Temporal changes in the emissions of CH 4 and CO from China estimated from CH 4 / CO 2 and CO / CO 2 correlations observed at Hateruma Island

In situ observation of the atmospheric CO 2, CH4, and CO mixing ratios at Hateruma Island (HAT, 24.05 ◦ N, 123.80 E) often show synoptic-scale variations with correlative elevations during winter, associated with air transport from the East Asian countries. We examine winter (November–March) trends in 1CH4 / 1CO2, 1CO /1CO2, and1CO /1CH4 observed at Hateruma over the period 1999 to 2010. To investigate the relationship between the East Asian emissions and the short-term variations in the atmospheric mixing ratios, we use the FLEXPART Lagrangian particle dispersion model (LPDM). The observed ratios 1CH4 / 1CO2 and1CO /1CO2 both show an overall gradual decrease over the study period due to a recent rapid increase in fossil fuel consumption in China. We note, however, that the decreasing rates of 1CH4 / 1CO2 and1CO /1CO2 show gradual decrease and increase, respectively, during the entire observation periods used in this study. The 1CO /1CH4 slope, on the other hand, shows an increasing trend during 1999–2004 but a decrease during 2005– 2010. Calculation of the concentration footprint for the atmospheric observation at HAT by using the FLEXPART LPDM indicates that most of the short-term variations are caused by emission variations from northern and eastern China. Combined with a set of reported emission maps, we have estimated the temporal changes in the annual CH 4 and CO emissions from China under the assumption that the estimate of the fossil-fuel-derived CO2 emissions based on the energy statistics are accurate. The estimated annual CH 4 emissions, corresponding to nonseasonal sources or anthropogenic sources without rice fields, show a nearly constant value of 39± 7 TgCH4 yr−1 during 1998–2002, and then gradually increase to 46± 8 TgCH4 yr−1 in 2009/2010. The estimated annual CO emissions increase from 134 ± 32 TgCO yr−1 in 1998/1999 to 182± 42 TgCO yr−1 in 2004/2005, level off after 2005, and then slightly decrease to less than 160 TgCO yr −1 in 2008–2010.


Introduction
Methane (CH 4 ) is the second-most-important anthropogenic greenhouse gas in the atmosphere after carbon dioxide (CO 2 ).Analyses of ancient air occluded in polar ice core samples have suggested that atmospheric CH 4 has more than doubled since the preindustrial period (e.g., Etheridge et al., 1998;Nakazawa et al., 1993).This long-term increase can be attributed to increases in anthropogenic sources (e.g., fossil fuels, domestic ruminants, and rice fields).Recent systematic measurements of atmospheric CH 4 have shown a slowing down in the rate of increase after the 1980s with a significant interannual variability (Steele et al., 1992;Dlugokencky et al., 1998), leveling off in the early 2000s (Dlugokencky et al., 2003), and an abrupt renewed increase after 2007 (Rigby et al., 2008;Dlugokencky et al., 2009;Terao et al., 2011).These temporal changes in atmospheric CH 4 are a result of an overall balance between the emission and loss processes that include soil absorption and chemical reactions with atmospheric hydroxyl radical (OH) and stratospheric O( 1 D) and Cl atoms.

Published by Copernicus
Carbon monoxide (CO), produced during incomplete combustion of fossil fuels and biomass and the oxidation of atmospheric CH 4 and nonmethane volatile organic compounds (NMVOC), is an atmospheric pollutant.CO also plays an important role in atmospheric chemistry.The CO reaction with OH, a dominant sink for atmospheric CO, removes about 75 % of OH in the atmosphere (Thompson, 1992), influencing the atmospheric oxidation capacity.Therefore, changes in the CO emission could indirectly affect the atmospheric levels of CH 4 and other species that are removed by OH (e.g., Logan et al., 1981).Depending on the levels of the mixing ratios of nitrogen oxides (NO x ), CO acts as a source for tropospheric ozone, which is also a greenhouse gas.Since both CH 4 and CO have the potential to affect directly and indirectly the global climate, it is crucial to understand their changing emissions with time by obtaining better spatiotemporal information about their surface fluxes.
In terms of regional emissions of greenhouse gases and air pollutants, East Asia is one of the most important regions in the world because of the positive association between anthropogenic emissions and the recent rapid economic growth in the region (e.g., Emission Database for Global Atmospheric Research (EDGAR) v4.2 (EC-JRC/PBL, 2011); Carbon Dioxide Information Analysis Center (CDIAC) (Boden et al., 2011); Regional Emission inventory in ASia (REAS) v2.1; Kurokawa et al., 2013).For example, fossilfuel-derived CO 2 emissions in China, based on the CDIAC database, have more than doubled over the recent decade (Fig. 1), and surpassed the United States as the world's largest fossil CO 2 emitter in 2006 (Gregg et al., 2008).The increases in the fossil CO 2 emissions from China will also likely lead to enhanced emissions of other chemical species related to fossil fuel combustion.The recently revised EDGAR v4.2 shows an increase in the estimated anthropogenic CH 4 and CO emissions from China after 2003 (Fig. 1b and c).These increases are mostly related to CH 4 from coal mining and CO from the sectors of industrial manufacturing and metal production.However, the CH 4 and CO emission inventories are considered to be more uncertain than the fossil CO 2 emission inventory because those emissions related to complete combustion are generally well estimated while the emissions related to incomplete combustion and agricultural activities are poorly constrained (Kurokawa et al., 2013).These emission increases will result in a corresponding elevation in the mixing ratios of CH 4 , CO, and CO 2 in the downwind region of the sources.
Downward from the source regions, a synoptic-scale variation (SSV) in the mixing ratio of one chemical species is usually associated with SSVs of other species that have similar regional emission distributions and atmospheric lifetime characteristics.Therefore, various combinations of specific chemical species with relatively well quantified sources or relationships can be used to constrain emission estimates of other species with poorly quantified source strengths.For example, atmospheric Radon-222 measurements have been used to constrain regional fluxes of such greenhouse gases as CO 2 and CH 4 (e.g., Levin et al., 2003;Schmidt et al., 2003;Wada et al., 2013).Worthy et al. (2009) estimated the interannual variation in anthropogenic CH 4 emissions from Europe and Siberia based on the CH 4 / CO 2 correlations observed at Alert, Canada.Yokouchi et al. (2006) estimated the emission strength of hydrofluorocarbons and halocarbons from East Asian countries based on their correlations with atmospheric CO observed at Hateruma Island.
The National Institute for Environmental Studies (NIES) has been carrying out in situ measurements of atmospheric greenhouse gases and pollutants including CO 2 , CH 4 , and CO at Hateruma Island (HAT; 24.05 • N, 123.80 • E), which is located off the coast of continental East Asia.During the period from late fall to early spring, HAT is mainly influenced by the air masses transported from East Asian countries, resulting in highly associated SSV events of enhanced mixing ratios of atmospheric CO 2 , CH 4 and CO.Using the observed CO 2 and CH 4 SSVs at HAT during winter (November-April), Tohjima et al. (2010) found that the ratio of the standard deviations of deseasonalized CO 2 and CH 4 (σ CO 2 / σ CH 4 ) gradually increased during 1996-2007, reflecting changes in the fossil CO 2 emissions from China.
In this study, in addition to revisiting the analysis of CO 2 and CH 4 at HAT, we have extended our study to include an analysis of the correlation between CO and CO 2 by using hourly data instead of daily averages, as was done in a previous study by Tohjima et al. (2010).Here we examine 24 h variations based on the hourly data because shorter-term variations are likely to reflect more directly the regional emissions from the East Asian countries.By calculating the temporal variations in the ratios of CH 4 / CO 2 , CO / CO 2 , and CO / CH 4 from the observed measurements at HAT over the period from 1997/1998 to 2010, and combining them with the footprint results for different gases from a Lagrangian particle dispersion model (FLEXPART) and the existing flux maps of CH 4 , CO and CO 2 , we are able to improve the estimate of the spatiotemporal distributions of these trace gases and their source strengths in continental Asia.

In situ observation at Hateruma Island
We have been monitoring atmospheric CO 2 , CH 4 and CO at HAT by using automated analytical systems, which have been described elsewhere by Mukai et al. (2001) for CO 2 , by Tohjima et al. (2002) for CH 4 , and Tanimoto et al. (2009) for CO.Therefore, we give only a brief description of these analytical systems.Two air intakes (inverted cylindrical beakers), one for CO 2 and the other for CH 4 and CO measurements, are placed at the top of a tower at a height of 36.5 m (46.5 m a.s.l.), and sample air is drawn in through a 3 / 8-inch stainless-steel tubing by diaphragm pumps placed inside the station building.
The sample air for CO 2 measurement is dried by passing it through 3 cold traps (2 • C, −40 • C and −70 • C).CO 2 is continuously measured by a nondispersive infrared analyzer (NDIR) (URA207, Shimadzu CO. Ltd., Japan) with an analytical precision of about 0.02 µmol mol −1 (ppm).CO 2 mixing ratios are determined against 4 working standard gases, each a mixture of cleanup air and pure CO 2 filled in a 49 L aluminum cylinder.The cleanup air is made from natural air by passing it through a cold trap to reduce water vapor, a heated Pt catalyst to oxidize CO and hydrocarbons into CO 2 and water vapor, and a molecular sieve 5A column to remove CO 2 and residual water vapor.The CO 2 mixing ratios of the 4 working standard gases span about 50 ppm, and fully cover the range of atmospheric variations observed at HAT.These 4 working standard gases are replaced every 2-2.5 yr, and their mixing ratios are calibrated against the NIES 09 CO 2 standard scale (Machida et al., 2011) in our laboratory before and after every use at HAT.Before and after differences are less than 0.05 ppm.
To reduce water vapor in the sample air for CH 4 and CO measurements, we use a Nafion ® tube dryer and a cold trap (−40 • C).The dried air is then introduced into the CH 4 and CO analytical systems, separately.CH 4 and CO are semicontinuously measured by a gas chromatograph equipped with a flame ionization detector (GC/FID) and a gas chromatograph/mercury oxide (GC / HgO) analyzer (RGA3, Trace Analytical Co. Ltd., USA), respectively.The analytical precisions are about 2 nmol mol −1 (ppb) for both CH 4 and CO measurements.The GC analysis interval for CH 4 and CO is 10 min, except during January 1996-November 1997, when the interval was 15 min for CH 4 .Thus the CH 4 data before November 1997 are not used in this study because of the lower analysis frequency.Both the CH 4 and CO mixing ratios are determined against the working standard gases, each made up of a mixture of cleanup air and pure CH 4 or CO gases filled in a 49 L aluminum cylinder.These CH 4 and CO working standards are traceable to the NIES 94 CH 4 standard scale (Terao et al., 2011) and the NIES 09 CO standard scale (Katsumata et al., 2011), respectively.
For CH 4 measurements, we use 3 working standard gases with mixing ratios of about 1700, 1850, and 2000 ppb.Each set of 3 working standard gases are replaced every 5 yr at HAT.After December 2009, we have changed the volume of the working gas cylinder to 9.4 L, and have replaced the working standard gases every year.We have yet to observe any significant drift in these CH 4 working standard gases during use at HAT.
For CO measurements, we use 3 working standard gases with mixing ratios of about 80 ppb, 200 ppb and 400 ppb for the period 1999 to 2004.After September 2004, we have added a 4th CO standard gas with a mixing ratio of about 800 ppb.The procedural protocol for the CO working gas is similar to that of the CH 4 working gas, and after December 2009 we have been using 9.4 L aluminum cylinders, instead of the 49 L cylinders, and are replacing them every year.Unfortunately, the working standards in all the 49 L aluminum cylinders except one showed significant drifts, ranging from 1 to 16 ppb yr −1 .The drifts in the CO working reference gases during use at HAT have been determined in our laboratory by linear or quadratic interpolation based on the temporal changes in the CO mixing ratios before and after use at HAT.In order to validate the interpolated CO mixing ratios of the working standard gases, we compare the in situ observation with flask sample measurements, which are collected primarily for O 2 / N 2 measurements at HAT (Tohjima et al., 2008).The CO mixing ratios from flask samples are measured by using a GC / HgO analyzer at our laboratory.The average and the standard deviation of the differences between the flask data and the corresponding in situ hourly data are 0.3 ppb and 7.0 ppb, respectively, with 89 % of the differences lying within a range of ±10 ppb (Fig. S1 in the Supplement); the slope of the correlation plot is 1.008 ± 0.003 (Fig. S2), giving confidence to the ability of our interpolation to accurately reflect the drift.
For this study, we use hourly averages of the atmospheric CO 2 , CH 4 , and CO mixing ratios observed at HAT for the correlation analysis.The study periods are from November 1997 to March 2010 for CO 2 and CH 4 , and from November 1998 to March 2010 for CO (Fig. 2).

Methods of correlation analysis
Our focus is on the temporal change in the correlational relationship between CO 2 , CH 4 , and CO associated with synoptic-scale variations (SSVs) observed at HAT.Our calculation procedure begins with obtaining a best-fit smooth curve to each of the time series by using the methods of Thoning et al. (1989) with a cut-off frequency of 4.6 cycles yr −1 , and then subtracting it from the original time series.The residual time series are denoted as CO 2 , CH 4 and CO.The hourly CO 2 , CH 4 , and CO time series from 20 January to 11 February in 2008 are shown in Fig. 3a as a typical example.We have also calculated the best-fit smooth curves by using cut-off frequencies of 2.3 and 9.2 cycles yr −1 and have found that the average CH 4 / CO 2 , CO / CO 2 , and CO / CH 4 slopes (see  Sect. 5.1) for the 3 different cut-off frequencies agree to within 1 %.
We then construct scatter plots of CH 4 vs.CO 2 , CO vs. CO 2 , and CO vs. CH 4 using the first 24 h of data.Reduced major axis (RMA) regression analysis is carried out on each scatter plot to obtain regression slopes ( CH 4 / CO 2 , CO / CO 2 , and CO / CH 4 slopes) (Hirsh and Gilroy, 1984).We repeat the RMA analysis as we successively shift the 24 h time window by 1 h over the entire data record.If (i) the absolute value of the correlation coefficient is less than 0.8 (|R| < 0.8), or (ii) the number of data is less than 5, or (iii) the standard deviation of CO 2 is less than 0.1 ppm (1σ CO 2 < 0.1), then we deem the regression slope to be statistically insignificant and do not include it in the calculation of monthly averages of the correlation slopes.The RMA analysis is a relatively robust method of calculating the slope of two variables which show some causative relationship.Although we arbitrarily choose the correlation coefficient of 0.8 as a rough criterion for selecting significant correlation slopes, it should be noted that the average correlation slopes of CH 4 / CO 2 , CO / CO 2 , and CO / CH 4 , as discussed in Sect.5.1, do depend slightly on the value of the criterion.These average slopes for |R| < 0.7, |R| < 0.8 and |R| < 0.9 are plotted in Fig. S3.The root mean squares of the overall differences in the average regression slopes for |R| < 0.7 and |R| < 0.9 from those for |R| < 0.8 are 2.5 %, 2.3 %, and 3.4 % for CH 4 / CO 2 , CO / CO 2 , and CO / CH 4 , respectively.
Figure 4 shows the average seasonal variation of the regression slopes.In the figure, each closed circle denotes the monthly average for the whole observation period, each error bar represents the standard deviation, and each black bar shows the number of regression slopes used to calculate the monthly average.The negative values of the average  1 2 3 4 5 6 7 8 9 10 11  CH 4 / CO 2 and CO / CO 2 slopes in the summer are due to contribution from the enhanced terrestrial CO 2 uptake.We also note that the seasonal variation in the average CO / CH 4 slope may be attributable mainly to the seasonality in the air mass transport.During the summer, air masses arriving at HAT are predominantly transported from the Pacific region and the contributions of the Southeast Asian emissions show a relative increase.(Fig. S4 shows the average footprint, which is discussed in Sect.4.1, during the summer period (May to September)).Thus the average CO / CH 4 slope is low in summer because the CO / CH 4 emission ratios for the Southeast Asian countries are lower than those for China, Japan and Korea (e.g., Kurokawa et al., 2013).However, it is possible that the seasonality in the emissions from East Asia, the maximum CH 4 emissions in summer (Yan et al., 2003) and the maximum CO emissions in winter (Streets et al., 2003;Zhang et al., 2009), and the significantly faster CO reaction with OH in summer could partially contribute to the seasonality in the average CO / CH 4 slope.
The number of regression slopes with high correlation coefficients (|R| > 0.8) decreases during the summer due to the predominant influence of maritime air masses from the Pacific at HAT (see Fig. S4).During the winter, fluxes from East Asia contribute to the short-term variations observed at HAT.But the monthly average slopes, especially CH 4 / CO 2 , are relatively constant from late autumn to early spring (from November to March), when the biotic activity is relatively dormant.These results seem to suggest that the emissions from East Asia during the 5-month winter period are relatively constant in time.
We now proceed to examine in detail the interannual variations in the correlation slopes observed at HAT during the 5-month winter period from November to March.Note that the 2006/2007 winter, for example, indicates the period from November 2006 to March 2007.The 5-month winter period used for this study is shorter than the 6-month winter period (November to April) used in a previous study by Tohjima et al. (2010).

Concentration footprint for the measurements at HAT
In order to investigate the relationship between regional emissions and the correlation slopes observed at HAT, we employ FLEXPART v8.0 (Stohl et al., 1998), a Lagrangian particle dispersion model.The model has been used to simulate the SSVs at HAT for CO 2 (Koyama et al., 2011), and for CO 2 and O 2 (Minejima et al., 2012).We compute concentration footprints that indicate the degree of sensitivity of each measurement to the surface fluxes upwind of the measurement site (HAT).To drive FLEXPART, we use 6-hourly meteorology data with a spatial resolution of 1.25 • × 1.25 • from the JMA Climate Data Assimilation System (JCDAS) provided by the Japan Meteorological Agency (JMA).In each model run, 10 000 particles are released from 36.5 m a.g.l. at HAT and transported back 8 days, giving enough time for particles to spread over East Asia.Footprints are prepared on a 0.5 • × 0.5 • grid for 1, 2, 3, . . ., 8 days back in time with a time resolution of 3 h from January 2006 to December 2010.Details of the model and simulation condition are described in Ganshin et al. (2012Ganshin et al. ( , 2013)).
The average footprint for HAT during the 5-month winter period is shown in Fig. 5.The distribution of the footprint clearly shows that the emissions from East Asia have significant influence on the observation at HAT.The average correlation slopes from the model are determined mostly by the emissions from the area with footprint larger than 1 × 10 −4 ppm (gC m −2 day −1 ) −1 .Therefore, the area confined by the 1 × 10 −4 ppm (gC m −2 day −1 ) −1 contour line is referred to hereafter as an effective footprint area (EFA) for the measurements at HAT. EFA includes northern China, eastern China, Korea, western Japan, and Taiwan.

Flux maps used in the model simulations
As part of the simulation procedure, we use fossil CO 2 emissions (fossil fuel burning and cement production) from  the EDGAR v4.2 CO 2 annual flux data (excluding shortcycle organic carbon from biomass burning) with a grid of 0.1 • × 0.1 • for the period 1997 to 2008.For our study, the spatial resolution of these annual fluxes is decreased to 0.5 • × 0.5 • .The emissions from China, Japan, Korea, Taiwan and the residual regions are then scaled to the national emission inventories from CDIAC (Boden et al., 2011) for the corresponding years.The fossil CO 2 flux maps for 2009 and 2010 are extrapolated from the 2008 EDGAR v4.2 emission map.
In addition to the fossil CO 2 fluxes, climatological monthly fluxes of the terrestrial biosphere from the optimized CASA (Carnegie-Ames-Stanford Approach) ecosystem model (Nakatsuka and Maksyutov, 2009) and of the oceanic air-sea exchange from Takahashi et al. (2009) are used.These data have a spatial resolution of 1 • × 1 • , and are applied year after year (no interannual variation) for simulating the atmospheric CO 2 SSVs at HAT.
For CH 4 , we use the monthly CH 4 emission maps gridded at 1 • × 1 • for 2007 taken from the emission scenario (E2 scenario) developed in Patra et al. (2009).Simulated atmospheric CH 4 mixing ratios in a transport model using these CH 4 emission maps generally agree with the observed global distribution and seasonality (Patra et al., 2009).We use the above CH 4 emission maps due to their technical ease of use and suitability, and not the CH 4 emission maps from the EDGAR v4.2 database because it only provides the annual flux maps.For CO, we use the EDGAR v4.2 CO annual flux maps gridded at 0.1 olution of the CO flux map is decreased to a resolution of 0.5 • × 0.5 • .In order to examine the influence of different emission databases on the concentration fields in a transport model, we also use anthropogenic CO 2 emission distributions from the Open-source Data Inventory of Anthropogenic CO 2 Emission (ODIAC) v3.0, which was originally developed by Oda and Maksyutov (2011) and modified by Maksyutov et al. (2013), and the recently revised REAS v2.1 (Kurokawa et al., 2013; http://www.nies.go.jp/REAS/), and CH 4 and CO emission distributions from REAS v2.1.Although both ODIAC v3.0 and REAS v2.1 provide monthly emission maps, annually averaged emission maps for CO 2 and CO are used for the simulation comparisons.The CO 2 emissions from China, Japan, Korea, and Taiwan contained in the ODIAC and REAS v2.1 databases are scaled to the CDIAC national emission inventories, as was done for the EDGAR CO 2 emission maps.Similarly, the annually averaged CO emission maps for 2007 from REAS v2.1 are scaled to match the EDGAR national emissions for China, Japan, Korea, and Taiwan for 2007.The REAS v2.1 monthly CH 4 emission maps for 2007 are also scaled to match the monthly national emissions of Patra et al. (2009) for China, Japan, Korea, and Taiwan.The influence of these different emission maps on the correlation slopes are discussed in Sect.5.2.

Simulation of SSVs of CO 2 , CH 4 , and CO and their correlation analysis
In this study, we have been able to compute concentration footprints from January 2006 to December 2010 using FLEXPART.We then multiply these concentration footprints with the abovementioned fluxes to produce simulated time series of SSVs corresponding to individual fluxes.Chemical destruction of CH 4 and CO by hydroxyl radical (OH) during transport from the source regions to the receptor (HAT) are estimated by using OH concentration values at various latitudinal locations of the back trajectory in the boundary layer.
The OH concentration data are obtained from the monthly climatological distributions of zonal mean OH (Spivakovsky et al., 2000).The reaction rate constants for CH 4 and CO with OH are obtained from the Jet Propulsion Laboratory database (Sander et al., 2006).Although the chemical reaction with OH does not significantly influence the simulated CH 4 concentration during its short transport time, it does reduce CO SSVs by about 5 % in the winter.The simulated CO 2 , CH 4 , and CO variations for the same period as shown Fig. 3a are depicted in Fig. 3b.As can be seen in Fig. 3a and b, the model results are generally able to capture the observed pollution events with elevated mixing ratios.However, many of the details in the observed variations of the mixing ratios are not well reproduced, and the simulated amplitudes are generally underestimated, consistent with the results of previous studies (Koyama et al., 2011;Minejima et al., 2012).These discrepancies between the simulation and observation can be attributed to errors in the transport and/or fluxes used in the model simulation.A lack of variations of several days due to the timescale of largescale air mass mixing might partially contribute to the apparent smaller-than-observed simulated variations because the filtering procedure mentioned in Section 3 cannot effectively remove such midrange variations.In this study, we have attempted to reduce the impact of transport errors by using correlation slopes to estimate regional emissions.It has been found that these correlation slopes are not strongly influenced by atmospheric transport under the assumption that the CO 2 , CH 4 and CO fluxes have similar spatial distributions in East Asia (cf.Tohjima et al., 2010).To examine the robustness of this assumption, we compare the flux maps prepared in Sect.4.2 (see Fig. S5).Since the strong emissions, primarily confined to the land areas, are generally distributed in the southern part of North China, east China, the Korean Peninsula, and Japan, we are confident that the spatial distributions of the CO 2 , CH 4 , and CO fluxes within EFA are roughly similar to each other.
In order to evaluate the relationship between the East Asian emissions and the correlation slopes during the winter period, we calculate average correlation slopes from the simulated 5 yr SSVs for various sets of the abovementioned emissions.The average correlation slopes are calculated as follows: time series of SSVs of CO 2 are obtained as a summed contribution from the fossil, oceanic, and land biotic CO 2 fluxes.From the simulated CH 4 , CO, and combined CO 2 time series, we obtain correlation slopes of CH 4 / CO 2 and CO / CO 2 , in the same way as the observed correlation slopes are calculated (see Sect. 3).We evaluate the winter average of the correlations slopes for the fossil emissions during 1998 to 2010 by using the corresponding time series of SSVs of the fossil CO 2 .In this calculation, any influence of the year-to-year variation in the meteorological transport on the average correlation slopes appears as variability in the correlation slopes from the 5 yr simulated SSVs.Temporal changes in the winter averages of the correlation slopes are depicted in Fig. 7. Tohjima et al. (2010) examined the interannual changes in the synoptic-scale variations of CO 2 and CH 4 at HAT during a 6-month period (November to April) of each year from 1996 to 2007, and found that the CO 2 variability gradually increased relative to that of CH 4 .They attributed the gradual increase to the recent rapid increase in the fossil-fuel-derived CO 2 emissions from China (Gregg et al., 2008).Thus, the increase in the CO 2 variability has caused the average CH 4 / CO 2 and CO / CO 2 correlation slopes shown in Fig. 7 to decrease gradually with time.The observed CH 4 / CO 2 and CO / CO 2 slopes decrease from about 12 to less than 10 ppb ppm −1 and from about 45 to 30 ppb ppm −1 during 1998-2010, respectively.But the rate of decrease is gradually getting smaller for the CH 4 / CO 2 slope, while it is gradually getting larger for the CO / CO 2 slope during the whole observation periods.Unlike these trends, the average CO / CH 4 slope shows gradual increase during 1999-

Simulated correlation slopes
As described in Sect.4, we obtain correlation slopes from the simulated CO 2 , CH 4 , and CO SSVs, using time-varying fossil CO 2 emissions.In Fig. 7a and b, we also depict the winter averages of the simulated CH 4 / CO 2 and CO / CO 2 slopes.Both of the simulated slopes show similar decreasing trends.The temporal change in the simulated CH 4 / CO 2 slopes agrees well with the observed change, especially for the early period.However, the temporal decrease in the decreasing rate is not well simulated for the latter period.This discrepancy may be explained by the previously noted increase in the CH 4 emissions from EFA, since the model CH 4 field is driven by the fixed 2007 CH 4 monthly emission maps, while the CO 2 field is driven by the temporally varying CDIAC emission inventories.The simulated CO / CO 2 slopes underestimate the observed slopes by about 30 % (Fig. 7b), suggesting that the CO / CO 2 emission ratio within EFA in the model is lower than the actual ratio.However, the overall decreasing trend pattern of the simulated CO / CO 2 slopes agrees generally with the observation.The simulated CO / CH 4 slope based on the fixed emissions is also about half of the observed slopes (Fig. 7c).
The winter averages of the CH 4 / CO 2 , CO / CO 2 , and CO / CH 4 ratios of the EFA emissions used in the simulations are also plotted in Fig. 7.There are general agreements between the emission ratios and the simulated correlation slopes for the CH 4 / CO 2 ratio, but the CO / CO 2 and CO / CH 4 emission ratios are about 20 % and 25 % lower than the simulated CO / CO 2 and CO / CH 4 correlation slopes, respectively.These higher simulated correlation slopes, despite the CO removal reaction with the OH radical in the model, can be explained by the more localized and heterogeneous distribution of CO emissions within EFA than CH 4 and CO 2 .
In order to examine the influence of different emission maps on the simulated CH 4 / CO 2 and CO / CO 2 slopes, we repeat the above calculation by replacing one emission map with the ODIAC or REAS emission map (see Sect. 4.2).The results are depicted in Fig. S7.Note that since the emission maps are scaled to match the national emissions from the East Asian countries, the differences in the correlation slopes should be attributed to the differences in the emission distributions.The ODIAC CO 2 emission maps systematically produce about 4 % and 5 % lower CH 4 / CO 2 and CO / CO 2 slopes, respectively, than the EDGAR CO 2 emission maps.The CH 4 / CO 2 slopes based on the REAS and EDGAR emission maps agree with each other to within ±1 %, while the CO / CO 2 slopes based on the REAS emission maps are systematically higher by about 2 % than those based on the EDGAR emission maps.Additionally, we simulate CH 4 / CO 2 and CO / CO 2 slopes with the 1998 EDGAR CO 2 emission map, although the national emissions had increased according to the CDIAC inventories.This caused the simulated slopes to gradually depart from those for the normal case, but the differences are less than +1 % even in 2010.The REAS CH 4 emission map systematically produces about 4 % higher CH 4 / CO 2 on average, while the REAS CO emission map produces about 13 % lower CO / CO 2 slopes.From these limited results, we take these differences in the correlation slopes as the uncertainties caused by the uncertain emission distributions: 5 % for CO 2 , 4% for CH 4 , and 13 % for CO.

Estimation of CH 4 and CO emissions from EFA
We estimate the year-to-year change in the emissions of CH 4 and CO from EFA so that the simulated CH 4 / CO 2 and CO / CO 2 correlation slopes match the observations.We do this by adjusting single multiplying factors for the CH 4 and CO emissions.In this estimation, we assume that the flux maps of fossil CO 2 based on the EDGAR v4.2 emission map and the CDIAC database are correct.Additionally, we also assume, for ease of interpretation, no interannual variations in the terrestrial biospheric and oceanic CO 2 fluxes.Since the simulated correlation slopes from the annual CO 2 emissions are based on a 5 yr (2006)(2007)(2008)(2009)(2010) concentration footprint calculated by FLEXPART, the annual emission estimates do not reflect the meteorological field of the corresponding years.The influence of the year-to-year variation in the meteorological condition on the simulated correlation slope is therefore part of the variability of the 5 yr simulation.
The optimized winter CH 4 and CO emissions are simply extended to annual emissions.We can do this because the CH 4 emissions from biotic sources like rice fields have large seasonality, with highest emissions occurring in the summer.In China, CH 4 emissions from rice fields occur from April to October, and more than 70 % of the total emission happens between June and August, with about 90 % of it occurring in southern China (between 23 • N and 33 • N) (Yan et al., 2003).Therefore, a large contribution to the observed CH 4 SSV at HAT during the winter comes from the nonseasonal CH 4 emissions from the anthropogenic sources, and not from biotic sources like the rice fields.
Interannual variations in the annual CH 4 and CO emissions are estimated using two scenarios: (1) S1 -changing the emissions from EFA; and (2) S2 -changing only the emissions from China within EFA.The estimated emissions are depicted in Fig. 8, where the results from S1 and S2 are denoted by the closed squares and open circles, respectively.The error bars in the figure represent estimated uncertainties, which consist of uncertainties associated with the observed and simulated average correlation slopes, the uncertainties associated with the correlation coefficient criteria (2.5 % for CH 4 / CO 2 and 2.3 % for CO / CO 2 ; see Sect.3), the uncertainty of the fossil-fuel-derived CO 2 emissions from China (15 %; Gregg et al., 2008), and the uncertainty derived from the uncertain emission distributions (5 % for CO 2 , 4 % for CH 4 , and 13 % for CO; see Sect.5.2).In Fig. 8, we also plot the Chinese part of the EFA emissions in S1 as open squares.The CH 4 emissions from China estimated in S1 and S2, shown as open circles and open squares in Fig. 8a, agree to within ±3 % of each other.Similar results can be seen in the CO emissions from China (Fig. 8b) although the Chinese CO emissions in S2 are, on average, systematically higher by about 4 % than those in S1.These results seem to suggest that the emissions from EFA outside China contribute little to the observed changes in the average correlation slopes at HAT and the emissions from the Chinese part of EFA can be robustly estimated by both scenarios.
The estimated annual CH 4 emissions are relatively stable during the first several years but increase thereafter; for CO, they seem to increase and decrease during the early part and the later part of the observation period, respectively.The annual CH 4 emission from China within EFA is about 21 TgCH 4 yr −1 during 1998-2005 and increases to about 27 TgCH 4 yr −1 in 2009/2010; but for CO it is about 100 TgCO yr −1 in 1998/1999, increases to about 140 TgCO yr −1 in 2004/2005, and then slightly decreases to about 120 TgCO yr −1 in 2009/2010.The simulated CO / CH 4 slopes based on these optimized CH 4 and CO emissions from China within EFA (S2), plotted in Fig. 7c as black open squares, show considerable agreement with the observations.

Comparison of the estimated emissions with other studies
The estimated CH 4 and CO emissions from China in the previous section are compared with the annual emissions from China estimated by other studies.Since we have optimized those Chinese emissions from EFA only (S2), we have added the emissions from China outside EFA (hereinafter referred to as the extended Chinese area (ECA)) to obtain an estimated national emission for the winter period.These additional emissions from ECA are computed from the CH 4 and CO flux maps prepared in Sect.4.2.We thus obtain additional emissions of 18 TgCH 4 yr −1 and 31.6 TgCO yr −1 , which we assume to be constant over the study period.Our estimated CH 4 and CO emissions from China are summarized in Table 1 and plotted in Figs. 9 and 10, respectively.We find that the estimated CH 4 and CO emissions from ECA contribute less than 47 % and 23 % to the total China emissions, respectively.

CH 4
Bottom-up estimates of the CH 4 emission from Chinese anthropogenic sources without the rice fields taken from the inventory databases EDGAR v4.2, REAS v1.1 (Ohara et al., 2007;  ( [2002][2003][2004][2005][2006][2007][2008], which is about 3 times larger than our estimates of 1.1 ± 0.2 TgCH 4 yr −2 (2002-2010).The REAS v2.1 estimates, being higher than our estimates, also show a faster increase of 3.6 ± 0.2 TgCH 4 yr −2 (2000-2008).About 70 % and 90 % of the increases in the Chinese emissions in the EDGAR v4.2 and REAS v2.1 estimates, respectively, are attributed to the emissions related to coal mining (fugitive emissions from solid fuels), and occur mostly within EFA.Note that the REAS v1.1 estimates are lower than our estimates and the differences from the REAS v2.1 estimate for 2000 are attributed to the fugitive emissions from fossil fuels (73 %) and the emissions from land disposal of solid waste (24 %).The possibility that the CH 4 emissions in the EDGAR v4.2 inventory are overestimated was also suggested by the following model studies.In a chemistry-transport model intercomparison experiment of CH 4 (TransCom-CH 4 ), the forward simulations of atmospheric CH 4 were conducted using several transport models and various sets of surface CH 4 emission scenarios (Patra et al., 2011) (Ohara et al., 2007) and REAS v2.1 (Kurokawa et al., 2013), respectively.
the global air sampling network and satellite sensor.The inversion result shows a significant increase in the anthropogenic CH 4 emissions from China but a smaller increase than that indicated by the EDGAR inventory.The increasing rate of 1.1 ± 0.3 TgCH 4 yr −2 estimated by Bergamaschi et al. (2013) for the period of 2000-2010 is in excellent agreement with our estimation.Therefore, we suspect that the EDGAR v4.2 and REAS v2.1 inventories are overestimating the recent increase in the CH 4 emissions related to the coal mining.

CO
The reported estimates of CO emission from China based on bottom-up and top-down approaches are plotted in Fig. 10, together with our estimates.The bottom-up estimates from REAS v1.1 and EDGAR v4.2 and other studies (Streets et al., 2003(Streets et al., , 2006;;Zhang et al., 2009) are plotted as circles in the figure.The top-down approach is divided into inverse modeling (Palmer et al., 2003;Wang et al., 2004;Heald et al., 2004;Yumimoto andUno, 2006, 2012;Tanimoto et al., 2008;Kopacz et al., 2009) and forward modeling (Heald et al., 2003;Allen et al., 2004), which are plotted as squares and diamonds in the figure, respectively.These top-down emissions were estimated by using CO observations from aircrafts, ground-based stations, satellites, or any combination thereof.In general, our estimated emissions agree with these reported bottom-up and top-down estimations except those taken from the EDGAR v4.2 database, which shows about 40 % lower estimates than the other estimates, including ours.
The top-down estimates, including ours, reflect not only the primary CO emissions but also the secondary CO production from the oxidation of NMVOC.However, we consider the contribution of the CH 4 oxidation to the top-down estimates of CO emissions based on the atmospheric observations in the downwind regions from China to be negligible because of the much longer life time of atmospheric CH 4 (about 10 yr; e.g., Patra et al., 2011) compared to its transit time.It is to be noted that the EDGAR database reports only the primary CO emissions.Duncan et al. (2007) estimated that the oxidation of NMVOC contributes nearly 50 % of the total primary CO emissions to the global CO emission.If this ratio is valid and can be applied to the EDGAR estimate for China, then the resulting net CO emissions with both primary and secondary sources can be applied to our top-down estimates.In addition, our winter emission estimates would of course be biased if the CO emission has a noticeable seasonality.For example, using monthly data for power generation and industry, as well as residential energy consumption, Zhang et al. (2009) developed a data set of monthly CO emissions from China.The result shows a significant seasonality, with 17 % larger average monthly emission for our 5-month winter than for an entire year.If in fact there is a strong seasonal variation in the CO emission, then our winter estimate needs to be reduced by 17 %, which also brings our estimate close to the EDGAR v4.2 estimate.The above discussion points to the importance of correct evaluation of the secondary CO emissions when comparing topdown and bottom-up emission estimates.Note that the REAS v2.1 estimates, in which the secondary CO emissions are not explicitly included, agree well with the top-down estimate.Kurokawa et al. (2013) attribute the differences in the CO emissions between REAS v2.1 and EDGAR v4.2 to the emission factors used in the estimations; the emission factors for REAS v2.1 might implicitly include the secondary productions.
The TRACE-P (Transport and Chemical Evolution over the Pacific) campaign, which was intensively conducted over the western Pacific in the spring of 2001, reported several CO emission estimates from China; these values are shown in Fig. 10.Those estimates range from 142 to 181 TgCO yr −1 , with lower values corresponding closely to the bottom-up estimates of REAS v1.1 and v2.1 and our estimates.The year-to-year variation in our emission estimates agree with the reported variation in the emission estimates during the period 1999-2010.The CO emission increase during 1999-2005 is easily understandable because fossil fuel consumption in China has steadily increased after 2000.The question is whether the CO emission from China has truly stopped increasing after 2005 in spite of the continued increase in the estimated fossil fuel consumption.
From the atmospheric CO 2 , CO and 14 CO 2 measurements at Tae-Ahn Peninsula in Korea (TAP) during 2004-2010, Turnbull et al. (2011) found that the emission ratio of CO to fossil-fuel-derived CO 2 from China showed a gradual decrease during 2004-2006 and then plateaued from 2007 till 2009.For the usual 3-month winter season (December-February), Wang et al. (2010) showed that an average of the CO / CO 2 slopes observed at the Miyun site (located about 100 km northeast of Beijing), also showed a decreasing trend from the 2004/2005 winter to the 2008/2009 winter, with the latter value being slightly larger than that for the 2007/2008 winter (Wang et al., 2010).They attributed these decreasing trends in the CO / CO 2 ratio to recent improvements in the fossil fuel combustion efficiency in China, corresponding in time to the implementation of air pollution reduction measures to improve the air quality in Beijing prior to the 2008 Summer Olympics and Paralympics Games.
Although the CO / CO 2 ratio values obtained by Wang et al. (2010)  This continued decrease seems to suggest that the overall combustion efficiency in China might still continue to improve even after the Olympics.However, as we noted above, the observed temporal changes in the CO / CO 2 slopes at HAT are mainly reflective of the year-to-year variation in the terrestrial CO 2 emission strength and atmospheric transport.Korea and Japan, which have lower CO / CO 2 emission ratios than China, could reduce the CO / CO 2 slopes at HAT.Such issues will be examined in our future work.

Conclusion
We have examined the year-to-year changes in the average correlation slopes of CH 4 / CO 2 , CO / CO 2 and CO / CH 4 for a 5-month winter from November to March observed at HAT over the last 10 yr.Consistent with the recent rapid increase in the fossil fuel CO 2 emissions from China, the observed CH 4 / CO 2 and CO / CO 2 slopes have correspondingly decreased.However, there are differences in the decreasing trends between CH 4 / CO 2 and CO / CO 2 slopes; the rate of decrease for the CH 4 / CO 2 slope is decreasing while that for the CO / CO 2 slope is increasing during the study period.These observational results may reflect differences in the interannual variations that are occurring between the emissions of CH 4 and CO from their source regions.Such different emission trends are also supported by the temporal changes observed in the CO / CH 4 slope, with an increasing trend during 1999-2004 and a decreasing trend during 2005-2010.
We have estimated the geographical distribution of the sensitivity of the winter measurements at HAT to the emissions from East Asia by using a Lagrangian particle dispersion model (FLEXPART) driven by meteorological data from 2006 to 2010.The average sensitivity distribution (average footprint) shows that the emissions from northern and eastern China mostly contribute to the observed shortterm variations at HAT.We have also evaluated the relationship between the emissions from East Asia and the correlation slopes at HAT by using FLEXPART and a set of previously reported emission maps.From the relationship, we have been able to obtain estimates of the CH 4 and CO emissions from China so that the simulated average CH 4 / CO 2 and CO / CO 2 slopes would match the observed slopes.In this calculation, we have assumed that there are no year-to-year variations in the land biotic and oceanic CO 2 emissions and that the fossil-fuel-derived CO 2 emissions change in accordance with the CDIAC database.The estimated CH 4 emissions from China, corresponding to emissions from nonseasonal sources or anthropogenic sources without rice fields, show a relatively constant value of about 39 ± 7 TgCH 4 yr −1 during 1998-2002, gradually increasing at a rate of 1.1 ± 0.2 TgCH 4 yr −2 during 2002-2010, and then reaching 46 ± 8 TgCH 4 yr −1 in 2009/2010.This increasing rate of the Chinese CH 4 emissions for the decadal period under study is much smaller than those taken from the EDGAR v4.2 and REAS v2.1 databases.However, these estimated CH 4 emission increases are not negligible in the context of the global CH 4 budget, and might substantially contribute to the recent atmospheric CH 4 increase (Rigby et al., 2008;Dlugokencky et al., 2009;Terao et al., 2011).
We have estimated the annual CO emissions from China to be 134 ± 32 TgCO yr −1 in 1998/1999, increasing to 182 ± 42 TgCO yr −1 in 2004/2005, plateauing during 2005-2008, and then decreasing to less than 160 TgCO yr −1 after 2008/2009.In spite of the recent continued increase in fossil fuel consumption in China, our results have shown that the recent stagnation or decrease in the CO emission from China points to the possibility of a continued improvement in the combustion efficiency in that country due to air pollution reduction measures that were implemented prior to the 2008 Olympics.Our CO emission estimates agree well with previously reported CO emission estimates, except for the EDGAR v4.2 database, which shows a value smaller than our estimates by about 30 TgCO yr −1 .However, this discrepancy may be attributed to the secondary CO production derived from the atmospheric oxidation of NMVOC, which is not included in the EDGAR bottom-up estimation.
Publications on behalf of the European Geosciences Union.Y. Tohjima et al.: Temporal changes in the emissions of CH 4 and CO

Fig. 2 .
Fig. 2. Time series of atmospheric (top) CO 2 , (middle) CH 4 , and (bottom) CO mixing ratios observed at HAT.Each dot represents hourly average.Black lines represent the smooth curve fits to the data.

Fig. 5 (
Fig. 5 (Tohjima et al.) Fig. 5. Average footprint (ppm (gC m −2 day −1 ) −1 ) for the measurements at HAT during the winter period (November to March).Meteorological data for 2006-2010 are used for the calculation.The location of HAT is indicated by the square.The area surrounded by the red thick contour lines of 1 × 10 −4 ppm (gC m −2 day −1 ) −1 is defined as an effective footprint area (EFA).

Figure 6
Figure 6 shows the histograms of the correlation slopes derived from the observations made during the winter periods of 1998/1999 and 2009/2010 for CH 4 / CO 2 , of 2001/2002 and 2009/2010 for CO / CO 2 , and of 2003/2004 and 2009/2010 for CO / CH 4 .These years are selected to highlight the extent of change in the distribution of the individual correlation slopes that has taken place over a decadal timescale.The correlation slopes of CH 4 / CO 2 , CO / CO 2 and CO / CH 4 for the 2009/2010 winter have narrower distributions and lower average values than those of the earlier years.These changes are attributable to changes in the relative strengths of the emissions from EFA.The histograms of the correlation slopes of the simulated CH 4 / CO 2 and CO / CO 2 (see Sect. 5.2) show similar temporal changes to those in the observation: the distributions become narrow in association with the increase in the fossil-fuel-derived CO 2 emissions in China (see Fig. S6, in which the histogram of the simulated CH 4 / CO 2 for 1998/1999 and 2009/2010 and CO / CO 2 for 2001/2002 and 2009/2010 are depicted).Note that the histograms of the simulated CH 4 / CO 2 and CO / CO 2 driven by the 1998 fossil CO 2 emission map show a similar temporal change in the distribution, suggesting that the contribution of the change in the emission distribution is relatively small.Temporal changes in the winter averages of the correlation slopes are depicted in Fig.7.Tohjima et al. (2010) examined the interannual changes in the synoptic-scale variations of CO 2 and CH 4 at HAT during a 6-month period (November to April) of each year from 1996 to 2007, and found that the CO 2 variability gradually increased relative to that of CH 4 .They attributed the gradual increase to the recent rapid increase in the fossil-fuel-derived CO 2 emissions from China(Gregg et al., 2008).Thus, the increase in the CO 2 variability has caused the average CH 4 / CO 2 and CO / CO 2 correlation slopes shown in Fig.7to decrease gradually with time.The observed CH 4 / CO 2 and CO / CO 2 slopes decrease from about 12 to less than 10 ppb ppm −1 and from about 45 to 30 ppb ppm −1 during 1998-2010, respectively.But the rate of decrease is gradually getting smaller for the CH 4 / CO 2 slope, while it is gradually getting larger for the CO / CO 2 slope during the whole observation periods.Unlike these trends, the average CO / CH 4 slope shows gradual increase during1999-2004 and decrease during 2005-2010.

Fig. 7 .
Figure 6 shows the histograms of the correlation slopes derived from the observations made during the winter periods of 1998/1999 and 2009/2010 for CH 4 / CO 2 , of 2001/2002 and 2009/2010 for CO / CO 2 , and of 2003/2004 and 2009/2010 for CO / CH 4 .These years are selected to highlight the extent of change in the distribution of the individual correlation slopes that has taken place over a decadal timescale.The correlation slopes of CH 4 / CO 2 , CO / CO 2 and CO / CH 4 for the 2009/2010 winter have narrower distributions and lower average values than those of the earlier years.These changes are attributable to changes in the relative strengths of the emissions from EFA.The histograms of the correlation slopes of the simulated CH 4 / CO 2 and CO / CO 2 (see Sect. 5.2) show similar temporal changes to those in the observation: the distributions become narrow in association with the increase in the fossil-fuel-derived CO 2 emissions in China (see Fig. S6, in which the histogram of the simulated CH 4 / CO 2 for 1998/1999 and 2009/2010 and CO / CO 2 for 2001/2002 and 2009/2010 are depicted).Note that the histograms of the simulated CH 4 / CO 2 and CO / CO 2 driven by the 1998 fossil CO 2 emission map show a similar temporal change in the distribution, suggesting that the contribution of the change in the emission distribution is relatively small.Temporal changes in the winter averages of the correlation slopes are depicted in Fig.7.Tohjima et al. (2010) examined the interannual changes in the synoptic-scale variations of CO 2 and CH 4 at HAT during a 6-month period (November to April) of each year from 1996 to 2007, and found that the CO 2 variability gradually increased relative to that of CH 4 .They attributed the gradual increase to the recent rapid increase in the fossil-fuel-derived CO 2 emissions from China(Gregg et al., 2008).Thus, the increase in the CO 2 variability has caused the average CH 4 / CO 2 and CO / CO 2 correlation slopes shown in Fig.7to decrease gradually with time.The observed CH 4 / CO 2 and CO / CO 2 slopes decrease from about 12 to less than 10 ppb ppm −1 and from about 45 to 30 ppb ppm −1 during 1998-2010, respectively.But the rate of decrease is gradually getting smaller for the CH 4 / CO 2 slope, while it is gradually getting larger for the CO / CO 2 slope during the whole observation periods.Unlike these trends, the average CO / CH 4 slope shows gradual increase during1999-2004 and decrease during 2005-2010.

Fig. 10 .
Fig. 10 (Tohjima et al.) and Turnbull et al. (2011) showed a slight increase after 2008, the CO / CO 2 slopes at HAT in 2009-2010 show a substantial decrease compared to those of the preceding years.Top-down estimates based on the satellite data from Yumimoto and Uno (2012), although showing a slight increase during 2005-2007, also show a distinct decrease from 2007 to 2008 (shown as open blue squares with line in Fig. 10) and maintain the lower values during 2008-2010.
For example, relative increases in the air mass transport from www.atmos-chem-phys.net/14/1663/2014/Y. Tohjima et al.: Temporal changes in the emissions of CH 4 and CO

2014 1672 Y. Tohjima et al.: Temporal changes in the emissions of CH 4 and COTable 1 .
Gregg et al. (2008)go.jp/frsgc/research/p3/emission.Summary of the estimated CH 4 and CO emissions from China a .Values for CO 2 are given in PgC yr −1 , for CH 4 in TgCH 4 yr −1 , and for CO in TgCO yr −1 .bFossilCO 2 emissions are taken from the CDIAC database.Each value is the average of the emissions for the consecutive two years described in the first column.The uncertainty is assumed to be 15 %, which is the lower limit of the estimation ofGregg et al. (2008).c Values represent the emissions from nonseasonal CH 4 sources (see text).d Uncertainties are calculated from the uncertainties of the fossil-fuel-derived CO 2 emissions in China, of the observed correlation slopes including the influence of the correlation coefficient criteria selection, and of the simulated correlation slopes including the influence of the uncertain emission distributions used in the simulation (see text).
htm) and REAS v2.1 are plotted for comparison with our results in Fig. 9.The EDGAR v4.2 emission estimates show good agreement with our estimates for the period 1998 to 2002.However, after 2002 the EDGAR v4.2 data show a much faster increase of 3.1 ± 0.1 TgCH 4 yr −2 www.atmos-chem-phys.net/14/1663/2014/Atmos.Chem.Phys., 14, 1663-1677, a . The forward CH 4 simulation based on the EDGAR v4.0 emissions, which are almost same as the EDGAR v4.2 emissions, shows a significantly faster growth rate during 2003-2007 than the observations.The Chinese emission increase contributes nearly 40 % to the global CH 4 emission increase in the EDGAR inventory.Recently, Bergamaschi et al. (2013) estimated global CH 4 emissions during the 2000s based on an inverse modeling constrained by atmospheric CH 4 data from Comparison of estimated nonseasonal CH 4 emissions from China.The values are expressed as annual emissions.Closed blue squares are the estimated emissions of this study.Green circles, light-blue squares, and light-blue circles represent the CH 4 emissions from anthropogenic sources (excluding rice fields) in China based on the emission inventories from EDGAR v4.2 (http://edgar.jrc.ec.europa.eu/),REAS v1.1