Observations of atmospheric ¹⁴CO₂ at Anmyeondo GAW station, South Korea: implications for fossil fuel CO₂ and emission ratios

To understand the Korean Peninsula's carbon dioxide (CO₂) emissions and sinks as well as those of the surrounding region, we used 70 flask-air samples collected during May 2014 to August 2016 at Anmyeondo (AMY; 36.53^∘ N, 126.32^∘ E; 46 m a.s.l.) World Meteorological Organization (WMO) Global Atmosphere Watch (GAW) station, located on the west coast of South Korea, for analysis of observed ¹⁴C in atmospheric CO₂ as a tracer of fossil fuel CO₂ contribution (C_ff). Observed ¹⁴C ∕ C ratios in CO₂ (reported as Δ values) at AMY varied from −59.5 ‰ to 23.1 ‰, with a measurement uncertainty of ±1.8 ‰. The derived mean value C_ff of (9.7±7.8) µmol mol⁻¹ (1σ) is greater than that found in earlier observations from Tae-Ahn Peninsula (TAP; 36.73^∘ N, 126.13^∘ E; 20 m a.s.l., 28 km away from AMY) of (4.4±5.7) µmol mol⁻¹ from 2004 to 2010. The enhancement above background mole fractions of sulfur hexafluoride (Δx(SF₆)) and carbon monoxide (Δx(CO)) correlate strongly with C_ff (r>0.7) and appear to be good proxies for fossil fuel CO₂ at regional and continental scales. Samples originating from the Asian continent had greater Δx(CO) : C_ff(R_CO) values, (29±8) to (36±2) nmol µmol⁻¹, than in Korean Peninsula local air ((8±2) nmol µmol⁻¹). Air masses originating in China showed (1.6±0.4) to (2.0±0.1) times greater R_CO than a bottom-up inventory, suggesting that China's CO emissions are underestimated in the inventory, while observed R_SF6 values are 2–3 times greater than inventories for both China and South Korea. However, R_CO values derived from both inventories and observations have decreased relative to previous studies, indicating that combustion efficiency is increasing in both China and South Korea.

1 Introduction

Carbon dioxide (CO₂) is the principle cause of climate change in the industrial era and has been increasing in the atmosphere at (2.4 ± 0.4) µmol mol⁻¹ a⁻¹ in the past decade globally (where 0.4 is the standard deviation of annual growth rates; http://www.esrl.noaa.gov/gmd/ccgg/trends/, last access: 6 December 2019). This increase is by release of CO₂ from fossil fuel combustion that has been demonstrated through ¹⁴C analysis of tree rings from the last two centuries (Stuiver and Quay, 1981; Suess, 1955; Tans et al., 1979). An atmospheric measurement program for the ratio ¹⁴C ∕ C in CO₂ was initiated in the 1950s and 1960s (Rafter and Fergusson, 1957; Nydal and Lövseth, 1996). Observed ¹⁴C ∕ C ratios are reported in delta notation (Δ(¹⁴CO₂)) as fractionation-corrected permil (or ‰) deviations from the absolute radiocarbon standard (Stuiver and Polach, 1977). Many studies show that the variation of Δ(¹⁴CO₂) is an unbiased and now widely used tracer for CO₂ emitted from fossil fuel combustion (Levin et al., 2003; Turnbull et al., 2006; Graven et al., 2009; Van der Laan et al., 2010; Miller et al., 2012). Therefore measurements of Δ(¹⁴CO₂) are important to test the effectiveness of emission reduction strategies to mitigate the rapid atmospheric CO₂ increase, since they can partition observed CO₂ enhancements, Δx(CO₂), into fossil fuel CO₂ (C_ff) and biological CO₂ (C_bio) components with high confidence (Turnbull et al., 2006).

When trace gases are co-emitted with C_ff, correlations of their enhancements with C_ff improve understanding of the emission sources of both C_ff and the co-emitted tracers. For example, CO and CH₄ emission inventories are typically more uncertain than the fossil fuel CO₂ emission inventory, since fossil fuel CO₂ emissions related to complete combustion are generally well estimated, while emissions related to incomplete combustion and agricultural activities are poorly constrained (Kurokawa et al., 2013). Temporal changes in the observed emission ratio of a trace gas to C_ff can be used to examine emission trends in the trace gas (Tohijima et al., 2014). Therefore the observed emission ratios of trace gases to C_ff can be used to evaluate bottom-up inventories of various trace gases (e.g., Miller et al., 2012). Here, we used two trace gases, carbon monoxide (CO) and sulfur hexafluoride (SF₆), for this analysis. CO is produced along with CO₂ during incomplete combustion of fossil fuels and biomass. CO enhancements above background (Δx(CO)) correlate well with C_ff and have been used as a fossil fuel tracer (Zondervan and Meijer, 1996; Gamnitzer et al., 2006; Turnbull et al., 2011a, b; Tohijima et al., 2014). SF₆ is an entirely anthropogenic gas and is widely used as an arc quencher in high-voltage electrical equipment (Geller et al., 1997). At regional to continental scales, persistent small leaks to the atmosphere of SF₆ are typically co-located with fossil fuel CO₂ sources and allow SF₆ to be used as an indirect C_ff tracer, if the leaks are co-located with C_ff emissions at the location and scale of interest (Turnbull et al., 2006; Rivier et al., 2006).

South Korea is a rapidly developing country with fast economic growth, and it is located next to China, which is the world's largest emitter of anthropogenic CO₂ (Boden et al., 2017; Janssens-Maenhout et al., 2017). The first Δ(¹⁴CO₂) measurements in South Korea were reported by Turnbull et al. (2011a) based on air samples collected during October 2004 to March 2010 at Tae-Ahn Peninsula (TAP; 36.73^∘ N, 126.13^∘ E; 20 m a.s.l.). This study showed that observed CO₂ at this site was often influenced by Chinese emissions, and the observed ratio of Δx(CO) : C_ff (R_CO) was greater than expected from bottom-up inventories. However South Korean Δ(¹⁴CO₂) data are still limited, and the ratio of the other trace gases to C_ff is barely discussed.

Here we use whole-air samples collected in glass flasks during May 2014 to August 2016 at Anmyeondo (AMY; 36.53^∘ N, 126.32^∘ E; 46 m a.s.l.) World Meteorological Organization (WMO) Global Atmosphere Watch (GAW) station, located on the west coast of South Korea and about 28 km SSE of TAP, where the first study was conducted. We decompose observed CO₂ enhancements into their fossil fuel and biological components at AMY to understand sources and sinks of CO₂. We also implemented cluster analysis using the NOAA Hybrid Single Particle Lagrangian Integrated Trajectory Model (HYSPLIT) to calculate back-trajectories for sample times and dates. Based on clusters of trajectories from specific regions, trace gas enhancement to C_ff ratios and correlation coefficients were analyzed, especially focusing on SF₆ and CO, to determine the potential of alternative proxies to Δ(¹⁴CO₂). Finally we compared our Δx(CO) : C_ff ratio with ratios determined from bottom-up inventories (EDGARv4.3.2 and Korea's National Inventory Report in 2018) to evaluate reported CO emissions and how they have changed since 2010.

2 Materials and methods

2.1 Sampling site and methods

The AMY GAW station is managed by the National Institute of Meteorological Sciences (NIMS) in the Korea Meteorological Administration (KMA). It has the longest record of continuous CO₂ measurement in South Korea, beginning in 1999. It is located on the west coast of South Korea, about 130 km southwest of the megacity of Seoul, whose population was 9.8 million in 2017. The semiconductor industry and other industries exist within a 100 km radius of the station. Also, the largest thermal power plants fired by coal and heavy oil in South Korea are within 35 km to the northeast and southeast of the station. The closest town, around 30 km to the east of AMY, is well known for its livestock industries. Local economic activities are related to agriculture, e.g., production of rice paddies, sweet potatoes, and onions, and the area is also known for its leisure opportunities that increase traffic and tourists in summer, indicating the complexity of greenhouse gas sources around AMY. On the other hand, air masses often arrive at AMY from the west and south, which is exposed to the Yellow Sea. Therefore AMY observes enhanced CO₂ compared to many other East Asian stations, due not only to numerous local sources but also the long-range transport of air masses from the Asian continent (Lee et al., 2019).

Two pairs of flask-air samples (four flasks in total, 2 L, borosilicate glass with Teflon O-ring sealed stopcocks) were collected about weekly from a 40 m tall tower at AMY, regardless of wind direction and speed from May 2014 to August 2016, generally between 14:00 and 16:00 local time (Table S1 in the Supplement), using a semi-automated portable sampler. A pair of flasks was flushed for 10 min at 5–6 L min⁻¹ then pressurized to 0.38 bar in less than 1 min. A second pair is collected shortly after the first (within 20 min). The portable sampler was checked for leaks after pressurizing by observing the pressure gauge before closing the stopcocks. Batches of sampled flasks were shipped to Boulder, CO, USA, every 2 months.

A total of 70 sets were collected and analyzed at the National Oceanic and Atmospheric Administration/Global Monitoring Laboratory (NOAA/GML) for CO₂, CO, and SF₆ and for Δ(¹⁴CO₂) by the University of Colorado Boulder, Institute of Arctic and Alpine Research (INSTAAR). NOAA/GML analyzed CO₂ using a nondispersive infrared analyzer, SF₆ using gas chromatography (GC) with electron capture detection, and CO using a vacuum UV resonance fluorescence instrument. All analyzers were calibrated with the appropriate WMO mole fraction scales (WMO-X2007 scale for CO₂, WMO-X2014A scale for CO, and WMO-X2014 for SF₆; https://www.esrl.noaa.gov/gmd/ccl/, last access: 4 December 2019). The measurement and analysis methods for these gases are described in detail (http://www.esrl.noaa.gov/gmd/ccgg/behind_the_scenes/measurementlab.html, last access: 4 December 2019). Measurement uncertainties for CO₂ and SF₆ are reported as 68 % confidential intervals. For CO₂, it is 0.07 µmol mol⁻¹ for all measurements used here. For SF₆, it is 0.04 pmol mol⁻¹. For CO, measurement uncertainty has not yet been formally evaluated but is estimated to be 1 nmol mol⁻¹ (68 % confidence interval). All CO₂, SF₆, and CO data at AMY can be downloaded from ftp://aftp.cmdl.noaa.gov/data/trace_gases/ (last access: 1 September 2020). When we compare NOAA's CO₂ measurements from flask air with quasi-continuous measurements by KMA at AMY, the difference was $-\mathrm{0.11}\pm \mathrm{2.32}$ µmol mol⁻¹ (mean ± 1σ), close to GAW's compatibility goal for CO₂ (± 0.1 ppm for Northern Hemisphere measurements; Lee et al., 2019).

The analysis methods for Δ(¹⁴CO₂) are described by Lehman et al. (2013). Measurement repeatability of Δ(¹⁴CO₂) in aliquots of whole air extracted from surveillance cylinders is 1.8 ‰ (1σ), roughly equating to 1 µmol mol⁻¹ C_ff detection capability from the measurement uncertainty alone. The Δ(¹⁴CO₂) data at AMY are tabulated in Table S1. Among four flasks, the air from two flasks, after analysis for greenhouse gas mole fractions, was combined and analyzed for Δ(¹⁴CO₂).

2.2 Data analysis method using Δ(¹⁴CO₂) data

2.2.1 Calculation of C_ff and C_bio

As Turnbull et al. (2009) suggested, the observed CO₂ (C_obs) at AMY can be defined as

$$\begin{array}{}\text{(1)}& C_{\mathrm{obs}}=C_{\mathrm{bg}}+C_{\mathrm{ff}}+C_{\mathrm{other}},\end{array}$$

where C_bg, C_ff, and C_other are the background, recently added fossil fuel CO₂, and the CO₂ derived from the other sources.

According to Tans et al. (1993), the product of CO₂ abundance and its isotopic ratio is conserved; the isotopic mass balance can be described as below:

$$\begin{array}{}\text{(2)}& {\mathrm{\Delta }}_{\mathrm{obs}}{C}_{\mathrm{obs}}={\mathrm{\Delta }}_{\mathrm{bg}}{C}_{\mathrm{bg}}+{\mathrm{\Delta }}_{\mathrm{ff}}{C}_{\mathrm{ff}}+{\mathrm{\Delta }}_{\mathrm{other}}{C}_{\mathrm{other}},\end{array}$$

where Δ is the Δ(¹⁴C) of each CO₂ component of Eq. (1).

Therefore we can calculate fossil fuel CO₂ by combining Eqs. (1) and (2) as

$$\begin{array}{}\text{(3)}& C_{\mathrm{ff}}={\displaystyle \frac{C_{\mathrm{bg}}({\mathrm{\Delta }}_{\mathrm{obs}}-{\mathrm{\Delta }}_{\mathrm{bg}})}{{\mathrm{\Delta }}_{\mathrm{ff}}-{\mathrm{\Delta }}_{\mathrm{bg}}}}-{\displaystyle \frac{C_{\mathrm{other}}({\mathrm{\Delta }}_{\mathrm{other}}-{\mathrm{\Delta }}_{\mathrm{bg}})}{{\mathrm{\Delta }}_{\mathrm{ff}}-{\mathrm{\Delta }}_{\mathrm{bg}}}}.\end{array}$$

Fossil-fuel-derived CO₂ contains no ¹⁴C because the half-life of ¹⁴C is (5700 ± 30) years (Godwin, 1962), while these fuels are hundreds of millions of years old. As we mentioned in Sect. 1, Δ(¹⁴CO₂) is reported as a per mil (‰) deviation from the absolute radiocarbon reference standard corrected for fractionation and decay with a simplified form; $\mathrm{\Delta }{(}^{\mathrm{14}}\mathrm{C})\approx [R_{\mathrm{sample}}{(}^{\mathrm{14}}\mathrm{C}/\mathrm{C})/R_{\mathrm{standard}}{(}^{\mathrm{14}}\mathrm{C}/\mathrm{C})-\mathrm{1}]\mathrm{1000}$ ‰, where R(¹⁴C ∕ C) is the ¹⁴C ∕ C ratio. Therefore Δ_ff is set at −1000 ‰ (Stuiver and Pollach, 1977). Background values (Δ_bg) in Eqs. (1) to (3) are determined from measurements from background air collected at Niwot Ridge, Colorado, a high altitude site at a similar latitude to AMY (NWR; 40.05^∘ N, 105.58^∘ W; 3526 m a.s.l.). Turnbull et al. (2011a) showed that the choice of background values did not significantly influence derived enhancements due to the large regional and local signal at TAP, 28 km from AMY. NWR Δ(¹⁴CO₂) and other trace gas background values are selected using a flagging system to exclude polluted samples (Turnbull et al., 2007) and then fitted with a smooth curve following Thoning et al. (1989).

The second term of Eq. (3) is typically a small correction for the effect of other sources of CO₂ that have Δ(¹⁴C) differing by a small amount from that of the atmospheric background, such as CO₂ from the (1) nuclear power industry, (2) oceans, (3) photosynthesis, and (4) heterotrophic respiration.

(1) The nuclear power industry produces ¹⁴C that can influence the C_ff calculation. South Korea has nuclear power plants along the east coast that may influence AMY air samples when air masses originated from the eastern part of South Korea (Fig. 1). It is also possible that Chinese nuclear plants could influence some samples. Here we did not make any correction for this since most nuclear installations in this region are pressurized water reactors, which produce mainly ¹⁴C in CH₄ rather than CO₂ (Graven and Gruber, 2011). (2) For the ocean, although there may also be a small contribution from oceanic carbon exchange across the Yellow Sea, we consider this effect small enough to ignore (Turnbull et al., 2011a). It was also demonstrated there is no significant bias from the oceans including the East China Sea (Song et al., 2018), even at coastal sites in the Northern Hemisphere (Turnbull et al., 2009). Larger scale ocean exchange and also stratospheric exchange affect both background and observed samples equally, so they can be ignored in the calculations. (3) For the photosynthetic terms, ¹⁴C in CO₂ accounts for natural fractionation during uptake, so we also set this observed value to be the same as the background value. (4) Therefore we only consider heterotrophic respiration. For land regions, where most fossil fuel emissions occur, heterotrophic respiration could be a main contributor to the second term of Eq. (3) due to ¹⁴C disequilibrium potentially. When this value is ignored, C_ff is consistently underestimated (Palstra et al., 2008; Riley et al., 2008; Hsueh et al., 2007; Turnbull et al., 2006). For this, corrections were estimated to be (−0.2 ± 0.1) µmol mol⁻¹ during winter and (−0.5 ± 0.2) µmol mol⁻¹ during summer (Turnbull et al., 2009, 2006).

Figure 1 A total of 70 air-parcel back-trajectories were calculated for 72 h periods at 3 h intervals from May 2014 to August 2016 using the HYSPLIT model in conjunction with KMA UM GDAPS data at a 25 km×25 km resolution. Station locations are as follows: WLG (Waliguan; 36.28^∘ N, 100.9^∘ E; 3816 m a.s.l.), SDZ (Shandianzi; 40.65^∘ N, 117.12^∘ E; 287 m a.s.l.), and AMY (Anmyeondo; 36.53^∘ N, 126.32^∘ E; 86 m a.s.l.). TAP (Tae-Ahn Peninsula; 36.73^∘ N, 126.13^∘ E; 20 m a.s.l.) is around 28 km northeast of AMY.

CO₂ enhancements relative to baseline CO₂ are defined as Δx(CO₂), with the excess signal of C_obs minus C_bg in Eq. (1). Partitioning of Δx(CO₂) into C_ff and C_bio is calculated simply from the residual of the difference between observed Δx(CO₂) and C_ff.

2.2.2 The ratio of trace gas enhancement to C_ff and its correlation

To obtain the correlation coefficient (r) between C_ff and other trace gas enhancements ($\mathrm{\Delta }x\left(x\right)=x_{\mathrm{obs}}-x_{\mathrm{bg}}$) and the ratio of any trace gas to C_ff (R_gas), we use reduced major axis (RMA) regression analysis (Sokal and Rohlf, 1981). The distributions of R_gas are normally broad and non-Gaussian, and RMA analysis is a relatively robust method of calculating the slope of two variables that show some causative relationship. Here, x_bg was derived from NWR with the same method described in Sect. 2.2.1. The relevant equations are presented from Eqs. (S1) to (S3) in the Supplement. Results for each species are given in Table 1.

2.3 HYSPLIT cluster analysis

HYSPLIT trajectories were run using Unified Model-Global Data Assimilation and Prediction System (UM-GDAPS) weather data at a 25 km×25 km horizontal resolution to determine the regions that influence air mass transport to AMY. A total of 70 air-parcel back-trajectories were calculated for 72 h periods at 3 h intervals matching the time of each flask-air sample taken at AMY from May 2014 to August 2016. We assign the sampling altitude as 500 m, since it was demonstrated that HYSPLIT and other particle dispersion back-trajectory models (e.g., FLEXPART) are consistent at 500 m altitude (Li et al., 2014). Cluster analysis of the resulting 70 back-trajectories categorized six pathways through which air parcels arrive at AMY during the time period of interest.

Among the calculated back-trajectories, 67 % indicate air masses originating from the Asian continent. Back-trajectories of continental background air (CB) originating in Russia and Mongolia occurred 13 % of the time. A total of 23 % of the trajectories originated in and traveled through northeast China (CN). The CN region includes Inner Mongolia and Liaoning, one of the most populated regions in China, with 43.9 million people in 2012. These CN air masses arrive in South Korea after crossing through western North Korea. A total of 17 % of the trajectories are derived from central eastern China around the Shandong area (CE). The CE region contains Shandianzi (SDZ; 40.65^∘ N, 117.12^∘ E; 287 m a.s.l.), located next to the megacities of Beijing and Tianjin, which are some of China's highest CO₂-emitting regions (Gregg et al., 2008). A total of 14 % have an ocean background (OB), derived from the East China Sea. Among them, a few of the trajectories passed over the eastern part of China (e.g., over Shanghai) at high altitude (1000 m). Flow from the Korean Peninsula also travels through heavily industrialized and/or metropolitan regions in South Korea (Korean Peninsula local air, KL; 19 %) and under stagnant conditions (polluted local region, PL; 14 %). Some of the KL air masses also passed over the East Sea and Japan.

Table 1 Means and standard deviations of C_ff (µmol mol⁻¹), CO (nmol mol⁻¹), and SF₆ (pmol mol⁻¹) (total N=50; without PL N=41). The correlations (r) and the ratio (R_gas) of enhancement between C_ff values were determined by reduced major axis (RMA) regression analysis on each scatter plot to obtain regression slopes. The uncertainty of R_gas refers to Eq. (S2). When r is less than 0.7, R_gas was not included here. N is the number of data. The unit of R_CO is nanomoles per micromole (nmol µmol⁻¹), and for $R_{{\mathrm{SF}}_{\mathrm{6}}}$ it is picomoles per micromole (pmol µmol⁻¹). A plot of R_CO and $R_{{\mathrm{SF}}_{\mathrm{6}}}$ is shown in Fig. S1 in the Supplement. CB represents continental background, CN northeast China, CE central eastern China, OB ocean background, KL Korean Peninsula local air, and PL polluted local air mass.

3 Results and discussion

3.1 Observed Δ(¹⁴CO₂) and portioning of CO₂ into C_ff and C_bio

AMY Δ(¹⁴CO₂) values are almost always lower than those observed at NWR, which we consider to be broadly representative of background values for the midlatitude Northern Hemisphere (Fig. 2). NWR Δ(¹⁴CO₂), which is based on weekly air samples, was in the range 10.0 ‰ to 21.2 ‰, with an average (16.6 ± 3) ‰ (1σ, standard deviation) from May 2014 to August 2016. Waliguan (WLG; 36.28^∘ N, 100.9^∘ E; 3816 m a.s.l.), an Asian background GAW station in China, also showed similar Δ(¹⁴CO₂) levels to NWR, with an average of (17.1 ± 6.8) ‰ in 2015 (Niu et al., 2016, measurement uncertainty ± 3 ‰, n=20). Δ(¹⁴CO₂) at AMY varied from −59.5 ‰ to 23.1 ‰ and had a mean value of (−6.2 ± 18.8) ‰ (1σ, n=70) during the measurement period (Table S1). This was similar to results from observations at SDZ, which is located about 100 km northeast of Beijing, in the range of −53.0 ‰ to 32.6 ‰, with an average (−6.8 ± 21.1) ‰ (1σ, n=32) during September 2014 to December 2015 (Niu et al., 2016).

Figure 2 Time series of (a) observed CO₂ dry air mole fraction (open circles) and observed CO₂ (C_obs) minus C_ff calculated from Δ(¹⁴CO₂) (closed circles). (b) Δ(¹⁴CO₂) at AMY (black circles) and at NWR (Niwot Ridge; line), baseline data. (c) Time series of C_ff and C_bio calculated from Δ(¹⁴CO₂) (left) and the frequency distribution at AMY (right).

Calculated C_ff at AMY ranges between −0.05 and 32.7 µmol mol⁻¹, with an average of (9.7 ± 7.8) µmol mol⁻¹ (1σ, n=70); high C_ff was observed regardless of season (Fig. 2a). One negative C_ff value of −0.05 µmol mol⁻¹ was estimated due to greater AMY Δ(¹⁴CO₂) than NWR on 30 July 2014. Although negative C_ff values are nonphysical, this value is not significantly different from zero and is reasonable given that this air originated from the OB sector. The range of C_ff in the AMY samples is similar to that observed at TAP from 2004 to 2010 (−1.6 to 42.9 µmol mol⁻¹ C_ff), but C_ff is on average about twice as high at AMY as in the 2004 to 2010 TAP samples (mean (4.4 ± 5.7) µmol mol⁻¹, n=202; Turnbull et al., 2011a). A more detailed comparison of results based on differences between samples derived from the Asian continent and Korean Peninsula local air is provided in Sect. 3.2.

Estimated C_bio, as defined in Sect. 2.2.1, varied from −18.1 to 15.7 µmol mol⁻¹ (mean (0.9 ± 5.8) µmol mol⁻¹) at AMY (Fig. 2c). C_bio showed a strong seasonal cycle, with the lowest values from July to September, when photosynthetic drawdown is expected to be strongest, in good agreement with the previous TAP study (Turnbull et al., 2011a). Even though C_bio was at times negative, mainly due to photosynthesis during summer, the largest positive C_bio was also observed in summer.

The largest C_ff by season was observed in order of winter (DJF; (11.3 ± 7.6), n=14) > summer (JJA; (10.7 ± 9.2), n=11) > spring (MAM; (8.6 ± 8.0), n=22) > autumn (SON; (7.6 ± 5.6), n=17), in units of micromoles per mole (µmol mol⁻¹). When we consider only positive contributions of C_bio samples, the order was summer ((4.6 ± 4.0), n=14) > autumn ((4.1 ± 2.5), n=9) > spring ((3.8 ± 2.6), n=13) > winter ((3.4 ± 2.5), n=11), in units of micromoles per mole (µmol mol⁻¹).

C_ff in summer was nearly as high as in winter. This is because lower wind speeds are observed at AMY during summer (Lee et al., 2019). When we analyzed seasonal boundary layer height for each sample by UM-GDAPS, it also showed a similar result; it was highest in winter (with a range from 150 to 1100 m) and lowest in summer (with a range from 100 to 500 m). This suggests that these high summer C_ff values may reflect emission from local activities, which were described in Sect. 2.1, more than in other seasons.

The highest C_bio value was also observed in the summer in the PL sector. The PL sector showed that positive C_bio correlates with CH₄, which is a tracer for agriculture when observed in TAP local air masses. Turnbull et al. (2011a) also showed similar results.

In winter, C_bio was relatively lower than in other seasons, while C_ff was highest. During winter, AMY is mainly affected by long-range transport of air masses from China due to the Siberian High (Lee et al., 2019). Therefore air samples were less affected by local activities in winter, but C_bio still contributed almost 23 % to Δx(CO₂). In the dry season (from October to March), forest fires, which contribute the largest portion of total CO₂ emissions from open fires at the national scale, are concentrated in northeastern and southern China (Yin et al., 2019). The highest CO was observed in winter ((449.1 ± 244.1) nmol mol⁻¹ (1σ) in winter while (236.8 ± 124.4) nmol mol⁻¹ (1σ) in summer), which also supports biomass burning and biofuels as being large contributors to observed CO₂ enhancements in winter. Turnbull et al. (2011a) also showed that 20 %–30 % of winter CO₂ enhancements at TAP were likely contributed by biofuel combustion, along with plant, soil, human, and animal respiration.

Regardless of the source, we find that C_bio contributes substantially to atmospheric CO₂ enhancements at AMY in air masses affected by local and long-range transport, so when only CO₂ enhancements above background are compared to bottom-up inventories, it can create a bias due to C_bio contributions.

3.2 C_ff comparison between Korean Peninsula local and Asian continent samples

To more clearly identify samples originating from the Asian continent (trajectory clusters CB, CN, CE, and OB) and Korean Peninsula local air (trajectory cluster KL) after cluster analysis of the 70 sets of measurements, we use wind speed data from the automatic weather system (AWS) installed at the same level as the air sample inlet at AMY. Among the data from CB, CN, CE, OB, and KL, when wind speed was less than 3 m s⁻¹, we assumed that those samples could be affected by local pollution. PL was also ruled out since it was affected by local pollution under stagnant conditions. Therefore we use only 41 sets of observations for this analysis (Table 1).

C_ff is highest in the order CE > CN > KL > CB > OB (Table 1). During the measurement period, the averages from the Asian continent (sectors CE and CN) were higher than KL without the baseline sector (CB and OB). The calculated mean C_ff using only CE, CN, CB, and OB, which sample substantial outflow from the Asian continent, was (7.6 ± 3.9) µmol mol⁻¹.

When we compared the KL samples ((8.6 ± 5.3) µmol mol⁻¹) with those from Korean Peninsula local air masses observed at TAP ((8.5 ± 8.6) µmol mol⁻¹, n=58; Turnbull et al., 2011a), mean C_ff was quite similar (Fig. 3). However, when comparing the C_ff values from CB air masses in this study and TAP long-range-transported samples (from China) (n=144; Turnbull et al., 2011a), C_ff almost doubled from (2.6 ± 2.4) to (4.3 ± 2.1) µmol mol⁻¹, even though they might be expected to have had similar air mass back-trajectories. We also compared the values at SDZ from 2009 to 2010 (Turnbull et al., 2011a) and in 2015 (Niu et al., 2016); they also increased, not only in the samples that were affected by Beijing and the North China Plain (SDZ-BN), which are comparably polluted, but also in the samples that were affected by northeast China (SDZ-NE). For SDZ-BN samples, C_ff increased from (10 ± 1) to (16 ± 7.6) µmol mol⁻¹ from 2009–2010 (n=32) to 2015 (n=32). The AMY samples from CE, which flow over Beijing, showed (11.2 ± 8.3) µmol mol⁻¹ of C_ff and were also slightly greater than the 2009–2010 SDZ-BN samples (Turnbull et al., 2011a). For SDZ-NE samples, C_ff was (3 ± 7) µmol mol⁻¹ in 2009 to 2010 and increased to (7.6 ± 6.8) µmol mol⁻¹ in 2015. Since the SDZ-NE samples are affected by northeast China according to Turnbull et al. (2011a) and Niu et al. (2016), we also see CN that originated from northeast China (NE) and its mean value of C_ff had increased around (10.6 ± 6.9) µmol mol⁻¹ compared to those values in 2009 to 2010.

Figure 3 Calculated C_ff (µmol mol⁻¹). Red bars are for KL, and blue bars are for South Korean long-range-transported samples (from China) (2004–2010 from Turnbull et al., 2011a). Black bars are for SDZ-BN samples that were affected by Beijing and the North China Plain. Gray bars for SDZ-NE indicate samples that were affected by regions northeast of SDZ. SDZ (2009–2010) is from Turnbull et al. (2011a), and SDZ (2015) is from Niu et al. (2016). Hatched red, blue, black, and gray bars are derived from this study during 2014 to 2016.

It has been suggested that inter-annual variability in observed mean C_ff in South Korea could reflect changing fossil fuel CO₂ emissions or could indicate inter-annual variability in the air mass trajectories of the (small) dataset of flask-air samples (Turnbull et al., 2011a). Even though the growth rate of C_ff emission has been decreasing slowly in East Asia since 2010 due to emission reduction policies (Labzovskii et al., 2019), reported emissions increased 16.7 % in China and 1.8 % in South Korea from 2010 to 2016 (Janssens-Maenhout et al., 2017). This is broadly consistent with the flat trend in observed C_ff in KL air masses and with the upward trend in C_ff observed in air masses flowing out from Asia. Therefore it is possible that AMY mean C_ff increased relative to the earlier TAP observations due to increased fossil fuel emissions from the Asian continent.

On the other hand, the values from this study showed large variability with small sample numbers due to a different sampling strategy, environment, and synoptic conditions such as boundary layer height at the sampling time from reference studies. Further study will be necessary to understand these increased values.

3.3 Correlation of C_ff with SF₆ and its emission ratios

We calculated correlation coefficients (r from Eq. S3) between SF₆ and CO enhancements with C_ff and their ratios from Eq. (S1) with the 50 samples that were described in Sect. 3.2 including the PL sector (n=9) and whose values are tabulated in Table 1.

The correlations of CO enhancements (Δx(CO)) with C_ff were strong (r>0.7) in all sectors except PL, while SF₆ enhancements (Δx(SF₆)) correlated strongly with C_ff (r>0.8) for CE and OB in outflow from the Asian continent and KL. R_CO and $R_{{\mathrm{SF}}_{\mathrm{6}}}$ were different between Korean Peninsula local air masses and outflows from the Asian continent. Here we discuss $R_{{\mathrm{SF}}_{\mathrm{6}}}$, and in Sect. 3.4 we discuss R_CO in more detail.

For SF₆, observed mean levels were high, in order of (KL, PL) > (CN, CE) > (OB, CB) (Table 1). SF₆ values in KL and PL were higher than from the Asian continent, since South Korea has larger SF₆ emissions than most countries (ranked fourth in 2010 according to EDGAR4.2.) because of liquid-crystal display (LCD) and electrical equipment production (Fang et al., 2014). Even though both KL and PL showed a higher SF₆ mole fraction than outflows of the Asian continent, the correlation is different between KL and PL (Table 1). Under stagnant conditions, emitted SF₆ is less diluted by mixing, so that in PL, Δx(SF₆) correlated weakly with C_ff. On the other hand, KL, CE, and OB showed strong correlations (r>0.8). Those three sectors are also larger SF₆ sources compared to other regions, according to SF₆ emission estimates for Asia (Fang et al., 2014). Because long-range transport allows time for mixing, SF₆ and C_ff emissions are effectively co-located at not only continental scales but also regional scales. Thus SF₆ can be a good tracer of fossil fuel CO₂ for those regions.

The correlation between Δx(SF₆) and C_ff was strong in CE, OB, and KL; however, $R_{{\mathrm{SF}}_{\mathrm{6}}}$ is different between South Korea and outflow from the Asian continent (Fig. S2). In a previous study, observed $R_{{\mathrm{SF}}_{\mathrm{6}}}$ was 0.02 to 0.03 pmol µmol⁻¹ at NWR in 2004 (Turnbull et al., 2006). Here, the ratio was (0.19 ± 0.03) and (0.17 ± 0.03) pmol µmol⁻¹ for CE and OB respectively. For KL, it was (0.66 ± 0.16) pmol µmol⁻¹, indicating much larger ratios than in outflow from the Asian continent. Further, observed $R_{{\mathrm{SF}}_{\mathrm{6}}}$ is 2 to 3 times greater for all air masses than predicted from bottom-up inventories based on a national scale. For this calculation, we use EDGAR4.3.2 for CO₂ and EDGAR4.2 for SF₆. We repeat the calculations for both CO₂ and SF₆ with Korea's National Inventory Report (KNIR; Greenhouse Gas Inventory and Research Center, 2018). Using SF₆ for 2010 from EDGAR4.2, we obtain $R_{{\mathrm{SF}}_{\mathrm{6}}}$ of 0.08 pmol µmol⁻¹ for China, while for South Korea it was 0.14 pmol µmol⁻¹. Especially for South Korea, this is much lower than the observed $R_{{\mathrm{SF}}_{\mathrm{6}}}$. When KL $R_{{\mathrm{SF}}_{\mathrm{6}}}$ was compared to ratios calculated from KNIR (0.27 pmol µmol⁻¹ for 2010 and 0.22 pmol µmol⁻¹ for 2014), it was closer to observed $R_{{\mathrm{SF}}_{\mathrm{6}}}$ than EDGAR but still underestimated (Figs. S3 and S2). This result suggests that the observed ratio could be used to re-evaluate the bottom-up inventories (Rivier et al., 2006), especially targeting the Asian continent. Even though KL $R_{{\mathrm{SF}}_{\mathrm{6}}}$ showed greater uncertainty than CE and OB, it is still greater than bottom-up inventories, such as KNIR and EDGAR. Therefore it would be useful to obtain more data to try and derive a more robust estimate to evaluate SF₆ emission inventories for South Korea.

3.4 Correlation of C_ff with CO and its emission ratios

High CO was mainly observed in outflow from the Asian continent in order of CE > CN > PL > (CB, KL) > OB (Table 1). The order of CO is quite different to that of SF₆. CO from KL and PL is lower than from outflow from the Asian continent, except for the OB sector, indicating that high CO can be a tracer of outflow from the Asian continent. Since CO is produced during incomplete combustion of fossil fuel and biomass, it is more closely related to fossil fuel CO₂ emissions than the other trace gases. Therefore in most cases the correlation between CO and C_ff was strong. R_CO was very different between Korean Peninsula local air masses ((8 ± 2) nmol µmol⁻¹) and those originating from the Asian continent ((29 ± 8) to (36 ± 2) nmol µmol⁻¹), due to differences in combustion efficiencies and the use of catalytic converters. The higher continental emission ratios may also result from some contribution of biofuel combustion and agricultural burning in the Asian continent, which have significantly higher CO emission than fossil fuel combustion (Akagi et al., 2011). For example, for CB the CO level is similar to KL, while R_CO is higher than KL with low C_ff.

Typically CO shows seasonal variations, with lower values in summer due to the atmospheric chemical sink, OH. Among the samples, the samples collected in summer were mainly rejected through wind speed cutoff (less than 3 m s⁻¹) since AMY has lower wind speed in summer (Lee et al., 2019). Only the OB sector includes four summer samples (of seven) because summer air masses are mainly from the southern part of the Yellow Sea (Lee et al., 2019). However, we assumed R_CO is less affected by the summer sink, since only two Δx(CO) samples were negative for OB (Fig. S1), and R_CO was consistent whether or not the negative Δx(CO) values were considered. To compare emission ratios derived from atmospheric observations with those from inventories for 2000 to 2012, we calculated the inventory emission ratio ($I_{\mathrm{CO}/{\mathrm{CO}}_{\mathrm{2}}}$) as

$$I_{\mathrm{CO}/{\mathrm{CO}}_{\mathrm{2}}}=E_{\mathrm{CO}}/E_{{\mathrm{CO}}_{\mathrm{2}}}\times M_{{\mathrm{CO}}_{\mathrm{2}}}/M_{\mathrm{CO}},$$

where E_CO and $E_{{\mathrm{CO}}_{\mathrm{2}}}$ are total CO and fossil fuel CO₂ emissions in gigagrams (Gg a⁻¹, 10⁹ g a⁻¹) from the bottom-up national inventory. M_X is the molar masses of CO and CO₂ in grams per mole (g mol⁻¹).

We use EDGAR4.3.2 (Janssens-Maenhout et al., 2017) and KNIR (Greenhouse Gas Inventory and Research Center, 2018) for inventory information for both CO and CO₂.

The uncertainty of EDGAR4.3.2 fossil fuel CO₂ emissions was reported as a 95 % confidence interval (Janssens-Maenhout et al., 2019), ± 5.4% for China and ± 3.6 % for South Korea (personal communication with Efisio Solazzo in the EDGAR team, 2020). The uncertainties of CO and SF₆ emissions were not reported by EDGAR. For KNIR, the CO₂ 2016 emission uncertainty in the energy sector was ± 3% (Greenhouse Gas Inventory and Research Center, 2018). KNIR does not provide uncertainties for other emission sectors of CO₂, nor for emissions of CO and SF₆.

In Fig. 4 we confirm that the CO to C_ff emission ratios (R_CO) derived from both observations and inventories for China and South Korea are decreasing. Since C_ff emissions appear to be flat (South Korea) or slightly increasing (China), this indicates that combustion efficiency and/or scrubbing of CO is improving.

Figure 4 R_CO for China (a) and for South Korea (b). Black circles: EDGARv.4.3.2 emission inventory. Black squares: Korea's National Inventory Report (KNIR; Greenhouse Gas Inventory and Research Center, 2018). Blue symbols are from other studies (Suntharalingam et al., 2004; Wang et al., 2010; Turnbull et al., 2011a; Tohjima et al., 2014; Niu et al., 2018; Tang et al., 2018). Red symbols: this study. The y axis error bars represent the uncertainty in the slope according to Eq. (S2). The x axis error bars represent the period for the mean value.

For South Korea, EDGAR4.3.2 indicated that CO emissions from the energy sector (98 % to 99 % of total emission) decreased by 47 % between 1997 and 2012. South Korean fossil fuel CO₂ emissions increased until 2011 and remained mostly constant from 2011 to 2016 ((603 901 ± 4315) Gg a⁻¹ CO₂) (Fig. S4). Therefore the decreased trend in the emission ratio seems to reflect recent decreases in CO emissions in South Korea. Turnbull et al. (2011a) determined an observed mean R_CO of (13 ± 3) nmol µmol⁻¹ during 2004 to 2010. Suntharalingam et al. (2004) estimated R_CO 15.4 nmol µmol⁻¹ for South Korea in 2001 from CO₂ and CO airborne observations (C_ff was not determined). Recently, the KORUS-AQ campaign, which was conducted over Seoul from May to June 2016, estimated R_CO to be 9 nmol µmol⁻¹ (Tang et al., 2018) based on CO₂ and CO observations (C_ff was not determined). Our study gives R_CO of (8 ± 2) nmol µmol⁻¹ for South Korea, slightly but not significantly lower than the KORUS-AQ result for Seoul. Different contributions of C_bio and C_ff to total CO₂ may have biased the R_CO calculation when total CO₂ was used in the KORUS-AQ study (e.g., Miller et al., 2012). The South Korean national R_CO from EDGAR4.3.2 in 2012 was 6.7 nmol µmol⁻¹, consistent with our observations. Using KNIR for 2016, we obtain R_CO of 2.1 nmol µmol⁻¹. KNIR suffers from a large number of missing CO emission sources compared to EDGAR, as indicated by their reported emissions, 638.3 and 2580.8 Gg a⁻¹ in 2012, respectively (Fig. S5). For example, CO emissions recently derived from fugitive emissions and residential/other sectors increased to 14 % and 11.5 % of total emission respectively in EDGAR but were not reported in KNIR.

For China the inventories estimate that CO emissions from the energy sector, (96.5 ± 0.2) %, were almost constant through the 1990s and then increased during the early 2000s from industrial processes (8.8 % of total emissions in 2012). Fossil fuel CO₂ emission in China also increased until 2013 and then stayed roughly constant at (10 461 890 ± 60 571) Gg a⁻¹ according to EDGAR4.3.2. Thus even though both emissions show an increase from 2000 to 2016 for fossil fuel CO₂ and to 2012 for CO, the emission ratio decreased (Figs. S4 and 4), which seems to indicate that combustion efficiency is improving. Many studies observed decreasing R_CO in China from 2000 to 2010 (Turnbull et al., 2011a; Wang et al., 2010). Suntharalingam et al. (2004) reported that R_CO was 55 nmol µmol⁻¹ in 2001 (C_ff was not determined). In the Beijing region, R_CO decreased from 57.80 to 37.59 nmol µmol⁻¹ during 2004 to 2008 (Wang et al., 2010). The overall R_CO was (47 ± 2) nmol µmol⁻¹ at SDZ for 2009–2010 and (44 ± 3) nmol µmol⁻¹ in air masses that originated from the Asian continent from 2005 to 2009 (Turnbull et al., 2011a). Tohjima et al. (2014) explained that surface-based R_CO decreased from 45 to 30 nmol µmol⁻¹ in outflow air masses from China from 1998 to 2010. Fu et al. (2015) also observed R_CO of 29 nmol µmol⁻¹ over mainland China in 2009. In Beijing, which is located along the path of CE, it was (30.4 ± 1.6) nmol µmol⁻¹ and (29.6 ± 3.2) nmol µmol⁻¹ for Xiamen in 2016, which is in the OB sector (Niu et al., 2018). During KORUS-AQ in 2016, R_CO of 28 nmol µmol⁻¹ was observed over the Yellow Sea. Some of those studies did not differentiate C_ff from the total CO₂ enhancement, so, although R_CO still includes uncertainties, it is continually decreasing.

In this study R_CO is (29 ± 8), (31 ± 8), (36 ± 2), and (31 ± 4) nmol µmol⁻¹ for CB, CN, CE, and OB, consistent with Tang et al. (2018) and Niu et al. (2018). On the other hand, R_CO in CE is higher than in other sectors in this study. The Shandong area, which is located in the path of CE, has been plagued with problems of combustion inefficiency and ranked as the largest consumer of fossil fuels in all of China (Chen and Li, 2009). The uncertainties in our observed R_CO for this region overlap with other sectors such as CB, CN, and OB, so further monitoring of the ratios will help to obtain more detailed information.

In South Korea and China, atmosphere-based R_CO values calculated by this study are (1.2 ± 0.3) times (with KL) and (1.6 ± 0.4), (1.7 ± 0.4), (2 ± 0.1), and (1.7 ± 0.2) times greater (with CB, CN, CE, and OB) than in the inventory, respectively (Fig. 4). This is in agreement with previous studies (Turnbull et al., 2011a; Kurokawa et al., 2013; Tohjima et al., 2014). One explanation is that EDGAR does not reflect secondary CO production, which can be a significant contributor to CO (Kurokawa et al., 2013). Also, CO derived from biomass burning and biofuels was not included in this inventory. Therefore, this indicates that top-down observations are necessary to evaluate and improve bottom-up emission products.

4 Summary and conclusions

To understand CO₂ sources and sinks in Korea as well as those of the surrounded region, we collected Δ(¹⁴CO₂) with 70 flask samples from May 2014 to August 2016. We summarize our results below.

Observed Δ(¹⁴CO₂) values at AMY ranged from −59.5 ‰ to 23.1 ‰ (a mean value of (−6.2 ± 18.8) ‰ (1σ)) during the study period, almost always lower than those observed at NWR, which we consider to be broadly representative of background values for the midlatitude Northern Hemisphere. This reflects the strong imprint of fossil fuel CO₂ emissions recorded in AMY air samples.

Calculated C_ff using Δ(¹⁴CO₂) at AMY ranges between −0.05 and 32.7 µmol mol⁻¹, with an average of (9.7 ± 7.8) µmol mol⁻¹ (1σ); this average is twice as high as in the 2004 to 2010 TAP samples (mean (4.4 ± 5.7) µmol mol⁻¹) (Turnbull et al., 2011a). We also observed high C_ff regardless of the season or source region. After separately identifying samples originating from the Asian continent and the Korean Peninsula, we determined that the mean C_ff increased relative to the earlier observations due to increased fossil fuel emissions from the Asian continent as showing by the consistent growth in reported emissions, which increased 16.7 % in China and only 1.8 % in South Korea from 2010 to 2016. Note, however, that our data span a relatively limited time period and are subject to different synoptic conditions during the sampling time from previous studies, so a longer time series would increase confidence in tracking this change.

Because Δx(CO) and Δx(SF₆) agreed well with C_ff but showed different slopes for Korean Peninsula local air and the Asian continent, these R_gas values can be indicators of air mass origin, and these gases can be proxies for C_ff. Overall, we have confirmed that R_CO values derived from both an inventory and observations have decreased relative to previous studies, indicating that combustion efficiency is increasing in both China and South Korea.

However, atmosphere-based R_gas values are greater than bottom-up inventories. For CO, our values are (1.2 ± 0.3) times and (1.6 ± 0.4) to (2.0 ± 0.1) times greater than in inventory values for South Korea and China, respectively. This discrepancy may arise from several sources including the lack of contribution of atmospheric chemical CO production such as oxidation of CH₄ and non-methane volatile organic compounds (VOCs). Observed $R_{{\mathrm{SF}}_{\mathrm{6}}}$ is 2 to 3 times greater than in inventories. Therefore those values in our study can be used for improving bottom-up inventories in the future.

Finally, we stress that because C_bio contributes substantially to Δx(CO₂), even in winter, Δ¹⁴C-based C_ff (and not Δx(CO₂)) is required for accurate calculation of both R_CO and $R_{{\mathrm{SF}}_{\mathrm{6}}}$.