg CH₄ yr⁻¹ for 2000–2009 and 32 [18–40] for 2010–2019 Tg CH₄ yr⁻¹ (Table 3).

3.3.5 CH₄ lifetime

The atmospheric lifetime of a given gas in steady state may be defined as the global atmospheric burden (Tg) divided by the total sink (Tg yr⁻¹) (IPCC, 2001). This value is different from what is called perturbation lifetime. Perturbation lifetime is used to determine how a one-time pulse emission may decay as a function of time as needed for the calculation of global warming potentials (GWPs) and as a result is related to a theoretical concept. For CH₄, the corresponding perturbation lifetime that should be used in the GWP calculation is 11.8 $\pm$ 1.8 years (Forster et al., 2021). In this section, we discuss the global atmospheric lifetime (also called “burden lifetime” or “turnover lifetime”) that characterises the time required to turn over the global atmospheric burden and defined as the burden divided by the removal flux.

Global models provide an estimate of the loss of the gas due to individual sinks, which can then be used to derive lifetime due to a specific sink. For example, the tropospheric lifetime of CH₄ is determined as the global atmospheric CH₄ burden divided by the loss from OH oxidation in the troposphere, sometimes called “chemical lifetime”. The total lifetime of CH₄ corresponds to the global burden divided by the total loss including tropospheric loss from OH oxidation, stratospheric chemistry, and soil uptake. The CCMI (Plummer et al., 2021) and CMIP6 (Collins et al., 2017) runs estimate the tropospheric methane lifetime at about 9.2 years (average over years 2000–2009), with a range of 7.5–11 years (see Table S5). This range agrees with previous values found in ACCMIP and CCMI (9.3 [7.1–10.6] years; Voulgarakis et al., 2013, 9 [7.2–10.1] years; Saunois et al., 2020). Adding 31 Tg to account for the soil uptake to the total chemical loss of the CMIP6 and CCMI models, we derive a total CH₄ lifetime of 8.2 years (average over 2000–2009 with a range of 6.8–9.7 years). The lifetime calculated over 2010–2019 based on CMIP6 simulations is similar (Table S5). These updated model estimates of total CH₄ lifetime agree with the previous estimates from ACCMIP (8.2 [6.4–9.2] years for year 2000; Voulgarakis et al., 2013) and Saunois et al. (2020)-based CCMI models. Reducing the large spread in CH₄ lifetime (between models, and between models and observation-based estimates) would (1) bring an improved constraint on global total methane emissions and (2) ensure an accurate forecast of future climate.

4 Atmospheric observations and top-down inversions

4.1 Atmospheric observations

Systematic atmospheric CH₄ observations began in 1978 (Blake et al., 1982) with infrequent measurements from discrete air samples collected in the Pacific at a range of latitudes from 67° N to 53° S. Because most of these air samples were from well-mixed oceanic air masses and the measurement technique was precise and accurate, they were sufficient to establish an increasing trend and the first indication of the latitudinal gradient of methane. Spatial and temporal coverage was greatly improved soon after (Blake and Rowland, 1986) with the addition of the Earth System Research Laboratory from the US National Oceanic and Atmospheric Administration (NOAA/GML) flask network (Steele et al., 1987; Lan et al., 2024; Fig. 1), and the Advanced Global Atmospheric Gases Experiment (AGAGE) (Cunnold et al., 2002; Prinn et al., 2018), the Commonwealth Scientific and Industrial Research Organisation (CSIRO, Francey et al., 1999), the University of California Irvine (UCI, Simpson et al.. 2012), and in situ and flask measurements from regional networks, such as ICOS (Integrated Carbon Observation System) in Europe (https://www.icos-ri.eu/, last access: 1 April 2025). The combined data sets provide the longest time series of globally averaged CH₄ abundances. Since the early 2000s, CH₄ column-averaged mole fractions have been retrieved through passive remote sensing from space (Buchwitz et al., 2005a, b; Butz et al., 2011; Crevoisier et al., 2009; Frankenberg et al., 2005; Hu et al., 2018). Ground-based Fourier transform infrared (FTIR) measurements at fixed locations also provide time-resolved CH₄ column observations during daylight hours and a validation data set against which to evaluate the satellite measurements such as the Total Carbon Column Observing Network (TCCON) network (e.g. Pollard et al., 2017; Wunch et al., 2011) or the Network for Detection of Atmospheric Composition Change (NDACC) (e.g. Bader et al., 2017).

In this budget, in situ observations from the different networks were used in the top-down atmospheric inversions to estimate CH₄ sources and sinks over the period 2000–2020. Satellite observations from the TANSO/FTS instrument on board the satellite GOSAT were used to estimate CH₄ sources and sinks over the period 2010–2020. Other atmospheric data (FTIR, airborne measurements, AirCore, isotopic measurements, etc.) have been used for validation by some groups, but not specifically in this study. However, further information is provided in Tables S7, S8, S9, S10, and S11 and a more comprehensive validation of the inversions is planned to use some of these data.

4.1.1 In situ CH₄ observations and atmospheric growth rate at the surface

We use globally averaged CH₄ mole fractions at the Earth's surface from the four observational networks (NOAA/GML, AGAGE, CSIRO, and UCI). The data are archived at the World Data Centre for Greenhouse Gases (WDCGG) of the WMO Global Atmospheric Watch (WMO-GAW) programme (https://gaw.kishou.go.jp/, last access: 1 April 2025), including measurements from other sites that are not operated as part of the four networks. The CH₄ in situ monitoring network has grown significantly over the last decade due to the emergence of laser diode spectrometers, which are robust and accurate enough to allow deployments with low maintenance, enabling the development of denser networks in developed countries (Stanley et al., 2018; Yver Kwok et al., 2015) and new stations in remote environments (Bian et al., 2015; Nisbet et al., 2019).

The networks differ in their sampling strategies, including the frequency of observations, spatial distribution, and methods of calculating globally averaged CH₄ mole fractions. Details are given in the supplementary material of Kirschke et al. (2013). The global average values of CH₄ abundances at Earth's surface presented in Fig. 1 are computed using long-term measurements from background conditions with minimal influence from immediate emissions. All measurements are calibrated against gas standards either on the current WMO reference scale or on independent scales with well-estimate differences from the WMO scale. The current WMO reference scale, maintained by NOAA/ESRL, WMO-X2004A (Dlugokencky et al., 2005), was updated in July 2015. NOAA and CSIRO global means are on this scale. AGAGE uses an independent standard scale (based on work by Tohoku University (Aoki et al., 1992) and maintained at Scripps Institution of Oceanography (SIO)), but direct comparisons of standards and indirect comparisons of atmospheric measurements show that differences are well below 5 ppb (Tans and Zwellberg, 2014; Vardag et al., 2014), and the TU-1987 scale used for AGAGE measurements is only a 0.5 ppb difference from WMO-X2004A at 1900 ppb level. UCI uses another independent scale that was established in 1978 and is traceable to NIST (Flores et al., 2015; Simpson et al., 2012) but has not been included in standard exchanges with other networks, so differences with the other networks cannot be quantitatively defined. Additional experimental details are presented in the supplementary material of Kirschke et al. (2013) and references therein.

The globally averaged CH₄ (shown in Fig. 1a) and its growth rate (derivative of the deseasonalised trend curve; shown in Fig. 1b) through to 2022 are plotted for the four measurement programmes using a procedure of signal decomposition described in Thoning et al. (1989). We define the annual $G_{\mathrm{ATM}}$ as the increase in the atmospheric concentrations from 1 January in one year to 1 January in the next year. Agreement among the four networks is good for the global growth rate, especially since $\sim$ 1990. The large differences observed mainly before 1990 probably reflect the different spatial coverage of each network. The long-term behaviour of globally averaged atmospheric CH₄ shows a positive growth rate (defined as the derivative of the deseasonalised mixing ratio) that slows down from the early 1980s through 1998, a near-stabilisation of CH₄ concentrations from 1999 to 2006, and a renewed period with positive persistent overall accelerating growth rates since 2007, slightly larger after 2014.

From 1999 to 2006, the annual increase in atmospheric CH₄ was remarkably small at 0.6 $\pm$ 0.1 ppb yr⁻¹. After 2006, the atmospheric growth rate increased to a level similar to that of the mid-1990s ($\sim$ 5 ppb yr⁻¹) and for 2014 and 2015 even to that of the 1980s ($>$ 10 ppb yr⁻¹). In the two recent years 2020 and 2021, the highest growth rates of 15 ppb yr⁻¹ and 18 ppb yr⁻¹ (see Sect. 6) were unprecedented starting in the 1980s. On decadal timescales, the annual increase is on average 2.2 $\pm$ 0.3 ppb yr⁻¹ for 2000–2009, 7.6 $\pm$ 0.3 ppb yr⁻¹ for 2010–2019, and 15.2 $\pm$ 0.4 ppb yr⁻¹ for the year 2020 (Table 3). Both climate variability and anthropogenic emission changes are responsible for variations in atmospheric CH₄ growth rates. Indeed, climate variations such as El Niño–Southern Oscillation induce changes in emissions such as biomass burning or wetland emission but also impact OH oxidation (e.g. Rowlinson et al., 2019; Zhao et al., 2020b; Peng et al., 2022).

4.1.2 Satellite data of column average CH₄

In this budget, we use satellite data from the JAXA satellite Greenhouse Gases Observing SATellite (GOSAT) launched in January 2009 (Butz et al., 2011; Morino et al., 2011) containing the TANSO-FTS instrument, which observes in the shortwave infrared (SWIR). Different retrievals of CH₄ based on TANSO-FTS/GOSAT products are made available to the community: from NIES (Yoshida et al., 2013), from SRON (Schepers et al., 2012), and from the University of Leicester (Parker et al., 2020; Parker and Boesch, 2020). The three retrievals are used by the top-down systems (Tables 4 and S6). Although GOSAT retrievals still show significant unexplained biases and limited sampling in cloud covered regions and in the high-latitude winter, they represent an important improvement compared to the first satellite measuring CH₄ from space, SCIAMACHY (Scanning Imaging Absorption spectrometer for Atmospheric CartograpHY), both for random and systematic observation errors (see Table S2 of Buchwitz et al., 2016).

Here, as in Saunois et al. (2020), only inversions using GOSAT retrievals are used.

4.2 Top-down inversions used in the budget

An atmospheric inversion is the optimal combination of atmospheric observations, of a model of atmospheric transport and chemistry, of a prior estimate of CH₄ sources and sinks, and of their uncertainties, to provide improved estimates of the sources and sinks and their uncertainty. The theoretical principle of CH₄ inversions is detailed in the Supplement (Sect. S3), and an overview of the different methods applied to CH₄ is presented in Houweling et al. (2017).

We consider an ensemble of inversions gathering various chemistry–transport models, differing in vertical and horizontal resolutions, meteorological forcing, advection and convection schemes, and boundary layer mixing. Including these different systems is a conservative approach that allows us to cover different potential uncertainties of the inversion: model transport, set-up issues, and prior dependency. General characteristics of the inversion systems are provided in Table 4. Further details can be found in the referenced papers and in the Supplement (Sect. S6). Each group was asked to provide gridded flux estimates for the period 2000–2020, using either surface or satellite data, but no additional constraints were imposed so that each group could use their preferred inversion set-up. Two sets of prior emission distributions were built from the most recent inventories or model-based estimates (see Sect. S4), but its use was not mandatory (see Tables S8 to S11 for the inversion characteristics). This approach corresponds to a flux assessment but not to a model intercomparison as the protocol was not too stringent. Estimating posterior uncertainty is time-consuming and computer-resource-consuming, especially for the 4D-Var approaches and Monte Carlo methods. Posterior uncertainties have not been requested for this study, but they were found to be lower than the ensemble spread in Saunois et al. (2020). Indeed, chemistry–transport models differ in interhemispheric transport, stratospheric CH₄ profiles, and OH distribution, limitations which are not fully considered in the individual posterior uncertainty. As a result, we report the minimum–maximum range among the different top-down approaches.

Seven atmospheric inversion systems using global Eulerian transport models were used in this study; they contributed to the previous budgets that included eight atmospheric inversion systems in Saunois et al. (2016) and nine in Saunois et al. (2020). Each inversion system provided one or several simulations, including sensitivity tests varying the assimilated observations (surface or satellite), the OH interannual variability, or the prior flux ensemble. This represents a total of 24 inversion runs with different time coverage: generally, 2000–2020 for surface-based observations and 2010–2020 for GOSAT-based inversions (Tables 4 and S7). In poorly observed regions, top-down surface inversions may rely on the prior estimates and bring little or no additional information to constrain (often) spatially overlapping emissions (e.g. in India, China). Also, we recall that many top-down systems solve for the total fluxes at the surface only or for some categories that may differ from the GCP categories. When multiple sensitivity tests were performed, the mean of this ensemble was used not to overweight one particular inverse system. It should also be noticed that some satellite-based inversions are in fact combined satellite and surface inversions as they use surface-based inversions to correct the latitudinal bias of the satellite retrievals against the optimised atmosphere measurements to correct for errors in the transport model especially in the stratosphere (e.g. Segers et al., 2022; Maasakkers et al., 2019). Nevertheless, these inversions are still referred to as satellite-based inversions. Most of the top-down models use the OH distribution from the TRANSCOM experiment (Patra et al., 2011) either as fixed over the period or with the interannual variability derived by Patra et al. (2021).

Each group provided gridded monthly maps of emissions for both their prior and posterior total and for sources per category (see the categories Sect. 2.3). Results are reported in Sect. 5. Atmospheric sinks from the top-down approaches have been provided for this budget and are compared with the values reported in Saunois et al. (2020). Not all inverse systems report their chemical sink; as a result, the global mass imbalance for the top-down budget is derived as the difference between total sources and total sinks for each model when both fluxes were reported.

5 Methane budget: top-down and bottom-up comparison

5.1 Global methane budget

5.1.1 Global total methane emissions

At the global scale, the total annual emissions amount inferred by the ensemble of 24 inversions is 575 [553–586] Tg CH₄ yr⁻¹ for the 2010–2019 decade (Table 3), with the highest ensemble mean emission of 608 [581–627] Tg CH₄ yr⁻¹ for 2020. Global emissions for 2000–2009 (543 Tg CH₄ yr⁻¹) are consistent with Saunois et al. (2016, 2020), and the range for global emissions (526–558 Tg CH₄ yr⁻¹) falls within the range in Saunois et al. (2016) (535–569) and Saunois et al. (2020) (524–560), although the ensemble of inverse systems contributing to this budget is different from Saunois et al. (2016, 2020). Changes in ensemble members contributing to the different budgets are a feature of each new GMB release and, therefore, introduce a source of variation (Table S7). The range reported gives the minimum and maximum values among studies and does not reflect the individual full uncertainties. In addition, most of the top-down models use the same OH distribution from the TRANSCOM experiment (Patra et al., 2011), which introduces less variability to the global budget than is likely justified, so it contributes to the rather low range (10 %) compared to bottom-up estimates (see below). We recall here that Zhao et al. (2020a) found an uncertainty of about 17 % in global methane emissions (518 to 611 Tg CH₄ yr¹ for the early 2000s) due to changes in OH burden and distribution (OH ranging from 10.3 to $\mathrm{12.6}\times {\mathrm{10}}^{\mathrm{5}}$ molecules cm⁻³)

The bottom-up estimates considered here differ substantially from the top-down results, with annual global emissions being about 15 % larger at 669 [512–849] Tg CH₄ yr⁻¹ for 2010–2019 (Table 3). Yet, thanks to the double counting corrections in this budget, bottom-up and top-down budgets are in better agreement compared to previous GMB releases. For the period 2000–2009, the discrepancy between bottom-up and top-down was about 30 % of the top-down estimates in Saunois et al. (2016, 2020) (167 and 156 Tg CH₄ yr⁻¹ respectively), a value that has been reduced significantly in this budget (now 95 Tg CH₄ yr⁻¹ ($<$ 17 %) for the same 2000–2009 period). This reduction is due to improvements from an important decrease in the estimate of emissions from natural and indirect anthropogenic emissions from bottom-up approaches and more specifically inland freshwater emissions. From the previous budget, the estimate for inland freshwater emissions (lakes, ponds, reservoirs, rivers, and streams) decreased from 159 Tg CH₄ yr⁻¹ to 112 Tg CH₄ yr⁻¹ (47 Tg decrease). Then, 23 Tg was removed in the total freshwater ecosystem emissions due to double counting between vegetated wetlands and mostly small ponds and lakes (Sect. 3.2.2). As a result, the combined wetland and inland freshwater emissions are estimated to be 242 Tg CH₄ yr⁻¹ for 2000–2009 (Table 3), compared with 306 Tg CH₄ yr⁻¹ in Saunois et al. (2020).

This budget is the first that reconciles bottom-up and top-down total emissions within the uncertainty ranges. However, the uncertainty in the global budget remains high because of the large range reported for emissions from freshwater systems. Still, the upper bound of global emissions from bottom-up approaches is not consistent with top-down estimates that rely on OH burden constrained by methyl chloroform atmospheric observations and is still likely overestimated.

Figure 6

Methane global emissions from five broad categories (see Sect. 2.3) for the 2010–2019 decade for top-down inversion models (left light-coloured boxplots) (in Tg CH₄ yr⁻¹) and for bottom-up models and inventories (right dark-coloured boxplots). For combined wetland and inland freshwaters three estimates are given: top-down estimates (left), bottom-up estimates (middle), and bottom-up estimates (right) for wetlands only. Median value and first and third quartiles are presented in the boxes. The whiskers represent the minimum and maximum values when suspected outliers are removed (see Sect. 2.2). Suspected outliers are marked by stars. Bottom-up quartiles are not available for bottom-up estimates, except for wetland emissions. Mean values are represented by “$+$” symbols; these are the values reported in Table 3.

[Figure omitted. See PDF]

5.1.2 Global methane emissions per source category

The global CH₄ emissions from natural and anthropogenic sources (see Sect. 2.3) for 2010–2019 are presented in Figs. 6 and 7 and Table 3. Top-down estimates attribute about 65 % of total emissions to anthropogenic activities (range of 55 %–70 %) and 35 % to natural emissions. Bottom-up estimates attribute 57 % of emissions to direct anthropogenic and the rest to natural plus indirect anthropogenic emissions. A current predominant role of direct anthropogenic sources of CH₄ emissions is consistent with and strongly supported by available ice core and atmospheric CH₄ records. These data indicate that atmospheric CH₄ varied around 700 ppb during the last millennium before increasing by a factor of 2.6 to $\sim$ 1800 ppb since pre-industrial times. Accounting for the decrease in mean lifetime over the industrial period, Prather et al. (2012) estimated from these data a total source of 554 $\pm$ 56 Tg CH₄ in 2010, of which about 64 % (352 $\pm$ 45 Tg CH₄) was of direct anthropogenic origin, consistent with the range in our top-down estimates.

Figure 7

The Global Methane Budget for the 2010–2019 decade. Both bottom-up (left) and top-down (right) estimates are provided for each emission and sink category (in Tg CH₄ yr⁻¹), as well as for total emissions and total sinks. Combined wetland and inland freshwaters are depicted as natural and indirect anthropogenic sources (darker green and pink hatches) to recall Fig. 4 (Sect. 3.2.2). Adapted from Saunois et al. (2016, 2020).

[Figure omitted. See PDF]

Natural and indirect anthropogenic emissions

Although smaller than in previous Global Methane Budget releases, the main remaining discrepancy between top-down and bottom-up budgets is found for the natural and indirect anthropogenic emission total (105 Tg), with 311 [183–462] Tg CH₄ yr⁻¹ for bottom-up and only 206 [188–225] Tg CH₄ yr⁻¹ for top-down over the 2010–2019 decade (Table 3). In the bottom-up estimates, this discrepancy comes first from the estimates in both inland freshwater sources (64 Tg) and second from other natural sources (20 Tg from geological sources, termites, oceans, and permafrost). The top-down approaches may be biased due to missing fluxes (mainly inland freshwaters) in their prior estimates.

For 2010–2019, the top-down and bottom-up derived estimates for wetlands emissions of 165 [145–214] Tg CH₄ yr⁻¹ and 159 [119–203] Tg CH₄ yr⁻¹ (Table 3) respectively are comparable within their range. Based on diagnostic wetland area values (see notes in Table 3), bottom-up mean wetland emissions for the 2000–2009 period are smaller in this study than those of Saunois et al. (2016) but larger than in Saunois et al. (2020). The changes in wetland emissions from bottom-up models may be related to updates on the wetland extent data set (WAD2M), the use of two different meteorological forcings for this study, and a different set of models (see Sect. 3.2.1). Conversely, the current 2000–2009 mean top-down wetland estimates are lower than those of Saunois et al. (2016) and Saunois et al. (2020) (Table 3). In the bottom-up estimates, the amplitude of the range of emissions of 116–189 is roughly similar to Saunois et al. (2016) (151–222) and Saunois et al. (2020) (102–179) for 2000–2009. Here, the larger range in bottom-up estimates of wetland emissions is due to the use of GSWP3-W5E5 and greater sensibilities of some models to the climate parameters, as discussed in Sect. 3.2.1. Bottom-up and top-down estimates for wetland emissions agree better in this study ($\sim$ 5 Tg yr⁻¹ for 2000–2009) than in Saunois et al. (2016, 2020) ($\sim$ 17 Tg yr⁻¹ and $\sim$ 30 Tg yr⁻¹ respectively). Natural emissions from inland freshwater systems were not included in the prior fluxes used in the top-down approaches, due to unavailable or uncertain gridded products at the start of the modelling activity. However, emissions from these inland freshwater systems may be implicitly included in the posterior estimates of the top-down models, as these two sources are close and probably overlap at the rather coarse resolution of the top-down models. This is the reason why the “wetland emissions” in the top-down budget in fact better correspond to the sum of combined wetland and inland freshwater emissions in the bottom-up budget. The double counting of 23 Tg CH₄ reduces the bottom-up budget for combined wetland and inland freshwaters from 271 Tg CH₄ yr⁻¹ to 248 Tg CH₄ yr⁻¹ (Sect. 3.2.2). Comparing the 2000–2009 decadal emissions from wetlands and inland freshwater ecosystems estimated by the bottom-up approaches across the last three Global Methane Budget articles shows an upward and then a downward revision with 305 (183 $+$ 122) Tg CH₄ yr⁻¹, 356 (147 $+$ 209) Tg CH₄ yr⁻¹, and 248 (159 $+$ 112 $-$ 23) Tg CH₄ yr⁻¹ (respectively from Saunois et al. (2016, 2020) and this work; the sum in brackets corresponds to the sum of vegetated wetland emissions and inland water emissions estimated through the different budgets). The combined wetland and inland freshwater emissions discrepancy between bottom-up and top-down approaches amounts to 105 Tg CH₄ yr⁻¹ for the 2010–2019 decade. From a top-down point of view, the sum of all the natural sources is more robust than the partitioning between wetlands, inland waters, and other natural sources. Including all known spatiotemporal distributions of natural emissions in top-down prior fluxes would be a step forward in consistently comparing natural versus anthropogenic total emissions between top-down and bottom-up approaches.

In the top-down budget, wetlands represent 28 % on average of the total methane emissions but only 24 % in the bottom-up budget (because of higher total emissions inferred) (see Table 3). Given the large uncertainties, neither bottom-up nor top-down approaches included in this study point to significant changes in wetland emissions between the two decades 2000–2009 and 2010–2019 at the global scale.

For the 2010–2019 decade, top-down inversions infer “other natural emissions” (Table 3) at 43 [40–46] Tg CH₄ yr⁻¹, whereas the sum of the individual bottom-up emissions is 63 [24–93] Tg CH₄ yr⁻¹, contributing to a 20 Tg discrepancy between bottom-up and top-down approaches. Atmospheric inversions infer the same amount over the decade 2000–2009 as over 2010–2019, which is almost half of the value reported in Saunois et al. (2016) (68 [21–130] Tg CH₄ yr⁻¹). This reduction in magnitude and uncertainty is due to (1) a more consistent way of considering other natural emissions in the various inverse systems (same prior estimate as in this budget) and (2) a difference in the ensemble of top-down inversions reported here compared to previous releases. It is worth noting that most of the top-down models include about the same ocean and onshore geological emissions and termite emissions in their prior scenarios. However, none include freshwater or permafrost emissions in their prior fluxes and thus in their posterior estimates.

Geological emissions are associated with relatively large uncertainties, and marine seepage emissions are still widely debated (Thornton et al., 2020). However, summing up all bottom-up fossil-CH₄-related sources (including anthropogenic emissions) leads to a total of 165 [135–190] Tg CH₄ yr⁻¹ in 2010–2019, which is about 29 % of the top-down global CH₄ emissions and 25 % of the bottom-up total global estimate. These results agree with the value inferred from ¹⁴C atmospheric isotopic analyses of 30 % contribution of fossil-CH₄ to global emissions (Etiope et al., 2008; Lassey et al., 2007b). These total fossil fuel emissions from bottom-up approaches agree well with the ¹³C-based estimate of Schwietzke et al. (2016) of 192 $\pm$ 32 Tg CH₄ yr⁻¹. In the bottom-up budget, the larger total emissions (due to uncertainties in bottom-up estimates of natural emissions) lead to a lower fossil fuel contribution compared to Lassey et al. (2007b).

Direct anthropogenic emissions

Total direct anthropogenic emissions for the period 2010–2019 were assessed to be statistically consistent between top-down (369 Tg CH₄ yr⁻¹, range 350–391) and bottom-up approaches (358 Tg CH₄ yr⁻¹, range 329–387); however, top-down approaches infer direct anthropogenic emissions larger by 11 Tg CH₄ yr⁻¹ on average compared to bottom-up approaches (Table 3). The partitioning of direct anthropogenic emissions between agriculture and waste, fossil fuels extraction and use, and biomass and biofuel burning also shows good consistency between top-down and bottom-up approaches, though top-down approaches still suggest less fossil fuel and more agriculture and waste emissions than bottom-up estimates (Table 3 and Figs. 6 and 7). For 2010–2019, agriculture and waste contributed an estimated 228 [213–242] Tg CH₄ yr⁻¹ in the top-down budget and 211 [195–231] Tg CH₄ yr⁻¹ in the bottom-up budget. Fossil fuel emissions contributed 115 [100–124] Tg CH₄ yr⁻¹ in the top-down budget and 120 [117–125] Tg CH₄ yr⁻¹ in the bottom-up budget. Biomass and biofuel burning contributed 27 [26–27] Tg CH₄ yr⁻¹ in the top-down budget and 28 [21–39] Tg CH₄ yr⁻¹ in the bottom-up budget. Biofuel CH₄ emissions rely on very few estimates currently (Wuebbles and Hayhoe, 2002). Although biofuel is a small source globally ($\sim$ 12 Tg CH₄ yr⁻¹), more estimates are needed to allow a proper uncertainty assessment. Overall for top-down inversions the global fraction of total emissions for the different source categories is 40 % for agriculture and waste, 20 % for fossil fuels, and 5 % for biomass and biofuel burning. With the exception of biofuel emissions, the uncertainty associated with global anthropogenic emissions appears to be smaller than that of natural sources but with an asymmetric uncertainty distribution (mean significantly different than median). The relative agreement between top-down and bottom-up approaches may indicate a limited capability of the inversion to separate emissions and a dependency to their prior fluxes; this agreement should therefore be treated with caution. Indeed, in poorly observed regions, top-down inversions rely on the prior estimates and bring little or no additional information to constrain (often) spatially overlapping emissions (e.g. in India, China). Also, as many top-down systems solve for the total fluxes at the surface or for some categories that may differ from the GCP categories, their posterior partitioning relies on the prior ratio between categories that are prescribed using bottom-up inventories.

5.1.3 Global budget of total methane sinks

Top-down estimates

The annual CH₄ chemical removal from the atmosphere is estimated to be 521 Tg CH₄ yr⁻¹ averaged over the period 2010–2019, with an uncertainty of about $\pm$2 % (range 485–532 Tg CH₄ yr⁻¹) (Table 3). All the inverse models account for CH₄ oxidation by OH and O(¹D), and some include stratospheric Cl oxidation (Table S8 to S11). Most of the top-down models use the OH distribution from the TRANSCOM experiment (Patra et al., 2011) either as fixed over the period or including interannual variability from Patra et al. (2021). This study shows no trend in OH and IAV below $\pm$4 %, in agreement with Thompson et al. (2024) (no significant OH trend and IAV $<$ 2 %). As a result, the range of the top-down sink estimates is rather low compared to bottom-up estimates (see below). Differences between transport models affect the chemical removal of CH₄, leading to different chemical loss rates, even with the same OH distribution. However, uncertainties in the OH distribution and magnitude (around $\pm$10 % at the global scale; Zhao et al., 2019) are not considered in our study, while they could contribute to a significant change in the chemical sink and then in the derived posterior emissions through the inverse process (Zhao et al., 2020a), around $\pm$17 % at the global scale, much larger than the model spread derived here. The chemical sink represents more than 90 % of the total sink, the rest being attributable to soil uptake (35 [35–36] Tg CH₄ yr⁻¹). The rather narrow range is due to the use of the same climatological soil sink provided within the modelling protocol, which is based on Murguia-Flores et al. (2018). This sink estimate used as a prior estimate in the inversions is a bit higher than the mean estimate of the soil sink calculated by bottom-up models (30 Tg CH₄ yr⁻¹; Sect. 3.3.4).

Bottom-up estimates

The total chemical loss for the 2010s reported here is 602 Tg CH₄ yr⁻¹ with an uncertainty of 21 % ($\sim$ 125 Tg CH₄ yr⁻¹). Differences in chemistry schemes in the models (especially in the stratosphere) and in the volatile organic compound treatment probably explain most of the discrepancies among models (Zhao et al., 2019).

5.2 Latitudinal and regional methane budgets

The latitudinal and regional breakdown of the bottom-up budget is based on crude assumptions that we acknowledge here. Natural and indirect anthropogenic emissions are based on wetland gridded products from land surface models and the combination of the maps from lakes and ponds from Johnson et al. (2022b), reservoirs from Johnson (2021), and streams and rivers from Rocher-Ros (2023) and Rocher-Ros (2023), the sum of those three scaled to 89 Tg CH₄ yr⁻¹ (shown in Fig. 5) to artificially include the double counting (estimated only at the global scale) and match the global estimate. However, we acknowledge that this procedure distributes the double counting relatively to the final emission distribution and not according to the freshwater ecosystems where the double counting probably occurs. Wild animal and permafrost maps do not exist and are missing from the calculation, leading to at least 3 Tg CH₄ yr⁻¹ of discrepancy. However, as mentioned previously (Sect. 3.2.5 and 3.2.7), this 3 Tg CH₄ yr⁻¹ estimate is probably underestimated in the bottom-up budget. Geological and ocean sources are based on Etiope et al. (2019) and Weber et al. (2019) gridded products scaled to 50 Tg CH₄ yr⁻¹ to be consistent with the reported global values. Finally, we use the termite emission map produced for this budget and used in the global budget. The latitudinal budget does not include the estimates from the FAO and the USEPA for the direct anthropogenic emissions as they are only provided at country scale.

5.2.1 Latitudinal budget of total methane emissions

The latitudinal breakdown of emissions inferred from atmospheric inversions reveals a dominance of emissions in the latitudinal band 90° S–30° N of 364 [337–390] Tg CH₄ yr⁻¹, representing 64 % of the global total (Tables 5 and 6). As emissions in the tropics (30° S–30° N) dominate this latitudinal contribution, we may refer to 90° S–30° N as the tropics in the following 32 % of the emissions are from the mid-latitudes (187 [160–204] Tg CH₄ yr⁻¹) and 4 % from high latitudes (above 60° N). The amounts of emissions depend on the land surface area of the region; however, the 90° S–30° N latitudinal band represents 53 % of global land surfaces and the boreal region 60–90° N around 13 %. Hence, the relative contribution of the emissions from the 90° S–30° N region is much larger (11 percentage points more) than the percentage of its land surface areas; on the contrary the boreal regions (60–90° N) emissions contribute significantly less than the surface area percentage of this region (9 percentage points less). The ranges around the mean latitudinal emissions are larger than for the global CH₄ sources. While the top-down uncertainty is less than $\pm$5 % at the global scale, it increases to $\pm$7 % for the tropics, to $\pm$12 % the northern mid-latitudes, and to more than $\pm$ 20 % in the northern high latitudes (for 2010–2019; Table 5). Both top-down and bottom-up approaches consistently show that CH₄ decadal emissions have increased by $+$21–27 Tg CH₄ yr⁻¹ in the tropics and by $+$5–16 Tg CH₄ yr⁻¹ in the northern mid-latitudes between 2000–2009 and 2010–2019 using the mean ensemble estimate.

Over 2010–2019, at the global scale, satellite-based inversions infer almost identical emissions to ground-based inversions (difference of $+$1 [$-$3 to 9] Tg CH₄ yr⁻¹, with GOSAT-based inversion a bit higher than surface measurements-based inversions), when comparing consistently surface versus satellite-based inversions for each system, similar to Saunois et al. (2020). This difference is much lower than the range derived between the different systems (range of 20 Tg CH₄ yr⁻¹ using surface- or satellite-based inversions). This result reflects that differences in atmospheric transport among the systems probably have more impact on the estimated global emissions than the types of observations assimilated.

As expected, considering the different coverage of observation data sets, regional distributions of inferred emissions differ depending on the nature of the observations used (satellite or surface). The largest differences (satellite-based minus surface-based inversions) are observed over the tropical region, between $-$10 and $+$43 Tg CH₄ yr⁻¹ (90° S to 30° N) and the northern mid-latitudes (between $-$36 and $-$2 Tg CH₄ yr⁻¹). Satellite data provide stronger constraints on fluxes in tropical regions than surface data, due to a much larger spatial coverage. It is therefore not surprising that differences between these two types of observations are found in the tropical band and consequently in the northern mid-latitudes to balance total emissions, thus affecting the north–south gradient of emissions. However, the regional patterns of these differences are not consistent through the different inverse systems. Indeed, some systems found higher emissions in the tropics when using GOSAT instead of surface observations, while others found the opposite. This difference between inversion systems may depend on whether or not a bias correction is applied to the satellite data based on surface observations and also on the modelled horizontal and vertical transports, in the troposphere and in the stratosphere.

5.2.2 Latitudinal methane emissions per source category

The analysis of the latitudinal CH₄ budget per source category (Fig. 8 and Table 6) can be performed both for bottom-up and top-down approaches but with limitations. Bottom-up estimates of natural and indirect anthropogenic emissions are based on assumptions as specified at the beginning of this Sect. 5.2. For top-down estimates, as already noted, the partitioning of emissions per source category has to be considered with caution. Indeed, using only atmospheric CH₄ observations to constrain CH₄ emissions makes this partitioning largely dependent on prior emissions. However, differences in spatial patterns and seasonality of emissions can be utilised to constrain emissions from different categories by atmospheric methane observations (for those inversions solving for different sources categories, see Sect. 2.3).

Figure 8

Methane latitudinal emissions from five broad categories (see Sect. 2.3) for the 2010–2019 decade for top-down inversion models (left light-coloured boxplots) (in Tg CH₄ yr⁻¹) and for bottom-up models and inventories (right dark-coloured boxplots). For combined wetland and inland freshwaters three estimates are given: top-down estimates (left), bottom-up estimates (middle), and bottom-up estimates (right) for wetlands only. Median value and the first and third quartiles are presented in the boxes. The whiskers represent the minimum and maximum values when suspected outliers are removed (see Sect. 2.2). Suspected outliers are marked by stars. Bottom-up quartiles are not available for bottom-up estimates, except wetland emissions. Mean values are represented by “$+$” symbols; these are the values reported in Table 6.

[Figure omitted. See PDF]

Agriculture and waste are the largest sources of CH₄ emissions in the tropics and the Southern Hemisphere (140 [121–150] Tg CH₄ yr⁻¹ in the bottom-up budget and 150 [135–168] Tg CH₄ yr⁻¹ in the top-down budget, about 40 % of total CH₄ emissions in this region) (Table 6). However, combined wetland and inland freshwater emissions are nearly as large with 151 [85–234] Tg CH₄ yr⁻¹ in the bottom-up budget and 128 [112–155] Tg CH₄ yr⁻¹ in the top-down budget (Table 6). Anthropogenic emissions dominate in the northern mid-latitudes, with the highest contribution from agriculture and waste emissions (40 % of total emissions in the top-down budget), closely followed by fossil fuel emissions (32 % of total emissions, top-down budget). Boreal regions are largely dominated by inland freshwater emissions (41 % and 54 % of total emissions, top-down and bottom-up budget respectively) (Table 6).

The largest discrepancies between the top-down and the bottom-up budgets are found in the mid-latitudes and boreal regions from the natural and indirect sources with bottom-up estimates twice as large as the top-down ones, especially in the inland freshwater category.

The uncertainty for wetlands and inland freshwater emissions is larger in the bottom-up models than in the top-down models (mostly wetlands), while uncertainty in anthropogenic emissions is larger in the top-down models than in the bottom-up inventories. The large uncertainty in tropical inland freshwater emissions (mostly wetlands) of $\pm$44 % results from large regional differences between the bottom-up land surface models. Although they are using the same forcings, their responses in terms of flux density show different sensitivities to temperature, water vapour pressure, precipitation, and radiation.

5.2.3 Regional budget for total emissions

The regional breakdown of emissions is provided for 18 continental regions (see map in Fig. S3 and Table S1 with the country aggregation in the Supplement).

At the regional scale and for the 2010–2019 decade (Table 7), total methane emissions are dominated by Southeast Asia with 63 [52–71] Tg CH₄ yr⁻¹, China with 57 [37–72] Tg CH₄ yr⁻¹, and South Asia with 52 [43–60] Tg CH₄ yr⁻¹ (top-down budget). These top three emitters contribute 30 % of total global CH₄ emissions. The following high-emitting regions are Brazil with 47 [41–58] Tg CH₄ yr⁻¹, equatorial Africa with 47 [39–59] Tg CH₄ yr⁻¹, the USA with 38 [32–46] Tg CH₄ yr⁻¹, southwestern South America with 38 [30–48] Tg CH₄ yr⁻¹, Russia with 36 [27–45] Tg CH₄ yr⁻¹, Europe with 31 [24–36] Tg CH₄ yr⁻¹, the Middle East with 31 [24–39] Tg CH₄ yr⁻¹, northern Africa with 25 [23–29] Tg CH₄ yr⁻¹, and Canada with 20 [17–24] Tg CH₄ yr⁻¹. Other regions contribute less than 20 Tg CH₄ yr⁻¹.

Table 7

Regional methane emissions (regions ranked by continent) (in Tg CH₄ yr⁻¹) for the last decade of 2010–2019, based on top-down and bottom-up approaches. Uncertainties are reported as [min–max] range of reported studies. Differences of 1 Tg CH₄ yr⁻¹ in the totals can occur due to rounding errors. For bottom-up approaches, natural and indirect anthropogenic sources are estimated based on available gridded data sets (see text in Sect. 5.2). As some emissions are missing gridded products (wild animals, permafrost, and hydrates), discrepancies may occur in terms of totals proposed in Table 3. Bottom-up direct anthropogenic estimates are based on all products (gridded and per country).

Region	Total emissions		Natural and indirect		Direct anthropogenic
			anthropogenic emissions		emissions
	Bottom-up	Top-down	Bottom-up	Top-down	Bottom-up	Top-down
USA	49 [27–77]	38 [32–46]	24 [7–43]	12 [7–22]	26 [19–34]	25 [16–31]
Canada	38 [14–71]	20 [17–24]	32 [11–63]	14 [11–22]	6 [3–8]	7[5–9]
Central America	18 [10–28]	17 [14–19]	8 [3–17]	5 [2–6]	10 [8–12]	12 [11–13]
Northern South America	19 [9–35]	16 [13–20]	10 [3–17]	9 [7–11]	9 [6–17]	7 [6–8]
Brazil	51 [26–79]	47 [41–58]	32 [11–57]	26 [22–36]	19 [16–22]	21 [17–26]
Southwestern South America	34 [16–51]	38 [30–48]	21 [6–35]	24 [16–34]	13 [10–16]	14 [12–17]
Europe	42 [29–57]	31 [24–36]	17 [6–30]	7 [5–9]	25 [22–27]	24 [20–31]
Northern Africa	24 [18–33]	25 [23–29]	7 [2–13]	6 [6–8]	18 [16–20]	19 [17–21]
Equatorial Africa	47 [28–83]	47 [39–59]	23 [10–49]	24 [20–30]	24 [19–34]	23 [19–29]
Southern Africa	21 [5–43]	19 [16–24]	11 [2–29]	8 [7–10]	10 [3–14]	11 [10–12]
Russia	48 [24–83]	36 [27–45]	25 [9–47]	14 [11–18]	23 [15–36]	21 [14–29]
Central Asia	15 [6–29]	10 [8–13]	8 [2–19]	1 [0–2]	8 [4–10]	9 [7–11]
Middle East	35 [21–47]	31 [24–39]	9 [3–15]	4 [1–6]	26 [18–31]	28 [20–34]
China	71 [55–99]	57 [37–72]	15 [4–33]	4 [3–7]	57 [51–66]	53 [34–66]
Korea–Japan*	6 [4–12]	5 [4–6]	3 [1–7]	1 [1–1]	4 [3–5]	4 [3–5]
South Asia	58 [49–72]	52 [43–60]	13 [5–25]	6 [5–6]	45 [44–47]	45[37–49]
Southeast Asia	64 [42–93]	63 [52–71]	32 [19–54]	27 [20–34]	32 [23–39]	35 [31–46]
Australasia	16 [9–26]	13 [10–17]	10 [4–19]	6 [4–7]	7 [6–7]	7 [6–7]

^* Korea collectively refers North Korea and South Korea in this table.

5.2.4 Regional budget per source category

In agreement with Stavert et al. (2021), natural and indirect anthropogenic emissions are dominated by Brazil, Canada, Russia, equatorial Africa, and Southeast Asia, contributing 126 Tg CH₄ yr⁻¹ in the bottom-up and 105 Tg CH₄ yr⁻¹ in the top-down budget (Table 7), i.e. 47 % and 50 % of the global natural and indirect anthropogenic emissions in these budgets respectively. At regional scale also, the ranges of uncertainty in natural and indirect anthropogenic emissions are much larger in the bottom-up budget than in the top-down budget (Fig. S5). Except for four regions (Canada, Brazil, northern South America, southwestern South America), direct anthropogenic emissions contribute more than half of the total regional emissions. Due to the large uncertainty and discrepancies in natural and indirect emissions estimates, the regional direct anthropogenic fractions may differ between the bottom-up and top-down budgets. However, in absolute values, the highest direct anthropogenic emitters are the same in the two budgets, with China and South Asia being the top two by far, contributing 56 [51–66] Tg CH₄ yr⁻¹ and 45 [44–47] Tg CH₄ yr⁻¹, respectively (bottom-up values; Fig. 9 and Table 7). These two regions contribute 28 % (26 %) of the global direct anthropogenic emissions in the bottom-up (top-down) budget. The ranks of direct anthropogenic emitters are similar to those presented in the last budget (Stavert et al., 2021). Southeast Asia, the USA, the Middle East, Europe, equatorial Africa, and Russia emit between 32 and 23 Tg CH₄ yr⁻¹ as direct anthropogenic emissions (bottom-up values; Fig. 8). Brazil, northern Africa, and southwestern South America emit between 10 and 20 CH₄ yr⁻¹, while the rest of the regions emit less than 10 CH₄ yr⁻¹ direct anthropogenic emissions (Table 7 and Fig. S5).

Figure 9

Regional anthropogenic emissions for the 2010–2019 decade from bottom-up estimates (in Tg CH₄ yr⁻¹). Regions are ranked by their total anthropogenic emissions. Note that each category has its own emission scale. Note that Korea collectively refers to North Korea and South Korea in this figure.

[Figure omitted. See PDF]

The sectoral partitioning at the regional scale has been derived from both bottom-up and top-down approaches. However, the top-down budget has more limitations, as the sectoral partitioning is usually based on the prior fluxes fractions at the pixel scale, and assimilating only total methane observations does not allow one to disentangle the different source sectors overlapping in a pixel grid. However, differences in spatial patterns and seasonality of emissions can still be constrained by atmospheric CH₄ observations for those inversions solving for different sources categories (see Sect. 2.3).

Bottom-up approaches allow deeper sectorial splitting, especially in terms of direct anthropogenic emissions (Fig. 9). Table 7 and Figs. 9 and 10 present the estimations of CH₄ emissions on average over 2010–2019. Figure 10 presents the budgets for three main categories (combined wetland and inland freshwaters, fossil fuels, and agriculture and waste); a more detailed figure and table including the five categories is available in the Supplement (Fig. S6 and Tables S13 to S18). Values for each individual data set for the decades 2000–2009, 2010–2019, and the last year 2020 are made available in a spreadsheet (see “Data availability”).

Figure 10

Regional emissions (in Tg CH₄ yr⁻¹) for three broad main emissions categories for the 2010–2019 decade: combined wetland and inland freshwaters, fossil fuel, and agriculture and waste from top-down estimates (left boxplot) and bottom-up estimates (right boxplots). The inner map shows the region's distribution (also see the Supplement, Table S1 and Fig. S3). More categories are presented in the Supplement in Fig. S6. Note that Korea collectively refers to North Korea and South Korea in this figure.

[Figure omitted. See PDF]

For most regions, “combined wetland and inland freshwater emissions” are the most uncertain in the bottom-up budget, and generally their range is larger than in the top-down budget. In the top-down budget for 2010–2019 (Table 7), this category contributes the most to the regional emissions in Brazil with24 [20–33] Tg CH₄ yr⁻¹, Southeast Asia with 24 [14–29] Tg CH₄ yr⁻¹ (though similar to their agriculture and waste emissions of 24 [21–31] Tg CH₄ yr⁻¹), equatorial Africa with 22 [19–28] Tg CH₄ yr⁻¹, southwestern South America with 22 [14–33] Tg CH₄ yr⁻¹, Canada with 12 [9–18] Tg CH₄ yr⁻¹, northern South America with 8 [6–10] Tg CH₄ yr⁻¹, southern Africa with 7 [4–9] Tg CH₄ yr⁻¹. The agriculture and waste emissions category dominates in South Asia with 39 [33–43] Tg CH₄ yr⁻¹, China with 30 [13–37] Tg CH₄ yr⁻¹, Europe with 19 [16–23] Tg CH₄ yr⁻¹, the USA with 13 [9–16] Tg CH₄ yr⁻¹, northern Africa with 13 [12–14] Tg CH₄ yr⁻¹, Central America with 9 [8–10] Tg CH₄ yr⁻¹, and North and South Korea and Japan with 3 [3–4] Tg CH₄ yr⁻¹. Fossil fuel emissions dominate in the Middle East with 18 [11–24] Tg CH₄ yr⁻¹ and Russia with 14 [8–23] Tg CH₄ yr⁻¹ (close to their combined wetland and inland freshwater emissions of 11 [8–13] Tg CH₄ yr⁻¹).

The four largest contributors to the fossil fuel sector remain China, the Middle East, Russia, and the USA. Altogether they contribute 67 (64) Tg CH₄ yr⁻¹ in the bottom-up (top-down) budget, around 55 % of the global fossil fuel emissions. The bottom-up and top-down approaches generally agree in terms of ensemble mean, except for China for which the top-down estimates suggest lower emissions than the inventories. While Chinese fossil fuel emissions occur mainly through coal mining activity (88 %), the Middle East, Russia, and the USA extract mainly oil and gas (100 %, 80 %, 72 %).

The three largest contributors to the agriculture and waste sector remain South Asia, China, and Southeast Asia. Together they contribute 88 (92) Tg CH₄ yr⁻¹ in the bottom-up (top-down) budget, around 40 % of the global agriculture and waste sector (Table 7). While the ensemble means tend to agree between bottom-up and top-down budgets, the uncertainty derived from the top-down approaches is larger, especially for these three regions. CH₄ emissions due to rice cultivation originate mostly from these same three regions (Southeast Asia, China, and South Asia). Livestock management emissions occur mainly in South Asia with 20 [18–22] Tg CH₄ yr⁻¹, Brazil with 12 [11–13] Tg CH₄ yr⁻¹, China with 11 [8–16] Tg CH₄ yr⁻¹, and Europe with 11 [10–12] Tg CH₄ yr⁻¹ (bottom-up estimates; Table 7). The USA, equatorial Africa, northern Africa, and southwestern South America emit between 7 Tg CH₄ yr⁻¹ and 10 Tg CH₄ yr⁻¹ in this subsector. Other regions emit less than 4 Tg CH₄ yr⁻¹ in the livestock management sector. The waste sector emissions are dominated by three regions: China with 11 [6–14] Tg CH₄ yr⁻¹, South Asia with 9 [4–11] Tg CH₄ yr⁻¹, and Europe with 8 [6–12] Tg CH₄ yr⁻¹ (bottom-up estimates; Table 7). These three regions contribute around 40 % of the global emissions of the waste sector. It is worth noting that the uncertainty in the inventory estimates at the regional scale is around 40 % (from the min–max range of the estimate, not including the uncertainty from each inventory).

6 Insights on the methane cycle from 2020–2022, during which there were unprecedented high growth rates of methane emissions

The mean emissions estimate for the last year of the budget (2020) was 608 [581–627] Tg CH₄ yr⁻¹ (top-down), with 65 % of the emissions from direct anthropogenic sources. This is 65 Tg CH₄ yr⁻¹ higher (11 %) than the mean emissions of the 2000–2009 decade and 6 % higher than 2010–2019. In Jackson et al. (2024), we estimated that total methane emissions increased by around 20 % between the early 2000s (2000–2002) and the late 2010s (2018–2020). Year 2020 was the second highest year in terms of atmospheric CH₄ growth rate ($+$15.2 ppb yr⁻¹) since systematic measurements began in the late 1980s, coming in just behind the highest in 2021 at 17.97 ppb yr⁻¹. A few studies analysed the large growth rate increase between 2019 ($+$9.7 ppb yr⁻¹) and 2020 ($+$15.2 ppb yr⁻¹) of $+$5.4 ppb yr⁻¹ (corresponding to $+$14.4 $\pm$ 2.0 Tg CH₄ yr⁻¹) (Peng et al., 2022; Stevenson et al., 2022). Peng et al. (2022) estimated that the 2019–2020 growth rate change was almost equally due to an increase in wetland emissions (6.9 $\pm$ 2.1 Tg CH₄ yr⁻¹) and a decrease in the OH chemical loss (7.5 $\pm$ 0.8 Tg CH₄ yr⁻¹) due to reduced OH precursor emissions during the COVID lockdown (Laughner et al., 2021). The COVID-19 lockdown resulted in decreased NO_x emissions and reduced fossil-fuel-related CH₄ emissions (Thorpe et al., 2023), leading to less OH production. At the global scale, Feng et al. (2023) calculated an emission increase of 27 Tg CH₄ yr⁻¹ between 2019 and 2020 considering constant OH and a smaller increase of 21 Tg CH₄ yr⁻¹ when including a 1.4 % decrease in OH. Increased emissions were mainly found in the northern tropics. Qu et al. (2022) also inferred a 31 Tg CH₄ yr⁻¹ increase in emissions, mostly in the tropics, half of it in Africa. Furthermore, Niwa et al. (2024) suggested emission increases by 10–18 Tg CH₄ yr⁻¹ in 15° S–10° N and by 20 Tg CH₄ yr⁻¹ in 10–35° N from 2016–2019 to 2020–2022. Such a result is compatible with wetland driven abnormal emissions during a consecutive 3-year La Niña event spanning from 2020 to 2022 (Zhang et al., 2023; Nisbet et al., 2023). The difference in terms of methodology and approaches between these three studies makes it difficult to compare them quantitatively but provides a robust understanding on the possible causes. Importantly, all the studies indicate, in various proportions, increasing CH₄ emissions in the tropics and in the boreal region, potentially driven by microbial emission from wetlands due to wetter and warmer climate, and a significant contribution of reduced OH concentrations due to the COVID lockdown.

Based on our ensemble of data, we find that top-down approaches infer a much larger change in CH₄ emissions (median [Q1–Q3] at $+$23 [10–31] Tg CH₄ yr⁻¹) than bottom-up approaches ($-$1 [$-$5 to 3] Tg CH₄ yr⁻¹) between 2019 and 2020 (Fig. S7). Bottom-up approaches suggest a very small increase in wetland emissions (around $+$1 [0–3] Tg CH₄ yr⁻¹), while top-down approaches suggest on average a larger increase for wetlands of $+$8 [5–11] Tg CH₄ yr⁻¹, mainly in the tropics and mid-latitudes. It is worth noting that large uncertainties exist for a given year and that the interannual variability is much lower than the ensemble spread. While bottom-up approaches suggest almost constant fossil fuel emissions and slight increase in agriculture and waste ($+$3 Tg CH₄ yr⁻¹), top-down approaches tend to derive higher emissions changes ($+$6 Tg CH₄ yr⁻¹ from the fossil fuel sector and $+$11 Tg CH₄ yr⁻¹ from agriculture and waste as the median over the ensemble). Biomass burning emissions decreased using both approaches by about 5 Tg CH₄ yr⁻¹, in agreement with Peng et al. (2022). Some inversions were run with IAV of OH from Patra et al. (2021) and others with constant OH. However, the inferred OH IAVs in 2019 and 2020 are rather low (0.3 % and 0.15 % on yearly average) in Patra et al. (2021), leading to a small impact in terms of emissions changes between 2019–2020, with $+$22 [9–31] (median [Q1–Q3]) based on the inversions with constant OH and 19 [7–28] based on the inversions with varying OH (Fig. S8).

This first analysis based on our ensemble shows how challenging it is to attribute CH₄ emissions changes to a specific sector or region between two years, because related uncertainties remain much larger than the targeted signal to explain. This calls again for further improvement of both approaches.

NOAA estimates 2021 and 2022 methane atmospheric growth rates of 17.8.0 $\pm$ 0.5 and 14.0 $\pm$ 0.8 ppb yr⁻¹ respectively (Lan et al., 2024). They show a continuation of very high growth rates, challenging again our understanding of the methane budget. The very high values of CH₄ growth rate over 2020–2022 have also been accompanied by a sharp decline in the stable isotopic signal, ${\mathit{\delta }}^{\mathrm{13}}$C_CH4, which suggests that this recent increase in methane growth rate is at least partly explained by increased emissions from microbial sources such as those found in wetlands, inland waters, agriculture, and waste systems (Nisbet et al., 2023; Michel et al., 2024). However, it is worth noting that almost all aforementioned published top-down studies include constraints only on CH₄ and do not discuss the consistency with the atmospheric isotopic signal.

At the time of submission of the manuscript, bottom-up estimates for anthropogenic emissions for 2021 and 2022 were only available from the EDGARv8 data set (https://edgar.jrc.ec.europa.eu/dataset_ghg80, last access: 1 April 2025; EDGAR, 2023). This research inventory suggests that anthropogenic emissions continued to increase from 2020 (374 Tg CH₄ yr⁻¹) to 2021 (379 Tg CH₄ yr⁻¹) and 2022 (386 Tg CH₄ yr⁻¹) with around 62 % of the increase due to the fossil fuel sources, 23 % from the waste sector, and 14 % from the agriculture sector (Table S19). The bottom-up estimate of wetland emissions for 2021–2023, derived from a single wetland model, indicates positive anomalies of 26 Tg CH₄ yr⁻¹ in 2020, 23 Tg CH₄ yr⁻¹ in 2021, and 21 Tg CH₄ yr⁻¹ 2022 relative to the 2000–2006 baseline (https://earth.gov/ghgcenter/data-catalog/lpjeosim-wetlandch4-grid-v1, last access: 1 April 2025; Zhang et al., 2023).

7 Future developments, missing elements, and remaining uncertainties

In this budget, robust features and uncertainties on sources and sinks estimated by bottom-up or top-down approaches have been highlighted as well as discrepancies between the two budgets. Limitations of the different approaches have also been highlighted. Four shortcomings of the CH₄ budget were already identified in Kirschke et al. (2013) and Saunois et al. (2016, 2020) and are revisited below, pointing to key research areas. Although much progress has been made, they are still relevant, and actions are needed. However, these actions fall into different timescales and actors. Here, we revisit the four shortcomings of the contemporary methane budget and discuss how each weakness has been addressed since Saunois et al. (2020). Each section ends by discussing remaining research needs with a list of suggestions, from higher to lower priority.

• 1.

  Shortcoming 1. The amount of methane emitted by wetland and inland water systems are still highly uncertain, and double counting may remain.

This first shortcoming has probably received the largest interest in the last few years with significant improvements. First, a community effort has been made based on more studies, documenting or modelling more inland freshwater systems and synthesising emissions from the complex and heterogeneous ensemble of emitting areas: wetlands, ponds, lakes, reservoirs, streams, rivers, estuaries, and marine systems. The range of wetland and inland water emissions has been narrowed down with improved wetland extent and refined estimates for inland freshwater systems. Double counting between inland freshwater systems has been estimated for the first time and accounted for in this budget. All these improvements decreased the discrepancy between top-down and bottom-up estimates of combined wetland and inland freshwater emissions from 156 Tg CH₄ yr⁻¹ in Saunois et al. (2020) down to 85 Tg CH₄ yr⁻¹ in this update for the 2000–2009 decade. Gridded maps for lakes, ponds, reservoirs, and streams and rivers freshwater emissions have been produced over the past years (Johnson et al., 2021, 2022a; Johnson, 2021; Rocher-Ros et al., 2023; Rocher-Ros, 2023), making the spatial distribution of CH₄ sources almost complete for the first time and allowing a better description of prior emissions in future top-down inversions.

Next steps in the short term are listed below from the highest to lowest priority:

• i.

  Integration of spatial distribution of inland waters in atmospheric inversion models is needed to reach a full description of prior methane sources and sinks.

• ii.

  Refinement is needed of double counting estimation and its possible reduction with more precise spatial and temporal distributions of the different systems contributing to inland freshwater emissions by using very high-resolution satellite data (down to metre resolutions) to properly separate them. The development is needed of a dynamical global high-resolution (typically few metres) classification of saturated soils and inundated surfaces based on satellite data (visible and microwave), surface inventories, and expert knowledge.

• iii.

  Ongoing efforts need to be continued to calibrate and evaluate land surface models for wetland emissions against in situ observations such as FLUXNET-CH₄ (Knox et al., 2019; Delwiche et al., 2021) or BAWLD-CH₄ (Kuhn et al., 2021) for boreal regions and to avoid dependence on top-down estimates. It is still critical to increase the limited number of tropical observations and to assimilate them in the inverse systems to help address the issue (e.g. Kallingal et al., 2024).

• iv.

  Ongoing efforts need to be continued to develop a diversity of modelling approaches (among them process-based model or machine learning approaches) to estimate wetland and inland freshwater CH₄ emissions, including lateral fluxes and reducing upscaling issues, as done by e.g. Zhuang et al. (2023) for lakes.

• v.

  There needs to be continuous integration of collected flux measurements such as in the FLUXNET-CH₄ activity (Knox et al., 2019; Delwiche et al., 2021) or in the BAWLD-CH₄ data set (Kuhn et al., 2021) to provide global flux maps based on machine learning approaches or other approaches (Peltola et al., 2019; McNicol et al., 2023).

In the long run, developing measurement systems will help to improve estimates of the diversity of wetland and inland freshwater sources and further reduce uncertainties:

• More systematic measurements of CH₄ fluxes and their isotopic signatures from sites reflecting the diversity of environment of wetlands and inland waters, complemented with environmental metadata (e.g. soil temperature and moisture, vegetation types, water temperature, acidity, nutrient concentrations, net primary productivity, soil carbon density for wetlands, lake morphologies) will allow us to better understand and estimate the processes of production and transport to the atmosphere (diffusive, ebullition, plants mediated, etc.) and to better constrain methane fluxes and their isotopic signatures in the different modelling approaches (Glagolev et al., 2011; Turetsky et al., 2014).

• 2.

  Shortcoming 2. The uncertainty in the global methane sink from both top-down and bottom-up approaches has not been reduced.

The inverse systems used here have similar caveats to those described in Saunois et al. (2016, 2020) (same OH field, same kind of proxy method to optimise it), leading to quite a constrained atmospheric sink and therefore total global CH₄ sources. Although we have used the latest release of CCMI-2022 (Plummer et al., 2021) and CMIP6 simulations (Collins et al., 2017), the uncertainty of derived CH₄ chemical loss from the chemistry–climate models remains at the same (large) level compared to the previous intercomparison project ACCMIP (Lamarque et al., 2013). The causes of uncertainties on the CH₄ loss and the differences between the different OH fields derived from chemistry–transport models (CTMs) and climate–chemistry models (CCMs) have been widely discussed (e.g. Nicely et al., 2017; Zhao et al., 2019, 2020a). These results emphasise the need to first assess and then improve atmospheric transport and chemistry models, especially vertically, and to integrate robust representation of OH fields in atmospheric models. Recently, numerous efforts based on satellite data have been made to constrain OH distribution, variability, and trends (e.g, Anderson et al., 2023, 2024; Pimlott et al., 2022; Zhao et al., 2023; Zhu et al., 2022). Finally, soil uptake estimates rely on very few studies, and interannual variations remain underconstrained.

Next steps, in the short term, could include developments by the modelling community in the following:

• The soil uptake needs to be estimated with different land surface models (creating an ensemble), and its variations over the past decade need to be discussed.

• The impact of using updated and varying soil uptake estimates needs to be assessed, especially considering a warmer climate in the top-down approach. Indeed, for top-down models resolving for the net flux of CH₄ at the surface integrating a larger estimate of soil uptake would allow larger emissions and then would reduce the uncertainty with the bottom-up estimates of total CH₄ sources.

• The reactivity of the air parcels in the chemistry–climate models needs to be studied further, and new diagnostics need to be defined to assess modelled CH₄ lifetimes such as in Prather et al. (2023).

• Benchmarking of CTMs and CCMs needs to be developed regarding simulated OH distribution and variability (as in Zhao et al. (2019) for example) to increase efforts to assess biases and improve atmospheric chemical schemes in CTMs and CCMs.

• Methods need to be developed to better constrain OH. Some have been proposed: satellite CH₄ observations (Zhang et al., 2018; Anderson et al., 2023, 2024) could afford this, but strategy is needed (see Duncan et al., 2024, and references therein), and using halogenated compounds beyond methyl chloroform (MCF), such as done in box models (Thompson et al., 2024) to derive a 3D dynamical OH. Such methods should be able to reach very low uncertainty for OH burden and trends ($<$ 2 %) in order to really better constrain the CH₄ budget. Duncan et al. (2024) discuss the existing satellite-based methods and propose a strategy to constrain OH from space-based approaches.

• The aforementioned different potential OH chemical fields need to be integrated, also including interannual variability, to assess the impact on the methane budget following Zhao et al. (2020a).

In the long run, other parameters should be (better) integrated into top-down approaches:

• The magnitude of the CH₄ loss through oxidation by tropospheric Cl is a process debated in the recent literature. More modelling (e.g. Thanwerdas et al., 2022b) and instrumental studies should be devoted to reducing the uncertainty of this potential additional sink before integrating it in top-down models. This would be especially critical if inversions using ¹³C–CH₄ observations were included in GMB in the future.

• 3.

  Shortcoming 3. The partitioning of methane sources and sinks by region and emission sectors using top-down approaches are lacking strong constraints.

In this work, we report inversions assimilating satellite data from GOSAT, which bring more constraints than provided by surface stations alone, especially over tropical continents. However, we still found that satellite- and surface-based inversions and the different inversion systems do not consistently infer the same regional flux distribution.

The estimates contributing to the Global Methane Budget are further used in more specific studies focusing on the comparison of the estimates from bottom-up and top-down approaches at national (Deng et al., 2022) and regional scales, including efforts from the GCP-REgional Carbon Cycle Assessment and Processes (RECCAP2) (Petrescu et al., 2021, 2023; Tibrewal et al., 2024; Lauerwald et al., 2023b; and other RECCAP2 publications to come, see https://www.globalcarbonproject.org/reccap/publications.htm, last access: 1 April 2025).

Next steps, in the short term, could integrate developments to be made by the top-down community:

• GOSAT 2 retrievals should be included (Noël et al., 2022; Imasu et al., 2023) for the GOSAT-based inversions, and TROPOMI-based inversions should be considered (as done in Tsuruta et al., 2023, Shen et al., 2023, Chen et al., 2022, Qu et al., 2021 or Yu et al., 2023) in the next releases once at least 8 years of data are available to provide a decadal estimate and biases are reduced for global-scale use (Lorente et al., 2023; Balasus et al., 2023). Indeed, recent satellite developments have provided higher temporal and spatial resolutions of CH₄ observations in regions with poor in situ measurements (Fig. S9; such as TROPOMI observations in northern Africa).

• Newly available updated gridded products need to be integrated for the different natural sources of CH₄ in their prior fluxes (e.g. inland freshwaters) to reach a full spatial description of sources and sinks and to be able to better compare the top-down budget with the bottom-up budget.

• Newly developed 4D variational inversion systems using isotopic species in the top-down budget need to be integrated (Basu et al., 2022; Thanwerdas et al., 2024; Drinkwater et al., 2023; Mannisenaho et al., 2023).

• The availability of in situ data at high temporal resolution needs to be improved for the scientific community, especially ones covering poorly documented regions such as China (Liu et al., 2021b; Guo et al., 2020), India (Nomura et al., 2021; Lin et al., 2015; Tiwari and Kumar, 2012), and Siberia (Sasakawa et al., 2010, 2017; Fujita et al., 2020; Winderlich et al., 2010), which have not been delivered so far to international databases or only at poor temporal resolution.

• Information from imagery satellites needs to be integrated (e.g. TROPOMI, Carbon Mapper, MethaneSAT, GHGSAT) for high to super-emitters to improve prior fluxes of anthropogenic emissions in terms of quantity and locations for each covered sector.

In the long run, integrating more measurements and regional studies will help to improve the top-down systems and further reduce the uncertainties:

• CH₄ surface networks need to be extended to poorly observed regions (e.g. tropics, China, India, high latitudes) and to the vertical dimension. Aircraft regular measurements (e.g. Filges et al., 2015; Brenninkmeijer et al., 2007; Paris et al., 2010; Sweeney et al., 2015), AirCore campaigns (e.g. Andersen et al., 2018; Membrive et al., 2017), and TCCON observations (e.g. Wunch et al., 2011, 2019) remain critical in complementing satellite data that do not observe well in cloudy regions and at high latitudes. Those observations are also crucial to evaluate and eventually correct satellite biases (Buchwitz et al., 2022).

• Continuous isotopic measurements of CH₄ need to be extended and developed to help partitioning methane sources and to be integrated in 4D variational isotopic inversions (e.g. Yacovitch et al., 2021).

• Global data from future satellite instruments with intrinsic low biases need to be integrated, such as active lidar techniques with MERLIN (Ehret et al., 2017), which are promising to overcome issues of systematic errors (Bousquet et al., 2018) and should provide measurements over the Arctic, contrary to the existing and planned passive missions.

• Other co-emitted species such as radiocarbon for fossil/non-fossil emissions (Lassey et al., 2007a, 2007b; Petrenko et al., 2017), CO (e.g. Zheng et al., 2019) for biomass burning emissions, and ethane for fugitive emissions (e.g. Ramsden et al., 2022) could bring additional information for partitioning emissions.

• 4.

  Shortcoming 4. The atmospheric transport models used in the top-down budget still present issues that have not been fully investigated.

The TRANSCOM experiment synthesised in Patra et al. (2011) showed a large sensitivity of the representation of atmospheric transport on CH₄ abundances in the atmosphere. In particular, the modelled CH₄ budget appeared to depend strongly on the troposphere–stratosphere exchange rate and thus on the model vertical grid structure and circulation in the lower stratosphere. Also, regional changes in the CH₄ budget depend on the characteristics of the atmospheric transport models used in the inversion (Bruhwiler et al., 2017; Locatelli et al., 2015). This axis of research is demanding important development from the atmospheric modelling community. Waiting for future improvements (finer horizontal and vertical resolutions, more accurate physical parameterisation, increase in computing resources, etc.), assessing atmospheric transport error and the impact on the top-down budget remain crucial and mostly rely on the use of an ensemble of models. Methodology changes should be integrated into the next methane budget releases.

Firstly, independent validation should be performed against aircraft measurements (available through the CH₄ GLOBALVIEWplus v6.0 ObsPack (Schuldt et al., 2023), the IAGOS data portal (https://iagos.aeris-data.fr/download/, last access: 1 April 2025), the NIES portal (https://db.cger.nies.go.jp/ged/en/datasetlist/index.html, last access: 1 April 2025) for CONTRAIL (e.g.Machida et al., 2008) and Siberian measurements (e.g. Sasakawa et al., 2017), or the WDCGG data portal (https://gaw.kishou.go.jp/, last access: 1 April 2025) for additional flights over three other Japanese airports and Orléans, France). Secondly, AirCore data sets (available through the NOAA Global Monitoring Laboratory website (https://gml.noaa.gov/aftp/data/AirCore/, last access: 1 April 2025, Baier et al., 2021) and the French AirCore programme for atmospheric sampling (https://aircore.aeris-data.fr, last access: 1 April 2025, Membrive et al., 2017)) should be used to evaluate the vertical profile of methane in the models. Then those evaluations against aircraft and AirCore measurements could be used to sort or weight the different estimates in the top-down budget.

Next steps, in the short term, could include some development to be addressed by the top-down community to reduce atmospheric transport errors:

• Further methodologies need to be developed to extract stratospheric partial column abundances from observations such as TCCON data (Saad et al., 2014; Wang et al., 2014), AirCore (e.g. Andersen et al., 2018; Membrive et al., 2017), or ACE-FTS (De Mazière et al., 2008) or MIPAS (Glatthor et al., 2024) satellite data.

• SWIR and TIR measurements need to be combined from space to better constrain the tropospheric column, from TROPOMI and IASI, for example in the MethanePlus ESA project (https://methaneplus.eu/#docs, last access: 1 April 2025, Buchwitz et al., 2024) or GOSAT (Kuze et al., 2020).

• Transport models codes need to be able to be run on graphics processing units (GPUs) to achieve subdegree-resolution global inversions (Chevallier et al., 2023).

In the long run, there need to be developments within the dynamical core of the atmospheric transport models through the implementation of hexagonal–icosaedric grid with finer resolution (Dubos et al., 2015; Niwa et al., 2017, 2022; Lloret et al., 2023), and improvements in the simulated boundary layer dynamics or troposphere–stratosphere exchanges are promising to reduce atmospheric transport errors.

8 Data availability

The data presented here are made available in the belief that their dissemination will lead to greater understanding and new scientific insights into the methane budget and changes to it and help to reduce its uncertainties. For research projects, if the data used are essential to the work to be published or if the conclusion or results largely depend on the data, co-authorship should be considered. Full contact details and information on how to cite the data are given in the accompanying database.

The accompanying database includes a NetCDF file defining the regions used, an archive with the maps of prior fluxes used in the top-down activity, an archive with data corresponding to Figs. 3 and 5, and one Excel file organised in the following spreadsheets.

The file Global_Methane_Budget_2000-2020_v1.0.xlsx includes (1) a summary, (2) the methane observed mixing ratio and growth rate from the four global networks (NOAA, AGAGE, CSIRO and UCI), (3) the evolution of global anthropogenic methane emissions (including biomass burning emissions) used to produce Fig. 2, (4) the global and latitudinal budgets over 2000–2009 based on bottom-up approaches, (5) the global and latitudinal budgets over 2000–2009 based on top-down approaches, (6) the global and latitudinal budgets over 2010–2019 based on bottom-up approaches, (7) the global and latitudinal budgets over 2010– 2019 based on top-down approaches, (8) the global and latitudinal budgets for year 2020 based on bottom-up approaches, (9) the global and latitudinal budgets for year 2020 based on top-down approaches, and (10) the list of contributors to contact for further information on specific data.

This database is available from ICOS Carbon Portal (10.18160/GKQ9-2RHT, Martinez et al., 2024).

9 Conclusions

We have built an updated global methane budget by using and synthesising a large ensemble of published methods and new results using a consistent, transparent, and traceable approach, including atmospheric observations and inversions (top-down models), process-based models for land surface emissions and atmospheric chemistry, and inventories of anthropogenic emissions (bottom-up models and inventories). For the 2010–2019 decade, global CH₄ emissions are 575 Tg CH₄ yr⁻¹ (range of 553–586 Tg CH₄ yr⁻¹), as estimated by top-down inversions. About 65 % of global emissions are anthropogenic (range of 63 %–68 %). Bottom-up models and inventories suggest larger global emissions (669 Tg CH₄ yr⁻¹ [512–849]) mostly because of larger and more uncertain natural emissions from inland freshwater systems, natural wetlands, geological seepage, and likely some unresolved double counting of these sources. It is also likely that some of the individual bottom-up emission estimates are too high, leading to larger global emissions from the bottom-up approach than the atmospheric constraints suggest. However, the important progress in this update is that for the first time, the bottom-up and top-down budgets agree within their uncertainty ranges. This is substantial progress toward defining more accurate global methane emissions.

The latitudinal breakdown inferred from the top-down approach reveals a dominant role of tropical emissions ($\sim$ 64 %) compared to mid-latitudes ($\sim$ 32 %) and high ($\sim$ 4 %) northern latitudes (above 60° N) emissions.

Our results, including an extended set of atmospheric inversions, are compared with the previous budget syntheses of Kirschke et al. (2013) and Saunois et al. (2016, 2020). They show overall good consistency when comparing the same decade (2000–2009) at the global and latitudinal scales. The magnitude and uncertainty of most natural or indirect anthropogenic sources have been revised and updated. In particular, this new budget benefits from large efforts and collaborations from the research community to provide improved estimates of the magnitude and uncertainty of the different freshwater sources and helps reduce the potential double counting at the global scale. Of note, newly available gridded data sets for lakes, ponds, reservoirs, streams, and rivers allow building latitudinal and regional estimates for all these sources for the first time in these estimates. In the next review, we hope to be able to reduce uncertainties in emissions from inland freshwater systems by better quantifying the emission factors of each contributing subsystems (streams, rivers, lakes, ponds) and estimating double counting at regional scale or avoiding double counting by better defining the surface areas of each ecosystem. Another important priority for improvements is the uncertainty on the chemical loss of CH₄, which still needs to be better assessed in both the top-down and the bottom-up budgets. Building on the improvement of the points detailed in Sect. 7, our aim is to update this budget synthesis as a living review paper regularly ($\sim$ every 3 or 4 years). Each update will produce a more recent decadal CH₄ budget, highlight changes in emissions and trends, and incorporate newly available data and model improvements.

It is still under debate why exactly there has been a sustained increase in atmospheric CH₄ (more than $+$5 ppb yr⁻¹) since 2007 (Nisbet et al., 2019; Nisbet et al., 2023; Turner et al., 2019). Some likely explanations, already introduced by Saunois et al. (2017) and further investigated by Jackson et al. (2020, 2024) and other studies, include, by decreasing order of certainty, the following: (1) a positive contribution from microbial and fossil sources (e.g. Nisbet et al., 2019, 2023; Schwietzke et al., 2016; Jackson et al., 2020), a negative contribution from biomass burning emissions before 2014 (Giglio et al., 2013; Worden et al., 2017); (2) a negligible role of Arctic emission changes (e.g. Nisbet et al., 2019; Saunois et al., 2017); and (3) a tropical dominance of the increasing emissions (e.g. Saunois et al., 2017; Jackson et al., 2020; Wilson et al., 2021; Drinkwater et al., 2023). Although the accelerated atmospheric methane growth rate in 2020 (15.2 ppb yr⁻¹) has found some explanation with the impact of the world pandemic in 2020, the sustained observed growth rates in 2021 (17.8 ppb yr⁻¹) and 2022 (14 ppb yr⁻¹) still challenge our understanding of the global methane cycle. While in Jackson et al. (2020, 2024) the increase in CH₄ emissions over the last 2 decades is attributed almost entirely to direct anthropogenic emissions, the uncertainty range from the GMB ensemble is large, and the contribution from natural emissions (wetlands) is still largely uncertain. Besides the decadal change in CH₄ emissions, large interannual variability can occur from these natural emissions. The recent high record of CH₄ growth rate highlights the potential of large variations from natural emissions from one year to another, in particular wetland emissions (e.g. Peng et al., 2022; Feng et al., 2023). These remain the challenges to be overcome in better quantifying global methane emissions.

Further investigation is needed in follow-up studies to (1) compare these results to the official UNFCCC declarations and to important assessment (as those of IEA) as done previously for example in Deng et al. (2022, 2024) or more specifically for fossil fuel emissions in Tibrewal et al. (2024) and (2) further discuss the trend and interannual variability of CH₄ sources and sinks at sectoral and regional scales as in Jackson et al. (2020, 2024), Stavert et al. (2021), or RECCAP2-related publications (e.g. Petrescu et al., 2021, 2023; Lauerwald et al., 2023b; Hugelius et al., 2024) and discuss the compatibility of the budget against the atmospheric isotopic signal such as in Saunois et al. (2017). The next budgets will be critical to assess whether the Global Methane Pledge is successful and assess methane mitigation efforts.

The GCP will continue to support and coordinate the development of improved flux estimates for all budget components and new underlying science to support improved modelling, acquisition of observations, and data integration. At regular intervals (3–4 years), we will continue to bring all flux components together to produce an improved and updated global CH₄ budget and provide a global benchmark for other CH₄ products and assessments.

Appendix A Table A1

Comparison of terminologies used in this study and previous reports for methane sources.

GCP terminology (this study)		IPCC AR6 (Canadell et al., 2021)	National GHG inventories (used by UNFCCC according to IPCC (2006) and IPCC (2019))	IPCC (2006, 2019) source sector numbering
Anthropogenic sources
Fossil fuels	Coal mining	Coal mining	Fugitive emissions from fuels/solid fuels	1B1
	Oil and gas	Oil and gas	Fugitive emissions from fuels/oil and natural gas	1B2
	Transport	Transport	Transport	1A3
	Industry	Industry	Mineral, chemical, metal industry, and others	2A, 2B, 2C, 2D, 2E
			Energy/fuel combustion activities	1A except 1A3 $+$ 1B3
Agriculture	Enteric fermentation and manure management	Enteric fermentation and manure management	Livestock	3A
	Rice cultivation	Rice cultivation	Rice cultivation	3C7
Waste	Landfills and waste	Landfills and waste	Waste	4
Biofuel and biomass burning	Biofuel burning	Biofuel burning	Biofuel burning	1A4b
	Biomass burning	Biomass burning	Biomass burning	3C1
Natural and indirect sources
Wetlands	Wetlands	Wetlands	–	–
Inland freshwaters	Reservoirs	Included in inland freshwaters	Land (including reservoirs)	in 3B
	Lakes, ponds, and rivers	Including in inland freshwaters	Only canal, ditches and ponds for human uses	in 3B
Other natural sources	Oceans	Oceans	–	–
	Termites	Termites	–	–
	Geological sources	Geological sources	–	–

Table A2

Summary of methodological changes since the previous budget (Saunois et al., 2020). No significant changes have been applied to the vegetation (Sect. 3.2.8), wild animal (Sect. 3.2.5), and terrestrial permafrost and hydrates (Sect. 3.2.7) estimates, though literature has been expanded and/or updated.

	Saunois et al. (2020)	This study
Region definitions (Table S1, Fig. S3)	18 continental regions $+$ ocean	same regions except the last region including only Australia and New Zealand and called Australasia
Anthropogenic global inventories (see Table 1, Sect. 3.1.1)	CEDS, EDGARv4.3.2, USEPA (2012), FAO, and GAINS ECLIPSE v6	CEDS, EDGARv6, and v7, USEPA (2019), FAO, IIASA GAINS v4 Add estimate of ultra emitters from Lauvaux et al. (2022)
Biomass burning data sets	FINNv1.5, GFASv1.3, GFEDv4.1s, QFEDv2.5	FINNv2.5, GFASv1.3, GFEDv4.1s, QFEDv2.5
Estimate of wetland emissions (see Tables 2 and S3 and Sect. 3.2.1)	13 land surface models involved, runs with either prescribed areas or based on hydrological scheme, single meteorological forcing	16 land surface models involved, runs with either prescribed areas or based on hydrological scheme, two sets of meteorological forcings
Estimate of reservoir emissions (Sect.3.2.2)	based on Deemer et al. (2016)	based on Johnson et al. (2021), Johnson (2021), Rosentreter et al. (2021), and Harrison et al. (2021)
Estimate of lake and pond emissions (Sect.3.2.2)	based on Bastviken et al. (2011), Wik et al. (2016b), and Tan and Zhuang (2015)	lakes $>$ 0.1 km2: based on Rosentreter et al. (2021), Zhuang et al. (2023), and Johnson et al. (2022a, b) lakes and ponds $<$ 0.1 km2: based on Rosentreter et al. (2021) and Johnson et al. (2022a, b)
Estimates of stream and river emissions (Sect. 3.2.2)	from Stanley et al. (2016)	based on Rosentreter et al. (2021), Rocher-Ros et al. (2023) and Rocher-Ros (2023)
Estimates of the anthropogenic perturbation component of inland freshwater emissions (Sect. 3.2.2)	–	based on several individual studies on the effect of eutrophication on emissions from lakes and ponds (see text in Sect. 3.2.2)
Estimate of the double counting in the aquatic systems (Sect. 3.2.2)	–	due to the counting of small lakes and ponds ($<$ 0.1 km2) in the vegetated wetlands areas used in land surface models and to lateral transport from vegetated wetland to rivers.
Geological sources (Sect. 3.2.3) – onshore and offshore	based on Etiope and Schwietzke (2019)	same as in Saunois et al. (2020)
Termite emissions (Sect. 3.2.4)	GPP: Zhang et al. (2017) termite biomass: Jung et al. (2011) EF: Kirschke et al. (2013) and Fraser et al., 1986)	GPP: Wild et al. (2022) termite biomass: based on different studies depending on regions (see text) EF: Sugimoto et al. (1998) Applied a correction factor for mound from Nauer et al. (2018)
Oceanic sources (Sect. 3.2.6)	modern biogenic: based on Wuebbles and Hayhoe (2002), Laruelle et al. (2013), and Rosentreter et al. (2018); geological: based on Etiope et al. (2019)	modern biogenic: based on Rosentreter et al. (2021, 2023) and Laruelle et al. (2025) geological: based on Etiope et al. (2019)
Tropospheric OH oxidation (Sect. 3.3.2) and stratospheric loss (Sect. 3.3.3) (see Supplement Table S4)	based on results from 11 models contributing to the Chemistry Climate Model Initiative (Morgenstern et al., 2017)	based on results from 11 models contributing to the Chemistry Climate Model Initiative 2022 (Plummer et al., 2021) and the CMIP6 simulations (Collins et al., 2017)
Tropospheric reaction with Cl	based on Hossaini et al. (2016), Wang et al. (2019b), and Gromov et al. (2018)	based on Hossaini et al. (2016), Sherwen et al. (2016), Wang et al. (2019b, 2021b), and Gromov et al. (2018)
Soil uptake (see Table S6)	based on Tian et al. (2016)	based on VISIT, JSBACH and MeMo surface models
Estimates through top-down approaches (see Tables S7 and S8 to S11)	9 inverse systems contributing, prior fluxes based on EDGARv4.2 or v4.3.2 for most inversions. Most inversion used constant OH.	7 inverse systems contributing, runs with constant and varying OH, prior fluxes based on either EDGARv6 or GAINS

Table A3

Funding supporting the production of the various components of the global methane budget in addition to the authors' supporting institutions (see also acknowledgements).

Funder and grant number (where relevant)	Authors/simulations/observations
Director, Office of Science, Office of Biological and Environmental Research of the US Department of Energy under contract no. DE-AC02-05CH11231 to Lawrence Berkeley National Laboratory as part of the RUBISCO Scientific Focus Area.	WJR, QZ, E3SM/ELM simulations
Funded by NASA's Interdisciplinary Research in Earth Science (IDS) programme and the NASA Terrestrial Ecology and Tropospheric Composition programmes	MSJ; lake and reservoir bottom-up methane emission data sets
Funded by Agence National de la Recherche through the project Advanced Methane Budget through Multi-constraints and Multi-data streams Modelling (AMB-M3) – (ANR-21-CE01-0030)	AM, MS
Funded by the National Science Foundation (grant no. 2143449)	MAH
The Environment Research and Technology Development Fund (JPMEERF21S20800) of the Environmental Restoration and Conservation Agency provided by Ministry of the Environment of Japan	YN, NISMON-CH4
Funded by the German Federal Ministry of Education and Research (BMBF) via the “PalMod” project (grant no. 01LP1921A) and EU H2020 (grant no. 951288 ERC Q-ARCTIC)	TK; CH4 emission modelling with JSBACH and LPJ-MPI
Funded by the Swedish Research Council VR (2020-05338) and Swedish National Space Agency (209/19)	WZ; LPJ-GUESS simulations
Funded by BELSPO (project FedTwin ReCAP), EU Horizon 2020 project ESM2025 (no. 101003536), and FRNS PDR project CH4-lake (T.0191.23)	PR; inland water, coastal and oceanic CH4 emission synthesis
EU H2020 (725546 ERC METLAKE and 101015825 TRIAGE), Swedish Research Councils VR (2022-03841) and Formas (2018-01794)	DB; inland waters – data and bottom-up estimation.
Supported by the Newton Fund through the Met Office Climate Science for Service Partnership Brazil (CSSP Brazil)	NG; JULES simulations
Funded by United Nations Environment Programme, Stanford University DTIE21-EN3143	RBJ; inversions and general budget support
The Joint Fund for Regional Innovation and Development of the National Natural Science Foundation (grant no. U22A20570); the Natural Sciences and Engineering Research Council of Canada (NSERC, grant no. 371706)	Changhui Peng/TRIPLEX-GHG
Computing Resources
LSCE computing resources	Marielle Saunois, Philippe Bousquet, Joël Thanwerdas and Adrien Martinez
NASA High-End Computing (HEC) programme through the NASA Advanced Supercomputing (NAS) Division at NASA Ames Research Center	Matthew S. Johnson (MSJ)
Deutsches Klimarechenzentrum (DKRZ), Hamburg, Germany	Thomas Kleinen (TK)
ALICE High Performance Computing Facility at the University of Leicester	GOSAT retrievals
FUJITSU PRIMERGY CX2550M5 at MRI and NEC SX-Aurora TSUBASA at NIES	Yosuke Niwa (YN)
Support for atmospheric observations
Australian Antarctic Division	CSIRO flask network
Australian Institute of Marine Science	CSIRO flask network
Bureau of Meteorology (Australia)	Kennaook/Cape Grim AGAGE, CSIRO flask network
Commonwealth Scientific and Industrial Research Organisation (CSIRO, Australia)	Kennaook/Cape Grim AGAGE, CSIRO flask network
Department of Climate Change, Energy, the Environment and Water (DCCEEW, Australia)	Kennaook/Cape Grim AGAGE
Meteorological Service of Canada	CSIRO flask network
NASA: grants NAG5-12669, NNX07AE89G, NNX11AF17G, NNX16AC98G and 80NSSC21K1369 to MIT with subawards to the University of Bristol (for Barbados and Mace Head) and CSIRO (for Kennaook/Cape Grim); grants NAG5-4023, NNX07AE87G, NNX07AF09G, NNX11AF15G, NNX11AF16G, NNX16AC96G, NNX16AC97G, 80NSSC21K1210 and 80NSSC21K1201 to SIO.	AGAGE calibrations and measurements at SIO, La Jolla and AGAGE station operations at Trinidad Head, Mace Head, Barbados, American Samoa, and Kennaook/Cape Grim
National Oceanic and Atmospheric Administration (NOAA, USA) contract RA133R15CN0008 to the University of Bristol	Barbados
NOAA USA	CSIRO flask network
Refrigerant Reclaim Australia	Kennaook/Cape Grim AGAGE
UK Department for Business, Energy & Industrial Strategy (BEIS) contract TRN1537/06/2018 and TRN 5488/11/2021 to the University of Bristol	Mace Head
National Oceanic and Atmospheric Administration (NOAA, USA)	Cape Matatula
Japanese Ministry of Environment	GOSAT data, Robert Parker
Japanese Aerospace Exploration Agency, National Institute for Environmental Studies	GOSAT data, Robert Parker
UKRI UK: grants NE/W004895/1, NE/R016518/1, NE/X019071/1 and MR/X033139/1	GOSAT data, Robert Parker
The Swedish Research Council VR (2022-04839), European Space Agency projects AMPAC-Net and CCI$+$ permafrost, European Union's Horizon 2020 Research and Innovation Programme to the Nunataryuk project (no. 773421)	Permafrost region, Gustaf Hugelius

Note on former version

A former version of this article was published on 15 July 2020 and is available at 10.5194/essd-12-1561-2020.

The supplement related to this article is available online at https://doi.org/10.5194/essd-17-1873-2025-supplement.

Author contributions

MS, AM, and JT gathered the bottom-up and top-down data sets and performed the post-processing and analysis.

MS, BP, PB, PeC, and RJ coordinated the global budget. MS, BP, PB, PeC, RJ, PP, and PCi contributed to the update of the full text, and all coauthors appended comments. AM, ED, and XL produced the figures. DJB, NG, PH, AI, AJ, TK, TL, XL, KMcD, JMe, JMu, SP, CP, WR, HT, YY, WZ, ZZ, Qing Z, Qiuan Z, and Qianlai Z performed surface land model simulations to compute wetland emissions. GA, DB, SC, BRD, GE, MAH, GH, MSJ, RL, SN, GRR, JAR, EHS, PRa, PRe, and TSW provided data sets useful for natural emission estimates and/or contributed to text on bottom-up natural emissions. LHI, SJS, TNF, GRvW, and MC provided anthropogenic data sets and contributed to the text for this section. AM, JT, PP, DBe, RJ, YN, AS, AT, and BZ performed atmospheric inversions to compute top-down methane emission estimates for sources and sinks. EJD, XL, DRB, PBK, JM, RJP, MR, MS, DWo, and YYo are PIs of atmospheric observations used in top-down inversions and/or contributed the text describing atmospheric methane observations. FD, MS, and JT contributed to the bottom-up chemical sink section by providing data sets, processing data, and/or contributing to the text. FMF provided data for the soil sink.

Competing interests

At least one of the (co-)authors is a member of the editorial board of Earth System Science Data. The peer-review process was guided by an independent editor, and the authors also have no other competing interests to declare.

Disclaimer

Publisher’s note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors.

Acknowledgements

This paper is the result of a collaborative international effort under the umbrella of the Global Carbon Project, a project of Future Earth and a research partner of the World Climate Research Programme (WCRP). We acknowledge all the people and institutions who provided the data used in the global methane budget as well as the institutions funding parts of this effort (see Table A3). We are very grateful for the help provided by Alex Vermeulen in publishing the Global Methane Budget data set on the Integrated Carbon Observation System (ICOS) website. We acknowledge the modelling groups for making their simulations available for this analysis, the joint WCRP Stratosphere-troposphere Processes And their Role in Climate/International Global Atmospheric Chemistry (SPARC/IGAC) Chemistry-Climate Model Initiative (CCMI) for organising and coordinating the model data analysis activity, and the British Atmospheric Data Centre (BADC) for collecting and archiving the CCMI model output. We acknowledge the long-term support provided by the Commonwealth Scientific and Industrial Research Organisation (CSIRO) and the National Environmental Science Program – Climate Systems Hub to coordinate and support activities of the Global Carbon Project. We are grateful to the Emissions Database for Global Atmospheric Research (EDGAR) team (Monica Crippa, Diego Guizzardi, Frederico Pagani, Manjola Banja, Edwin Schaaf, Marilena Muntean, William Becker and Fabio Monforti-Ferrario) for the work needed to publish the EDGAR greenhouse gas emission data sets used in this work (https://edgar.jrc.ec.europa.eu/, last access: 1 April 2025). We are particularly indebted to the dedicated station/instrumental operators/scientists that have gathered the data and ensured their high quality.

More specifically, we acknowledge Adriana Gomez-Sanabria, for her work on waste sector emissions in the GAINS model; Katherine Jensen, for her contribution to the Surface Water Microwave Product Series SWAMPS; Fortunat Joos, for his contribution to simulations with the Land surface Processes and eXchanges model (LPX Bern); Ray Langenfelds and Elise Guerette, for their contribution to the CSIRO network; Paul Miller, for his contribution to simulations with the Lund–Potsdam–Jena General Ecosystem Simulator (LPJ-GUESS); Peng Shushi, for his contribution to simulations with the Organising Carbon and Hydrology In Dynamic Ecosystems (ORCHIDEE) model; Shamil Maksyutov, for his contribution to simulations with the inverse model at the National Institute for Environmental Studies (NIES); Isobel Simpson, for her contribution to the University of California Irvine (UCI) network; Paul Steele, for his former contribution to the CSIRO network; Ray Weiss, for his contribution to the Advanced Global Atmospheric Gases Experiment (AGAGE) network; Christine Widenmeyer, for her contribution to the Fire INventory from the National Center for Atmospheric Research (FINN) database; Xiaoming Xu, for his contribution to simulations with the Integrated Science Assessment Model (ISAM); Yuanzhi Yao, for his contribution to simulations with the Dynamic Land Ecosystem Model (DLEM); Diego Guizzardi, for his contribution to EDGAR; Maria Tenkanen, for her contribution to the Carbon Tracker – Europe (CTE) outputs; and Giulia Conchedda, for her contribution to the Food and Agriculture Organization (FAO) database. FAOSTAT data collection, analysis, and dissemination are funded through FAO regular budget funds. The contribution of relevant experts in member countries is gratefully acknowledged. We acknowledge Juha Hatakka from the Finnish Meteorological Institute (FMI) for making methane measurements at the Pallas station and sharing the data with the community. We thank Ariana Sutton-Grier and Lisamarie Windham-Myers for reviewing an earlier version of the manuscript. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the US Government. We warmly thank the three anonymous reviewers for their time and their feedback, which greatly improved the article.

Financial support

This research has been supported by the Biological and Environmental Research programme (grant no. DE-AC02-05CH11231), the Agence Nationale de la Recherche (grant no. ANR-21-CE01-0030), the Environmental Restoration and Conservation Agency (grant no. JPMEERF21S20800), the Bundesministerium für Bildung und Forschung (grant no. 01LP1921A), Vetenskapsrådet (grant nos. 2020-05338 and 2022-03841), the Swedish National Space Agency (grant no. 209/19), the Belgian Federal Science Policy Office (grant no. FedTwin ReCAP), EU Horizon 2020 (grants nos. 101003536, 951288 ERC Q-ARCTIC, 725546 ERC METLAKE, and 101015825 TRIAGE), the Fonds De La Recherche Scientifique – FNRS (grant no. T.0191.23), the Svenska Forskningsrådet Formas (grant no. 2018-01794), the United Nations Environment Programme (grant no. DTIE21-EN3143), and the Natural Sciences and Engineering Research Council of Canada (grant no. NSERC, no. 371706). Atul K. Jain is supported by the NASA LCLUC programme (award no. 80NSSC24K0920). All funding and support are detailed in Table 3.

Review statement

This paper was edited by Yuyu Zhou and reviewed by three anonymous referees.