In-situ measurements of atmospheric hydrofluorocarbons (HFCs) and perfluorocarbons (PFCs) at the Shangdianzi regional background station, China

Atmospheric hydrofluorocarbons (HFCs) and per- fluorocarbons (PFCs) were measured in-situ at the Shang- dianzi (SDZ) Global Atmosphere Watch (GAW) regional background station, China, from May 2010 to May 2011. The time series for five HFCs and three PFCs showed occa- sionally high-concentration events while background condi- tions occurred for 36 % (HFC-32) to 83 % (PFC-218) of all measurements. The mean mixing ratios during background conditions were 24.5 ppt (parts per trillion, 10 12 , molar) for HFC-23, 5.86 ppt for HFC-32, 9.97 ppt for HFC-125, 66.0 ppt for HFC-134a, 9.77 ppt for HFC-152a, 79.1 ppt for CF4, 4.22 ppt for PFC-116, and 0.56 ppt for PFC-218. The background mixing ratios for the compounds at SDZ are consistent with those obtained at mid to high latitude sites in the Northern Hemisphere. North-easterly winds were as- sociated with negative contributions to atmospheric HFC and PFC loadings (mixing ratio anomalies weighted by time associated with winds in a given sector), whereas south- westerly advection (urban sector) showed positive loadings. Chinese emissions estimated by a tracer ratio method us- ing carbon monoxide as tracer were 3.6± 3.2 kt yr 1 for HFC-23, 4.3± 3.6 kt yr 1 for HFC-32, 2.7± 2.3 kt yr 1 for HFC-125, 6.0± 5.6 kt yr 1 for HFC-134a, 2.0± 1.8 kt yr 1 for HFC-152a, 2.4± 2.1 kt yr 1 for CF4, 0.27± 0.26 kt yr 1 for PFC-116, and 0.061± 0.095 kt yr 1 for PFC-218. The lower HFC-23 emissions compared to earlier studies may be a result of the HFC-23 abatement measures taken as part of Clean Development Mechanism (CDM) projects that started in 2005.


Introduction
Hydrofluorocarbons (HFCs) and perfluorocarbons (PFCs) are greenhouse gases with global-warming potential (GWP) values for time horizon of 100 yr ranging from 140 to 11 700 (IPCC, 2007) and they are included in the Kyoto Protocol. Most HFCs are used in refrigeration, air conditioning, in aerosol spray cans, and as foam blowing compounds (Barletta et al., 2011;McCulloch, 1999) while HFC-23 (CHF 3 ) is mostly a by-product of the HCFC-22 (CHClF 2 ) production (McCulloch and Lindley, 2007;Miller et al., 2010;Oram et al., 1998). PFCs are emitted from aluminum smelters (CF 4 ,, and are used and released in the semiconductor industry . HFCs are replacing chlorinated ozone-depleting gases such as chlorofluorocarbons (CFCs) and hydrofluorochlorocarbons (HCFCs). HFC and PFC emissions are expected to have been increased in China within the last years in accordance with the high industrialization and from the substitutions of CFCs and HCFCs. Some Chinese emissions have been previously estimated based on measurements at stations downwind of China (Li et al., 2011;Kim et al., 2010; HFC-152a (CH 3 CHF 2 ), HFC-125 (CHF 2 CF 3 ), CF 4 , PFC-116 (C 2 F 6 ), and PFC-218 (C 3 F 8 ) at the World Meteorological Organization/Global Atmosphere Watch (WMO/GAW)'s regional background station Shangdianzi (SDZ) in northern China from May 2010 to May 2011. Mixing ratios are presented and characterized for both "background" and "polluted" conditions. Additionally, the impact of local surface horizontal advection on the observed compounds has been investigated. Chinese HFC and PFC emissions are estimated by a tracer ratio method using carbon monoxide as a reference tracer.

Description of the site
The Shangdianzi station (40 • 39 ′ N, 117 • 07 ′ E, 291.3 m asl) is located in a mountainous area approximately 100 km northeast of urban Beijing in the North China Plain. SDZ is one of the atmospheric background stations run by China Meteorological Administration (CMA) and also an affiliated station of Advanced Global Atmospheric Gases Experiment (AGAGE) network. Previous studies showed that this site experienced both pollution events influenced by air masses from Beijing and other industrialized areas in the North China Plains and clean air masses from remote areas, such as Siberia, Mongolia, and the Chinese province Inner Mongolia (Lin et al., 2008;Vollmer et al., 2009;Zhang et al., 2010). The wind rose of the observation period is shown in Fig. 1. Influenced by monsoon and local valley topography, the prevailing wind sectors for the whole year are NE/ENE/E (background sectors) and SW/WSW/W (urban sectors) during the observation period, with total frequencies of approximately 40 % and 32 % respectively.

Experimental methods
Air samples are analyzed in-situ using the "Medusa" automated custom-built gas chromatographic system with mass spectrometric detection (Miller et al., 2008), with relative measurement precisions typically < 5 % for HFC-32 and PFCs and < 2 % for other HFCs. The system was installed in an air-conditioned container in May 2010. The sample inlet was installed at 8 m on a 10 m tower located at a distance of 30 m from the container, resulting in a total inlet line < 80 m. The air is continuously drawn through a 10 mm OD Synflex-1300 tubing by means of a membrane pump, resulting in a response time < 5 min. The Medusa was added to the gaschromatographic measurement system that had been in operation since 2006 Zhang et al., 2010) to add HFC and PFC measurements to the sampling program. Carbon monoxide (CO) has been measured in-situ since 2007 by a non-dispersive infrared (NDIR) method (Horiba AMPA-360CE) with a resolution of 1 min and with a measurement precision < 10 ppb (parts per billion, 10 −9 , molar). measurements are linked to the WMO-2000 calibration scale and are reported accordingly. All the instruments mentioned above share the same inlet. Horizontal surface wind data with a resolution of 1 h are obtained from the top of the same tower.
Each Medusa air sample measurement (every second hour) is bracketed by a reference gas (termed working standard or quaternary standard) measurement to detect and correct for drift in the detector sensitivity. Our measurements are linked to AGAGE (Prinn et al., 2000;Mühle et al., 2010), and are reported as dry air mole fractions on the calibration scales developed at the Scripps Institution of Oceanography (SIO), and the University of Bristol (UB): SIO-2005 (CF 4 , HFC-134a, HFC-152a), , UB-98 (HFC-125).

Baseline estimation
For the analysis of pollution events, baseline conditions were distinguished from pollution events by applying the "robust extraction of baseline signal" filter (REBS; Ruckstuhl et al., 2012). The filter iteratively fits a local regression curve to the data using robustness weights for values above the baseline and iteratively excluding data points outside a 3.5 sigma range around the current baseline. A window width of 60 days was chosen for the REBS. This allows the baseline fit to follow relatively quick changes in baseline levels due to the change in advection from mostly northerly directions during the colder season to mostly southerly advection during the warm season. The fit converged after 5 iterations. The filter was applied separately to all halocarbon and to the CO time series.

Calculation of source receptor relationships
To identify the area from which SDZ received significant emission contribution to the observed mole fractions, backward trajectory calculations were performed using the Lagrangian particle dispersion model FLEXPART (V9.01; Stohl et al., 2005). For each 3-hourly time interval 50 000 model particles were released at the location of SDZ and traced backwards in time for 10 days. FLEXPART simulates advection by the main wind, turbulent and convective dispersion of the model particles. The model was driven by the operational analysis of the European Centre for Medium range Weather Forecast (ECMWF) with a global resolution of 1 • by 1 • and higher resolution nest of 0.2 • by 0.2 • over northeastern China.
The derived source receptor relationships (SRR) (Seibert and Frank, 2004) present a measure how sensitive the observations at SDZ are to emissions on the grid. The SRR (in s kg −1 m 3 ) for a certain time can directly be multiplied with the grid cell emissions (in kg s −1 ) and divided by the grid cell volume (in m 3 ) at the source to derive mass mixing ratios at the receptor. Large SRR values indicate strong sensitivities to emissions. In Fig. 2 the total SRR is shown, calculated as the sum of all SRRs during the entire year for which CO observations at SDZ were classified as pollution events.

Emission estimation using a tracer ratio method
The tracer ratio method (TRM) has been used for the emission estimate of halocarbons in a number of previous studies (Barletta et al., 2011;Kim et al., 2010;Li et al., 2011;Palmer et al., 2003;Reimann et al., 2004;Saito et al., 2010;Shao et al., 2011;Yokouchi et al., 2006). The method is based on the assumption that the target halocarbon and the chosen tracer behave similar in the atmosphere and thus their emission ratio is conserved during the transport from the source to the receptor. The increase of the tracers above their baseline concentrations as observed at any receptor can then be used to estimate the tracer ratio and consequently the emissions of a target tracer if the emissions of the reference tracer are known with a sufficient degree of certainty. The method requires that there are no significant atmospheric loss or production processes of the tracers during the period of transport. Furthermore, the method is supposed to work best if emission sources of different tracers are collocated and temporal emission profiles are similar. The distance between emission sources and receptor will also influence the applicability of the method in that it determines the degree of atmospheric mixing (spatial integration) from different emission sources. Receptors that are located relatively close to the sources may observe relatively well defined ratios from more specific sources (source areas) because atmospheric turbulence did not yet disperse and mix individual signals. Receptors at larger distances may observe larger spread in ratios, since signals from different sources were mixed within the atmosphere prior to arrival at the site. Hence such sites may be more suitable to up-scale emission estimates to larger areas and the collocation requirement may be relaxed. For remote receptors the definition of above baseline signals becomes more and more challenging because the atmospheric mixing (integration) dilutes the original emission ratio and observed pollution peaks may not significantly differ from the baseline. These requirements and limitations should be kept in mind when interpreting the emission estimates as derived with the TRM.
In this study, we chose CO as the reference tracer because most of the observed HFCs and PFCs were statistically significantly (p < 0.01) correlated with in-situ CO measurements, showing Pearson correlation coefficients larger than 0.45 (see Fig. 3). The correlation was not significant for HFC-134a (r = 0.26). The use of CO as reference tracer is in contrast to recent studies by Kim et al. (2010), Li et al. (2011) and Saito et al. (2010) who used HCFC-22 as a reference tracer. However, in the case of SDZ the correlations of HCFC-22 with the observed HFCs and PFCs were less consistent as for CO (Pearson correlation coefficient larger 0.5 for HFC-134a and not significantly different from 0 for HFC-23, HFC-152a, PFC-116 and PFC-218). Using CO as a reference tracer has the advantage that CO emissions from China were extensively studied in the recent past (Ohara et al., 2007;Streets et al., 2003;Zhang et al., 2009) and are relatively well known. One disadvantage of using CO as a reference tracer may be that its emissions may be less well collocated with halocarbon emissions, owing to the fact that CO emissions do not only result from the same processes as halocarbon emissions but may, for example, be caused by residential use of coal and biofuels in more rural areas of China. The fraction of residential CO emissions was estimated by  Zhang et al. (2009) and the EDGAR global emission inventory (http://edgar.jrc.ec.europa.eu/) to be about a third of the Chinese total emissions. This fraction is considerably larger than that for the USA (6 %, EDGAR V4.2) but similar to that of European countries (France 21 %, Germany 29 %, Italy 22 %) for which CO was successfully used as a reference tracer for emission estimation of halogenated greenhouse gases (Reimann et al., 2004). In contrast to European CO emissions, Chinese CO emissions undergo a rather pronounced annual cycle as shown by Streets et al. (2003). This may lead to a larger spread in the observed tracer ratio and an increased uncertainty in halocarbon emission estimates. A further limitation of using CO as a reference tracer is the fact that CO is both consumed (oxidation by hydroxyl radicals) and produced (from oxidation of methane and other volatile organic compounds) in the atmosphere. However, consider-ing the relatively short transport times (days) from the major emissions sources in North-East China towards SDZ and CO lifetimes in the order of weeks this presents a minor source of error.
Finally, the emissions of the target halocarbon E x [kt yr −1 ] can be calculated from its mixing ratio enhancement over baseline conditions ( C x ) and the mixing ratio enhancement of CO ( CO) by: Here E CO is the total emission of CO, M CO is its molecular weight, and M x is the molecular weight of the target halocarbon. The uncertainty of the emission estimate is given by error propagation.
Total Chinese CO emissions during the observation period were estimated to 193.3 Mt yr −1 using the estimate by Zhang et al. (2009) of 166.9 Mt yr −1 for 2006 and a growth rate of 3.4 % yr −1 , assuming that the growth rate for 2006 to 2011 remains the same as the one reported by Zhang et al. (2009) for 2001 to 2006. The uncertainty of the total Chinese CO emissions was estimated by Zhang et al. (2009) to ± 35 % (1σ ). Assuming an uncertainty of the growth rate of ± 100 % we estimate the uncertainty for the period of investigation to be ± 37.8 %.
The observations at SDZ may not be representative for the whole of China. The area influencing SDZ during pollution events covered most of the densely-populated North China Plains, and also extends towards southern parts of China such as Yangtz river region. Sensitivities towards Korea were relatively small and were neglected during further analysis. The site would have been sensitive to emissions from Mongolia. However, these were considered to be relatively small and were not further considered. The total CO emissions from the provinces to which SDZ is most sensitive (green areas: Anhui, Beijing, Hebei, Henan, Jiangsu, Jilin, Liaoning, Inner Mongolia, Shandong, Shanxi, Tianjin) were 88 Mt yr −1 in 2006 ) and assuming the same country-wide growth rate as above 102 Mt yr −1 in 2010/2011. In a first step, we estimate the emissions for the mentioned Provinces (called North China Plains in the following) using the stated uncertainty of ± 37.8 %, while in a second step total Chinese emissions were extrapolated using an additional uncertainty of 10 % to account for representativeness uncertainties.
The tracer ratio b has been calculated with different methods in the past. While some authors estimate it as the ratio between the average enhancement over the baseline (e.g., Reimann et al., 2004), others have calculated the median of the ratio from different pollution events (e.g., Li et al., 2011), and yet others have used regression analysis, fitting a straight line to the data (Shao et al., 2011). Here the tracer ratio b and its uncertainty was estimated using the weighted total least square (WTLS) regression of a straight line with uncertainties in both coordinates (Krystek and Anton, 2007). This procedure assures that the calculated slope is invariant to the selection of the parameter on the x-and y-axis, which is not guaranteed when using the ordinary least square regression. Furthermore, in contrast to the orthogonal distance regression used by Shao et al. (2011), the WTLS regression allows for proper weighting of uncertainties in both variables, which is important in the case of different average uncertainties between parameters, and also for individual data points with increased uncertainty.
Prior to the regression analysis "above baseline" mixing ratios were aggregated to 5-day averages to remove autocorrelation in the data. Aggregates were only kept for further analysis if at least 3 valid "above baseline" values were present within the aggregation interval. Without this data aggregation the calculated uncertainties of b would be underestimated, because the regression assumes independent uncertainties of the input data, which is not the case for autocorrelated time series. The uncertainties of the 5-day aggregates were governed by non-random uncertainties (accuracy) due to calibration scale and propagation uncertainty, while the aggregation diminishes the random part (precision) of the uncertainty. Combined uncertainties of the aggregates were 3 % for CO and 5 % for the HFCs and PFCs with the exception of HFC-23 were, due to smaller precision, the uncertainties were 10 %.
As an alternative estimate of the tracer ratio b (b r in the following), we calculated the median of the ratio of all 5-day aggregates (called direct ratio calculation in the following), while the median absolute deviation was used as an estimator of the uncertainty of b r .

Time series and background data selection
One-year time series for HFC and PFC mixing ratios from 6 May 2010 to 5 May 2011 are shown in Fig. 4 (data gaps were due to instrument malfunction). For most HFCs and CF 4 , mixing ratios at SDZ show large episodic events, with elevated concentrations from polluted air masses most likely arriving from urban areas. Furthermore, 1 month of measurements is shown for HFC-125 and CO to show the concurrent behavior of the two substances.
Using the pollution filter, 36 % (HFC-32) to 83 % (PFC-218) of all valid measurements have been selected as background data (Table 1). Pollution-classified mixing ratios (grey dots) exhibit large fluctuations compared to the background measurements. The mean mixing ratio difference between pollution and background can be attributed to recent emissions. Of all compounds measured, HFC-134a has the greatest mixing ratio difference between pollution and background. Compared to HFCs, pollution events for the PFCs are less frequent and smaller in size, with the exception of CF 4 .

Comparison of background mixing ratio and trends
Since the HFC and PFC working standards at SDZ are traceable to the standards scale of the AGAGE network, the data are directly comparable to results from the AGAGE stations.
Here we compare our results to the following four AGAGE stations: Jungfraujoch, Mace Head, Trinidad Head, and Cape Grim. The geographic information of these 4 stations is listed in Table 2. Cape Grim, located in the Southern Hemisphere  Table 2). The differences for the background HFCs mixing ratios between SDZ and Trinidad Head, which are almost located at the same latitude, are within 2.9 % for all compounds listed in Table 2.
There are differences for the background mixing ratios of HFCs and PFCs up to several ppt (parts per trillion, 10 −12 , molar) between SDZ and Cape Grim, especially for HFC-134a and HFC-152a. These larger differences are expected as they are typically observed for all long-lived gases emitted in substantial quantities from human activities. Of all the compounds listed in Table 2, HFC-152a shows the biggest relative difference of all stations, due to its shortest lifetime of all measured compound.
There is no recent report on HFC-32 NH background levels. Global background mixing ratios are reported as 2.7 ppt for 2008 (WMO Ozone Assessment report 2011: see Montzka et al., 2011). Assuming that the growth rate remained the same as of 2008 (0.6 ppt yr −1 ), the global background mixing ratio from May 2010 to May 2011 is calculated as 4.2 ppt which is smaller than in our study at SDZ. The difference (1.7 ppt) might due to the difference between NH and SH, or bigger growth rate of HFC-32 since 2008.
The annual mean growth rates of HFC-32, HFC-134a, and HFC-125 are calculated by linear curve fitting for the compounds with R 2 > 0.7. Figure 5 presents HFC-32 and HFC-125 as examples. For HFC-23, HFC-152a, and three PFCs, annual mean growths are obtained from monthly mixing ratio means difference between May 2010 (6 May to 31 May) and May 2011 (1 May to 31 May) because the linear cure fitting for them were with poor R.

Impact of local surface horizontal winds on the observed HFCs and PFCs
The combination of in-situ measurements with meteorological data can improve the understanding of the relationship between enhanced concentrations and emissions (Dlugokencky et al., 1993;Zhou et al., 2004). In this study, the impacts of local winds on atmospheric HFC and PFC mixing ratios were investigated based on one year of observations. Synchronous hourly surface wind data were calculated based on meteorological measurement from SDZ. The surface wind data was calculated based on the sampling time of the in-situ measurement and was then separated into 16 wind directions. Mixing ratio anomalies (differences between mixing ratios in each wind sector and their means) and loadings (mixing ratio anomalies weighted by time (8760 h) associated with winds in a given sector) were calculated based on the method described in the previous studies Zhou et al., , 2004. Both anomalies and loadings are performed on the full set of measurements. Figure 6 displays statistical anomalies and loadings of the HFCs and PFCs from the 16 wind sectors.
Atmos. Chem. Phys., 12, 10181-10193    For PFC-218, anomalies for all wind directions are within ± 0.01 ppt indicating that there are no distinct emission sources in the region influencing SDZ. For other compounds, the mixing ratios are most enhanced from WSW and SW, which are the directions towards downtown Beijing. These two sectors also contribute most to loadings. Winds originating from the northeast sector always exhibited negative mixing ratio anomalies as concentrations are lowest during times of air mass origins from N/NNE/NNW. However, the wind frequencies of these sectors are relatively small so that their loadings do not contribute much. The biggest negative loadings are in sectors NE/ENE which have almost same amounts as position contribution of SW/WSW. Of all HFCs and PFCs, the mixing ratio of HFC-134a is most enhanced (> 5 ppt) in SSW/SW/WSW and loading for SW/WSW is approximately 4000 ppt h. For other HFCs and CF 4 , the maximal anomalies and loading are around 0.5 ∼ 2. ppt and 700 ∼ 2900 ppt h, while for PFC-116, the maximal anomalies and loading are less than 0.1 ppt and 120 ppt h.

Emission estimates by the tracer ratio method
The results of the regression analysis of all halocarbons versus CO are shown in Fig. 7.
The estimated HFC-23 emissions (see Table 3) were 1.9 ± 1.7 kt yr −1 and 1.4 ± 1.1 kt yr −1 for the North China Plains and 3.6 ± 3.2 kt yr −1 and 2.6 ± 2.1 for the whole of China by regression analysis and by direct ratio calculation of TRM. Our emission estimate for HFC-23 is much smaller than that for 2004 by Yokouchi et al. (2006) (10 ± 4.6 kt yr −1 ). Considering that HFC-23 is mainly a byproduct in the HCFC-22 production, these reductions could be explained by the fact that some Chinese companies have executed the Clean Development Mechanism (CDM) since 2005, which may have resulted in reduced HFC-23 emissions. Based on the data reported by the United Nations Framework Convention on Climate Change (UNFCCC, http://cdm.unfccc.int), the achieved reductions in emission Table 3. Emission estimates for hydrofluorocarbons (HFCs) and perfluorocabons (PFCs) for the North China Plain and the whole of China. Results are given for the two methods to estimate the tracer ratio b: weighted total least square (WTLS) regression analysis (Regression) and median of direct ratio calculation (Ratio). All numbers are given in kt yr −1 . The uncertainty estimates represent the 95 % confidence limits. The emission estimates for HFC-134a may be questionable due to insignificant correlation between CO and HFC-134a so its estimates are given braces.  2008). Due to the release of HFC-23 at a few point sources the collocation requirement with CO emissions may be less well fulfilled for HFC-23 than for other HFCs and PFCs. However, the correlation coefficient between CO and HFC-23 was not smaller than for a number of other HFCs/PFCs. This might be explained by the fact that none of the HCFC-22 producing factories is close to SDZ and HFC-23 emissions from these point sources will be mixed with regional scale CO emissions on their way to SDZ. HCFC-22 factories are distributed throughout the whole of China. Although individual production numbers are not known to us, it is possible to deduce that about half of these factories are situated within the area from which SDZ receives emissions. This is similar to our estimate concerning total versus regional CO emissions. Therefore, no strong bias of our HFC-23 emissions estimate due to the geographical location of the factories is expected. HFC-32, HFC-125 and HFC-152a are mainly used as foam blowing agents, aerosol propellants, and in refriger-ation mixtures (Velders et al., 2009). Emissions of HFC-32, HFC-125, HFC-152a were 4.3 ± 3.6, 2.7 ± 2.3 and 2.0 ± 1.8 kt yr −1 , respectively, for the whole China by regression analysis of TRM and the differences between emissions estimated by two methods were within 32 % for the three HFCs. Our estimates agree with previous studies Stohl et al., 2010) considering the uncertainties (see Table 4).
The correlation coefficient of HFC-134a and CO was not significantly different from 0. The intercept estimated by the regression was significantly different from 0, suggesting that the assumption of collocated CO and HFC-134a emission does not hold well in this case and our HFC-134a emission estimates should carefully be interpreted. The estimated HFC-134a emissions (see Table 3) differ strongly between the two estimates of the tracer ratio 3.2 ± 2.9 kt yr −1 and 7.6 ± 5.9 kt yr −1 for the North China Plains and 6.0 ± 5.6 kt yr −1 and 14 ± 11 kt yr −1 for the whole of China. Compared to the other parameters and considering the weak correlation our uncertainty estimate of HFC-134a emissions seems to be too small. Of all HFCs, HFC-134a has the largest emission. Two thirds of the produced HFC-134a was used for mobile air conditioners to replace CFC-12 since 2001. With the rapid increase of the Chinese automobile industry and the phasing out of CFC-12, Chinese HFC-134a emission probably will be continuously increasing. Bottom-up emission estimate based on difference assumption varies from 13.6 kt yr −1 to 21.1 kt yr −1 for 2010 with a growth rate of 30 % yr −1 (Hu et al., 2009;. The correlation coefficients of three PFCs and CO were significantly different from 0 and the intercepts estimated by the regression were not significantly different from 0 of CF 4 and PFC-116. However, for PFC-218, the intercept estimated by the regression was not significantly different from 0 and the estimated emissions differ strongly between the two estimates of TRM (see Table 3). Emission estimates for CF 4 , PFC-116, and PFC-218 were 2.4 ± 2.1, 0.27 ± 0.26, and 0.061 ± 0.095 kt yr −1 for the whole of China by regression analysis. PFC-116 emission is approximately half of the emission estimated by Saito et al. (2010) and  for 2008 and close to EDGAR report for 2008 (0.263 kt yr −1 ). For CF 4 and PFC-218, there are small differences between our estimates and other studies considering the respective uncertainties.

Conclusions
Based on the 1-yr measurements at Shangdianzi, we observe similar baseline levels compared to those AGAGE stations at similar latitudes, and positive trends were found for HFC-23, HFC-32, HFC-125, HFC-134a, HFC-152a, CF 4 , PFC-116, and PFC-218. Compared to a previous study (Yokouchi et al., 2006), HFC-134a had replaced HFC-23 with largest emission of HFCs in China. This is likely to become more Table 4. Chinese emission estimates for hydrofluorocarbons (HFCs) and perfluorocarbons (PFCs). Uncertainty estimates for this study represent the 95 % confidence limits. Numbers from this study refer to the results obtained with the regression method. The emission estimates for HFC-134a may be questionable due to insignificant correlation between CO and HFC-134a so its estimates are given braces. pronounced with the execution of CMD projects as well as the rapid increase of China's automobile industry (using HFC-134a as air conditioning fluid). Our emission estimates using the CO ratio method are tagged with considerable uncertainties due to limited spatial and temporal coverage. A network with more stations, especially observations from south China, would improve spatial resolution and reduce the uncertainties in the emission estimates.