An integrated analysis of contemporary methane emissions and concentration trends over China using in situ and satellite observations and model simulations

. China, being one of the major emitters of greenhouse gases, has taken strong actions to tackle climate change, e.g., to achieve carbon neutrality by 2060. It also becomes important to better understand the changes in the atmospheric mixing ratios and emissions of CH 4 , the second most important human-inﬂuenced greenhouse gas, in China. Here we analyze the sources contributing to the atmospheric CH 4 mixing ratios and their trends in China over 2007–2018 using the GEOS-Chem model simulations driven by two commonly used global anthropogenic emission inventories: the Emissions Database for Global Atmospheric Research (EDGAR v4.3.2) and the Community Emissions Data System (CEDS). The model results are interpreted with an ensemble of surface, aircraft, and satellite observations of CH 4 mixing ratios over China and the Paciﬁc region. The EDGAR and CEDS estimates show considerable differences reﬂecting large uncertainties in estimates of Chinese CH 4 Chinese CH 4 emission estimates based on EDGAR and natural sources increase from 46.7 Tg per annum (Tga − 1 ) in 1980 to 69.8 Tga − 1 in 2012 with an increase rate of 0.7 Tga − 2 , and estimates with CEDS increase from 32.9 Tga − 1 in 1980 and 76.7 Tga − 1 in 2014 (a much stronger trend of 1.3 Tga − 2 over the period). Both surface, aircraft, and satellite measurements indicate CH 4 increase rates of 7.0–8.4 ppbva − 1 over China in the decade. We ﬁnd that the model simulation using the CEDS inventory and interannually varying OH levels can best reproduce these observed CH 4 mixing ratios and trends over China. Model results over China are sensitive to the global OH level, with a 10 % increase in the global tropospheric volume-weighted mean OH concentration presenting a similar effect to that of a 47 Tga − 1 decrease in global CH 4 emissions. We further apply a tagged tracer simulation to quantify the source contributions from different emission sectors and regions. We ﬁnd that domestic CH 4 emissions 14.0 % of the mean surface mixing ratio and drive 66.7 % of the surface trend (mainly via the energy sector) China over We emphasize that intensive CH 4 measurements covering eastern China will help us better assess the driving factors of CH 4 mixing ratios and support the emission mitigation in China.


Introduction
Atmospheric methane (CH 4 ) is the second most important anthropogenic greenhouse gas, contributing more than a quarter of the human-induced radiative imbalance since 1750 (IPCC, 2013). It also plays an important role in atmospheric chemistry as an essential precursor for tropospheric ozone and stratospheric water vapor (Turner et al., 2019). Global mean atmospheric CH 4 surface mixing ratios increased from about 1650 ppbv in the mid 1980s to about 1770 ppbv in the late 1990s, then stabilized around this level in the early 2000s, and started increasing again from 2007 Nisbet et al., 2019). The regrowth of the atmospheric CH 4 mixing ratio has drawn worldwide attention and led to many different or even contradictory explanations Yin et al., 2021;Zhao et al., 2019;Turner et al., 2019;Maasakkers et al., 2019). Difficulties in the attribution of the trends are mainly associated with large uncertainties in changes in the CH 4 emissions from various sources, as well as the chemical loss via oxidation by hydroxyl radical (OH) (Turner et al., 2019). A better understanding and quantification of the interannual variability in CH 4 emissions and the drivers of the concentration growth in the recent decade is important to support its mitigation. CH 4 has both important anthropogenic and natural sources. It can be emitted from human activities including coal mining, oil and gas exploitation, livestock, rice cultivation, waste deposit, and wastewater treatment. It also has a large natural source from wetlands, with small sources from forest fires, termites, and geological seeps. Global bottom-up estimates of CH 4 emissions based on statistics of source activities or process-based models have reported a wide range of total CH 4 emissions of 542-852 Tg a −1 in the 2000s (Kirschke et al., 2013). Atmospheric top-down analyses constrained by surface, satellite, and aircraft observations of CH 4 mixing ratios tend to suggest lower total CH 4 emissions of 526-569 Tg a −1 in the period (Kirschke et al., 2013) and find even greater uncertainties in the relative contributions from different CH 4 emission sectors (Kirschke et al., 2013;Saunois et al., 2016Saunois et al., , 2020. Over 90 % of atmospheric CH 4 is lost via oxidation by OH in the troposphere, leading to a lifetime of 9.14 (±10 %) years against this sink (IPCC, 2013). Additional minor sinks include soil absorption, loss in the stratosphere, and reactions with chlorine radicals (IPCC, 2013). The contemporary growth of atmospheric CH 4 levels reflect an imbalance between its global sources and sinks.
China is one of the most significant methane producers, especially for anthropogenic sources such as coal mining (Saunois et al., 2016). Anthropogenic sources in China contributed about 13 % of the global anthropogenic CH 4 emissions in the 2000s (Kirschke et al., 2013). The recent bottomup emission inventory of Peng et al. (2016) found that the total Chinese CH 4 emissions increased from 24.4 Tg a −1 in 1980 to 45.0 Tg a −1 in 2010, with the largest source sector being rice cultivation in 1980 which was replaced by coal mining after 2005. However, large uncertainties exist in our understanding of the contemporary changes in CH 4 emissions over China (Saunois et al., 2020), e.g., whether the Chinese CH 4 emissions from coal mining have decreased due to the mitigation policy in recent years (Miller et al., 2019;Sheng et al., 2019). Atmospheric inversion analyses are typically applied at global scales due to very limited in situ CH 4 measurements over this region in the 2000s. The increases in spatiotemporal observations (from satellite or aircraft) and the development of atmospheric transport models would be helpful in constraining methane sources over China, but different datasets and methods could provide discrepant information (Thompson et al., 2015;Miller et al., 2019). China has pledged to peak the carbon dioxide emissions by 2030 and to reach carbon neutrality by 2060 for tackling climate change. As CH 4 is the second most important anthropogenic greenhouse gas, it also becomes crucial to quantify its emissions and concentration trends in China.
In this study, we aim to better understand the recent trends in CH 4 emissions and mixing ratios in China using the GEOS-Chem (Goddard Earth Observing System-Chemistry) chemical transport model driven by two commonly used global anthropogenic emission inventories: the Emission Database for Global Atmospheric Research (EDGAR, version 4.3.2) (Janssens-Maenhout et al., 2019) and the Community Emissions Data System (CEDS, version 2017-05-18) (Hoesly et al., 2018). We use an ensemble of surface, aircraft, and satellite observations to assess the CH 4 mixing ratios and trends from the surface to the troposphere and conduct a series of model simulations to examine their driving factors, as well as the influence of the interannual variability in global volume-weighted OH concentrations. An improved tagged CH 4 tracer simulation (with 100 region-and sector-specific tracers) is applied to identify and quantify the contributions to the spatial patterns of CH 4 mixing ratios and trends over China in the recent decade of 2007-2018.

Surface and aircraft measurements
We use the surface CH 4 mixing ratio measurements from the Global Monitoring Division (GMD) of the Earth System Research Laboratory (ESRL) at the National Oceanic and Atmospheric Administration (NOAA). The CH 4 mixing ratios are measured by gas chromatography with flame ionization detection (Dlugokencky, 2005). The measurement database (https://www.esrl.noaa.gov/gmd/dv/data/, last access: 3 March 2021) includes 95 sites globally providing monthly averages of mixing ratios (ppbv). The database has been widely used in assessing regional and global CH 4 mixing ratios and budgets Fraser et al., 2013;Cressot et al., 2014;Turner et al., 2016;Miller et al., 2019). Here we focus on four sites located in China, as summarized in Table 1 (Wofsy et al., 2011). ATom consists of four campaigns from July 2016 to May 2018 (Wofsy et al., 2018). Figure 1 shows the flight tracks from the two campaigns. Both HIPPO and ATom datasets provide the merged 10 s data products for all flights (Wofsy et al., 2017(Wofsy et al., , 2018, which cover the four seasons temporally and the regions over the Pacific Ocean and North America spatially. Both campaigns provide global-scale measurements of atmospheric composition in all seasons and conduct continuous profiling between ∼ 0.15 and 8.5 km altitude with many profiles extending to nearly 14 km. Here we sample the model results at the hourly resolution along flight tracks as shown in Fig. 1 and average them in 2 • latitude bins for the comparison.

GOSAT satellite observations
The TANSO-FTS ((Thermal And Near infrared Sensor for carbon Observation -Fourier Transform Spectrometer) instrument on board the Greenhouse Gases Observing Satellite (GOSAT) launched in early 2009 measures the backscattered solar radiation from a sun-synchronous orbit at around 13:00 LT (Butz et al., 2011;Kuze et al., 2016). The observations have a pixel resolution of around 10 km diameter and are separated by about 250 km along the observing track with a global coverage every 3 d . GOSAT retrieves column-averaged dry-air CO 2 and CH 4 mixing ratios from the shortwave infrared (SWIR) spectrum with near-unit sensitivity down to the surface (Butz et al., 2011). We use the University of Leicester version 7.2 GOSAT XCH 4 proxy retrieval over China from January 2010 to December 2017. The glint data over the oceans are not used in this study due to the sparse data coverage. The CH 4 product has been validated by Parker et al. (2015) against the Total Carbon Column Observing Network (TCCON) and MACC-II (Monitoring Atmospheric Composition and Climate) model XCH 4 data, and a precision of 0.7 % is suggested.
To compare with the GEOS-Chem model results as described below, the GOSAT CH 4 observations and satellite averaging kernels are averaged over the 2 • × 2.5 • or 4 • × 5 • model grid. We use the satellite observations which pass the criteria that the grid has more than 12 months of valid observations which have passed their quality control. The simulated vertical profiles (VMR mod ) are applied with the satellite averaging kernels (AKs) and a priori estimates (VMR apr ) using Eq. (1) following Parker et al. (2020).
where AK i is the retrieval averaging kernel and h i is the pressure weight for the vertical level i. This provides column mean CH 4 mixing ratios (XCH mod 4 ), with the vertical sensitivity of satellite retrievals accounted for.

The GEOS-Chem model description and simulation design
We use the GEOS-Chem global chemical transport model v11-02 release candidate (http://geos-chem.org, last access: 3 March 2021) driven by MERRA-2 meteorological fields from the NASA Global Modeling and Assimilation Office (GMAO). The MERRA-2 dataset has a native horizontal resolution of 0.5 • latitude × 0.625 • longitude and is degraded to 4 • × 5 • or 2 • × 2.5 • resolutions for input to GEOS-Chem. We use the CH 4 simulation that calculates the CH 4 sinks using prescribed global distributions of OH concentrations or loss frequencies. The model has been applied in a number of studies to understand the global and regional CH 4 emissions and mixing ratios (Wecht et al., 2014;Turner et al., 2015;Maasakkers et al., 2019Lu et al., 2021;Zhang et al., 2021). All the simulations are initiated in the year 1980, and we focus on the model results in the period of 2007-2018. We find that changes in the initial CH 4 conditions in January 1980 would not affect simulation results after January 2000, indicating that a spin-up time of over 20 years is sufficient for our analyses.
We use and compare two global anthropogenic CH 4 inventories: the Emissions Database for Global Atmospheric Research (EDGAR v4.3.2) covering 1970 and the Community Emissions Data System (CEDS, version 2017-05-18) (Hoesly et al., 2018(Hoesly et al., ) covering 1970(Hoesly et al., -2014. A detailed comparison of the two emission estimates will be presented in Sect. 3. The EDGAR CH 4 emissions do not account for seasonal variations. Here we have applied seasonal scalars to CH 4 emissions from manure management based on a temperature dependence described by Maasakkers et al. (2016) and to those from rice cultivation following Zhang et al. (2016) in the EDGAR inventory. The CEDS inventory as used in this study provides gridded emission estimates with monthly variations.  (Darmenov and Da Silva, 2013). Termite and seepage emissions are, respectively, from Fung et al. (1991) and Maasakkers et al. (2019).
The oxidation of CH 4 by tropospheric OH is calculated in the model using 3-D monthly averaged OH concentrations archived from a standard GEOS-Chem tropospheric chemistry simulation in Wecht et al. (2014). Global uniform scalers are then applied to account for the interannual variability in OH concentrations during 1980-2010 as simulated by the CESM model in Zhao et al. (2019). As shown in Fig. S1 in the Supplement, the resulting global volumeweighted mean OH increases by 0.20 % a −1 in 1980-2000 and 0.37 % a −1 in 2000-2010, finally reaching to 10.9 × 10 5 molec. cm −3 . Other minor sinks include tropospheric oxidation by chlorine atoms using monthly chlorine concentration fields of Sherwen et al. (2016), stratospheric loss computed with monthly loss frequencies of Murray et al. (2012), and soil uptake of Fung et al. (1991) with a temperaturedependent seasonality (Ridgwell et al., 1999).
We have conducted a series of model simulations over 1980-2018 as summarized in Table 1 to investigate the impacts of OH concentrations and model resolution. For all the datasets of emissions (using EDGAR and CEDS) and sinks as described above, the closest available year will be used for simulation years beyond their available time ranges as recent studies suggested weak trends in Chinese CH 4 emissions after 2010 Liu et al., 2021). Since CH 4 has a long lifetime of about 9 years, model results in the later years (e.g., after 2012 for EDGAR and after 2014 for CEDS) are strongly affected by the emissions in earlier years. Evaluations of these model results with the NOAA surface measurements at the four Chinese sites indicate that the simulation with CEDS and interannually varying OH at 2 • × 2.5 • resolution (GCC in Table 1) relatively better captures the measured mixing ratios and trends from 2007 onwards, as will be discussed in Sect. 3.2.
We further apply a tagged CH 4 tracer simulation to quantify the sources contributing to CH 4 mixing ratios and trends in China over 2007-2018. The tagged CH 4 tracer approach has been recently applied in GEOS-Chem to quantify source contributions in the US Midwest (Yu et al., 2021) and GFDL-AM4.1 with focuses on the global CH 4 budget (He et al., 2020). We implement 100 tracers that tag CH 4 emissions from different source types (agriculture, energy, industry, transportation, wastewater, residents, shipping, biomass burning, wetlands, seeps, and termites) and different regions (China, India, Europe, South America, North America, Africa, Oceania, etc.). The regions used for the tagged simulation are shown in Fig. 2, mainly based on Bey et al. (2001) with additional tagged regions for China and India in Asia. Global soil uptake is also tagged as a sink of CH 4 . We run the tagged CH 4 simulation using the model settings of GCC (i.e., CEDS and interannually varying OH) for the period of 1980-2018. The results allow us to quantify the detailed source contributions to CH 4 mixing ratios and trends over China.  Table S1 in the Supplement compare the anthropogenic emissions of EDGAR and CEDS, natural emissions, and sinks in our model simulations (GCE (GEOS-Chem with EDGAR emissions) and GCC (GEOS-Chem with CEDS emissions) as shown in Table 1) with the estimates in the literature summarized by Saunois et al. (2020). The emissions in the two decades of 2000-2009 and 2008-2017 from both bottom-up and top-down studies are reported in Saunois et al. (2016Saunois et al. ( , 2020 and are thus compared with corresponding estimates in this study. The anthropogenic emission source categories are different in the EDGAR and CEDS inventories, and we organize all sources into five main cate- gories (agriculture and waste, biomass burning, fossil fuels, wetlands, and other sources) following Saunois et al. (2020), as also summarized in Table S2 in the Supplement.

CH 4 emissions and sinks over the globe and China
As shown in Fig. 3, the global total (anthropogenic and natural) emissions over 2000-2009 are 520 Tg a −1 for GCE and 533 Tg a −1 for GCC. These total emissions are in the low end of the top-down estimates of 547 Tg a −1 with a range of 524-560 Tg a −1 and are smaller than the bottom-up estimates of 703 (566-842) Tg a −1 . The bottom-up estimates summarized by Saunois et al. (2020) included EDGAR and CEDS, and we can see that the differences from our emissions are largely driven by the underestimates of some natural emissions (e.g., geological, termite, and freshwater emissions), which are substantially reduced in the top-down estimates. In the 2008-2017 period, global total CH 4 emissions in GCE and GCC have increased to 556 Tg a −1 in GCE and to 574 Tg a −1 in GCC and are within the topdown emission range of 576 (550-594) Tg a −1 . The contributions of anthropogenic sources on total CH 4 emissions are about 63 % (2000-2009) and 65 % (2008-2017) in GCE and 65 % (2000GCE and 65 % ( -2009GCE and 65 % ( ) and 67 % (2008GCE and 65 % ( -2017 in GCC, which are slightly larger than 60 % and 62 % in the top-down estimates of Saunois et al. (2020). The global CH 4 chemical losses simulated in GCE and GCC are also consistent with the top-down estimates for both periods, while the sink of soil uptake might be underestimated in the model. Table 2 Saunois et al. (2020). The natural sources (e.g., wetlands, biomass burning) and the soil uptake in our study are relatively low compared with the estimates in Saunois et al. (2020). For the period of 2008-2017, the CH 4 emissions from fossil fuels increase to 22.8 Tg a −1 in GCE and 38.4 Tg a −1 in GCC, which are also at the averaged level and the high end of the bottomup estimate (16.6-39.6 Tg a −1 ) in Saunois et al. (2020).   Table S2 including emissions from agriculture and waste (AW), fossil fuels (FF), wetlands (WL), biomass burning (BB), others (OT), and sinks due to soil uptake (SU) and chemical loss (CL). The bar charts show bottom-up (dark-colored bars) and top-down (light-colored bars) estimates in previous studies as summarized by Saunois et al. (2020). The global CH 4 sources and sinks in the GCE (black circles) and GCC (black triangles) model simulations are also shown. Table S1 summarizes the values presented in the figure.  Table 2. the years before, which are largely driven by the emissions from the fossil fuels or energy sector. The largest differences between GCE and GCC, as also discussed in Fig. 4, come from the sectors of fossil fuels and agriculture. Agricultural sources in GCE account for 54.7 % of the total CH 4 emissions in 1980 and gradually decrease to 37.0 % in 2018, which mainly results from decreases in emissions from rice cultivation with some offset due to increases in the livestock emission. The contributions of agricultural sources in GCC are much smaller with values of 36.3 % in 1980 and 21.5 % in 2018. The energy or fossil fuels sector becomes the largest contributor of Chinese CH 4 emissions in recent years in GCC, accounting for 52.2 % of the total emissions in 2018, and largely drives the larger positive trend in GCC than GCE.
The comparisons above indicate large uncertainties in the Chinese CH 4 emission estimates, as to some extent covered by the EDGAR and CEDS anthropogenic emission inventories. The magnitude and temporal variations of methane budgets over the past decades are known to have large uncertainties (Kirschke et al., 2013;Turner et al., 2019;Saunois et al., 2020). Relative uncertainties are about 20 %-35 % for anthropogenic emissions such as fuel exploitation and agriculture and waste, about 50 % for biomass burning and wetlands, and up to 100 % or greater for other natural sources (Saunois et al., 2020). Uncertainties in the methane sinks are about 10 %-20 % by proxy methods such as using  Table S2. Annual mean emission totals and trends over 1980-2018 (with asterisks denoting the statistical significance of p value < 0.05) are shown inset. methyl chloroform and are 20 %-40 % by atmospheric chemistry models (Saunois et al., 2016). More detailed regional methane datasets can help improve assessing the global budget (Xu and Tian, 2012;Valentini et al., 2014;Saunois et al., 2016). We will further discuss the uncertainties in CH 4 emissions in the last section.

Observed and simulated methane mixing ratios and trends in China
Based on the emissions described above, we have conducted a series of model simulations as summarized in Table 1 and evaluated the model results with surface CH 4 measurements at the four Chinese sites. We find that when using the interannually fixed OH (global tropospheric volume-weighted mean of 10.6 × 10 5 molec. cm −3 as shown in Fig. S1), both model simulations with the EDGAR and CEDS emissions overestimate the observed CH 4 trends from 2007 onwards by 0.8-6.2 ppbv a −1 with EDGAR (Run1) and by 4.0-10.9 ppbv a −1 with CEDS (Run2). The model simulated CH 4 mixing ratios and trends over China are rather sensitive to the global OH levels. In the sensitivity simulations with global OH decreasing 10 % (Run5) or increasing 10 % (Run6) relative to the fixed levels (global mean of 10.9×10 5 molec. cm −3 ) over 2010-2018, CH 4 mixing ratios would, respectively, increase by 2.0 %-3.4 % or decrease by 1.9 %-3.2 % at the four Chinese sites (Fig. S2 in the Supplement). Increasing OH levels by 10 % would lead to negative trends in CH 4 mixing ratios at all four sites over 2010-2018 (Fig. S2). Such effects are also found in the simulation with global CH 4 emissions decreasing by 50 Tg a −1 over the same period (Run 7 in Table 1 and Fig. S2). The uses of interannually varying OH (Fig. S1) in model simulations (Run3 and Run4 in Table 1) overall correct the high biases in simulated CH 4 trends in simulations with fixed OH (Run1 and Run2) at the Chinese sites. We find that changing model horizontal resolution from 4 • × 5 • to 2 • × 2.5 • does not significantly affect the simulated surface CH 4 trends. Hereafter, we will focus our analyses on the model simulations at 2 • × 2.5 • resolution and with interannually varying OH (i.e., GCE and GCC in Table 1). Figure 6 shows the measured and simulated time series of monthly CH 4 mixing ratios at the four Chinese sites. Both GCE and GCC model results are shown, and distinct differences in CH 4 mixing ratios can be seen between the two simulations. Among the four Chinese sites, the largest CH 4 mixing ratio is observed at the SDZ site, a rural site near Beijing surrounded by high anthropogenic emissions, compared with the other three Chinese background sites (DSI, LLN, and WLG). GCC with high anthropogenic emission estimates simulate on average 1.0 %-4.7 % higher CH 4 mixing ratios than GCE results and are 0.3 %-6.5 % higher than measurements at the four Chinese sites. Measured CH 4 mixing ratios at the four sites have been increasing at the rates  of 7.0-7.9 ppbv a −1 since 2007. The GCC model results reproduce the trends in CH 4 mixing ratios at the DSI, LLN, and WLG sites, while they overestimate the 2009-2015 trend measured at SDZ by a factor of 2. The GCE model results in general underestimate the measured trends except for that at the SDZ site. These results can be explained by the higher CH 4 emission estimates and increases in CEDS than EDGAR since 2007 and may also reflect that the regional CH 4 emissions around SDZ (i.e., North China) are too high in CEDS. Further evaluations of the two model simulations with CH 4 column mixing ratio measurements (since 2011) at six TCCON sites in Asia  show similar results, with small biases of 0.2 %-1.0 % in CH 4 mixing ratios for GCC and negative biases of 2.6 %-3.7 % for GCE (Fig. S3 in the Supplement). This again reflects the higher Chinese CH 4 emission estimates in years around 2012 in CEDS than EDGAR, which then affect the model simulations afterwards by using their emissions of the latest available years.
Comparisons with satellite and aircraft observations further provide spatially and vertically resolved evaluations of the model simulations. Figures 7 and 8 show, respectively, the GOSAT-observed and model-simulated spatial distributions of seasonal mean CH 4 mixing ratios and trends over 2010-2017. The latitude-dependent biases between simulations and observations have been found to be noticeable at the 4 • ×5 • resolution but are significantly smaller at 2 • ×2.5 • (Stanevich et al., 2020). The GOSAT-observed CH 4 column mixing ratios over China peak in autumn (1825.6 ppbv on average) and reach a minimum in spring (1797.4 ppbv). There is a stronger seasonality in the CH 4 mixing ratio in the South China (1856.9 ppbv in autumn vs. 1826.8 ppbv in spring) likely attributed to the seasonal variation in agriculture emissions. The GOSAT-observed 2010-2017 trends show small spatial and seasonal variations over China with values of 7.67-8.43 ppbv a −1 . Both GCE and GCC modelsimulated CH 4 mixing ratios present similar spatial patterns with high correlation coefficients (r > 0.90), while GCEsimulated mixing ratios are on average biased low by 23.5-32.4 ppbv (∼ 1.6 %), and GCC results are overestimated by 25.6-36.8 ppbv (∼ 1.7 %). This discrepancy between the two simulations is mainly due to the CH 4 emissions from fossil fuels, which are 23.5 Tg a −1 for the GCE and 39.9 Tg a −1 for the GCC in China over 2010-2017. As for the CH 4 trends during 2010-2017 over China, both GCC and GCE show similar spatial patterns as those observed by GOSAT with moderate correlations of 0.2-0.5, while GCC model results have smaller biases of −1.7-0.4 ppbv a −1 , compared to GCE results that in general underestimate the trends by 2.6-4.7 ppbv a −1 . Figure 9 shows the latitudinal distribution of annual mean CH 4 mixing ratios as observed by HIPPO and ATom aircraft campaigns at three altitude layers (1-2, 4-5, and 7-8 km). Model results sampled along the flight tracks at their observing time are also shown. Both aircraft measurements Figure 6. Comparison of GCE-simulated (with EDGAR anthropogenic emissions and interannually varying OH; red lines) and GCCsimulated (with CEDS and interannually varying OH; blue lines) monthly mean CH 4 mixing ratios with NOAA in situ observations (black lines) in China. The observed mean mixing ratios (in unit of ppbv), trends (ppbv a −1 ), and corresponding model biases are shown inset. and model results are then averaged in 2 • latitude bins. As shown in Fig. 9, large latitudinal gradients in the tropospheric CH 4 mixing ratios between the northern and southern hemispheres, in particular in the lowest 2 km of the tropics, are observed by the aircraft measurements and are captured by the model results with the two emission inventories. Similar to the comparison with GOSAT observations, GCE modelsimulated CH 4 mixing ratios tend to be lower than those in GCC due to the lower estimate of global emissions in GCE (556 Tg a −1 ) than GCC (574 Tg a −1 ) since 2008 (Table S1). GCE model results underestimate the aircraft measurements with mean negative biases of 27.5-31.1 ppbv at the three altitude layers for HIPPO and even larger negative biases of 61.5-73.7 ppbv for ATom. By contrast, GCC model results are in general too high with biases of 18.4-22.8 ppbv for HIPPO and −1.7-9.4 ppbv for ATom. The biases in GCC are overall smaller than those in GCE.
The changes in the model bias for the comparisons with HIPPO and ATom measurements reflect their simulated trends in the CH 4 mixing ratios. Since both HIPPO (2009) and ATom (2016 provide measurements over the Pacific (black box in Fig. 1), we calculate the differences between HIPPO and ATom measurements as the observed CH 4 concentration trends over this region, and these trends also largely reflect the influences from upwind Asian CH 4 sources and levels. Figure 10 shows aircraft-observed and corresponding model-simulated trends separated for four seasons. The HIPPO (2009) and ATom (2016 CH 4 trends as estimated by the aircraft measurements range from 5.8 to 10.7 ppbv a −1 for the different seasons and altitudes, with typically higher increasing rates in boreal summer and autumn than those in boreal spring and winter. Both GCE and GCC model results tend to underestimate the trends, but the biases in GCC are much smaller than GCE. A distinct feature that can be seen from aircraft observations is the high CH 4 increasing rates over the tropics in boreal summer and autumn (reaching 15 ppbv a −1 ), while both model results do not capture it and show weak latitudinal gradients in the CH 4 trends. These tropical CH 4 increases are likely driven by the increasing tropical microbial emissions either from wetlands or livestock shown in some recent papers (Nisbet et al., 2016;Saunois et al., 2017;Worden et al., 2017;Maasakkers et al., 2019;Yin et al., 2021;Zhang et al., 2021), which have not been found in the model simulations.
Summarizing the comparisons of model results with all available measurements over China and the Pacific, we find that the surface, aircraft, and satellite CH 4 measurements have indicated rather consistent increase rates of CH 4 mixing ratios over China with values ranging from 7.0 to 8.4 ppbv a −1 in recent years. As CH 4 has a lifetime of about 9 years, such increases reflect changes in not only domestic emissions but also global emissions. The GCE and GCC model simulations with the interannually varying OH levels both capture the main features of the observed CH 4 mixing ratios and trends over China, and the GCC results show much smaller model biases than GCE. We will thus use the GCC model simulation to quantify the domestic and global sources contributing to the CH 4 mixing ratios and trends over China.

Source attribution of CH 4 mixing ratios and trends in 2007-2018
Here we apply the GCC model configuration (i.e., the CEDS inventory and interannually varying OH) in the tagged CH 4 simulation. The GCC model results can generally reproduce the spatial distribution of GOSAT-observed CH 4 levels and trends as shown in Fig. S4 in the Supplement, with mean biases of 27.4 ppbv (observed 1805 ppbv vs. simulated 1833 ppbv) in the global CH 4 mixing ratio and −0.8 ppbv a −1 (observed 7.08 ppbv a −1 vs. simulated 6.26 ppbv a −1 ) in the trend. As described in the Sect. 2.3, our tagged CH 4 simulation includes 100 region-and sectorspecific CH 4 tracers. The tagged CH 4 simulation is conducted over 1980-2018, and we analyze the results for 2007-2018. Figure 11 shows contributions of CH 4 emissions from different source regions and different sectors on the mean surface mixing ratios and trends in China during this time period, and the values are also summarized in Table 3 for mixing ratios and Table 4 for trends. As for mixing ratios, we find that the largest contributor of the Chinese CH 4 mixing ratio averaged over 2007-2018 is the wetland emission in South America, accounting for 10.5 % due to the large emission magnitude. Together with other sources, emissions in South America contribute 20.2 % of the surface CH 4 levels over China, followed by the sources from Africa (17.0 %) and Europe (15.0 %). The Chinese domestic emissions account for 14.0 % of the CH 4 mixing ratio. The emission contributions to the mixing ratio are generally proportional to their emission magnitudes because of the CH 4 lifetime of about 9 years, and seasonal variations in the percentage contributions are small, as can be seen in Fig. 11 (the top left panel). Figure 11 and Table 4 also show the source contributions to the 2007-2018 trends in the surface CH 4 mixing ratio over China. Based on the emission inventory in GCC, the simulated mean trend in the surface CH 4 mixing ratio is 9.75 ppbv a −1 over the land of China. The domestic energy sector is identified as the largest driver of the trend in China contributing an increase rate of 5.54 ppbv a −1 . Accounting for the trends driven by emissions from the agricultural and wastewater sectors, domestic contributions can reach 6.50 ppbv a −1 (67 % of 9.75 ppbv a −1 ). The remain-    (Table S2) including agriculture (AGR), energy (ENE), wastewater (WST), residents (RCO), biomass burning (BBN), wetlands (WTL), seeps (SEE), termites (TER), and others (OTH) including industry (IND), transportation (TRA), and shipping (SHP), as well as from different regions (Fig. 2). Values are estimated using the tagged CH 4 tracer simulation. ing trends of 3.25 ppbv a −1 are then contributed by emission changes outside China. We find that the anthropogenic sources (mainly from energy, agricultural, and wastewater sectors) in Africa and other Asian regions (India and Rest of Asia) contribute, respectively, trends of 3.25 and 2.45 ppbv a −1 over China, highlighting the strong CH 4 emission increases in these regions such as large emission increases from livestock sources over South Asia and tropical Africa in 2010-2018 . On the contrary, Europe is the only region where CH 4 emissions from nearly all sectors have been decreasing (Jackson et al., 2020), which leads to a negative trend of −1.81 ppbv a −1 over China. We find strong spatial variation in the contribution values over different regions of China with standard deviations up to 11 % for the contributions to CH 4 mixing ratios and up to 0.4 ppbv a −1 to the trends (Fig. 11). Not only near the surface, we find similar results for the CH 4 mixing ratios throughout the troposphere over China with slightly smaller growth rates in the upper troposphere (Fig. S5 in the Supplement).
Our results indicate that trends in China are dominated by energy emissions from coal, oil, and gas, with significant contributions from the wastewater and agricultural sectors. This is consistent with the top-down emission inversion results by Miller et al., (2019)  The analyses above demonstrate strong foreign source contributions to the CH 4 mixing ratios, as well as CH 4 trends over China. We further find large spatial heterogeneity in the domestic vs. foreign contributions. Figure 12 shows the spatial distributions of domestic emission contributions to Chinese CH 4 surface mixing ratios and trends over 2007-2018 calculated as the percentages of sums of all Chinese-tagged tracers to the total levels. We can see that the domestic contribution to the CH 4 surface mixing ratio ranges from 12.4 % in the western China to 15.1 % in central China, and to the trends ranges from 62.6 % over the Tibet Plateau to 70.1 % in the central China. The largest domestic contributions for both surface mixing ratios and trends are found in the central eastern China, so that measurements over this region would most reflect the CH 4 emission changes in China.

Conclusions and discussion
In summary, we have investigated the sources contributing to the CH 4 mixing ratios and trends over China in the recent decade (2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) using the GEOS-Chem global model. The CH 4 model simulations are conducted considering two different commonly used anthropogenic emission inventories (EDGAR v4.3.2 and CEDS) and are evaluated with available    surface, aircraft, and satellite measurements of CH 4 mixing ratios over China and the Pacific region. The surface, aircraft, and satellite measurements have shown CH 4 concentration increase rates of 7.0-8.4 ppbv a −1 over China in recent years. We find that model results are sensitive to the selection of anthropogenic emission inventories and OH levels. By using the CEDS anthropogenic emission inventory and interannually varying OH levels (Fig. S2), the model can generally reproduce the measured CH 4 mixing ratios and trends over China. This corresponds to mean Chinese anthropogenic CH 4 emissions of 69.4 Tg a −1 (with an increase rate of 1.2 Tg a −2 ) and global tropospheric volume-weighted mean OH concentrations of 10.8 × 10 5 molec. cm −3 (with an increase rate of 0.25 % a −1 ) over 2007-2018.
We apply a tagged CH 4 tracer simulation that implements region-and sector-specific tracers to diagnose and to understand their emission contributions. Using the model simulation with CEDS and interannually varying OH, we find strong influences from foreign sources on both CH 4 mixing ratios and recent increases over China due to the long lifetime of CH 4 . For the mean surface CH 4 mixing ratio over China (1873China ( .0 ppbv over 2007China ( -2018, domestic CH 4 emissions account for 14.0 %, and contributions from the sources outside China reach 86.0 %, including 20.2 % from South America, 17.0 % from Africa, 15.0 % from Europe, 13.0 % from North America, and 12.8 % from the Rest of Asia group. For the mean CH 4 concentration trend over China (9.75 ppbv a −1 over 2007-2018), the largest driver is estimated to be the domestic energy source contributing 5.54 ppbv a −1 , and other important domestic source contributions include emissions from wastewater (0.68 ppbv a −1 ) and agriculture (0.30 ppbv a −1 ); natural sources such as wetland emissions have insignificant trend contributions. Emission changes in foreign sources are also significant. The increase rate of 3.14 ppbv a −1 in the Chinese surface CH 4 mixing ratio can be attributed to sources in other Asian countries (India and Rest of Asia), 1.64 ppbv a −1 to Africa, and 0.95 ppbv a −1 to South America (Table 4).
It should be noted that our source attribution results can be biased by the use of CEDS and the uncertainty in the interannual variations in OH levels. The Chinese anthropogenic CH 4 emissions in the CEDS inventory are higher and have increased more rapidly than EDGAR v4.3.2 in the past decade. The two emission inventories significantly differ in the sectors of fossil fuels and agriculture. CEDS estimates higher CH 4 emissions from fossil fuels while lower emissions from agriculture compared with EDGAR v4.3.2. A number of top-down emission inversion studies using surface and satellite observations have found that the EDGAR v4.3.2 (Maasakkers et al., 2019;Miller et al., 2019) and previous EDGAR versions (Alexe et al., 2015;Thompson et al., 2015;Turner et al., 2015;Pandey et al., 2016) overestimated the CH 4 emissions from coal production in China likely due to the CH 4 emission factors for coal mining being too high in the region (Peng et al., 2016). A recent bottom-up es-timate suggested that Chinese coal mining CH 4 emissions have been decreasing since 2012 driven by China's coal mine regulation , but the interannual trend in Chinese coal emissions still has large uncertainties among studies (Miller et al., 2019;Sheng et al., 2019;Lu et al., 2021).
We also find that the interannual variability in OH concentrations can strongly affect the simulated CH 4 concentration trends. Using interannually fixed OH concentrations, the model would overestimate the observed CH 4 growth from 2007 onwards in China with both the EDGAR and CEDS anthropogenic emissions. The influence of a 10 % increase in the global volume-weighted mean OH concentration (from 10.9 × 10 5 molec. cm −3 to 12.0 × 10 5 molec. cm −3 ) on the simulated Chinese CH 4 mixing ratios is equivalent to that of a 47 Tg a −1 decrease in global CH 4 emissions. The use of interannual variability in OH provided by Zhao et al. (2019) improves the model-simulated Chinese CH 4 mixing ratios and trends. However, large discrepancies exist in the different model OH simulations that would lead to a wide range (> ±30 ppbv) of simulated CH 4 mixing ratios (Zhao et al., 2019). Despite these uncertainties, our study emphasizes the importance of emission changes in both domestic and foreign and anthropogenic and natural sources on the Chinese CH 4 concentration trends. Future work with more intensive CH 4 measurements covering eastern China will help us better assess the driving factors of Chinese CH 4 mixing ratios and recent growth. Data availability. NOAA surface observations are available at https://doi.org/10.15138/VNCZ-M766 . The GOSAT Proxy XCH 4 data can be accessed through the Copernicus C3S Climate Data Store (https://cds.climate.copernicus.eu, Parker et al., 2015). TC-CON data were obtained from the TCCON Data Archive hosted by CaltechDATA through https://data.caltech.edu/records/266 (Goo et al., 2017), https://data.caltech.edu/records/1092 , https://data.caltech.edu/records/957, https://data.caltech. edu/records/1090, https://data.caltech.edu/records/958 (Morino et al., 2017a, b, c), and https://data.caltech.edu/records/288 (Shiomi et al., 2017). The HIPPO data used in this study can be obtained at https://doi.org/10.3334/CDIAC/hippo_010 (Wofsy et al., 2017). The ATom data are available at https://doi.org/10.3334/ORNLDAAC/1581 (Wofsy et al., 2018). Modeling dataset can be accessed by contacting the corresponding author.
Author contributions. HT and LZ designed the study. HT conducted the modeling and data analyses with contributions from LZ, XL, YZ, and BY. RJP and HB provided the GOSAT CH 4 data and contributed to the interpretation and discussion of their use in the study. HT and LZ wrote the paper with input from all authors.
Competing interests. The contact author has declared that neither they nor their co-authors have any competing interests.
Disclaimer. Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.