Evaluation of anthropogenic emissions of carbon monoxide in East Asia derived from the observations of atmospheric radon-222 over the western North Pacific

We used the observed CO/ 222Rn ratio in the Asian outflows at Minamitorishima (MNM), Yonagunijima (YON), and Ryori (RYO) in the western North Pacific from 2007 to 2011, together with a three-dimensional chemical transport model (STAG), in order to estimate anthropogenic emissions of CO in East Asia. The measurements captured highfrequency synoptic variations of enhanced 222Rn (ERN) events associated with the long-range transport of continental air masses. 222Rn and CO showed high correlation during the ERN events observed at MNM and YON in the winter and spring, but not at RYO. The STAG transport model reproduced well the concentrations of observed 222Rn when forced with a constant and uniform flux density of 1.0 atom cm−2 s−1, but underestimated the associated enhancement of synoptically variable CO caused by the underestimated flux values in the EDGAR ver. 4.1 emission database used in the model for East Asia. Better estimates for the East Asian emission were derived using a radon tracer method based on the difference in the enhancement ratio of CO/222Rn between the observation and the model. The anthropogenic emissions of CO for China, Japan, and Korea were estimated to be 203 Tg CO yr −1, 91 % of which originated in China. When compared with other estimated emissions of CO, our estimated result showed consistency with those of the inverse method, whereas the emission database of EDGAR was about 45 % smaller than our anthropogenic estimation for China.


Introduction
Recently, East Asia has become one of the large source regions of anthropogenic pollutants, along with North America and Europe.The emissions from East Asia continue to increase due to the growing economy and population (Akimoto, 2003).The continental outflow of pollutants from East Asia has had a significant impact on the environment in the downwind regions of the western North Pacific (Tanimoto et al., 2005).Several studies have also provided evidence of various influences on the air quality over certain regions of North America due to the long-range transport of polluted air masses from Asia (e.g.Andreae et al., 1988;Parrish et al., 1992;Husar et al., 2001).
The anthropogenic emission of trace gases from East Asia is not well quantified.In particular, emission estimates of carbon monoxide from incomplete combustion are characterized by large uncertainties due to the variability in the burning of fossil fuel, biofuel, and biomass in space and time.Bottom-up estimates based on the socioeconomic data have produced several emission databases, such as those of Streets et al. (2003), the Intercontinental Chemical Transport Experiment-Phase B (INTEX-B) emission inventory (Zhang et al., 2009), the Regional Emission inventory in Asia (REAS) (Ohara et al., 2007), and the Emission Database for Global Atmospheric Research (EDGAR) (Olivier et al., 2005).Another method to estimate surface emissions is to use a top-down approach, such as an inverse method (Heald et al., 2004;Kopacz et al., 2009;Hooghiemstra et al., 2011).However, there are significant disagreements between the bottom-up and top-down emission estimates for East Asia (Heald et al., 2004;Palmer et al., 2006).Those disagreements are caused by substantial uncertainties associated with the insufficient knowledge about the source regions and the emission factors used to convert the socioeconomic data into emission values in the bottom-up approach, and by uncertainties in the atmospheric transport mechanism and the lack of sufficient observational data to constrain the model in the inverse method.In this study we propose a third, and independent, method to assess and evaluate the disagreement.
The radon tracer method has been used to estimate continental emissions of methane (CH 4 ) (Thom et al., 1993;Schmidt et al., 1996;Levin et al., 1999), nitrous oxide (N 2 O) (Schmidt et al., 2001), carbon dioxide (CO 2 ) (Schmidt et al., 2003;Hirsch, 2007), and chlorofluorocarbons (CFCs) (Biraud et al., 2000(Biraud et al., , 2002) ) in Europe.Radon-222 ( 222 Rn) has been used to parameterize transport and dilution of ground level emissions of trace gases in the atmospheric surface layer (Kuhlmann et al., 1998;Schmidt et al., 2001Schmidt et al., , 2003)). 222Rn is a radioactive noble gas with a half life of 3.824 days (e-folding time τ = 5.52 days) and is produced by the decay of radium-226 distributed uniformly in the soil. 222Rn exhalation from soils has been found to be rather homogeneous in a restricted region and almost constant with time (Dörr and Münnich, 1990).The exhalation rate from the ocean is about two orders of magnitude less than that from soils (Wilkening, 1975).Thus, 222 Rn is a useful tracer for air masses originating from the land.Our study is the first to utilize the radon tracer method to obtain emission estimates of trace gases for East Asia.
The National Institute of Advanced Industrial Science and Technology (AIST) and the Meteorological Research Institute (MRI) have developed a high precision 222 Rn measuring system (Wada et al., 2010) and used it to measure atmospheric concentration of 222 Rn on the ground for more than three years at remote sites in the western North Pacific.In addition, the Japan Meteorological Agency (JMA) has been simultaneously measuring atmospheric concentrations of trace gases such as carbon monoxide (CO), carbon dioxide (CO 2 ), methane (CH 4 ), and ozone (O 3 ) at the same sites for more than a decade (Tsutsumi et al., 2006;Wada et al., 2007Wada et al., , 2011)).Wada et al. (2011) showed that the combustion-related enhancement ratio of CO to CO 2 observed at these sites depended on the emission ratio of CO to CO 2 in East Asia.In addition, Wada et al. (2010) demonstrated that the enhanced 222 Rn peaks observed at MNM showed a high correlation with synoptic scale variation of CO observed simultaneously at this station.These results indicated that the emission ratios of the trace gases emitted from East Asia are conserved in the enhancement ratios reflected at these sites.Thus, emissions of the trace gases in East Asia can be estimated from the observed data over the western North Pacific by the enhancement ratio of CO and 222 Rn.In this study, we show observational results of atmospheric 222 Rn and CO at the remote sites in the western North Pacific and resolve the relationship between 222 Rn and CO.Next, an estimation technique based on the radon tracer method is developed to obtain East Asian CO emission estimates based on the observed atmospheric 222 Rn.Finally, we compare our estimated results with other estimated values by the bottomup and top-down methods.
Minamitorishima (MNM) is a remote coral island in the western North Pacific, about 2000 km southeast of Tokyo.At this site, easterly winds associated with the maritime air masses prevail much of the year, except in the winter season, when the continental air masses from East Asia are transported by the synoptic scale weather systems.Since this island is very small, about 1.4 km 2 with a population of about fifty people, the influence of local sources and sinks on the trace gas measurement is negligible as reported by previous trajectory analysis (Wada et al., 2007(Wada et al., , 2011)).
Yonagunijima (YON) is a remote island located about 111 km east of Taiwan.At this site, northeasterly winds associated with the continental high pressure system prevail, except in the summer when southerly wind associated with the maritime air masses dominates the site.Although the island is located at the same latitude as MNM, YON is situated immediately downwind of the Asian continent (Fig. 1), thus frequently influenced by the polluted air masses from Asia (Tsutsumi et al., 2006).Since the island is small, lightly populated and has little traffic, the local influence on the measurements is not significant as reported by the trajectory analysis (Wada et al., 2011).
Ryori (RYO) is located on the east coast of the northeastern part of Japan.Northwesterly winds from the Siberian high pressure system dominate the site throughout the year.In the summer season, however, southerly winds associated with the Pacific subtropical high pressure system are often observed.The station is located on the ridge of a hill along the sawtooth coastline facing the Pacific Ocean, at an elevation of 260 m above the sea level.Due to its geographical location and complex terrain around the station, RYO is influenced by local sinks and sources, along with air pollutants associated with the long-range transport of maritime and continental air masses.

Measuring method
The 222 Rn analyzer used in this study was developed by AIST and MRI (Wada et al., 2010) for precise measurements of atmospheric 222 Rn.It is based on the electrostatic collection method that collects positively charged progenies of 222 Rn in the sampling air on the light sensitive photodiode (PIN photodiode) for an alpha particle detector charged high negative voltage.It automatically allows continuous measurement, to obtain hourly mean data from the data obtained every 10 minutes.Calibration of this analyzer was made at Nagoya University, with a detailed description of the calibration method given elsewhere (Iida et al., 1991(Iida et al., , 1996)).The detection limit of this measuring system is 0.16 Bq m −3 for hourly mean data.
The measurement by this 222 Rn analyzer has been made at MNM since September 2007, and at YON since November 2008.Our current sampling method is a commonly used one for greenhouse gas observation, and is described in Wada et al. (2007Wada et al. ( , 2010)).Air is sampled at the top of a 10 m observational tower.After drying by the electric cooling units and membrane tube dryers, the sample air is introduced into the 222 Rn analyzer at a flow rate of 3.0 L min −1 at all the sites.The measurement of 222 Rn at RYO was started in March 2009 by using a commercially available 222 Rn monitor with a detection limit of about 0.5 Bq m −3 (Wada et al., 2010).
The concentrations of CO were measured continuously by using a nondispersive infrared analyzer (NDIR: GA-360S, Horiba, Ltd., Japan) with an overall analytical precision of ± 5 ppb.The measuring system was changed to a gas chromatograph (TRA-1 Round Science Inc.), with an overall precision of ± 2 ppb, since 2008, 2009, and 2010 at YON, RYO, and MNM, respectively.The standard gases to calibrate the analyzers are traceable to the WMO mole fraction scale.

3-D chemical transport model
A 3-D global chemical transport model named Simulator of Trace Atmospheric constituent on a Global scale (STAG) developed at the AIST (Taguchi, 1996;Taguchi et al., 2002a) was used in this study.This model, which consists of 28 sigma vertical levels with a horizontal resolution of 1.875 • ×1.875 • , was driven every 6 hrs using the meteorological reanalysis data provided by the National Oceanic and Atmospheric Administration/the Cooperative Institute for Research in Environmental Sciences (NOAA/CIRES) Climate Diagnostics Center at their website (http://www.cdc.noaa.gov/).The boundary layer height at each time step in the model was determined by the bulk Richardson number (Troen and Mahrt, 1986) and a uniform distribution of trace gas concentrations was assumed in the boundary layer.The basic performance of the model was evaluated by using 222 Rn concentrations (Taguchi et al., 2002b).
The emanation rate of 222 Rn was given a value of 1.0 atoms cm −2 s −1 for all the land areas between 60 • S and 60 • N, and 0.5 atoms cm −2 s −1 for the land areas between 60 • N and 70 • N, excluding Greenland, with no temporal variation (Jacob et al., 1997;Taguchi et al., 2002bTaguchi et al., , 2011)).The emanation rate of 222 Rn from the ocean was set to 0 atoms cm −2 s −1 , since our study focused on the short term variations in atmospheric 222 Rn.
The CO emissions from all anthropogenic sources used in the model were obtained from the Emission Database for Global Atmospheric Research (EDGAR) version 4.1 (EC-JRC/PBL, 2010).Biomass burning, however, was not included.The CO from photochemical production via the oxidation of CH 4 and nonmethane hydrocarbons (NMHCs) were regarded as constant with a value of 40 ppb, a value based on the result of previous simulation studies (Liang et al., 2004;Takigawa et al., 2005).The photochemical destruction of CO with hydroxyl (OH) radical was included in the model as the only CO sink.The rate coefficient of this photochemical destruction was obtained from Atkinson et al. (2006).The concentration of OH radical was obtained from the climatological seasonal fields published by Spivakovsky et al. (2000).

Observational data of the atmospheric CO and 222 Rn
Figure 2 shows the hourly mean concentrations of atmospheric CO observed at MNM, YON, and RYO located in the western North Pacific.As reported by Wada et al. (2011), all three stations reveal a distinct seasonal cycle accompanied by high-frequency synoptic variations, although the annual mean CO level is station dependent.The seasonal cycle shows a minimum during the summer season from July to August and a maximum during the winter season from January to March.The synoptic variations consisting of episodically enhanced CO peaks on timescales less than one week are clearly revealed at all three stations, although their frequencies and magnitudes are different from each other.It has already been demonstrated by various modeling studies that major high CO concentration peaks observed at these stations are usually caused by the transport of Asian pollutions emitted mainly from China, Korea and Japan (Liang et al., 2004;Takigawa et al., 2005;Sawa et al., 2007).
Figure 3 shows the hourly mean concentrations of atmospheric 222 Rn observed at MNM, YON, and RYO.The temporal variations of 222 Rn at all three stations show, as was the case with CO, a distinct seasonal cycle accompanied by highfrequency synoptic variations.No significant long-term trend was found at any of the stations.The difference in the magnitude of the 222 Rn concentration among the three stations is related to their respective geographical locations with different distances from the continental 222 Rn source regions.The seasonal cycles observed at MNM and YON are similar, with a maximum in the winter season around December-January and a minimum in the summer season around July-September.In contrast, a different pattern of seasonal cycle at RYO was found, reflecting a strong influence from local 222 Rn source emissions in a complex terrain surrounding the station.
No significant diurnal cycle was found at both MNM and YON, indicating no impact of local emissions on the 222 Rn measurements at these island stations.Thus, the numerous synoptic variations observed at MNM and YON were likely influenced by the long range transport of air masses from the Asian continent.In contrast, the distinct diurnal cycle was observed at RYO with a maximum in the early morning and a minimum in the afternoon.This behavior is consistent with the diurnal cycle in the venting rate associated with the planetary boundary layer (PBL) height over inland regions (Moses et al., 1960;Servant, 1966;Chambers et al., 2009).These results support a strong local influence on the 222 Rn measurements at RYO from the nighttime inversion.

Enhanced 222 Rn events
The observed concentrations of 222 Rn at three stations reveal numerous synoptic variations consisting of episodically enhanced 222 Rn peaks on timescales less than one week (Fig. 3).These peaks are consistent with the transport of continental air masses since no local sources were found that would produce similar peaks, although RYO measurements are influenced by local sources.Based on the preliminary results obtained at MNM, these 222 Rn peaks showed a high correlation with synoptic scale variation of CO observed simultaneously at this station (Wada et al., 2010).
Figure 4 shows the detailed variations of 222 Rn observed at MNM, YON, and RYO during January 2010, compared with those of observed concentrations of atmospheric CO.During the month, 10, 9, and 4 enhanced 222 Rn events were identified at MNM, YON, and RYO, respectively.Corresponding increase in CO during each enhanced 222 Rn event is also shown in the figure.This 222 Rn-CO relationship was also observed at all three stations for other months during our study period.Each event corresponded to a cold frontal passage over the station, in a similar process described by Sawa et We identified the enhanced 222 Rn peaks as deviations from a fitted curve obtained by applying the 26th Butterworth filter with a 30-day low pass filter cut-off, a procedure described in Wada et al. (2011).The differences between the hourly observed data and the smoothed seasonal curve were calculated as residual 222 Rn ( Rn), and are plotted in Fig. 4. Positive Rn peaks that were identified as enhanced 222 Rn (ERN) events were chosen based on the following criteria: (1) Rn ≥ 2σ from the fitted curve, where σ is the standard deviation calculated from the negative Rn in each month, and (2) Rn lasted for more than four hours.Following these

Correlation of observed CO with 222 Rn
In order to determine the enhancement ratios of CO and 222 Rn, slopes of Rn with CO were calculated for each ERN event.CO was determined by the same procedure that was used to determine Rn. Figure 5 shows some examples of the enhanced CO correlations and their slopes for the ERN events observed at MNM, YON, and RYO in January 2010.Since high correlations between CO and 222 Rn were found for the ERN events, their slopes were obtained by using the reduced major axis (RMA) regression (Hirsch and Gilroy, 1984).The slope of the enhanced values is denoted as CO/ Rn.Note that the units of CO/ Rn are in ppb (Bq m −3 ) −1 .CO/ Rn obtained at MNM varied with each ERN event.These differences were found to be dependent on the transit time of the air masses from the Asian continent to MNM and the 222 Rn decay during the transport (Wada, 2011).The variation in CO/ Rn from ERN event to ERN event at MNM was about 10.8 % after correcting for the 222 Rn decay during transport.These results indicate that the air masses reaching MNM from the continental Asia are well mixed by the time they reach the station.On the other hand, CO/ Rn at YON varied noticeably even though the 222 Rn decay correction was negligible due to short transit times of the air mass transport (Wada et al., 2011).The variation, therefore, is likely attributable to the variation in the sources themselves.In comparison, the ERN events at RYO were not that many and the CO-Rn correlation was significantly low, due likely to the influence of local sources and the complex terrain around the station.

Model simulation of the atmospheric 222 Rn and CO
222 Rn concentration was simulated by using the global chemical transport model STAG. Figure 6 shows a comparison of simulated 222 Rn concentration in the surface layer with the observed data of 6 hourly means at MNM, YON, and RYO for 2010.The model was generally successful in reproducing the seasonal variations observed at all three stations.The simulated summer 222 Rn at MNM showed low values of less than 0.2 Bq m −3 , consistent with the observed concentrations at the station.At YON, the observed 222 Rn values in the winter and spring seasons were well simulated by the model, althogh the simulated 222 Rn was little lower than the observation in the summer due likely to subgrid scale emissions of 222 Rn along the east coast of China.The simulated seasonal cycle at both YON and MNM agreed well with the observation.In this study, the model calculations were carried out by using a time invariant and spatially uniform 222 Rn flux density, although it has been reported that the 222 Rn fluxes over the continent do vary with time and space mainly due to changes in the soil water content (Dörr and Munnich, 1990;Genthon and Armengaud, 1995).The consistency between the model and the observation in this study indicated that the 222 Rn flux density employed in the model was adequate for reproducing the observed 222 Rn concentration at these stations.However, in order to estimate the impact of uncertainty in the 222 Rn flux density on our CO emission estimates, we conducted an additional simulation using the simulated results of the monthly mean flux density of 222 Rn (60 Hirao et al. (2010).As we will show in Sect.3.7, the impact was minimal.
The model was also able to identify all the major enhanced 222 Rn events in the synoptic scale variation observed at the stations, particularly in terms of the timing of the occurrences of these ERN events.However, the model slightly  The simulation of CO was carried out by using the same model used for the simulation of 222 Rn. Figure 7 shows a comparison of simulated CO concentration in the surface air with the observed data of 6 hourly means at MNM, YON, and RYO for 2010.The model was successful in reproducing the observed seasonal cycle, the synoptic variations, and the enhanced CO peaks at these stations.Although the model underestimated the observed 222 Rn peaks, its underestimation of the observed CO peaks proved to be much more significant.

Relationship between observed data and simulated results for CO and 222 Rn
Figure 8 shows the relationships between the observed 6 hourly mean concentration values and their simulated results for 222 Rn and CO for 2010 at the three stations.The correlation plots for 222 Rn show that the overall ratios of model/observation are about 0.72-1.00,with a 95 % confi-dence interval of 0.02 calculated by the bootstrap method (e.g.Xiao et al., 2004), resulting mainly from the model underestimation of large 222 Rn peaks within subgrid scale variations.The data at MNM showed a compact relation with a high correlation coefficient of 0.84, while lower correlation coefficients of 0.64 and 0.47 were found at YON and RYO.Most of the major ERN events observed at all the stations throughout the year were captured by this model simulation.However, the correlation plots for CO at MNM and YON indicate that the model underestimated the observed CO, more significantly than for 222 Rn, due to the underestimated CO emissions in the EDGAR database that was used to drive the model CO field.This problem in the EDGAR database was also pointed out by several previous modeling studies (e.g.Kiley et al., 2003;Arellano, 2004;Ma and van Aardenne, 2004).Previous studies on the combustion ratio of CO/CO 2 observed around the western North Pacific indicated that YON is influenced mostly by the emissions from China where anthropogenic emissions have been significantly underestimated (e.g.Palmer et al., 2003;Heald et al., 2004).On the other hand, RYO is influenced mostly by the regional sources in Japan, while MNM is influenced by the sources distributed over a much greater areas in East Asia (Wada et al., 2011).Therefore, the emission of CO from the regions detected at the stations MNM and YON are significantly underestimated when compared with that of RYO.Among all the stations, MNM is located farthest from the continental East Asia.It therefore "sees" CO emissions from a wider geographical region in East Asia than the other stations.Combined with the fact that CO and 222 Rn are highly correlated at MNM, we used the observed and simulated data at MNM to obtain a more realistic estimate of the CO emission from East Asia than before.

Estimation procedure based on the radon tracer method
The radon tracer method has been mainly applied to inland sites in European.We applied the method to our sites in the western North Pacific region.The basic concept of our method is that the enhancement ratio of CO/ 222 Rn associated with each ERN event shows agreement between the model and the observation when the emission database used in the model is realistic.Error in either the CO or 222 Rn emission in the model can produce disagreement with the observed CO/ 222 Rn ratio.In our study, the model simulated the observed 222 Rn well.Therefore, any difference between the modeled and the observed CO/ 222 Rn ratio can be traced to errors in the estimated CO emission in the model, which can be adjusted to bring the model simulation results closer to the observed.This way, we were able to obtain a more realistic estimate of the CO emission in East Asia by using the CO/ 222 Rn ratio.It should be noted that the uncertainty in the model transport was viewed as not being a significant issue because the same transport model was used to simulate both CO and 222 Rn.We conducted a tagged tracer experiment to adjust the simulated CO so that the simulated CO/ 222 Rn ratio would match the observed ratio.The adjustment is expressed as a cost function by which the CO emission is optimized (e.g.Suntharalingam et al., 2004).Figure 9 shows the regions used for the tagged tracer experiment.The global emission area is divided into 14 regions.China was divided into 3 regions of North China (NC), South China (SC), and West China (WC), with their boundaries defined by 100 • E longitude and 35 • N latitude.Japan and Korea were combined into one region (JpK).
The adjusted simulated CO was obtained by the following equation; where X CO represents the adjusted simulated CO at MNM, n is the number of regions (equal to 14 in our study), C i is the simulated CO at MNM with influence from region i obtained by the tagged tracer experiment, and k i is the corrective coefficient for region i.These corrective coefficients were determined by minimizing the cost function which was defined in this study as a residual sum of squares between the observed and the simulated ratios associated with each ERN event.The cost function is described by the following equation; where S is the residual sum of squares and m is the number of ERN events.Through changing the corrective coefficients of k i in Eq. ( 1), optimized pairs of k i are obtained by minimizing S in Eq. (2).

Estimates of CO emission in East Asia
Figure 10 shows the results of the tagged tracer experiment for CO.The shaded areas represent simulated ERN events for the peaks 1, 4, 7, 8, 9, and 10 of the observed ERN events shown in Fig. 4. Other peaks were not well reproduced, probably because these events were relatively small.The figure shows that the synoptic scale variation of atmospheric CO at MNM was influenced by the three regions of NC, SC, and JpK, while other regions were negligible.The CO emission at MNM from biomass burning in Southeast Asia was also deemed negligible based on the results of an additional CO simulation by using biomass burning emission of GFED ver.
3.1 from 2005 to 2010 (van der Werf et al., 2010) as a driver for the CO field in STAG.Consequently, the corrective coefficients in Eq. ( 1) for the regions NC, SC and JpK were adjusted to optimize the CO emissions in the model, whereas the corrective coefficients for other 11 regions were fixed to be 1.To perform calculations using Eq. ( 2), the ERN events observed from November to April were selected in order to ensure the influence of cold-front transport mechanism for all the events.In addition, the ERN events were characterized by high correlation between enhanced CO and 222 Rn for both the observed and simulated results, ensuring pollution events with well mixed air masses.Twenty ERN events with correlation coefficients of more than 0.7 were found, constituting about 11.2 % of the identified ERN events during November to April.With these 20 ERN events, the corrective coefficients for the regions of NC, SC, and JpK were determined by changing each corrective coefficient from 0.5 to 3.0.From the results of the optimization, the corrective coefficients for NC, SC, and JpK were determined to be 2.8, 1.4, and 1.1, respectively.Figure 11  emission estimate of 203 Tg CO yr −1 for China, Japan, and Korea, with 185 Tg CO yr −1 attributed to China.The original emission given in the EDGAR ver.4.1 database for China was 102 Tg CO yr −1 , which is about 45 % smaller than our emission estimate.Although the criterion for selecting an ERN event was based arbitrarily on the correlation coefficient being greater than 0.7, changing the criterion from 0.6 to 0.8 resulted in one standard deviation variability of only 8.3 % in emission estimates.
As mentioned earlier, in order to estimate the impact of uncertainty in the 222 Rn flux density on our CO emission estimation, we conducted a simulation using the 222 Rn flux density obtained by Hirao et al. (2010) to drive our transport model, instead of the constant 222 Rn flux of 1.0 atom cm −2 s −1 .The difference in the estimated results for the CO emission in China between using the constant 222 Rn flux and the heterogeneous flux of Hirao et al. (2010) came out to be only 3.2 %.
Considering the variation in the observed enhancement ratio, in the distribution of the 222 Rn flux density, and in the selection process of an ERN event, we estimated an overall error of our method to be about 14 %.

Comparison of CO emission estimates
Table 1 shows a comparison of the CO emission for East Asia obtained in this study with the emission estimates from several inversion and bottom-up methods.Although some estimated values include biomass burning emission, they all can be compared with each other because the biomass burning is not a significant source in China.The contribution of the biomass burning emission in China was reported as 6.7 % (Palmer et al., 2003) and 7.9 % (Heald et al., 2004) of the total CO emission.
Our study shows that 91 % of the total emission for East Asia comes from China.The remainder comes from Japan and Korea.This is consistent with the results of Palmer et al. (2003), Kopacz et al. (2009), and three bottom-up emissions.There are however, large variations in the emission estimates within each method, as well as between methods.Among the bottom-up methods, the EDGAR database shows the greatest underestimation.The variation among the inverse methods could be attributable to an increase in the emission during 2001 to 2005.Tanimoto et al. (2008) reported that an increase in the CO emission in China from 2001 to 2005 was estimated to be about 16 %.We used this rate of increase to adjust all the emission estimates listed in Table 1 to a reference year of 2005. Figure 12 shows the estimated emissions of CO in China listed in Table 1 (solid bars) and their adjusted values to the reference year 2005 (open bars).From the figure, it can be seen that the EDGAR database shows a significant underestimation by about 45 %.All other estimates, including the one from this study, agree reasonably well with each other.Ohara et al. (2007) reported that the REAS database was consistent with the inverse estimates and those from the forward modeling, whereas Zhang et al. (2009) reported that the INTEX-B database was partially consistent with a forward modeling result (Chen et al., 2009) at a suburban area near Beijing but underestimated in the Beijing urban area.

Conclusions
Atmospheric concentrations of CO and 222 Rn were observed continuously at Minamitorishima (MNM), Yonagunijima (YON), and Ryori (RYO) from 2007 to 2011.These measurements showed a seasonal cycle with a winter maximum and a summer minimum in both CO and 222 Rn.In addition, high-frequency synoptic variations associated with enhanced 222 Rn (ERN) events due to long-range transport of continental air masses were well captured by the station measurements.
The enhancements of CO associated with increased 222 Rn were found in most of the ERN events at MNM, YON, and RYO.These simultaneous elevations indicate that the concentrations of 222 Rn and CO were well mixed in the pollutant air masses over the continent and were exported to www.atmos-chem-phys.net/12/12119/2012/ the western North Pacific region by the way of cold frontal transport.The correlation of 222 Rn and CO during each ERN event observed at MNM was very high, with very little variation in the CO/ 222 Rn ratio from event to event during the winter.The correlation at YON was also high during the winter and spring, but the ratio varied from event to event due to the station's vicinity to the source regions, whereas the ratio at RYO was significantly low due to influence from local sources located in a complex terrain.
The chemical transport model STAG reproduced well the observed 222 Rn when forced with a constant and uniform flux density of 1.0 atom cm −2 s −1 .The simulated seasonal cycles and synoptic scale variations were in good agreement with the observed.The model also reproduced well the timing of the synoptic scale variation of the observed CO, but underestimated the magnitude of the variation.The ratios of the simulation/observation of CO were 0.52 and 0.42 at MNM and YON, respectively, whereas those of 222 Rn were 0.72 and 0.81.The relatively low ratio values resulted from using in the model the EDGAR emission database that is known to have a significant underestimation of CO emission values over China.
A tagged tracer experiment of CO at MNM indicated that the synoptic variation of CO at MNM was influenced by the emissions in North China (NC), South China (SC), and Japan and Korea (JpK), with emissions from other regions found to be negligible in the winter and spring.
An estimation procedure based on the radon tracer method was developed and applied to the measurements from MNM.By optimizing the a priori emission of the EDGAR ver.4.1 database for the ERN events observed at MNM during the winter and spring, the total anthropogenic CO emission from China, Japan, and Korea was estimated to be 203 Tg CO yr −1 , of which 91 % originated in China.When compared with other estimated emissions of CO based on top-down and bottom-up methods, our estimated result was found to be in general agreement, except for the emission database of EDGAR ver.4.1 which came out to be about 45 % smaller than our emission estimate for China.These results strongly indicate that an independent method, such as the radon tracer method used in this study, to estimate emissions is important in East Asia where there is still large uncertainty associated with various estimates.

Fig. 5 .
Fig. 5. Typical scatter plots of CO as a function of Rn observed at MNM (a), YON (b), and RYO (c) in January 2010.The solid line represents the RMA regression for each event.The peak numbers correspond to those in Fig. 4. Note that the x and y axis are different in each figure.

Fig. 6 .
Fig. 6.Simulated and observed atmospheric concentration of 222 Rn at MNM (a), YON (b), and RYO (c) in 2010.The black dots represent 6 hourly mean observed 222 Rn and the red solid line represents simulated 222 Rn.

Fig. 7 .
Fig. 7. Simulated and observed atmospheric concentration of CO at MNM (a), YON (b), and RYO (c) in 2010.The black dots represent 6 hourly mean observed CO and the red solid line represents simulated CO.

Fig. 8 .
Fig. 8. Correlations between simulated results and observed data for 222 Rn and CO in 2010 for MNM, YON, and RYO.The solid lines represent the RMA regressions.The slopes are calculated by the RMA regression with 95 % confidence interval obtained by the bootstrap method.The r values represent correlation coefficients.

Fig. 10 .
Fig. 10.Simulated results of the tagged tracer experiment for atmospheric CO at MNM.The gray shaded areas represent the enhanced CO events for peak 1, 4, 7, 8, 9, and 10 of the observed data in Fig. 4.

Fig. 11 .Fig. 12 .
Fig. 11.An example of correlations between CO and 222 Rn for the enhanced 222 Rn event on 23 November in 2009 from the observation (a) and simulations (b) with a priori (open circles) and a posteriori (solid triangles) emissions, respectively.The solid lines represent the RMA regressions and the r values represent correlation coefficients.
Regions divided into 14 sectors for the tagged tracer model experiment.JpK; Japan and Korea, NC; North China, SC; South China, WC; West China, SEAs; Southeast Asia, In; India, OC; Oceania, ERu; East Russia, WRu; West Russia, ME; Middle East, EU; Europe, Af; Africa, nAm; North America, sAm; South America.

Table 1 .
Comparison of estimated CO emissions in East Asia.
* These values were summed in each region defined in this study.* * FF; fossil fuel BF; Biofuel combustion BB; biomass burning.