Regional trends and drivers of the global methane budget

Abstract The ongoing development of the Global Carbon Project (GCP) global methane (CH4) budget shows a continuation of increasing CH4 emissions and CH4 accumulation in the atmosphere during 2000–2017. Here, we decompose the global budget into 19 regions (18 land and 1 oceanic) and five key source sectors to spatially attribute the observed global trends. A comparison of top‐down (TD) (atmospheric and transport model‐based) and bottom‐up (BU) (inventory‐ and process model‐based) CH4 emission estimates demonstrates robust temporal trends with CH4 emissions increasing in 16 of the 19 regions. Five regions—China, Southeast Asia, USA, South Asia, and Brazil—account for >40% of the global total emissions (their anthropogenic and natural sources together totaling >270 Tg CH4 yr−1 in 2008–2017). Two of these regions, China and South Asia, emit predominantly anthropogenic emissions (>75%) and together emit more than 25% of global anthropogenic emissions. China and the Middle East show the largest increases in total emission rates over the 2000 to 2017 period with regional emissions increasing by >20%. In contrast, Europe and Korea and Japan show a steady decline in CH4 emission rates, with total emissions decreasing by ~10% between 2000 and 2017. Coal mining, waste (predominantly solid waste disposal) and livestock (especially enteric fermentation) are dominant drivers of observed emissions increases while declines appear driven by a combination of waste and fossil emission reductions. As such, together these sectors present the greatest risks of further increasing the atmospheric CH4 burden and the greatest opportunities for greenhouse gas abatement.

(See Patra et al., 2014 and references therein). The large-scale geographical breadth of many of the chemical and physical drivers in OH spatial variability associated with intra-and interhemispheric transport of CH 4 over 1 year (Anderson et al., 2021) make the OH sink incompatible with a geo-politically defined surface-region-based study (the focus of this paper). As a result, only sources of CH 4 are discussed at regional scale in this study.
A fourth sink is the biological consumption of CH 4 by methanotrophic microbes in soils and other environments. It is estimated to be responsible for 5% of the global total CH 4 sink .
This process occurs in aerobic soils and inland waters where methanotrophic bacteria consume CH 4 and produce CO 2 (Le Mer & Roger, 2001). The oxidation of CH 4 can occur at the same site as the production of CH 4 either concurrently (e.g., aerobic soils capping landfills) or temporally shifted (e.g., sediments shifting from anoxic to oxic with seasonal waterflow patterns). As such, the oxidative process can play a role in controlling CH 4 emissions to the atmosphere. In this paper, rather than separating production and consumption we focus on the policy relevant net surface emissions and exclude discussion of some spatially diffuse non-anthropogenic fluxes (e.g., upland soil sinks).
This paper analyzes the recently published global CH 4 budget Saunois et al. (2020) and conducts an in-depth examination of CH 4 (total and sectoral) emission trends in 19 regions. Here we examine regions defined using geo-political boundaries and focus on net surface emissions (rather than separating production and consumption) because they are more directly relevant to policy analyses and the development of effective regional mitigation strategies. We use, as in Saunois et al., 2020, multiple emissions estimates from top-down (TD) approaches, based on atmospheric constraints, and bottom-up (BU) methods, which use extensive data inventories, terrestrial biospheric modelling, and the statistical upscaling of empirical data.
In presenting data from a range of BU and TD methods this paper reflects the current state of science in relation to global-scale estimates of regional CH 4 emissions. By also clearly and concisely articulating their current limitations we build the groundwork for future research which will incorporate improvements in TD and BU methodologies. Using multiple methodologies allows us to estimate not only the uncertainty of these regional emissions (using the spread in emissions estimates) but also the uncertainty in the regional trends (using the spread in emissions trends). Although not totally independent of BU methods (see Section 2.2), TD methods rely on CH 4 atmospheric mole fraction measurements which are an independent observational constraint. As such, commonalities between multiple BU and TD estimates can identify robust trends and patterns in regional CH 4 emissions and differences suggest areas requiring further investigation.
A regional decomposition (rather than latitudinal as in Saunois et al. (2020)) is essential as the drivers of CH 4 emissions vary widely in their spatial distribution and relative importance with climate, ecosystem type, anthropogenic activities and environmental policies. We also examine both decadal and annual data for the 2000-2017 period allowing the identification of trends and policy impacts.
We then examine four key questions pertinent to the design and implementation of emission reduction strategies: 1. Which regions are the largest contributors to global CH 4 emissions? 2. Which activities or sources are driving these emissions?-natural vs. anthropogenic sources and specific source sector types.
3. Which regions are the dominant contributors to the increasing trend in global CH 4 emissions and which sectors are driving this rise?
4. Are any regions seeing a decline in CH 4 emissions and which sectors are responsible for this decline?
We also examine the key differences in TD and BU estimates, using them to identify methodological limitations and to highlight uncertainties and areas requiring further investigation.

| ME THODS AND MATERIAL S
The spatial distribution of sources and sinks of CH 4 can be quantified using TD or BU approaches. BU approaches include processbased models, inventories, satellite-based products, and other data sets, which collate and scale-up local (or regional) direct flux measurements (e.g., flux towers and chamber-based approaches). Most commonly, BU-based estimates of anthropogenic CH 4 emissions are determined using inventory methods (e.g., EDGAR v4.3.2, Janssens- . Here, known regionally distributed activity information and statistics are combined with technology-specific emission factors and abatement factors which account for recapture and emissions mitigation technology. Satellite-derived biomass burning products combine estimates of the area burned, biomass loading, combustion completeness and biome-specific emission factors to estimate CH 4 emissions (e.g., van der Werf et al. (2017)).
Process-based models, for example wetland models (e.g., Poulter et al. (2017)), can also be used to estimate sector-specific regionally distributed CH 4 emissions. Lastly, a variety of methods have been used to statistically scale up small-scale or sparse empirical measurements of CH 4 flux to regional and global totals (e.g., Etiope et al., 2019). Regional studies, which typically use region specific data sets and methodologies, are not included in this analysis as they may bias cross-region comparisons. Instead, where appropriate, regional studies are used to inform the discussion of trends identified in the global-scale regionally distributed data sets.
TD approaches take observed atmospheric mole fractions of CH 4 and combine them with global (or regional) transport models and an inversion framework to calculate an optimum set of spatially and temporally distributed source and sink fluxes. Studies using this approach are plentiful and use a variety of transport models, inversion methods, atmospheric data sets (in situ, flask, total column ground-based and satellite) and prior flux estimates. Examples include Bergamaschi et al. (2018), Maasakkers et al. (2019) and Yin et al. (2021). Some methods aim to separate certain emission sectors based on differences in their spatial and temporal distributions (e.g., Bergamaschi et al. (2013)), while others only solve for net emissions at the surface (Chandra et al., 2021). Then the partitioning of TD posterior (output) fluxes between specific source sectors (e.g., Fossil vs. BB&F) is carried out with various degrees of uncertainty depending on the methods and the degree of refinement of sectors.
Here, we use a subset of the ensemble of TD and BU data sets gathered in Saunois et al. (2020) (Hoesly et al., 2018). For ease of reading, these will henceforth be referred to as EDGAR, GAINS, USEPA, and CEDS. It is important to note that while included in the analysis, the CEDS inventory (Hoesly et al., 2018) is based partly on an earlier data set (EDGARv4.2) which is known to overestimate Chinese coal emissions (Peng et al., 2016). As FAO does not provide Waste, Fossil and Biofuel emission estimates we, unlike Saunois et al. (2020), do not include these in our analysis.

| Wetlands
Regional BU Wetl estimates were examined using 13 models from the wetland model intercomparison conducted as part of the 2020 GCP CH 4 budget project (GCP, 2021). These model runs use a common wetland extent map (diagnostic)  while nine prognostic runs (which were not used for regional estimates but do inform the discussion) used a range of internally generated wetland extent maps (see supplement Section 4 of Saunois et al. (2020)).
Mean regional BU Wetl estimates were calculated using the data from the diagnostic wetland model runs.

| Non-wetland sources including oceanic
Four BU estimates of total other natural non-wetland (NonWetl) sources were calculated in this study. They ranged between 177-186 Tg CH 4 yr −1 and were determined as the sum of the three main sources: aquatic biogenic (including inland waters, oceanic, estuaries and blue carbon, i.e., mangroves, seagrasses and marsh systems), geological (land and ocean), and termite.
Here, we use a new estimate of global inland water and estuarine fluxes (Section S2). It is the first global-scale regionally distributed data set of this nature and was calculated as the sum of lake and reservoir, river and stream and estuarine fluxes. It was developed specifically for this study and not available at the time of publication for Saunois et al. (2020).
The biogenic oceanic estimate calculated by Saunois et al. (2020), 6 Tg CH 4 yr −1 (range 4-10 Tg CH 4 yr −1 ), included oceanic, estuarine and blue carbon fluxes. But estuarine and blue carbon fluxes occur at the interface between land and marine waters (i.e., straddling both land and ocean regions). In this study we treat the global oceans as a single region, hence to spatially distribute the estuarine and blue carbon fluxes they have been allocated to the adjacent land, rather than the ocean region. As such, the ocean biogenic component used in this study (3 Tg CH 4 yr −1 ) was calculated as the difference between the Saunois et al. (2020) total oceanic biogenic flux (6 Tg CH 4 yr −1 ) and the total of the new independent regionally distributed estimate of estuarine flux (3 Tg CH 4 yr −1 ).
Only a single spatially distributed estimate of geological CH 4 emissions, 37 Tg CH 4 yr −1 , is available (Etiope et al., 2019). This estimate is lower than the mean global total reported by Saunois et al. It is important to note that although these four sectors represent the four largest natural non-wetland CH 4 emissions, other natural flux types, such as wild animals, are excluded as their regional distribution was either unknown or poorly characterized. Global estimates of these omitted fluxes are very small, ranging between 1 and 4 Tg CH 4 yr −1 , <0.6% of total emissions .
Considering the magnitude of the uncertainties in the inland waters, oceanic, and termite estimates (>100%), it is expected that this omission is easily encompassed in the inherent uncertainty of the NonWetl estimates. There are three key details to note with TD inversions. Firstly, they fit to the observed atmospheric CH 4 mole fraction (i.e., the difference between CH 4 sources and sinks) rather than directly fitting to the individual sources or sinks. As part of this process all but two inverse models (TOMCAT and PYVARLMDz, Table S1) used a spatially varying but 12-month climatological prior estimate of the global CH 4 sink. This assumption is a significant limitation as this sink can vary interannually (Zhao et al., 2019). As such, for TD methods, some temporal variability in the chemical sink may instead be assigned to temporal variability in the source estimates. Similarly, variability in the spatial distribution of the chemical sink may also drive variability in the spatial distribution of the TD-derived regional CH 4 sources.

| Top-down data
However, as five estimates of the chemical sink spatial distribution were used in our suite of inversions this source of uncertainty should be reflected in the spread of the ensemble of the inversions.
Secondly, rather than optimizing for individual source sectors (e.g., Fossil or Wetl fluxes) some inversions optimize for the total flux for each region or pixel and then partition it to source sectors using a  (Table S1). Thirdly, TD methods generally use BU estimates as prior knowledge and as such BU and TD methods cannot be considered as truly independent.

| Data processing
Data sets were provided on a variety of grids, at different temporal resolutions and over a range of time periods (Supplementary Section S1 and references within Table 1). These data sets were regridded, projected to a common grid and interpolated or extended to a common temporal resolution and time period (Supplementary Section F I G U R E 1 The locations of the surface (triangle) and ground-based profile (circle) observation sites and GOSAT XCH 4 data density and extent (number of data points per grid cell per month for NIES full physics retrievals) for 2014. Grid cells are 2.5 × 2.5°. Note that many of the in situ sites are not operational or are not reporting data to the Global Atmospheric Watch repository. Also note that many of the GOSAT grids do not have uniform data coverage in all months as the instrument cannot see through the cloud covered areas and during polar nights [Colour figure can be viewed at wileyonlinelibrary.com] S1). Regions (18 land and 1 ocean) were constructed using geopolitical boundaries and chosen on the basis of size, geopolitical importance, and vegetation type ( Figure 4 and Saunois et al. (2020) Table   S1). Box plots of emissions estimates were calculated for the period 2008-2017 (Supplementary Section S3). Rates of change in regional and global emissions estimates were calculated over the 2000-2017 period as described in the Supplementary (Section S4). Comparisons of TD and BU regional emissions were made for the total regional CH 4 emission, the total natural and total anthropogenic emissions and sectoral emissions. In an effort to reduce the number of BU total emissions estimates, the mean of the NonWetl estimate (i.e. the sum of inland waters, termites, oceans, and geological fluxes) and the mean Wetl flux were combined with the five anthropogenic estimates to produce five BU estimates of total emissions. However, comparisons between the TD and BU estimates of Wetl fluxes were made using the full suite of 13 BU estimates and the mean. Similarly, TD and BU comparisons of NonWetl fluxes were made using five BU estimates (one for each termite estimate) and the mean.

| Top-down estimates
A total of 22 different gridded TD total source estimates were provided from the nine inversion systems included in Saunois et al. (2020). Of these, half were based solely on surface CH 4 data (TD SURF) and half included GOSAT satellite data (TD GOSAT). Separate TD SURF and TD GOSAT annual and decadal means were calculated globally and for each region. Unlike Saunois et al. (2020), here the decadal mean, median, and box plot calculations are based only on the SURF data. In some cases, multiple simulations were provided for a given inversion system. To avoid overweighting these inversion systems, the mean of each inversion system's simulations, rather than each individual simulation, was used to calculate the mean timeseries. However, to demonstrate the spread in the TD estimates the individual simulations were included in the time series figures.
See Table S2 for further information.
Total TD anthropogenic emissions were calculated separately for each inversion simulation as the sum of the Fossil, BB&F, and Ag&Waste emissions. Similarly, total TD natural emissions were calculated separately for each inversion simulation as the sum of the NonWetl (including ocean emissions) and Wetl emissions.

| RE SULTS
We first focus on commonalities between the BU and TD estimates, identifying robust features in the spatial distribution and temporal trends of CH 4 emissions with the aim of addressing the four key questions outlined earlier (Section 3.1). Second, we highlight key TD and BU differences (Section 3.2).
In  2003-2012, and 2008-2017 decadal means show these regional rankings to be fairly consistent over time (Table S3). Information in relation to latitudinal flux distribution can be found in Table 5 (Figure 3). In comparison, regional anthropogenic emissions can be divided into three groups, those >40 Tg CH 4 yr −1 , those 20 to 40 Tg CH 4 yr −1 and those <20 Tg CH 4 yr −1 (Figure 3 and Figure S2a).  Wetl emissions are also a significant proportion of total emissions in these regions with 45% (58%), 39% (38%), and 41% (46%) of the total regional emissions coming from Wetl for Brazil, Southeast Asia, and

| Total emission rates and temporal trends
The overall temporal trends in TD and BU CH 4 emission means are very similar within each region, with all but one region (Canada) showing agreement in the direction (positive or negative) of trends Only Southeast Asia and Russia show similar patterns in TD and BU mean IAV. A sectoral decomposition of these regions ( Figure 6) shows that this IAV is driven by a combination of Wetl and BB&F.
Although the absolute magnitude of the TD and BU total emissions estimates differ significantly for most regions, there are some regions, the United States, Southeast Asia, and the three African regions, which appear to agree well (average absolute difference 2000 to 2017 < 15%, SURF data only). On average, absolute differences between regional TD and BU estimates decrease by 3% with the addition of the GOSAT data. Agreement tends to improve in equatorial regions with a paucity of long-term ground-based observations

| Total CH 4 emissions
As identified in Saunois et al. (2020)  For 18 of the 19 regions TD total CH 4 posterior emission estimates are closer to their prior than the BU estimate ( Figure 2).
However, this trend is not consistent across the subsectors (e.g., Fossil, Figure S3b). For 16 of the 19 regions examined median regional BU estimates of total CH 4 flux are higher than posterior TD estimates ( Figure 2). However, only four regions (Canada, Central Asia, Russia, and Oceania, Figure 2) do not agree within the 25th-75th percentile range, with all these regions showing total BU ≫TD.
As with global estimates regional BU-TD discrepancies are driven by BU NonWetl emissions estimates ( Figure S3a-d cf. S3e), primarily elevated Inland Water emissions ( Figure S9).
Although TD sectoral partitioning can be problematic (Section 2.2), it is interesting that for the two land regions with higher TD  Table 2 Saunois et al. (2020) for further information on individual wetland models). In addition, the mean of BU wetland models run using a prognostic extent mode (where wetland extent is determined by the model) is significantly closer to those of the TD estimates for these two regions, 40 and 11 Tg CH 4 yr −1 for Brazil and South Asia, respectively. While the higher TD Wetl estimates may not be correct there are known issues with wetland mapping, particularly in flooded forest regions where closed canopies make remote sensing of water-saturated soils challenging (Prigent et al., 2020). As such, the TD-BU Wetl discrepancy along with the difference in prognostic and diagnostic BU wetland models suggests that there may be room for improvement in the diagnostic wetland map in the Brazilian and South Asian regions. It is also interesting to note that the addition of the GOSAT data to the TD inversion estimate decreases the TD Wetl estimate in the South Asia region by approximately 5 Tg CH 4 yr −1 . This result suggests that where surface site numbers are limited ( Figure 1) the TD-BU discrepancy may be partially driven by data coverage, emphasizing the importance of satellite data.

| Sectoral CH 4 emissions
As discussed earlier (Section 3.2.1), high BU Inland Waters estimates drive discrepancies in TD and BU total emissions estimates. These also lead to higher BU estimates of natural emissions ( Figure 3a) and in turn (for all regions other than South Asia), higher TD estimates of anthropogenic proportion (Figure 3c). The discrepancy in the Central Asia region is particularly large with a median TD anthropogenic proportion of 89% and a BU estimate of 37%. This difference is driven by the far higher (~15 times) BU estimate of natural emissions, 86% of which is Inland Waters ( Figure S9). Unlike land region estimates, in the ocean region a higher TD estimate of anthropogenic proportion, is driven by a higher (~3.5 times) TD estimate of anthropogenic Fossil emissions. and Fossil emissions for many regions. While there can be significant methodologically driven uncertainties with TD sectoral estimates (Section 2), when optimising for total CH 4 fluxes seven of the nine TD estimates (Table S1)  However, limiting the analysis to the remaining three BU inventories only reduces the annual mean TD and BU difference by 20%. In fact, the mean BU estimate (CEDS excluded) is consistently higher than all but two of the individual TD estimates with only one of the BU estimates, USEPA, within the range of the majority of the TD estimates. A closer examination of the sub-sectoral breakdown of the BU inventories suggests that the EDGAR and GAINS Fossil estimates are higher than the other inventories due to a combination of higher Oil&Gas emissions (10 times the USEPA inventory, Figure S4) and Coal emissions (up to 50% higher than the USEPA inventory, Figure   S4).
There are large discrepancies between the TD and individual BU inventory estimates of Northern South American Fossil emissions ( Figure S10). The CEDS estimate is consistently (> 4 times) higher than the other three BU inventories and the TD estimates ( Figure   S4 and S10). Interestingly, a closer examination of the gridded CEDS data shows the majority of these emissions occurring from a single point source in Venezuela. This assessment agrees with other inversion studies (Maasakkers et al., 2019;Y. Zhang et al., 2020), which suggest an under-estimate in Venezuelan Oil&Gas emissions in the EDGAR and UNFCCC estimates. However, Maasakkers et al. (2019) only revised these estimates to ~5 Tg CH 4 yr −1 , that is, far less than the ~12 Tg CH 4 yr −1 allocated by the CEDS emission inventory.

| Emissions and policy relevance
Four

This growth is driven predominantly by increases in Coal and
Ag&Waste emissions, with BU inventories suggesting that the increase in Chinese Coal emissions alone accounts for >15% of the global total emissions increase. Increases in Chinese coal mining activities have been linked to rapid economic growth, which relies predominantly on coal-based energy sources (Jackson et al., 2018;Miller et al., 2019;Peters et al., 2017). However, recent Chinese inventory studies suggest a decline in coal mining emissions since 2012 (Gao et al., 2020;Sheng et al., 2019). Similarly, increases in Chinese Waste emissions have been driven by economic growth and urbanization. This sector, however, has great mitigation potential as previous work reports that changes in landfill procedures could lead to a 30%-50% reduction in landfill emissions (Cai et al., 2018). The increases in agricultural CH 4 emissions are being driven by diet changes associated with the rising standard of living and the promotion of the agricultural economy (Tian et al., 2014;Zhang & Chen, 2010). Tian et al. (2014) also emphasize that given China's economic dependence on agriculture, decreases in agricultural emissions will only come from changes in production methods not a reduction in agricultural capacity. and future reductions may instead need to come from changes in farm management, feed composition, and additives and selective breeding (EIP-AGRI, 2017;Pellerin et al., 2014;Roque et al., 2021). Several TD studies have found that USA anthropogenic (in particular Oil&Gas and Livestock) emissions are underestimated in BU inventories (e.g., Maasakkers et al., 2021;Williams et al., 2021;Y. Zhang et al., 2020). Considering that the United States is the third largest contributor to global anthropogenic emissions and that BU, rather than TD, estimates are used to track and assess policy outcomes, resolving this discrepancy would be valuable. However, while the median BU USA anthropogenic emissions estimate (28 Tg CH 4 yr −1 for 2008-2017) is lower than the TD estimate (31 Tg CH 4 yr −1 ), they do agree within the 25th-75th percentile range ( Figure 3) and as such our data suggest that the methods may well be consistent.

| Methodological and science implications
In this paper, we explored a range of TD and BU emission estimates with the aim of providing policy-relevant information with sectorspecific emissions and trend data that are necessary for developing Other key developments in TD approaches include improvements in atmospheric transport models (especially vertical transport, stratospheric transport and the simulation of boundary layer dynamics), methodological improvements in TD sector-specific flux estimation using isotopic tracers and co-emitted compounds (e.g., ethane) as additional constraints and the use of the latest satellite data (e.g., TROPOMI http://www.tropo mi.eu). The Coal, Waste, and Livestock subsectors are common drivers of rapid increases and decreases in regional CH 4 emissions. As such, these sectors present the best opportunities for CH 4 emissions mitigation and the largest risks (drivers) of on-going emission increases.

| CON CLUS IONS
Regions with significant natural emissions (e.g., Brazil with 22-34 Tg CH 4 yr −1 from Wetl) are also key, with the influence of climate change on these sources an area of concern. Some studies (e.g., Zhang et al., 2018) suggest that changes in temperature and precipitation patterns may have already led to increases in CH 4 Wetl emissions growth rates.
However, as these flux types have no or limited mitigation potential the focus when implementing emission reduction schemes should be on regions with a large proportion of anthropogenic emissions.
Considering the large contribution of Global CH 4 emissions to the Earth's radiative budget, the predominance of increasing (rather than decreasing) CH 4 emissions is a large concern. However, the emissions decline observed in Europe and Korea and Japan provides evidence that a decrease in emissions in response to government policies and/or economic forces is possible. Bruxelles, Brussels, Belgium who, with P. Regnier, developed the regionally distributed estuarine flux data set.

CO N FLI C T O F I NTE R E S T
The authors declare that there are no competing financial interests.

DATA AVA I L A B I L I T Y S TAT E M E N T
A number of data sets were used to support the findings of this study. These data sets are either openly available in repositories as listed in the text and references, available on request from the original developers (see text and references there in) or are available from the corresponding author upon reasonable request.