Comparison of GC-μECD and OA-ICOS Methods for High-Precision Measurements of Atmospheric Nitrous Oxide (N2O) at a Korean GAW Station

Nitrous oxide (N2O) is a powerful greenhouse gas and is the largest remaining anthropogenic source of stratospheric ozone-depleting substances as halocarbons return towards preindustrial levels. To verify the N2O emission inventory using inverse analysis, precise and reliable measurements are necessary. In this study, we compared the conventional gas chromatography with the microelectron capture detector method (GC-μECD, Agilent 7890A) with advanced off-axis integrated cavity output spectroscopy (OA-ICOS, Los Gatos, EP-30) for atmospheric N2O measurements at the Jeju Gosan Suwolbong Station (JGS, 126.16◦ E, 33.30◦ N, 71.47 m a.s.l) in South Korea. The measurement uncertainties from linearity, repeatability, and reproducibility derived from the two instruments were compared. The values derived from GC-μECD were 2.4 to 8.7 times greater than that of OA-ICOS in all factors at the station. Since these factors affect the measurement quality, the calibration strategy should be well-established to reduce the measurement uncertainty. These uncertainties resulted in biases from the measurement of atmospheric N2O. The parallel inter-comparison experiment was implemented at JGS for 22 months, and the difference in atmospheric N2O was 0.17 ± 0.9 ppb between the two instruments. The significant differences were observed in the nonlinear range of the GC-μECD. Finally, these differences resulted in the over/underestimation of N2O characteristics locally and seasonally. Overall, OA-ICOS has a more robust performance with a lower measurement uncertainty than GC-μECD. Based on this study, we also suggest a calibration strategy for both instruments to achieve precise N2O measurements.


Introduction
Nitrous oxide (N 2 O) is a third powerful greenhouse gas with 0.202 W·m −2 radiative forcing (in 2019) expressed by global abundance changes relative to 1750, the preindustrial level [1] (The global warming potential of N 2 O is 300 times more than that of carbon dioxide (CO 2 ) over a 100 year time horizon due to its long lifetime [2,3]. N 2 O is also involved in stratospheric ozone depletion [4], and its emissions weighted by ozone depletion potential currently exceed those of all other substances [5].
The major sources of N 2 O are microbial denitrification and nitrification, which lead to N 2 O production in soils [6]; ocean waters [7,8]; and in streams, rivers, and lakes [9,10]. Atmospheric N 2 O increased globally by 1 ± 0.3 ppb yr −1 from 2010 to 2018 [11]. This increment resulted from the increased use of inorganic fertilizers and manure [12][13][14], and those sources are all subject to significant uncertainty. The significant increasing trend in agricultural soil emissions from Southern Asia, which includes China and India, is mostly due to the rise in nitrogen fertilizers in these developing economies [13][14][15].
Recently, to better quantify these emissions, the importance of verifying the bottom-up inventories using top-down constraints provided by atmospheric N 2 O measurements is increasing [15,16]. is linked to the Yellow Sea. Next to JGS is the widest plain in Jeju, where potatoes, garlic, and onions are harvested ( Figure 1). Therefore, when the wind speed is less than 5 m/s under stagnant conditions, high N2O concentrations are observed at the station.

Measurement System
GC-μECD (Agilent 7890A, CA, USA) and OA-ICOS (Los Gatos EP-30, CA, USA) collect ambient air from an intake at 12 m height at JGS. Before the air is injected into the instruments, all the sampled air is dried cryogenically to −50 ℃ [19]. We do not apply the water vapor correction provided by the OA-ICOS manufacturer. Therefore, the water vapor effects on the N2O observations are negligible [21]. The analysis condition for GC-μECD and instrument specification for OA-ICOS are presented in Table 1. Table 1. The analytical conditions for the microelectron capture detector method (GC-μECD) and the performance specification for off-axis integrated cavity output spectroscopy (OA-ICOS).  GC-μECD has been a widely used instrument for atmospheric N2O measurements. Two Porapak-Q columns, used as a precolumn and a main separation column, were plumbed with a 5-mL sample loop through two 6-port gas sample valves (Valco Instrument, TX, USA). Those valves allow automated sample injections, backflush of the precolumn, and the separation of N2O from

Measurement System
GC-µECD (Agilent 7890A, CA, USA) and OA-ICOS (Los Gatos EP-30, CA, USA) collect ambient air from an intake at 12 m height at JGS. Before the air is injected into the instruments, all the sampled air is dried cryogenically to −50°C [19]. We do not apply the water vapor correction provided by the OA-ICOS manufacturer. Therefore, the water vapor effects on the N 2 O observations are negligible [21]. The analysis condition for GC-µECD and instrument specification for OA-ICOS are presented in Table 1. Table 1. The analytical conditions for the microelectron capture detector method (GC-µECD) and the performance specification for off-axis integrated cavity output spectroscopy (OA-ICOS). GC-µECD has been a widely used instrument for atmospheric N 2 O measurements. Two Porapak-Q columns, used as a precolumn and a main separation column, were plumbed with a 5-mL sample loop through two 6-port gas sample valves (Valco Instrument, TX, USA). Those valves allow automated sample injections, backflush of the precolumn, and the separation of N 2 O from other components in the air. Carrier gas is 5% CH 4 in Ar (P5). The main and post-columns for the backflush prevent other ECD-sensitive species such as CFCs (Chlorofluorocarbons) from reaching the detector. Therefore, N 2 O can Atmosphere 2020, 11, 948 4 of 12 be measured without adjusting the oven temperature. This results in a shorter total run time than a conventional valve system [22,23]. The retention time of the N 2 O peak is normally~around 13 min in this system, and the total run time is less than 20 min. Therefore, this method allows three measurements per hour. Due to drift, one of the three samples every hour is air from a reference cylinder (one-point calibration).

GC-µECD (Agilent
OA-ICOS is a recently commercialized instrument. This instrument, which is used in this study, is made by Los Gatos Research (LGR, CA, USA). For OA-ICOS, the laser light enters the high-finesse cavity at a non-zero angle, generating a high density of transverse cavity modes [24]. This off-axis arrangement allows a range of beam incidence conditions to be used, making the analyzers less susceptible to changes in the alignment of the mirrors and allowing sensors to be simpler to operate and more robust for operation in the field. Here, we used the LGR-EP30 with a 434 mL cell, operated at 85 ± 0.1 Torr and 45 ± 0.005 • C. We calibrated the instrument weekly with one standard. When we tested the time for the OA-ICOS output to stabilize after gas switching, the stabilization takes at least 250 s (around 4 min) from high N 2 O (351.09 ppb) to zero air. After 250 s, the deviation ranges between 0.01 to 0.09 ppb for high N 2 O and 0 to 0.02 ppb for zero, respectively. For the laboratory test, OA-ICOS takes from 2 to 3 min, which is similar to this study [18]. Therefore, we carried out the experiments for linearity, repeatability, and reproducibility or calibration for OA-ICOS after the sample cell was flushed enough for 10 min.
We used tertiary standards on the WMO-X2006A N 2 O scale certified by the National Oceanic and Atmospheric Administration (NOAA), as in the WMO/GAW/CCL. The scale was propagated with an uncertainty of ±0.09 ppb at the 68% confidence level [25].
All data were collected and stored at the KMA/National Institute of Meteorological Sciences (NIMS) in Jeju, South Korea. Our data selection was implemented in two steps, automated (AQC) and manual (MQC), according to the instrument. (1) AQC: Here, the measurements were automatically flagged for GC-µECD when N 2 O followed an outlier over the range 0-600 ppb, the difference in retention times of consecutive N 2 O measurements was more than 1 min, and the ratio of the consecutive sample peak area was less than 0.8. For OA-ICOS, auto-flagging occurred when N 2 O was an outlier, as with GC-µECD, and measured cell pressure and temperature were outside the acceptable ranges provided by the manufacturer. (2) MQC: For both instruments, the manual flags were assigned by technicians at JGS according to the logbook based on the inlet filter exchange, diaphragm pump error, low flow rate, dehumidification system error, and calibration periods, such as participation in comparison experiments [19]. After these flagging steps, the data were designated Level 1 (L1). Data with flags were reviewed by scientists at NIMS, and valid data were selected as Level 2 (L2). We used L2 hourly data for the continuous comparison experiment in Section 4.2.

Linearity
The test is based on the premise that responses of the instrument change linearly with changes in the abundance of target species, or, to reiterate, the drift-corrected instrument response is the same within the range of the N 2 O of interest (313.85 to 351.09 ppb).
The GC-µECD peak areas of four standards were normalized due to the instrument drift based on injections from a reference gas between aliquots of standard gas. Therefore, an A-B-A' sequence was used for this test, where A was a tertiary standard and B was a reference gas to correct the drift. However, the OA-ICOS did not need the drift correction, since the drift in an hour was not significant (see Section 4.2).
The uncertainty of linearity is defined as the residuals from the regression curve and calculated by the equation below: Atmosphere 2020, 11, 948

of 12
Here, N is the number of cylinders (here, 4 cylinders), and R i the residuals of each cylinder. R i is the difference of the laboratory standard minus the value from the regression curve in Figure 2. Such differences were observed in past inter-comparison experiments through audits implemented by the Swiss Federal Laboratories for Materials Science and Technology (Empa) [26] and the Round Robin Test hosted by WMO/GAW/CCL with travelling cylinders, which are traceable to the WMO scale [25]. In those inter-comparison experiments, one-point calibration was used, which cannot cover the target range, and hence, it showed a significant difference in the experiments. The nonlinear response of the GC-μECD is well-known [23,25], and this could bias atmospheric measurements. Therefore, nonlinearity should be assessed when developing a calibration strategy to monitor atmospheric N2O using GC-μECD.
For OA-ICOS, we sampled one standard and used the calibration function provided by the manufacturer for the linearity test. We used 335 ppb for this calibration, and this covered the range of 310 ppb to 345 ppb. However, this can be out of range of the WMO/GAW compatibility when N2O is more than 345.7 ppb, as indicated by the increasing residuals as the N2O increased. This suggests that the monitoring stations in the regions that observe a wide range of atmospheric N2O should conform to the calibration function provided by the manufacturer with a one-point standard that covers the anticipated range of atmospheric N2O.
The Ur from OA-ICOS is 0.35 ppb. For GC-μECD, the Ur from the linear function is 0.85 ppb; however, we confirmed that it can decrease to 0.48 ppb when the second-order polynomial regression curve is used to describe the detector response to N2O (Table 2). To decrease the Ur, bracketing standards are recommended, at least when using GC-μECD.

Repeatability
Repeatability (U p ) is determined from the standard deviation (1σ) of the same standard gases in an hour and expressed by a pooled standard deviation: where S i is the standard deviation of 1 h averages of the same laboratory standard measurements, N i the number of data over 1 h (based on 10 s intervals for OA-ICOS and 3 peaks for GC-µECD), and N t is the total number of cylinders: 4. While testing the linearity for two instruments, we implemented the repeatability test simultaneously.

Reproducibility
Long-term reproducibility (U L ) was defined as the drift with the surveillance cylinder during 1 week with the same N 2 O-level standard gas and tracing the values for two months for both GC-µECD and OA-ICOS. Since positive and negative values were found for U L , we used the following equation: Atmosphere 2020, 11, 948 6 of 12 where x is the drift between weekly injections, and N is the total number of data (weekly data for two months).

Measurement Uncertainties
GC-µECD was nonlinear in the target range, meaning that one-point calibration was not sufficient, while OA-ICOS showed a linear response against the reference gases ( Figure 2). Even if a second-order polynomial was applied as the regression curve for GC-µECD, the residual (differences between the theoretical value from the function and observed values from the instrument) was still more than the WMO/GAW compatibility goal (the horizontal shaded area in Figure 2) in a certain range.
Such differences were observed in past inter-comparison experiments through audits implemented by the Swiss Federal Laboratories for Materials Science and Technology (Empa) [26] and the Round Robin Test hosted by WMO/GAW/CCL with travelling cylinders, which are traceable to the WMO scale [25]. In those inter-comparison experiments, one-point calibration was used, which cannot cover the target range, and hence, it showed a significant difference in the experiments. The nonlinear response of the GC-µECD is well-known [23,25], and this could bias atmospheric measurements. Therefore, nonlinearity should be assessed when developing a calibration strategy to monitor atmospheric N 2 O using GC-µECD.
For OA-ICOS, we sampled one standard and used the calibration function provided by the manufacturer for the linearity test. We used 335 ppb for this calibration, and this covered the range of 310 ppb to 345 ppb. However, this can be out of range of the WMO/GAW compatibility when N 2 O is more than 345.7 ppb, as indicated by the increasing residuals as the N 2 O increased. This suggests that the monitoring stations in the regions that observe a wide range of atmospheric N 2 O should conform to the calibration function provided by the manufacturer with a one-point standard that covers the anticipated range of atmospheric N 2 O.
The U r from OA-ICOS is 0.35 ppb. For GC-µECD, the U r from the linear function is 0.85 ppb; however, we confirmed that it can decrease to 0.48 ppb when the second-order polynomial regression curve is used to describe the detector response to N 2 O (Table 2). To decrease the U r , bracketing standards are recommended, at least when using GC-µECD. Table 2. Estimated uncertainty factors in the measurements of N 2 O from each instrument. Units are ppb. All terms have a 68% confidence interval. The values in brackets are derived from the GC-µECD applied by the second-order polynomial regression curve. U r : uncertainty of linearity, U p : repeatability, and U L : long-term reproducibility.

Uncertainty Factors
Analyzer For the gap between the minimum and maximum standard deviations over an hour, the value from the GC-µECD (0.20 ppb to 0.36 ppb) was almost five times greater than OA-ICOS (0.02 ppb to 0.04 ppb), while U p was ±0.26 ppb and ±0.03 ppb for the GC-µECD and OA-ICOS, respectively ( Table 2). The previous study reported that U p derived from the GC-µECD was less than 0.4 ppb with a 20 min injection of standard gases [27] and 0.1 ppb with a 1 h average [18]. The short-term continuous measurement repeatability incorporating the GC-µECD was 0.1 ppb to 0.3 ppb [28][29][30]. The U p from OA-ICOS was similar to that of the previous study at ±0.05 ppb with a model EP-40 in the lab test [18].

OA-ICOS GC-µECD Relevant Equation
The U L from OA-ICOS and the GC-µECD were ±0.46 ppb and ±1.19 ppb, respectively ( Table 2). The long-term drifts for 10 days from OA-ICOS with models EP38 and EP40 were reported as ±0.76 ppb Atmosphere 2020, 11, 948 7 of 12 and ±0.31 ppb, respectively, inside the laboratory [18]. Considering that our model is different from the reference and long-term drift is defined as one week, the U L in this study is at a similar level to that of previous studies. On the other hand, the U L is behind the WMO/GAW compatibility goal for N 2 O, suggesting that the calibration frequency should be adjusted. When we track the drift everyday with the same cylinder over a week, the drift increases to ±0.23 ppb after three days. Note that a previous study suggested twice a day [18], such that the estimation of the U L is necessary for deciding the calibration frequency of OA-ICOS. For the GC-µECD, the other study showed that the drift during two-four days was 0.2 ppb to 0.3 ppb [28], which are smaller than the value obtained from this study. It was suggested that the U p and U L from the GC-µECD can be decreased to 0.05 ppb (1σ) and 0.1 ppb (1σ), respectively, under ideal conditions with no other factor influencing those values other than the electronic noise [25]. However, the GC-µECD in a field cannot be controlled only by electronic noise. At the JGS, 23.3 ± 0.92 and 1008.5 ± 7.7 hPa in a year, the GC-µECD can be affected by these laboratory conditions. The variation of the GC-µECD sensitivity is related to variations in the temperature and pressure in the laboratory, indicating that 2.15% of sensitivity variations, with ±1.5 of temperature and ±10 hPa of pressure, respectively [31].

Inter-Comparison Experiment between OA-ICOS and the GC-µECD
As explained in Section 2.2, we applied one-point calibration for both instruments. The frequency of the reference gas injection to correct the drift was 1 h for the GC-µECD and a week for OA-ICOS.
We implemented parallel measurements on the two instruments for 22 months, with measurements every 10 sec for OA-ICOS and two peaks for the GC-µECD, respectively, at the JGS and the compared the hourly data (see Section 2.2). N 2 O from the two instruments showed R = 0.8. The value from the GC-µECD was lower than that from OA-ICOS during the experimental period, as indicated by the mean difference (OA-ICOS minus the GC-µECD) of 0.17 ± 0.9 ppb (Figure 3). This mean value was in the WMO/GAW extended the compatibility goal of ±0.3 ppb, though the standard deviation was quite large.
Atmosphere 2020, 11, x FOR PEER REVIEW 8 of 12 The differences might be related to the calibration methods for each instrument. As we described above (UL), the calibration frequency should be shorter than a week for OA-ICOS. The difference between the two instruments increased (up to around 3 ppb) for N2O by less than around 332 ppb, mainly. This might be derived from the nonlinear characteristic of the GC-μECD, while OA-ICOS performed well over the same range, as seen in Figure 2. Since the level of standard gas The differences might be related to the calibration methods for each instrument. As we described above (U L ), the calibration frequency should be shorter than a week for OA-ICOS. The difference between the two instruments increased (up to around 3 ppb) for N 2 O by less than around 332 ppb, mainly. This might be derived from the nonlinear characteristic of the GC-µECD, while OA-ICOS performed well over the same range, as seen in Figure 2. Since the level of standard gas was 334 ppb for the one-point calibration of the GC-µECD, as described in Section 3.2, a nonlinearity of less than 332 ppb might not be corrected. Therefore, it was confirmed that the calibration strategy is important for the precise measurements. For the GC-µECD, it also could be affected by laboratory conditions such as the variations of temperature and pressure that cause the differences from OA-ICOS.
When we compared the standard deviation of an hour with the observed ambient N 2 O during the period, the median value of OA-ICOS was 0.09 ppb, while for the GC-µECD, it was 0.25 ppb. Even though the same ambient samples were sampled by both instruments, the standard deviation from OA-ICOS was lower than that from the GC-µECD. This is reflected in the U p factor from the instrument and the number of samples under the same condition, indicating that the precision of OA-ICOS was better.

Enhancement N 2 O Values
The enhancements of N 2 O (exN 2 O) observed by the GC-µECD and OA-ICOS were compared to see the local characteristics in each season. The exN 2 O was expressed as the difference from the zonal mean over 17.5 • N to 53.1 • N (temperate Northern Hemisphere, TNH) observed by the NOAA (Figure 4). This is to see local and seasonal variations of the JGS N 2 O as removing, representing the zonal growth rate and seasonal characteristics from 17.5 • N to 53.1 • N equally from the values measured by the two instruments. The amplitude of the TNH is~0.5 ppb, with the lowest in August and highest in March. Even when this amplitude is considered, the JGS is almost always higher than the TNH.
Atmosphere 2020, 11, x FOR PEER REVIEW 9 of 12 lowest in August and highest in March. Even when this amplitude is considered, the JGS is almost always higher than the TNH. The JGS monthly mean was calculated as follows: (1) to select for the background conditions at the JGS, we considered the differences between the consecutive hourly averages of the L2 data, which was described in Section 2.2 as less than 0.5 ppb (1σ). (2) The selected hourly data were used to calculate the monthly means with the method by [32] to represent the regional baseline, thereby reducing the noise due to synoptic-scale atmospheric variability and measurement gaps.  The exN2O from OA-ICOS in the winter (December to February) is similar or lower than the monthly mean of the TNH through 2018 to 2019, even though the data for two months, from January to February of 2018, were missing (Figure 4, left). This might result from seasonal wind characteristics and local agricultural activities. The JGS could collect well-mixed air in the winter rather than in the other seasons due to a strong Siberian high from the continent. As described in Section 2, there is a large plain around the JGS and Jeju that raises two crops a year due to its warm weather. Therefore, high N2O episodes were observed, regardless of the season. Those environments The JGS monthly mean was calculated as follows: (1) to select for the background conditions at the JGS, we considered the differences between the consecutive hourly averages of the L2 data, which was Atmosphere 2020, 11, 948 9 of 12 described in Section 2.2 as less than 0.5 ppb (1σ). (2) The selected hourly data were used to calculate the monthly means with the method by [32] to represent the regional baseline, thereby reducing the noise due to synoptic-scale atmospheric variability and measurement gaps.
The exN 2 O from OA-ICOS in the winter (December to February) is similar or lower than the monthly mean of the TNH through 2018 to 2019, even though the data for two months, from January to February of 2018, were missing (Figure 4, left). This might result from seasonal wind characteristics and local agricultural activities. The JGS could collect well-mixed air in the winter rather than in the other seasons due to a strong Siberian high from the continent. As described in Section 2, there is a large plain around the JGS and Jeju that raises two crops a year due to its warm weather. Therefore, high N 2 O episodes were observed, regardless of the season. Those environments allow atmospheric N 2 O to be greater than the TNH in the spring to autumn, while a similar level of N 2 O to the TNH was observed in the winter.
However, the characteristics of exN 2 O from the GC-µECD are different each year, as shown by the N 2 O decrease in the summer of 2018 (even lower than the TNH) and winter of 2019. This contrasts the pattern from OA-ICOS.
On the other hand, the difference in the annual mean was 0.6 ppb between the two instruments (OA-ICOS minus the GC-µECD). The differences from the monthly mean were observed continuously from 2018 to 2019. It increased when the monthly N 2 O level decreased, and, especially, significant differences occurred for N 2 O of less than 322 ppb (June to August 2018), which was quite similar to the parallel comparison, indicating the bias reflected the monthly variation. For example, the GC-µECD N 2 O was similar to the TNH N 2 O in the summer of 2018, and this level might be underestimated compared to the values from OA-ICOS due to the nonlinear characteristic of the GC-µECD.
Therefore, the measurement uncertainty described in Section 4.1 finally resulted in the over/underestimations of large-scale characteristics of atmospheric N 2 O.

Conclusions
To understand the sources and sinks of atmospheric N 2 O and to reduce the uncertainty in the inverse assessments based on atmospheric observations, precise measurements are important. In this study, we investigated the conventional GC-µECD and the advanced OA-ICOS analytical methods for atmospheric N 2 O measurements.
Overall, the linearity, repeatability, and reproducibility tests showed that the OA-ICOS has a more robust performance than that of the GC-µECD in the in-situ station. Since these factors affect the measurement uncertainty, the calibration strategy should be well-established to reduce it.
For the OA-ICOS, one-point calibration is sufficient to monitor the atmospheric N 2 O. However, the calibration should be implemented on a less-than-weekly basis. When the drift was tested by injecting the same cylinder every day, the value showed that the drift increased to ±0.23 ppb after three days (WMO/GAW compatibility goal, ±0.1 ppb and extended goal, ±0.3 ppb). This means that the calibration frequency should be shorter than three days. Note, however, the environment is different according to the in-situ stations, so that the linearity, repeatability, and reproducibility tests should be performed at each station prior to the routine measurements for high-quality measurements.
It is suggested that the GC-µECD needs at least two reference cylinders to set the target levels within an hourly calibration. This is to correct its nonlinear characteristics and large drifts, even within an hour. In this study, we used one reference cylinder. Since three peaks are observed in an hour, if we apply the two-point calibration method, we can get only one atmospheric data among the three. According to the uncertainty factors, the value derived from reproducibility was greater than the other factors not only for the GC-µECD but, also, for OA-ICOS, indicating that the drift correction is the most important. Therefore, we put the priority on the drift correction. This calibration strategy resulted in biases in the nonlinear range, as described in Sections 4.2 and 4.3. Therefore, for the GC-µECD, both drift and nonlinear range corrections are necessary to get more accurate data.
For both instruments, the uncertainty value was high in the order of the U L > U r > U p . This means that, when we adjust the calibration frequency, we can reduce the total measurement uncertainty. For the GC-µECD, however, even after correcting the short/long-term drift with an hourly reference gas injection, the uncertainty from linearity resulted in the bias of the quasi-continuous measurements for atmospheric N 2 O.
Hourly L2 N 2 O data were compared between two instruments for 22 months, and the difference was 0.17 ± 0.9 ppb, with a correlation R of 0.8. When the monthly exN 2 O were compared, OA-ICOS showed a constant seasonal pattern, while the GC-µECD varied year by year. Even though the two instruments observed the same atmospheric N 2 O, the hourly data and monthly background levels were quite different, especially in the range related to the nonlinear characteristics of the GC-µECD. Finally, we confirmed that the seasonal and regional background levels can be underestimated or overestimated because of the instrument performance.
Through this paper, the OA-ICOS showed better performance in monitoring atmospheric N 2 O, with lower measurement uncertainty in the in-situ station, especially in regions with large fluctuations that reflect not only synoptic conditions but, also, local activities. Additionally, the appropriate calibration method for each instrument is strongly recommended for the high-quality measurements data.