Interannual variability on methane emissions in monsoon Asia derived from GOSAT and surface observations

In Asia, much effort is put into reducing methane (CH4) emissions due to the region’s contribution to the recent rapid global atmospheric CH4 concentration growth. Accurate quantification of Asia’s CH4 budgets is critical for conducting global stocktake and achieving the long-term temperature goal of the Paris Agreement. In this study, we present top-down estimates of CH4 emissions from 2009 to 2018 deduced from atmospheric observations from surface network and GOSAT satellite with the high-resolution global inverse model NIES-TM-FLEXPART-VAR. The optimized average CH4 budgets are 63.40 ± 10.52 Tg y−1 from East Asia (EA), 45.20 ± 6.22 Tg y−1 from Southeast Asia (SEA), and 64.35 ± 9.28 Tg y−1 from South Asia (SA) within the 10 years. We analyzed two 5 years CH4 emission budgets for three subregions and 13 top-emitting countries with an emission budget larger than 1 Tg y−1, and interannual variabilities for these subregions. Statistically significant increasing trends in emissions are found in EA with a lower emission growth rate during 2014–2018 compared to that during 2009–2013, while trends in SEA are not significant. In contrast to the prior emission, the posterior emission shows a significant decreasing trend in SA. The flux decrease is associated with the transition from strong La Ninña (2010–2011) to strong El Ninño (2015–2016) events, which modulate the surface air temperature and rainfall patterns. The interannual variability in CH4 flux anomalies was larger in SA compared to EA and SEA. The Southern Oscillation Index correlates strongly with interannual CH4 flux anomalies for SA. Our findings suggest that the interannual variability in the total CH4 flux is dominated by climate variability in SA. The contribution of climate variability driving interannual variability in natural and anthropogenic CH4 emissions should be further quantified, especially for tropical countries. Accounting for climate variability may be necessary to improve anthropogenic emission inventories.


Introduction
In order to achieve the long-term temperature goal 'well below 2 • C above pre-industrial levels and pursuing efforts to limit the temperature increase to 1.5 • C above pre-industrial levels' (Schleussner et al 2016), a 5 year global stocktake, to summarize reported national emission inventories and compare them to the observed greenhouse gas (GHG) atmospheric trends, will take place starting 2023. The global stocktake is an essential part of the Paris Agreement for countries to take a look at how their efforts are stacking up against the Paris Agreement temperature goal and provide inputs to parties to submit more ambitious climate pledges (Work Resources Institute 2020).
Compared to carbon dioxide (CO 2 ), atmospheric methane (CH 4 ) has a relatively shorter lifetime (∼10 years) and a higher global warming potential over 100 years, which makes it a suitable target for implementing rapid and achievable mitigation strategies (Dlugokencky et al 2011, Nisbet et al 2020. The concentration level of CH 4 in the atmosphere is over 150% higher than pre-industrial times (i.e. year 1750), and the current sharp rise of CH 4 concentration makes it as the key GHG threatening achievement of the Paris Agreement mitigation targets (Bousquet et al 2006, Nisbet et al 2019. The exact reasons for the CH 4 concentration accelerated are so far not well understood, in relevance to anthropogenic and natural emission activities, climate change, and tropospheric oxidant changes (Ghosh et al 2015, Saunois et al 2016b, Nisbet et al 2019, Turner et al 2019. Greenhouse gas Observing SATellite (GOSAT) observations detected significant enhancements of column-averaged dry-air mole fraction of methane (XCH 4 ) in Asia in the course of its operation since 2009 (Kivimaki et al 2019). The CH 4 concentration showed significantly increasing trends across China, which would bring a potential rise in surface temperature response over China (Wu et al 2019). Asia, with 60% of the population in the world, shares around 30% of global total CH 4 emissions, and around 40% of global anthropogenic emissions (Saunois et al 2016a). Previous studies reported increasing CH 4 emission in Asia during 2000-2012 due to increases of population, coal and natural gas consumption, paddy fields, and livestock (Patra et al 2016, Ito et al 2019, Saunois et al 2020. The observed CH 4 increase in Asia is attributed to emissions from big emitting countries with rapid economic growth (Bergamaschi et al 2013, Ito et al 2019. Asia, with 87% of the global rice paddy area and 90% of rice production over the world, is globally the main contributor to CH 4 emissions from agriculture (FAOSTAT 2017). As recent studies suggest that the paddy area was decreasing in Asia since 2007 (Zhang et al 2020a), other possible drivers should be addressed to explain the recent global CH 4 increase. CH 4 mitigation in the frame of the Paris Agreement, also requires a sound analysis of past emissions since the determination of national emission reduction targets and policies is directly affected by CH 4 emission budgets produced by individual countries (Feng et al 2017). Top-down methods are efficient tools for evaluation of emissions, as in the bottom-up inventories remain large uncertainties (Kirschke et al 2013, Lyon et al 2015, Peischl et al 2016, Peng et al 2016, Saunois et al 2016a, especially in non-CO 2 emission inventories the uncertainty around 50% at the higher end (Kirschke et al 2013, Wang et al 2018, Saunois et al 2020. Due to the improvement of inverse modeling and various observations available (Jacob et al 2016, Houweling et al 2017, especially the GOSAT's 10 year record of CH 4 retrievals (Parker et al 2020), there is an increasing number of top-down GHG budget estimations by inverse models (e.g. Brown 1993, Bergamaschi et al 2010, 2018, Fraser et al 2013, Maksyutov et al 2013, Miller et al 2013, Alexe et al 2015, Turner et al 2015, Tan et al 2016, Sheng et al 2018a, 2018b. In this paper, we estimated top-down CH 4 emissions for monsoon Asia at the country scale for 2009-2018 by a global high-resolution (0.1 • × 0.1 • ) inverse model using surface observations and GOSAT retrievals. We analyzed the spatiotemporal variations of anthropogenic and natural CH 4 fluxes in East, Southeast and South Asia during 2009-2018, took stock of two 5 year (2009-2013 and 2014-2018) emission budgets of top-emitting countries in Asia, and explored aspects that should be taken into account for coming global stocktake and further guidelines for the nationally determined contributions to emission reduction in Asia.

Inverse model
To estimate the top-down CH 4 emissions, we used the joint Eulerian three-dimensional transport model National Institute for Environmental Studies Transport Model (NIES-TM) coupled with a Lagrangian particle dispersion model named FLEXible PARTicle dispersion model (FLEXPART) (Ganshin et al 2012, Belikov et al 2016. The coupled model NTFVAR combines NIES-TM v08.1i with a horizontal resolution of 2.5 • and 32 hybrid-isentropic vertical levels (Belikov et al 2013) and FLEXPART model v.8.0 (Stohl et al 2005). The design of the inverse modeling system NTFVAR and the inverse method were described in details by Maksyutov et al (2020). The model adjoint, performance test and validation were described by Belikov et al (2016). The inverse modeling problem is formulated and solved to find the optimal value of x-vectors of corrections to prior fluxes at the minimum of the cost function J (x) in equation (1): where r is the residual (the difference between the observed concentration and the forward simulation made with prior fluxes without correction), R is the covariance matrix of observations, and B is the covariance matrix of fluxes. H is a matrix representing transport model. Flux corrections were estimated independently for two categories of emissions (i.e. anthropogenic and natural). Variational optimization was applied to obtain flux corrections as two sets of scaling factors to prior uncertainty fields on a monthly basis at a 0.1 • × 0.1 • resolution separately for anthropogenic and natural wetland emissions with bi-weekly time steps. The inverse model operates at the resolution of coupled transport model of 0.1 • × 0.1 • and applies spatial flux covariance length of 500 km. Sensitivity tests for the inversion setup have been performed using randomly perturbed observations and perturbed fluxes for different regions , and inversion results have been validated against aircraft measurement data, which were not used in the inverse simulations as described in Janardanan et al (2020).
More details about the model setup and initial condition test are described in supplementary material (available online at stacks.iop.org/ERL/16/024040/ mmedia).

Data sources
Prior CH 4 fluxes used in the model include anthropogenic emissions, natural emissions from wetlands, soil sink, emissions from biomass burning and other natural sources from the ocean, geological reservoirs and termites. Annual anthropogenic emissions were taken from the latest Emissions Database for Global Atmospheric Research (EDGAR v5.0). Monthly variation of anthropogenic emissions was incorporated based on the monthly emissions by EDGAR for the year 2015. Beyond 2015, we used the report from PBL Netherlands Environmental Assessment Agency (Olivier and Peters 2018) to extend EDGAR values of 2015 using equation (2), where t is year from 2016 to 2018: .
(2) Wetlands emissions (Cao mechanism (Cao et al 1996)) and soil sink (Curry mechanism (Curry 2007)) were from improved VISIT (Vegetation Integrative SImulator for Trace gases) model simulations (Ito and Inatomi 2012). Biomass burning emissions were from the daily GFASv1.2 (Global Fire Assimilation System) (Kaiser et al 2012). The oceanic, geological, and termite emissions were also included (Fung et al 1991, Lambert andSchmidt 1993). The uncertainties of the prior CH 4 fluxes were set for 30% of EDGAR and 50% of VISIT in the inversion.
From GOSAT, we used in our inversion the XCH 4 data from the thermal and near-infrared sensor for carbon observation-Fourier transform spectrometer. The GOSAT data were corrected as described in the supplementary material. In addition to the GOSAT retrievals (NIES, Level 2 retrievals, v.02.81) (Yoshida et  The meteorological data such as three-dimensional wind fields, temperature, and humidity at a temporal resolution of 6 h used for the transport model were obtained from the Japanese Meteorological Agency Climate Data Assimilation System and the Japanese 55 year Reanalysis (JRA-55) (Onogi et al 2007, Kobayashi et al 2015.

Results and discussion
3.1. Regional CH 4 fluxes and spatial distribution There are 27 countries included in the study area, which we divided into three subregions: East Asia (EA), South Asia (SA), and Southeast Asia (SEA). Figure 1 shows the spatial distribution of the aver-  prior estimates and the posterior estimates are within the model uncertainty range. The estimate of CH 4 emissions from SEA in this study is consistent with the estimate (43.5 ± 3.2) of Fraser et al (2013) and the estimate (41.5 ± 6.8) of Tan et al (2016). Note the study domain does not totally match that of this study and the periods of the estimations are different, which would cause the differences. For evaluation of the results from our inversion, we conducted a comparison of the modeled CH 4 concentrations using prior and posterior emissions and the measured concentrations. The estimation was evaluated by comparing the simulated and observed measurements at 13 surface sites in the study area (table S1 in supplementary material). The estimates with posterior emissions show decreased bias, root mean square error (RMSE) and improved correlations between observed and simulated concentrations, which indicates that our inverse model obtains optimized emissions that fit to the observations better. The improvement of our inversion optimizations is comparable to other studies or even better with lower RMSE (e.g. Alexe et al 2015, Zhang et al 2020b.

CH 4 budgets and trends in 2009-2013 and 2014-2018
According to the Paris Agreement, the global stocktake will be conducted in 2023, and a 5 year review will be taken regularly (Work Resources Institute 2020). National reports to the UNFCCC are currently rare for most Asian countries. Only Japan and Korea submitted national inventories yearly up to 2018; China reported values of 2010, 2012, and 2014; India, Pakistan, Vietnam, Thailand, and Malaysia reported only once during the study period 2009-2018(UNFCCC 2020. Here, we evaluate the last two 5 year CH 4 emission budgets in monsoon Asia for countries that emit more than 1 Tg y −1 using topdown inverse estimates (table 1). These 13 countries emit more than 97% of the total CH 4 emissions in the study area, and anthropogenic emissions are dominant in most countries except Bangladesh and Cambodia. The posterior total emissions stay stable from the first 5 year period to the second, while the prior total emissions show a 4% increase in the whole study area. The optimized total emissions for China, Japan, and South Korea show larger increases compared to    the prior estimates since the posterior emissions during the first 5 year period are lower than the prior emissions. The national report to UNFCCC of China is around 55 Tg y −1 during 2010-2014, closer to the posterior estimates (53.72 Tg y −1 , 62.16 Tg y −1 ) than to the prior estimates (64.00 Tg y −1 , 66.16 Tg y −1 ). The Japan national report to the UNFCCC shows values between 1.2 and 1.4 Tg y −1 , lower than both prior (2.13 Tg y −1 , 2.10 Tg y −1 ) and posterior (1.84 Tg y −1 , 2.15 Tg y −1 ). The optimized total emissions in other countries show less increase compared to the prior estimates, in which India, Pakistan, Thailand, Bangladesh, and Myanmar obtain higher optimized emissions during the first 5 year period compared to the prior emissions. The model estimates for India aresimilar to the studies by Saunois et al (2016a), Miller et al (2019), but higher than the study of Ganesan et al (2017) and the Indian national report to the UNFCCC (of 19.78 Tg y −1 in 2010). The national report of Pakistan with 5.55 Tg y −1 in 2015 is also lower than model estimates (table 1). The trends reveal geographical features; thus, we further explored the difference between the aggregated emissions for two 5 year periods for three subregions. Figure 3 shows prior and posterior emissions during two 5 year periods from subregions. In the first and second 5 year periods, the mean posterior emissions in EA are 58.64 Tg y −1 and 68.16 Tg y −1 , both lower by 16% and 6% than the mean prior emissions (69.98 Tg y −1 , 72.13 Tg y −1 ), respectively. The mean posterior emissions in SEA and SA are higher than the mean prior emissions by 5% and 13% during 2009-2013, but lower by 2% and 4% during 2014-2018. The mean posterior emission in SA decreases from 69 Tg y −1 to 59.7 Tg y −1 by 16% from the first 5 year period to the second 5 year period. The mean posterior estimates in SEA are similar around 45 Tg y −1 in the two 5 year periods.

Interannual flux variability
Ten years of monthly fluxes of prior and posterior estimates were aggregated for three subregions to investigate the interannual flux variabilities. Figure 4 shows the anomalies for prior and posterior total emissions in subregions EA, SA and SEA. The anomalies are constructed by subtracting the long-term monthly mean from the raw time series and constructing the 11 month running mean on the resultant time series. The long-term gradual trend was excluded by using anomalies.  anomaly of SEA shows relatively smaller variability except for an anomalous positive peak in 2015 coherence with prior anomaly due to the intensive fires in Indonesia that year, which emitted globally significant quantities of GHGs to the atmosphere, and significant enhancement of CH 4 total column values were observed by GOSAT (Parker et al 2016).
We investigated the trends in flux anomalies using the non-parametric Mann-Kendall approach (Hirsch et al 1982). The null hypothesis of trend test is no significant trend existing in a flux series, and a trend in flux series is considered 'significant' if the test probability P ⩽ 0.05. The results show statistically significant increasing trends in prior anomalies of EA, SEA, SA. The posterior anomalies show a statistically significant increasing trend in EA and a decreasing trend in SA. There is no statistically significant trend detected in the posterior anomaly of SEA. The anticorrelation between increasing prior flux and decreasing posterior flux from SA is not noticed before and causes interest to probe.

Connection between CH 4 flux variability and SOI
According to the EDGAR v5.0 estimate, the anthropogenic CH 4 emission sources in SA include mainly livestock (47%), rice (22%), waste (16%), and energy (15%), and the CH 4 emissions from livestock, rice and waste sectors keep rising in the study period due to increasing production by ruminants, paddy field harvest areas and waste, etc. We found in this study that the anomaly of anthropogenic emission in SA has a similar trend and variation with the anomaly of total emission in SA (r = 0.93 using Pearson's correlation test) (figure 5). We probed into the possible climate factors that might drive the flux variabilities, which are not calculated in the activitybased bottom-up estimates. The dominant interannual variability of the global climate system is ENSO (El Ninño-Southern Oscillation) (Trenberth et al 1998, Messie andChavez 2011), which plays a central role in seasonal to decadal global climate and impacts temperature and precipitation in its whole lifecycle. In Southern Asia, surface air temperature and precipitation can be significantly modulated by ENSO Qian 2019, Gehlot et al 2020). We investigated the connection between the smoothed posterior anomaly of SA emissions and the smoothed Southern Oscillation Index (SOI). The SOI is a measure of the intensity or strength of the Walker Circulation and the SOI is defined as the difference in surface air pressure between Tahiti and Darwin (Troup 1965). The trends of SOI (5 month forwards shift) and the flux anomalies in SA are in good agreement (figure 5). We used the Pearson test to examine the correlation and good correlations were found between SOI and the total flux and the anthropogenic flux anomalies with r = 0.6 (p-value = 2.1 × 10 −9 ) and r = 0.5 (p-value = 7.2 × 10 −10 ), respectively. The decrease of CH 4 fluxes from 2010 to 2016 corresponds to a cold-to-warm transition from a strong La Ninña in 2010 to a strong El Ninño in 2016 in SA. Recently, Sun et al (2020) conducted long-term field CH 4 flux measurement in the Yangtze River delta, China, and found that higher temperature and less rainfall in 2013 induced CH 4 emissions increase from rice paddy by 58%-294%, while rice grain yield did not increase and straw biomass increased by 15%-40% compared to other years. A previous study also found that Indian rice production dropped by 7% during El Ninño events (Selvaraju 2003). Another study found that rising temperature results in an increase in CH 4 production per unit of livestock products (Broucek 2015). The source for CH 4 production by ruminant enteric fermentation is a dietary carbohydrate, which influences the rate of fermentation, rate of rumen passage, and animal intake, contributing around 90% CH 4 emissions of livestock. In a high-temperature environment, the contents of the cell wall, acid detergent fiber, and lignin tend to increase, causing lower digestibility of feed and higher energy loss, which results in an increase in CH 4 production per unit of product through the decrease in the efficiency of animal production in tropics. Forage quality increase with decreasing temperatures induces a decrease of enteric CH 4 production (Lee et al 2017). With a 70% of CH 4 emissions from rice paddy and livestock, the SA flux variability is closely associated with climate variability, which might be not accounted for in the prior estimates.

Conclusions
This study presents the first detailed CH 4 budgets in East, Southeast, and South Asia inferred from a decade of satellite and ground-based atmospheric observations from 2009 to 2018 on a country scale. We used the global high-resolution inverse model NIES-TM-FLEXPART-VAR to optimize prior anthropogenic emissions from EDGAR v5.0, wetland emissions from VISIT model, and biomass emissions from GFASv1.2. The inverse corrections vary geographically, we found evident downwards anthropogenic emission corrections in eastern China, Korea and Japan, and upwards natural emission corrections in Tibet of China and northeast of India. The optimized average CH 4 budgets are 63.40 Tg y −1 from EA, 45.20 Tg y −1 from SEA, and 64.35 Tg y −1 from SA in a ten-year period. Emission budgets are estimated in two 5 year periods 2009-2013, 2014-2018 for countries with an emission budget of more than 1 Tg yr −1 (13 countries).
We compared the optimized CH 4 flux variations with the prior inventories. The prior estimates show significant increasing trends in EA, SEA, and SA, while posterior estimates release trends with geographical features. In EA the increasing trend is statistically significant, and in SEA no significant trend is singled out. The posterior emission in SA releases a significant decreasing trend on the contrary to the prior emission, and the flux is more variable compared to that in EA and SEA. We further probed the opposite trends found in prior and posterior emissions in SA and found that the SOI correlates well with the interannual anomalies of CH 4 flux. The flux decrease is associated with the transition of strong La Ninña ) to strong El Ninño (2015-2016 events and its accompanying effect of the Asian Monsoon system, which drives the patterns of temperature and rainfall. Recent field studies as discussed in section 3.4 found that decreasing temperature induce decreasing CH 4 fluxes from the paddy field, livestock, and waste, and these climate-CH 4 feedbacks explaining the decreasing trend in the posterior flux anomalies in SA may not be calculated in the prior flux estimates. The abnormally low anthropogenic emissions in SA and high fire emissions in SEA were found related to the strong El Ninño event in 2015. Our results also demonstrate the importance of the information about interannual variations in CH 4 fluxes in Asia is strongly influenced by ENSO climate variations. Strong negative anomalies in CH 4 fluxes occur during El Niño events in SA and notable positive anomalies in CH 4 fluxes are detected during La Ninña events in SA and SEA. Climate variability drives agriculture and wastes emission changes, a recent decrease of CH 4 emission in SA more likely due to climate variability (e.g. ENSO) rather than production yield decrease. Climate variability affects CH 4 emissions from paddy fields and livestock should be considered in the coming global stocktake under the Paris Agreement, in order to increase the accuracy, better explain changes in CH 4 budgets between 5 year stocktake periods. Accordingly, more studies are required to address climate-anthropogenic CH 4 feedbacks and to quantify the contribution of climate variability to interannual variabilities in CH 4 fluxes.

Data availability statement
The data that support the findings of this study are available upon reasonable request from the authors.