In situ measurement of atmospheric CO 2 at the four WMO/GAW stations in China

. Atmospheric carbon dioxide (CO 2 ) mole fractions were continuously measured from January 2009 to December 2011 at four atmospheric observatories in China using cavity ring-down spectroscopy instruments. The stations are Lin’an (LAN), Longfengshan (LFS), Shangdianzi (SDZ), and Waliguan (WLG), which are regional (LAN, LFS, SDZ) or global (WLG) measurement stations of the World Meteorological Organization’s Global Atmosphere Watch program (WMO/GAW). LAN is located near the megacity of Shanghai, in China’s economically most developed region. LFS is in a forest and rice production area, close to the city of Harbin in northeastern China. SDZ is located 150 km northeast of Beijing. WLG, hosting the longest record of measured CO 2 mole fractions in China, is a high-altitude site in northwestern China recording background CO 2 concentration. The CO 2 growth rates are 3.7 ± 1.2 ppm yr − 1 for LAN, 2.7 ± 0.8 ppm yr − 1 for LFS, 3.5 ± 1.6 ppm yr − 1 for SDZ, and 2.2 ± 0.8 ppm yr − 1 (1 σ ) for WLG during the period of form of atmospheric CO 2 underlying ﬂuxes


2542
S. X. Fang et al.: In situ measurement of atmospheric CO 2 at the four WMO/GAW stations in China over North America and western Europe, which allow us to constrain regional fluxes using inverse models (e.g., Broquet et al., 2013;Gourdji et al., 2012). But northeastern Asia, a region of fast economic growth with high emissions, is not adequately sampled, and global inversion results give divergent results for its CO 2 budget (e.g., Peylin et al., 2013;Sirignano et al., 2010).
With the rapid development of its economy, China has become the number one fossil fuel CO 2 emitter in 2006 and emitted 1.8 Pg C in 2011 (LeQuéré et al, 2013;. The emissions of China deduced from energy statistics are uncertain, and could be higher than currently reported by up to 0.38 g C yr −1 (Guan et al., 2012). Peters et al. (2011) also reported that more than half of the growth in global CO 2 emissions from 1990 to 2008 took place in China. China is a late starter of atmospheric greenhouse gas observations. In the past years, there have been lots of short-term CO 2 measurement campaigns and research programs. Most of these focused on emissions from agricultural sources or urban areas (Fu et al., 2009;Lei and Yang, 2010;Liu et al., 2009;Tang et al., 2006;Xing et al., 2005). Long-term atmospheric CO 2 observations in China have been relatively sparse. The China Meteorological Administration (CMA) has been responsible for background greenhouse gas measurements. The first station installed in China is Mt. Waliguan (WLG) in Qinghai province, at 3816 m a.s.l. Since May 1991, CMA has collected weekly air samples in glass flasks. These samples were shipped to the National Oceanic and Atmospheric Administration (NOAA) Earth System Research Laboratory (ESRL) in Boulder, Colorado, United States, and analyzed for a suite of greenhouse gases. In 1994, the station was established as a global World Meteorological Organization (WMO) Global Atmosphere Watch (GAW) measurement site, and was equipped to measure CO 2 mole fractions by means of a LI-COR nondispersive infrared (NDIR) analyzer. This instrument produced nearly 14 years of high-quality data, which is the longest continuous atmospheric CO 2 record in China (Zhou et al., 2003(Zhou et al., , 2005. In 2009, the aging instrument was replaced by a cavity ringdown spectroscopy (CRDS) analyzer (G1301, Picarro Inc.). Three greenhouse gases measurement stations were established as regional WMO/GAW sites at Lin'an (LAN) in Zhejiang province (near Shanghai), Longfengshan (LFS) in Heilongjiang province (near Harbin, in the northeast of China), and Shangdianzi (SDZ) near Beijing. The information of the station is accessible at http://gaw.empa.ch/gawsis/default. asp. Initially, those were automatic weather stations only, and in situ CO 2 /CH 4 measurement systems began in 2009 using CRDS instruments (G1301). In the previous study, we presented and analyzed the first 3 years CH 4 mole fractions observed at LAN, LFS, and WLG (Fang et al., 2013). Similarly, here we study the observed CO 2 mole fractions at the four stations.

Sampling sites
Locations of Lin'an (LAN), Longfengshan (LFS), Shangdianzi (SDZ), and Mt. Waliguan (WLG) are shown in Fig. 1. Information of the four stations is listed in Table 1. LAN is located at the center of the Yangtze Delta area, China, and is about 50 km from Hangzhou (the capital of Zhejiang province) and 150 km from Shanghai (the largest economic center in China). This station is approximately 6 km northeast of the town of Lin'an, which has a population of ∼ 100 000. North of the LAN station (1.4 km away) is a small factory where charcoal is manufactured from bamboo wood. The station is built on the top of a small hill and is surrounded by a few pieces of paddy rice field.
LFS is 175 km southeast of Harbin city. Wuchang, the nearest city, has ∼ 200 000 inhabitants and is 40 km to the northwest of the site. The station is located on the northwest edge of the Longfengshan water reservoir, which has an area of 20 km 2 . The reservoir's dam is located 100 m to the north of LFS. Right beyond the dam to the north, there is a small area of paddy rice field and several small villages (∼ 100 inhabitants). The station is located in a forest park with mainly pine trees.
SDZ is located at 150 km northeast of Beijing. The observatory is built on a mountainside (with the highest peak to the north). There is a small village (∼ 300 inhabitants) at about 0.8 km to the south of the station. A railway used by dieseldriven trains runs from southwest to northeast, ∼ 0.6 km away from the station. The local vegetation is mainly shrubs and corn.
WLG is situated in western China at 3816 m a.s.l., and is far from industrial and populated centers. Measurements from WLG provide essential information on sources and sinks over the Eurasian continent (Zhou et al., 2004(Zhou et al., , 2005. At SDZ, LAN, and LFS, the sampling inlets were initially fixed on the top of wind poles (10 m a.g.l.). Near each sampling inlet, a wind direction and speed sensor were also installed. The distances from the wind pole to the nearby observatory buildings are 60, 25, and 65 m for SDZ, LAN, and LFS, respectively, to ensure that the air sample is minimally affected by human activities in the buildings. In 2010, new sampling towers of 50 and 80 m height were erected at LAN and LFS, respectively. In 2011, an 80 m sampling tower was built at SDZ. A second sampling inlet was installed at the top level of each tower, in addition to the previous 10 m high inlet. The air intake into each Picarro analyzer then switched between the 10 m and the top-level intake every 5 min at the three stations. Because of the short records available from the highest level at the three regional stations, we mainly discuss measurements from the 10 m intake. At WLG station, the air sample inlet is fixed at an 80 m height of an 89 m sampling tower located 15 m from the laboratory.

Measurement system
CRDS systems (G1301, Picarro Inc.) are used for continuous measurements of atmospheric CO 2 and CH 4 . This type of instrument has been proven suitable for making precise measurement of CO 2 and CH 4 mole fraction since its response is both highly linear and very stable (Chen et al., 2010;Crosson, 2008). Air sample is delivered to the instrument by a vacuum pump (N022, KNF Neuberger, Freiburg-Munzingen, Germany) via a dedicated 10 mm o.d. sampling line (Synflex 1300 tubing, Eaton, OH, USA). Then the ambient air is filtered and pressurized at 1 atm. The ambient air is dried to a dew point of approximately −60 • C by passing it through a glass trap submerged in a −70 • C methanol bath (MC480D1, SP Industries, PA, USA). An automated sampling module equipped with a VICI 8 port multiposition valve is designed to sample from separate gas streams (standard gas cylinders and ambient air). The detailed schematic of the measurement setup is described in Fang et al. (2013). The residence time of air sample from the top of the inlets to inlet of Picarro is less than 30 s. The individual systems started service on 1 January 2009 at LAN, LFS, SDZ, and WLG and are still running. In this study, the analysis is restricted to the period of January 2009 to December 2011 period.

Calibration, quality control, and data processing
Carbon dioxide mole fractions are referenced to a working high standard (WH) and a working low standard (WL). Additionally, a calibrated cylinder filled with compressed ambient air is used as a target gas (T) to check the precision and stability of the system routinely. All standard gases are pressurized in 29.5 L treated aluminum alloy cylinders (Scott-Marrin Inc.) fitted with high-purity, two-stage gas regulators. The standard gases are calibrated with cylinders assigned by the WMO/GAW CO 2 Central Calibration Laboratory operated by NOAA/ESRL. All data are reported on the WMO X2007 scale (Zhao and Tans, 2006;Zhao et al., 1997). Calibration of the working standards and target gases is done at CMA's central laboratory in Beijing before and after use at the four stations. The two standards and target gas are analyzed by the system for 5 min every 12 h.
The discrete flask samples collected weekly at WLG have been measured by the Carbon Cycle Greenhouse Gases group (CCGG) of NOAA/ESRL in Boulder, CO, USA. Two flask samples were collected in series at 08:00 LT (local time) using glass flasks and a portable battery-powered sampling apparatus with a 5 m (a.g.l.) intake height (Dlugokencky et al., 1994). The samples were measured at NOAA using a non-dispersive infrared (NDIR) analyzer with a repeatability of approximately 0.1 ppm (Zhou et al., 2005). The average time between sampling and analysis is about 2 months. All CO 2 measurements are also tied to the WMO CO 2 scale X2007. The co-located NOAA flask program at WLG ensures that the long-term CMA in situ measurements can be routinely compared with an independent record.
Because of the dead volumes in the in situ sampling system, the response of the analyzer is usually not stable until 1 min after switching to a different gas source using the multiposition valve. Every gas cylinder (WH, WL, or T) was measured for 5 min and the data processing routine used the last 3 min of each 5 min segment to compute CO 2 mole fractions. The same is done for the ambient air data after switching in 5 min intervals from one sampling height to the other and after the reference gas measurements. The average of the last 3 min is used to represent the CO 2 concentration in current 5 min segment. The ambient measurements were calibrated using a linear two-point fit through the most recent standard gas measurements (WH and WL). Ambient CO 2 measurements were retained only during periods when the measurement of the most recent target gas (T) using this same calibration procedure was within ±0.1 ppm of its assigned value. More than 97 % of the total 5 min average data points were retained for meeting this criterion at the four stations and treated as valid data.
After computing the CO 2 mole fractions, the valid data were manually inspected. Occasionally, analytical or sampling problems -including local influences such as nearby fires, vehicles, and cattle identified according to the station logbook entries made by the local operators -affected the observations. Those events were eliminated prior to the next processing steps. More than 97 % of valid 5 min data remained for each station after this filtering step. The 5 min data selected in that manner were then aggregated to hourly averages. To evaluate the seasonal cycle and peak-to-peak amplitude, we used the curve-fitting method described by Thoning et al. (1989). The data were first fitted to a function with three polynomial terms and four harmonic terms. Residuals from the function were then filtered (smoothed) in the time domain using two different low-pass filters (a Butterworth filter, which computes weighted averages over 1.07 years at full width at half maximum (FWHM) of the weighting function, and the second filter 1.5 months at FWHM). The smoothed residuals are then added to the function to obtain smoothed time series with the seasonal cycle removed (1.07 years FWHM) or representative of approximately monthly values for the second filter, as seen in Fig. 2. Except for special notes, the averaged values in this study are reported with 95 % confidence intervals (CI). The CO 2 mole fractions in this study are atmospheric CO 2 dry air mole fractions. Figure 2 shows the hourly averages of the processed ambient air CO 2 concentration at the four stations. The data gaps at WLG are due to a malfunction of the instrument. Valid flask sample pair measurements at WLG during the same period are also shown. To test the measurement quality of the Picarro system, differences between discrete flask measurements from NOAA ESRL and corresponding hourly CO 2 mole fractions from the Picarro instrument at WLG are studied. For the greater part of the time, differences between flasks and in situ systems vary by ±1 ppm. The mean value of the difference (flask minus in situ) is 0.1 ± 0.1 ppm (95 % CI) for 168 sampling events (336 flasks). This result suggests that NOAA flask and CMA in situ measurements at WLG are in good agreement and the CMA calibration and quality control procedures are well implemented without generating systematic biases. The differences between the two data sets are likely due to the sampling height and method (Zhou et al., 2005). The discrete flask sampler collects air 5 m above the ground, while the in situ system is measuring air at 80 m. Due to unique topography at WLG station (located at the top of Mt. Waliguan in the Tibetan Plateau), the flask samples are collected at 08:00 LT (local time) before the updraft in the daytime may bring the local contaminations from the valley. The NOAA flask sampler collects air samples almost instantaneously (sampling period < 1 min), whereas the Picarro system generally has a continuous coverage during each hour. Thus, a part of the differences can be most likely explained by the atmospheric CO 2 fluctuations.

Mean diurnal cycles
The mean diurnal cycles of CO 2 variations in April, July, October, and January in all 3 years are used to represent the average variations in spring, summer, autumn, and winter, respectively. Only days containing 24-hourly averaged CO 2 mole fractions are used, because the day-to-day variations of CO 2 can be quite large. The mean diurnal variations are shown in Fig. 3. Generally, diurnal CO 2 variation is affected by two factors: local sources/sinks and short/medium-range transport (Artuso et al., 2009;Gerbig et al. 2006). In summer, the CO 2 diurnal variation has a similar phase between the sites, with peak values in the early morning and minima in afternoon. After sunrise, photosynthetic CO 2 uptake and mixing of near-surface air with lower concentrations aloft makes the CO 2 mole fractions decrease gradually and reach a stable minimum at about the period of 14:00 to 16:00 LT. In the late evening, when respiration dominates and the boundary layer becomes neutral or stable, CO 2 increases near the surface, and reaches a maximum at 06:00-07:00 LT in the morning, except at LFS, where the maximum concentration is observed at ∼ 05:00 LT. The earlier maximum at LFS is likely due to an earlier sunrise time.
The LAN station is located in the subtropics. Compared with the other three stations, variations of air temperature (yearly average: 14 • ) and solar radiation are relatively small during the year. As a result, photosynthetic/respiration fluxes are active (Kerang et al., 2004) in all seasons and diurnal CO 2 variations are thus significant throughout the year. The peakto-peak mean diurnal amplitudes are 20.7 ± 5.0, 43.2 ± 6.4, 20.0 ± 4.0, and 8.3 ± 4.6 ppm for spring, summer, autumn, and winter, respectively. The larger amplitude in summer is the four stations. The mean standard deviations for hourly CO 2 concentration are 9.5, 21.5, 15.9, and 12.7 ppm for spring, summer, autumn, and winter, respectively, which is much higher than for the other stations. These large values imply that the day-to-day variations of CO 2 are large compared to the periodic diurnal cycle. It can also be inferred that this site may be influenced by very nearby emissions. In spring and autumn, the diurnal variations have amplitudes of 5.5 ± 3.7 and 8.8 ± 6.7 ppm, respectively. The larger diurnal variation in autumn may be due to the straw burning in the northeastern China plain (Cao et al., 2008;Li et al., 2008). In summer and winter, the standard deviations of the hourly averages are so large that the uncertainty is larger than the discrepancy between peak and valley. As a result, the peak-topeak daily amplitude cannot be calculated in these seasons.
The WLG high mountain station receives well-mixed air masses from the lower troposphere, with very small diurnal variability (see also Keeling et al., 1976). CO 2 are systematically lower than at the three regional near-surface stations. Diurnal cycles are small throughout the year with the largest intensity in summer, when a diurnal amplitude of 3.4 ± 1.4 ppm is recorded. Observed seasonal and diurnal variations are similar to those observed from 1994 to 2000 reported by Zhou et al. (2002).

Comparison of CO 2 mole fractions between different levels
Similar to the previous study (Fang et al., 2013), to understand the effect of local sources and sinks on observed CO 2 concentration, the diurnal variations of hourly discrepancy between 10 m and the top of the sampling tower (50 m for LAN and 80 m for LFS) during the two-level parallel observation period are shown in Fig. 4. The record for SDZ is too short (six months) and is not drawn. Generally, CO 2 mole fractions at 10 m are higher than the top levels. The CO 2 differences at LAN and LFS show distinct diurnal variations, with large positive values at night and smaller values in the daytime, partly driven by the coupling between respired CO 2 emitted at night that usually accumulates in a shallow stable nocturnal boundary layer and net CO 2 uptake and efficient vertical mixing during daytime. At night, the maximum differences are 4.8 ± 0.5 ppm for LAN and 7.6 ± 1.3 ppm for LFS. The larger difference at LFS reflects there are stronger sources near the LFS observatory. Nevertheless, the height of top level at LAN (50 m) may also contribute to the smaller difference. At midday, differences at the two regional stations are the smallest and relatively stable. The values are less than 0.2 ± 0.2 ppm from 10:00 to 16:00 LT at LAN and 0.3 ± 0.2 ppm from 09:00 to 16:00 LT at LFS. Given the CO 2 diurnal variations at the two regional stations, the observed CO 2 mole fractions during these periods should be the least influenced by local sources and sinks. These significant differences between 10 and 50/80 m due to the covariance of local vegetation fluxes with vertical mixing prevents inversion models to use nighttime data at 10 m for regional flux estimates. Data during daytime at 10 m could be used in inversions with a bias correction of 0.2 to 0.3 ppm relative to 50/80 m.

Impact of local or regional sources on CO 2 concentration
In this study, we mainly focus on CO 2 concentration that is not obviously affected by very local sources and sinks (≤ ∼ 10 km). As the three regional stations are located in an area of complex terrain, the observed CO 2 mole fractions are likely affected by very local sources for certain wind directions. At LAN, the town of Lin'an is located to the southwest (SW-SSW sector). As a result, observed CO 2 mole fractions are higher than the other directions in the four seasons (data not shown). Especially in autumn, the maximum enhancement relative to the seasonal average (407.5 ± 0.4 ppm) is 7.4 ± 1.6 ppm. Additionally, the charcoal factory located at 1.4 km north of the station may emit a large amount of CO 2 during manufacturing. At LFS, the small villages located to the north (∼ 200 m) will affect the observed CO 2 mole fractions of the northern wind. This phenomenon is very obvious in winter, when local fossil fuel emission (coal for domestic heating) is stronger, with 3.6 ± 0.7 ppm above the average winter mean (404.6 ± 0.2 ppm). The SDZ station is located in an area with strong local sources in the vicinity of the site. Besides the small village located at about 0.8 km to the south, the railway used by diesel-driven trains runs from southwest to northwest, ∼ 0.6 km away from the station, may contribute to the CO 2 mole fractions during the whole year. Similar to the previous study by Zhou et al. (2004), the surface wind is used to identify the possible local and regional sources at WLG. From the wind-rose distribution patterns of CO 2 mole fractions on 16 directions (Fig. 5), it can be seen that surface winds originating from the NE-ENE sector are associated with CO 2 enhancements, with a maximum value of 3.6 ± 1.0 ppm above the seasonal average (391.4 ± 0.1 ppm) in winter. This likely reflects air masses exposed to regions with higher population density. Furthermore, Tang et al. (1999) studied the relationship of black carbon (BC) with surface winds and long-range transport and also found that the highest BC concentrations occur with air masses originating from the NE-NNE sector. They attributed these elevated BC concentrations to remote emissions from the Yellow River canyon industrial area (including cities of Xining, Lanzhou, etc.), 500 km to the northeast of WLG. This phenomenon is more obvious in winter, when the uptake by vegetation on the transporting path is very weak. In winter, in addition to NE-ENE sector, CO 2 concentration on E-ESE-SE sectors is also obviously higher. This is probably due to the emissions from coal burning for heating in Guide County, which is 5 km away to east of the station.
In order to further identify the effect of local sources and sinks, hourly CO 2 mole fractions during the 3 years were calculated for different wind speeds (Beaufort scale): scale 0: < 0.3 m s −1 ; scale 1: 0.3 to 1.5 m s −1 ; scale 2: 1.6 to 3.3 m s −1 ; scale 3: 3.4 to 5.4 m s −1 ; scale 4: 5.5 to 7.9 m s −1 ; scale 5: 8.0 to 10.7 m s −1 ; scale 6: 10.8 to 13.8 m s −1 , ≥; and scale 7: ≥ 13.9 m s −1 . Results show that the observed CO 2 concentration is strongly dependent on the wind speed at the four stations. At LAN, LFS, and SDZ, lower CO 2 mole fractions are always accompanied with higher wind speed in the four seasons, even in summer when the local photosynthesis is very active. These results prove that higher wind speed (and longer range transport) may dilute the very local emissions. From the yearly average, the average CO 2 mole fractions are 416.2 ± 1.0, 402.7 ± 3.9, and 411.0 ± 1.3 ppm, respectively, in calm conditions (scale 0). When the wind speed is Beaufort scale 5, the average CO 2 concentrations are 397.1 ± 3.5, 397.2 ± 1.3, and 398.9 ± 1.4 ppm for LAN, LFS, and SDZ, respectively. For WLG, in winter and spring, decreasing CO 2 concentration is also observed with increasing wind speed. However, during summer and autumn, an opposite behavior is observed. CO 2 increases with increasing wind speed, which may reflect that the local photosynthesis dilutes the anthropogenic signals. For example, in summer, the average CO 2 mole fraction is 384.5 ± 1.5 ppm for calm conditions (scale 0) and 388.2 ± 0.9 ppm at wind scale 7. (385.2 ± 0.1 ppm). Secondly, the "regionally or local" influenced sectors there were the NNE to NE sector for spring, the NE to ENE and SSW to NW sector for summer, the NE to ENE sector for autumn, and the NE to SE sector for winter (Fig. 5). The corresponding CO 2 mole fractions were flagged when surface winds were from these directions. Finally, as discussed above, the higher wind speeds might also transport emissions from the broader region and induce higher CO 2 mole fractions. Thus the CO 2 mole fractions were excluded to be background when surface wind speeds were below scale 1 or above scale 6.
After the three steps filter, one subset for each station is considered to be minimally influenced by local sources, measured during so-called "regional" events (for WLG, they are "background" events). For the three regional stations, about 16.6, 23.9, and 14.4 % of hourly CO 2 mole fractions are selected as regionally for LAN, LFS, and SDZ, respectively. On the other hand, the large ratios of local events (possibly affected by local sources and sinks) prove that local sources and sinks strongly affect the observed CO 2 concentration at 10 m intakes. The background events at WLG account for about 62.6 % of the total hourly mole fractions, which reflects that the observations at WLG (at 80 m intake height) represent well-mixed air masses from the lower troposphere.

Yearly growth rates
We used the curve-fitting method described by Thoning et al. (1989) to extract the trends at the four stations. The filtered CO 2 mole fractions at the four stations all show positive trends in the last 3 years. The growth rates of regionally representative CO 2 at the three regional stations are 3.7 ± 1.2, 2.7 ± 0.8, and 3.5 ± 1.6 ppm yr −1 (1σ ) for LAN, LFS, and SDZ respectively, which are all higher than the global values (2.3 ppm from 2009 to 2010, and 2.0 ppm from 2010 to 2011) (WMO Greenhouse Gas Bulletin, 2011 and also higher than the average of the past decade (∼ 2.0 ppm yr −1 ) (WMO, 2012). Especially at LAN and SDZ station, the values almost double the global average, which implies that the emissions at the Yangtze Delta area and the North China Plain are growing rapidly in recent years. Using a similar method, the growth rates of the locally representative CO 2 mole fractions are calculated to be 3.2 ± 3.0, 3.4 ± 1.0, and 4.2 ± 1.9 ppm yr −1 (1σ ) for LAN, LFS, and SDZ, respectively, which are also obviously higher than the global average.
At WLG, the average growth rate of CO 2 in the last 3 years is 2.2 ± 0.8 ppm yr −1 (1σ ). Figure 6 illustrates the monthly CO 2 mole fractions in the last 15 years (from 1997 to 2011) and the deseasonalized long-term trend at this station. The variation shows that CO 2 concentration reached a new high of 389.5 ± 1.9 ppm in 2010, which is about 0.5 ppm higher than the global average in 2010 and 25.6 ppm higher than the value at WLG in 1997. The long-term trend displayed a positive growth in the last 15 years with a value of ∼ 2.0 ppm yr −1 . As a WMO/GAW global station, the growth rate variation at WLG is similar to the global average with four distinct peaks (Fig. 6) in the periods 1998(WMO, 2012. The former two peaks are very likely linked to the El Niño-Southern Oscillation  (Conway et al., 2012b). LAN compares to those of sine of latitude value 0.5, WLG with 0.6, SDZ with 0.65, and LFS with 0.7. The closed circles and solid line denote the monthly averaged CO 2 concentration at the four stations. The open circles and dashed lines denote the monthly average CO 2 values from the MBL reference. The data in this figure are smoothed from the filtered CO 2 mole fractions (regionally representative for LFS, SDZ, LAN, and background for WLG) by the curve-fitting method of Thoning et al. (1989). (ENSO), which are very commonly seen at other stations (Artuso et al., 2009).

Seasonal variations of regional CO 2 mole fractions
We also used the curve-fitting method described by Thoning et al. (1989) to get the seasonal variations (Figs. 2 and 7) of the filtered CO 2 mole fractions (regionally representative for LFS, SDZ, LAN, and background for WLG) at the four stations. For comparison, simulated surface CO 2 concentration at similar latitude to the four stations from the marine boundary layer (MBL) reference computed by NOAA/GMD (LAN; sine of latitude value: 0.5; WLG, 0.6; SDZ, 0.65; and LFS, 0.7) (Conway et al., 2012b) are also overlaid. Seasonal cycle of CO 2 in the Northern Hemisphere is mainly dominated by exchange with the land biosphere and thus characterized by rapid decreases from June to August and large returns from September to December (Nevison et al., 2008;WMO, 2012). At the four stations, minimum CO 2 concentration all occur in August, which are similar to the Northern Hemisphere (MBL), and the time of occurrence at WLG is consistent with the previous study by Zhou et al. (2005). However, the seasonal CO 2 maximum occurrence times vary substantially, ranging from December for LFS and LAN, to March for SDZ, and to May for WLG. This difference is thought to be caused not only by regionally different terrestrial ecosystems and human activities but also by local meteorological conditions . The high CO 2 mole fractions observed at LFS in winter is likely due to regional coal burning (heating) and biomass burning (agricultural fires), which begins in October in the northeastern China plain. LAN is located in the Yangtze Delta area, which is the economically most developed region in China. In winter, in addition to the lower boundary layer, the increase of fossil fuel consumption may also partly contribute to the higher CO 2 mole fractions. The maximum CO 2 mole fraction occurrence time at WLG, a global WMO background site, agrees with the MBL and those observed at sites of similar latitudes (Nakazawa et al., 1993;Riley et al., 2005;Thoning et al., 1989).
At LAN, the monthly CO 2 mole fractions are obviously higher than the similar MBL references during the whole year, with an average monthly difference of 11.3 ± 4.3 (1σ ) ppm, which means that the LAN area (Yangtze Delta area) is a very strong source of regional atmospheric CO 2 . Similarly, the CO 2 mole fractions at SDZ are also generally higher than the concentration at similar latitude (sine of latitude value: 0.65) with a mean of 7.4 ± 2.7 ppm (1σ ). Both stations display a larger difference to the MBL in the autumn and winter period and smaller values in summer. The CO 2 mole fractions at LFS are strongly affected by regional vegetation canopy in summer. As a result, CO 2 mole fractions are apparently lower than the MBL reference during the same period, with the maximum difference of −15.1 ± 5.8 ppm (1σ ), which indicates that the LFS area may act as a sink of regional CO 2 in the summer. In the other seasons, similar to LAN and SDZ, CO 2 concentration is also obviously higher than the MBL references. Because WLG is located at a high altitude and in a remote area, the observed CO 2 concentration is close to the MBL references.
Because of the strong uptake by the rice plants and forest during growing season in the northeastern China plain, CO 2 mole fractions at LFS are extremely low in the summer and the peak-to-peak amplitude reaches to 35.5 ± 1.1 ppm (Table 2), which is the highest among the four stations. The amplitude at LFS in this study is also much higher than that (19.3 ± 4.1 ppm) by Zhang et al. (2008) during the period of 2003 to 2006. However, they used discrete flask sampling method, which might underestimate the value. CO 2 concentration at WLG is generally lower than those observed at the three regional stations and the seasonal peak-to-peak amplitude is 10.4 ± 0.4 ppm, which is close to similar MBL value (∼ 11.4 ppm; sine of latitude value: 0.6) and the previous study (10.5 ppm) by Zhou et al. (2005). Table 3 illustrates the annual CO 2 concentration at the four stations. For comparison, we also present the CO 2 mole fractions from Mauna Loa in the United States (MLO) (Conway et al., 2012a), Jungfraujoch in Switzerland (JFJ), and Ryori in Japan (RYO) during the same period. MLO and JFJ are WMO/GAW global measurement stations. RYO is GAW regional measurement station. The average regional CO 2 mole fractions at LAN, LFS, and SDZ are obviously higher than the global mean (∼ 389 ppm in 2010 and 390.9 ± 0.1 ppm in 2011) (WMO Greenhouse Gas Bulletin, 2011 and other WMO/GAW global stations such as MLO and JFJ. Compared with the CO 2 concentration observed at other regional stations (such as RYO) close to China, CO 2 mole fractions at the three regional stations are also higher, proving that the stations in China are influenced by (domestic) CO 2 emissions. The highest annual mean CO 2 concentration is observed at LAN of 404.2 ± 3.9 ppm in 2011. This is due to the megacities (Shanghai and Hangzhou) in this region as well as the many industrial sources such as coal-fired power plants (Wang et al., 2006;Zeng et al., 2008). The strong uptake by local vegetation in summer obviously decreases the atmospheric CO 2 mole fractions at LFS. As a result, the annual mole fractions at this station are relatively lower than those at LAN and SDZ.

Conclusions
Atmospheric CO 2 measurements made at four WMO/GAW stations in China have been reported. The diurnal patterns, atmospheric trends, and seasonal variations at each site were discussed. The results indicate that CO 2 mole fractions from 10 m a.g.l. at the three regional stations (LAN, LFS, and SDZ) are affected by local sources. Measurements from WLG are representative of the Tibetan Plateau. The CO 2 mole fractions at the three regional stations are generally higher than the values at respective similar latitudes from MBL references. CO 2 concentration at LAN is representative of the Yangtze Delta area, China, and is the highest among the four stations. Atmospheric CO 2 at LFS are strongly affected by local vegetation (paddy rice and forest) in summer. As a result, it has the lowest CO 2 concentration in summer and the highest seasonal amplitude among the four stations. The SDZ station is located in an area with complex topography. The day-to-day variation of atmospheric CO 2 is quite large. The seasonal CO 2 variations at WLG are small and similar to other GAW global stations (such as MLO in the USA, and JFJ in Switzerland) with peak concentration observed in spring.
In this study, we present continuous atmospheric CO 2 measurements from 2009 to 2011. The relatively short 3year record may introduce bias in our analysis of the seasonal variations and estimates of trends. However, the present analysis gives a first very valuable insight into the evolution and distribution of CO 2 mole fractions across China. To better understand atmospheric CO 2 in China, a more extensive CO 2 observing network and longer period of in situ measurements is required.