Stratospheric Air Sub-sampler (SAS) and its application to analysis of Delta O-17(CO2) from small air samples collected with an AirCore

. We present the set-up and a scientiﬁc application of the Stratospheric Air Sub-sampler (SAS), a device to collect and to store the vertical proﬁle of air collected with an AirCore (Karion et al., 2010) in numerous sub-samples for later analysis in the laboratory. The SAS described here is a 20 m long 1/4 inch stainless steel tubing that is separated by eleven valves to divide the tubing into 10 identical segments, but it can be easily adapted to collect smaller or larger samples. In the collection phase the SAS is directly connected to the outlet of an optical analyzer that measures the mole fractions of CO 2 , CH 4 and CO from an AirCore sampler. The stratospheric part (or if desired any part of the AirCore air) is then directed through the SAS. When the SAS is ﬁlled with the selected air, the valves are closed and the vertical proﬁle is maintained in the different segments of the SAS. The segments can later be analysed to retrieve vertical pro-ﬁles of other trace gas signatures that require slower instrumentation. As an application, we describe the coupling of the SAS to an analytical system to determine the 17 O excess of CO 2 , which is a tracer for photochemical processing of stratospheric air. For this purpose the analytical system described by (Mrozek et al., 2015) was adapted for analysis of air directly from the SAS. The performance of the coupled system is demonstrated for a set of air samples from an Air-Core ﬂight in November 2014 near Sodankylä, Finland. The standard error for a 25 mL air sample at stratospheric CO 2 mole fraction is 0.56 ‰ (1 σ ) for δ 17 O and 0.03 ‰ (1 σ ) for both δ 18 O and δ 13 C. Measured (cid:49) 17 O(CO 2 ) values show a clear correlation with N 2 O in agreement with already published data.


Introduction
Monitoring and studying the distribution of greenhouse gases throughout the atmospheric column is an important constituent of understanding contemporary climate change.Carbon dioxide (CO 2 ) is the most important molecule of the atmospheric carbon cycle and the increase of its abundance in the atmosphere is the primary factor of recent radiative forcing (IPCC, 2013).In the stratosphere the 17 O excess of CO 2 (expressed as 17 O(CO 2 )) is a valuable long-lived tracer for stratospheric chemistry and atmospheric circulation patterns.Measurement of the mole fraction and oxygen isotopic composition of CO 2 provides information on both transport times and photochemical lifetimes of CO 2 (Boering et al., 2004;Wiegel et al., 2013).Over the past decades several sampling campaigns have been carried out to observe and to understand the oxygen isotope enrichments of stratospheric CO 2 (Thiemens et al., 1995;Alexander et al., 2001;Lämmerzahl et al., 2002;Boering et al., 2004;Kawagucci et al., 2008;Wiegel et al., 2013).Unfortunately, measurement campaigns with rockets, balloons and aircraft platforms are expensive and therefore it is in general difficult to obtain stratospheric samples.
Published by Copernicus Publications on behalf of the European Geosciences Union.
The simple and lightweight sampling system AirCore (Karion et al., 2010) provides new opportunities to sampling the high altitude atmosphere at relatively low cost.The Air-Core device consists of a long (usually 100 m of longer) piece of coiled stainless steel tubing that is lifted to the stratosphere on a balloon with one end open and the other end closed.During ascent, the AirCore empties because of the decrease in pressure; during descent, ambient air successively fills the AirCore coil again as pressure increases.The atmospheric profile information in the coil is preserved because of limited gas diffusion inside the long tube.This means that the altitude profiles of various trace gases such as CO 2 , CH 4 and CO can be determined by processing the content of the AirCore through a fast analytical system quickly after recovery of the sampler.
Previously, the air from an AirCore was vented after analysis with a real time gas analyzer and not used for other, more sophisticated and slower, analyses.We developed a Stratospheric Air Sub-sampler (SAS) that collects (the stratospheric fraction of) air from an AirCore directly after online analysis and stores it in different segments of the SAS so that the profile is preserved.The SAS can be easily transported and processed, for example for the isotopic composition of trace gases or for halocarbon analysis.The limitation of the method is that the analytical system must be capable of analysing very small air samples, since the total stratospheric fraction of the AirCore profile is only of the order of 250 mL at ambient temperature and pressure, which is split into multiple segments in the SAS.
For the work presented here we apply the SAS concept to measurement of the 17 O excess of CO 2 in the stratospheric sub-samples.The analytical system described in Mrozek et al. (2015) was modified to allow air from the SAS to be flushed directly by the reference air into a new sample introduction unit at ambient pressures.The change of the oxidation reagent from CeO 2 powder to CuO wires reduced peak broadening and allowed detection of the equilibrated CO 2 peak without focusing on liquid nitrogen trap.The improved system is fully automated; only the change of the individual SAS segments is made manually.The successful coupling of the SAS and the 17 O analysis system was demonstrated by CO 2 stable isotope measurements on stratospheric air obtained from an AirCore flight near Sodankylä, Finland, in November 2014.

Method
The stratospheric air samples are collected in a two-step procedure.First, air samples from the surface up to the ceiling altitude of a stratospheric balloon flight (typically 30 km) are obtained with an AirCore system.Second, the stratospheric part of the collected air samples is recovered into an SAS after online analysis of trace gas concentrations.The details of the first step (Chen et al., 2016) are here briefly described in Sect.2.1.Section 2.2 and 2.3 provide the information relevant to the stratospheric air samples: the sub-sampling method and the description of the analytical system used for CO 2 isotope measurement.We include a schematic illustration for describing all the instruments and procedure in Fig. 1.

Stratospheric air sampling with AirCore
The AirCore device is made of stainless steel tubing coated with SilcoNert ® 1000 and has a total length of 100 m (40 m at 1/4 inch and 60 m at 1/8 inch, with a wall thickness of 0.01 inch).The payload that includes the AirCore and a radiosonde (Vaisala, type RS92-SGPL) weighs about 3.6 kg.Before each flight the AirCore is filled with a standard dry fill-gas with known CO 2 , CH 4 and CO mole fractions (CO 2 = 386.10± 0.09 ppm; CH 4 = 1880 ± 2 ppb; CO = 7972 ± 5 ppb) and closed with a shut-off valve at the inlet (Swagelok, part number SS-1GS4).A stratospheric balloon (Totex, type Tx3000) is used to launch the payload high into the stratosphere near Sodankylä, Finland.Just before launching the AirCore, the shut-off valve is opened, so that during the ascent the fill-gas leaves the AirCore coil due to the drop in pressure.After reaching an altitude of approximately 30 km, the balloon bursts, and the descent of the payload on a parachute begins.Ambient air flows into the Air-Core and the air from higher altitudes is continuously compressed and pushed towards the closed end of the AirCore by air from lower altitudes.The shut-off valve is closed automatically about 10 s after the landing, and the AirCore is quickly recovered and transported to the Finnish Meteorological Institute (FMI) laboratory for analysis.

Sub-sampling into SAS
The vertical profiles of CO 2 , CH 4 and CO mole fractions are measured within 2-3 h after landing of the AirCore at FMI using a gas analyzer (Picarro, model G2401), as presented in Fig. 2. The originally closed end of the AirCore coil is connected to the analyzer, while the end with the shut-off valve is connected to the fill-gas cylinder.By opening both ends of the AirCore, the air in the AirCore is flushed into the analyzer by the fill-gas.The flow rate is controlled at the Atmos.Meas. Tech., 9, 5607-5620, 2016 www.atmos-meas-tech.net/9/5607/2016/outlet of Picarro instrument and set to 38.2 mL min −1 .Calibration air can be analysed before and after measurement of the AirCore air.After the trace gas analysis, the top (i.e.stratospheric) part of the air collected with the AirCore (less than 20 % of the total collected air) is transferred into an SAS that is sent to Utrecht University for CO 2 isotope analysis.
The timing of the sub-sampling process is based on the flow rate of the air through the analyzer and the pump.The timing of the sub-sampling procedure has to be precisely established to ensure that the desired fraction of gas is collected with the SAS.To establish the timing for the sub-sampling procedure, we injected a spike of highly CO 2 -enriched gas www.atmos-meas-tech.net/9/5607/2016/Atmos.Meas.Tech., 9, 5607-5620, 2016 into the inlet of the Picarro analyzer and then connected the outlet of the pump to the inlet of the Picarro analyzer.Thus a closed loop without the SAS was established, and the CO 2 spike was measured multiple times as it circulated through the analyzer and the pump.This experiment allowed us to determine the timing of air travelling from the inlet of the analyzer to the inlet of the SAS.The flow rate was measured with a flow meter, and the timing for the sub-sampling procedure was established.Also, the membrane pump (Picarro Inc.) that is used for the sub-sampling was carefully tested for leaks to avoid contamination of stratospheric air with ambient laboratory air during the sub-sampling process.The outlet of the pump was connected to the gas analyzer (Picarro Inc., model G2401-m) to form a closed loop, which was filled with air with high mole fractions of CO 2 , CO and CH 4 .The flow rate was set to 35 mL min −1 and the air enclosed in this closed system was circulated nine times while measuring the CO 2 , CO, CH 4 and H 2 O mole fractions.The CO 2 mole fractions stayed between 940.4 and 941.0 ppm and a small initial fluctuation quenched after nine cycles.The rate of change of CO 2 mole fraction based on these measurements with a very high CO 2 mole fraction difference between the sample and ambient air was 0.1 ppm of CO 2 per minute.The effect of such a small contamination on the isotopic composition of CO 2 is therefore negligible (< 0.001 ‰ given the maximum 17 O(CO 2 ) of 5 ‰; see below).We note, however, that CO gets contaminated in the sub-sampling process at a rate of 17 ppb min −1 for CO (from a starting value of 500 ppb), so CO measurements would be compromised.
The SAS used for the CO 2 isotope measurements in Utrecht is made of 10 2 m long pieces of 1/4 inch diameter stainless steel tubing, which are connected by 11 Swagelok valves (part number SS-3CXS4) to form the 20 m long SAS.The tubes are bent to form identical rings to facilitate easy handling and transport.In the following, we refer to the rings as "SAS segments" and the valves as "three-way valves".The three-way valves are open and connect the segments when the stratospheric part of the AirCore air fills the SAS and are kept closed when the sub-sampling process is finished.Each SAS segment contains about 25 mL of the Air-Core air.Segment 1 corresponds to the highest altitude of the AirCore flight, and segment 10 contains air from the lowerstratosphere.
The idea of the SAS that is described here in detail has already been successfully implemented to enable the radiocarbon analysis of stratospheric CO 2 at the Centre for Isotope Research in Groningen, the Netherlands (Paul et al., 2016).
2.3 Continuous flow system to measure the isotopic composition of CO 2 The analytical system to measure the isotopic composition (δ 13 C, δ 18 O and 17 O) of CO 2 in each SAS segment is based on the principle presented in Mrozek et al. (2015).The isotopic composition is measured after gas chromato-graphic separation of the CO 2 from other air constituents.
The 17 O content cannot be determined directly from the measurement at m/z = 45, because of the isobaric interferences of 13 C and 17 O.Therefore, the 17 O content is obtained from isotope measurements on CO 2 before (PreCO 2 ) and after oxygen isotope exchange (PostCO 2 ) with a large reservoir of oxygen.Instead of using cerium (IV) oxide (CeO 2 ) as exchange material (Assonov and Brenninkmeijer, 2001;Mrozek et al., 2015), we use in our new system copper oxide (CuO) (Kawagucci et al., 2005).A single measurement requires two independent injections of 1 mL of air: one for direct measurement (PreCO 2 ) and one for measurement after oxygen isotope exchange (PostCO 2 ), plus 0.7 mL for flushing the injection lines (all at 1 bar pressure).To reduce the measurement uncertainty statistically we perform multiple measurements on one air sample.
In addition to the requirement of overpressure in the injection unit, another disadvantage of the method presented in Mrozek et al. (2015) was the severe peak broadening that was introduced by the strong flow resistance of the CeO 2 powder in the oxygen exchange unit, which required re-focusing of the CO 2 after equilibration.In the new system the powdered CeO 2 in the oxygen exchange unit was replaced with CuO wires.As a result the peak broadening was dramatically reduced by a factor of 7.5 (from 450 to 60 s) and the equilibrated CO 2 could be analysed without re-focusing.In addition, we developed a custom-made sample injection unit for samples provided by the SAS, where both sample and reference gas are injected via the SAS.
The improved continuous flow isotope ratio mass spectrometry analytical system (CF-IRMS) analytical system consists of: a sample injection unit to attach the SAS segments, a gas chromatographic column to separate the CO 2 from a 1 mL air aliquot, an oxygen isotope exchange unit for CO 2 equilibration with CuO and open split interface and an IRMS for isotope measurement (Fig. 3).Similar to the method described by Mrozek et al. (2015), three 6-port, 2position Valco valves (VICI, C6UWM) direct the gas flows through the analytical system.The four main components are described in the following subsections.

Sample injection
The 10 individual segments of the SAS are measured separately (see Sect. 2.4.1).After connecting a segment, the volume between the mass flow controller (MFC) and the segment is evacuated with a low vacuum (LV) pump and filled with reference air.All connections in the injection sub-unit are made of 1/8 inch o.d.stainless steel tubing.When the valves of a connected SAS segment are opened, as demonstrated in Fig. 3, the reference air flushes the stratospheric air sample from the SAS into the analytical system.The flow rate of the MFC is set to 1 mL min −1 during sample gas injection.We use the reference air itself as the carrier for the sample air.
Atmos.Meas.Tech., 9, 5607-5620, 2016 www.atmos-meas-tech.net/9/5607/2016/A single 17 O(CO 2 ) measurement requires two independent injections of the sample gas.The first injection is used for direct isotope measurement of CO 2 (PreCO 2 ), the second injection is for measuring the isotopically equilibrated CO 2 (PostCO 2 ).We inject the gas (reference air or sample air) into the GC column via a 1 mL sample loop in V1.The sample loop is always filled to ambient pressure because it is open to the outside air via the vent.We use extended flushing during sample loop loading (the 1 mL sample loop is flushed with the sample gas at a flow rate of 1 mL min −1 for 80 s) to avoid interference from outside air that may diffuse through the vent tubing into the sample loop.
It takes less than 1 h to measure an individual SAS segment, however, in the present set-up reference air is measured for several hours between the different segments (see Sect. 2.4.2).Usually two SAS segments are measured in one day.The CF-IRMS system operates fully automated and the only manual step required is the connection of the segments.

CO 2 separation from air
This sub-unit consist of a gas chromatography (GC) capillary column (ParaPLOT Q 25 m × 0.53 mm, Varian).As a carrier we use helium at a flow rate of 4 mL min −1 , supplied to the GC through one of the ports in valve V1.The GC is kept inside a heated stainless steel box at 40 • C to ensure uniform conditions for gas separation.On the GC column the differ-ent air constituents are separated.The air peak (mainly O 2 and N 2 ) elutes at 120 s after sample injection, CO 2 at 160 s and N 2 O at 190 s.The air peak leaves the analytical system through the vent in valve V3.Both CO 2 and N 2 O are directed either to IRMS or to the isotope exchange sub-unit.Nitrous oxide in our system does not interfere with CO 2 during the isotope ratio measurement because it is fully separated from CO 2 and gets destroyed in the CuO oven.This is discussed in detail in Sect.3.1.

Oxygen isotope exchange with CuO
This sub-unit consists of Valco valve 2 (V2) and an optional oxygen equilibration oven, the same as (Mrozek et al., 2015).The Valco valve V2 directs the CO 2 aliquot either directly to the IRMS or first into the equilibration oven before entering IRMS.The oven is an assembly of a quartz glass reaction tube (1.4 mm i.d., 3.0 mm o.d., 300 mm length), a tube furnace and a temperature controller.Inside the tube there are oxygenated Cu wires and a Ni catalyst.We refer to this assembly as CuO oven.At a temperature of 900 • C, the wires act as a fast and highly efficient oxygen equilibration medium.This is similar to Kawagucci et al. (2005), who for the same purpose used twisted wires of CuO and Pt catalyst, and different from (Assonov and Brenninkmeijer, 2001) and Mrozek et al. (2015), who used CeO 2 powder.The important advantage of oxygenated Cu / Ni wires over CeO 2 powder www.atmos-meas-tech.net/9/5607/2016/Atmos.Meas.Tech., 9, 5607-5620, 2016 in our system is the much smaller flow resistance.Therefore, the new exchange unit induces a much smaller peak broadening, and the equilibrated CO 2 does not require focusing anymore.As a result, the single analysis time shortened by 250 s in comparison to the method described in (Mrozek et al., 2015) (650 s instead of 900 s) and no liquid nitrogen is required.Before first use, and then on a weekly basis, the CuO oven is conditioned with O 2 (Ultra High Purity (UHP), Air Products) at a flow rate of 20 mL min −1 and a temperature of 600 • C, following the procedure of (Kawagucci et al., 2005).Under these conditions, the copper metal forms a coating of copper (II) oxide on the surface of the Cu wires according to After oxygenation, the IRMS needs at least 50 measurements for the signal to stabilize.During routine measurements we monitor the δ 18 O(CO 2 ) RefAir vs. VSMOW signal before and after equilibration to make sure that the equilibration reaction is quantitative.

IRMS interface and mass spectrometric analysis
This sub-unit of the analytical system (Fig. 3d) consists of a Valco valve 3 (V3), a Nafion ™ dryer, a custom-made open split system (Röckmann et al., 2003) and an IRMS (Thermo Fisher Scientific Delta V Advantage).The loop in valve V3 is used to insert an additional volume of approximatively 1 mL volume (1/4 inch o.d.tube connected with Swagelok fittings) into the flow path, which smooths and ensures a compact shape of the PostCO 2 peak in the IRMS.The Nafion dryer removes traces of water before the gas stream enters the IRMS via the open split system, which is also used to inject the pure CO 2 working gas.The IRMS measures ion current ratios 45/44 and 46/44 that originate from CO 2 isotopologues with masses 44, 45 and 46.

Reference air and reference CO 2
Our reference air cylinder (referred to as RefAir in the following) was filled with tropospheric air collected at an altitude of 20 m from the sixth floor of the Buys Ballot building on the Utrecht University campus in July 2014.The isotopic composition of the CO 2 in the reference air cylinder was calibrated against air cylinders provided by an intercomparison program of the World Meteorological Organization (WMO) and assigned the following isotope values: δ 13 C RefAir/VPDB = −8.09± 0.10 ‰ and δ 18 O RefAir/VSMOW = 41.05 ± 0.20 ‰.Following a mass dependent fractionation relation of tropospheric CO 2 (Kaiser, 2009), we assign δ 17 O RefAir/VSMOW = 21.5 ‰, so that 17 O RefAir = 0.0 ‰.The isotopic composition of the reference air is measured continuously in between the samples to monitor the stability and to correct for long-term trends of the CF-IRMS system.We start the CO 2 isotope analysis always from the SAS segment with the highest segment number, here segment number 10.The common port of valve no.11 (the end of segment 10) is connected to the injection line, which leads to V1, and the free port of valve no. 10 (the beginning of segment 10) to the MFC delivering reference air (see Fig. 3).Note that the beginning of the segment 10 is also the end of segment 9.
The injection lines are open to the laboratory air when the SAS segments are being exchanged.To avoid mixing of the precious sample air with laboratory air we evacuate the volume between the MFC and the SAS segment with the LV pump and flush this volume with reference air (LV pump exchanged to vent).It is important to depressurize the volume behind the MFC before opening the SAS segment.Overpressure in the injection lines must be avoided so that the small air sample is not pushed out of the inlet system.
After connecting an SAS segment as described above, the three-way valve no. 10 is opened, so that reference air starts slowly flowing through segment 10. Next, the threeway valve no.11 is opened, so that air is flushed to the sample loop of the injection system (see Fig. 3).The sample admission procedure is described in detail in the next section.
As the sample air is flushed into the system by the reference air, we simply continue measuring the reference air that is then flowing through the SAS segment.A disadvantage is that the last injections of the sample air are actually a mixture of sample and reference air (see below).The measurements of reference air after the sample measurement are later used for referencing.After the reference air measurements are completed, we close the three-way valves and disconnect segment 10.As the common port of valve no. 10 has to be connected to the injection line when measuring segment 9, segment 10 has to be physically disconnected from the SAS.Segment 9 is then treated the same as segment 10 before.We continue measuring and exchanging SAS segments one by one.For each SAS segment we run a sequence of 35 individual measurements (of two injections each).The results show that 5 of these measurements represent pure sample gas (SA), 10 of them contain sample-reference air mixtures, and 20 are pure reference air (RefAir) measurements (see Fig. 5).Theoretically, we should be able to analyse nine SA aliquots from each SAS segment (25/2.7 mL), but as described above the reference air is used as carrier gas and mixes with the sample air.Since we know the isotopic composition of the reference air, the method can potentially be improved by extracting SA information also from the "mixed" SA/RefAir peaks, which may decrease measurement uncertainty statistically (see Table 1).

Isotope analysis procedure
3 Performance of the continuous flow isotope analysis system

Separation and destruction of N 2 O
Nitrous oxide (N 2 O) interferes with the mass spectrometric analysis of CO 2 because the isotopologues of both CO 2 and N 2 O fall on the mass 44, 45 and 46 collectors and cannot be separated with conventional IRMS instruments (Mook and van der Hoek, 1983).These mass interferences can lead to www.atmos-meas-tech.net/9/5607/2016/Atmos.Meas.Tech., 9, 5607-5620, 2016 Figure 5. Raw-δ 45 CO 2 data (‰ vs RefCO 2 ) from a complete measurement sequence of an SASsegment.SA is the air sample stored in the SAS segment.RefAir before SA refers to the reference air passing through an empty SAS segment and defines IRMS stability before the new SA is connected.After exchanging the SAS segments we obtain five "pure" sample air measurements.Subsequent to the SA measurement, the reference air first mixes into the sample air; it takes about 10 measurements until pure reference air is processed.RefAir after SA refers to the reference air passing through the SAS segment after the SA has been completely flushed out.For the sample air presented here the "before equilibration' points overlap with the "after equilibration" points and cannot be distinguished.The bold values describe the uncertainty for a single measurement and the error expected for the five repeated measurements on one air sample stored in the SAS.SD: standard deviation.
To demonstrate successful GC separation and N 2 O decomposition inside the CuO oven at 900 • C we connected a dilution of 400 ppm N 2 O in synthetic air to the injection sub-unit of our analytical system.The N 2 O enriched gas was placed at the position of a flask, as shown in Fig. 3.The flask was opened and the N 2 O enriched gas filled the injection lines and the 4 mL inner volume of MFC.Next, we changed the position of the three-way valve in front of the MFC so that the reference air was flowing towards the MFC.As a consequence, the reference air mixed with the N 2 O-rich gas and both N 2 O and CO 2 were observed on the chromatogram.We monitored the N 2 O peak before and after the isotope exchange reaction: for the non-heated aliquot, the GC column separated completely the N 2 O from the CO 2 ; for the heated aliquot, the N 2 O was destroyed completely in the CuO oven (see Fig. 6).For the experiment shown in Fig. 6, the chromatogram was extended to 750 s instead of the normal length of 650 s because N 2 O elutes after CO 2 .Additionally, the fifth working RefCO 2 peak was omitted in order to observe the N 2 O peak.The test shows that N 2 O can be effectively removed on CuO / Ni wires at 900 • C and confirms the results of (Kawagucci et al., 2005).This removal method can potentially be applied to other trace gas measurement techniques.

CuO-CO 2 equilibration efficiency
To quantify the efficiency of oxygen isotope equilibration in the CuO oven we analysed two samples containing CO 2 with very different isotopic composition.The first one was our RefAir and the second one was a synthetic mixture of RefCO 2 diluted to 400 ppm CO 2 in synthetic air.The samples were injected via a stainless tube that is similar to a segment of the SAS, but longer (4 m length 1/4 inch o.d.).First, the reference air was injected multiple times through this tube.Next, we filled the injection tube with the RefCO 2 dilution and continued the measurements.Figure 7 presents the results.The isotopic difference between the two CO 2 samples was about 36 ‰ before isotope exchange.After the isotopic exchange reaction both gases were equilibrated to δ 18 O = (19.03± 0.18) ‰.From the difference in the oxygen isotopic composition between RefAir and RefCO 2 dilution before and after oxygen isotope exchange, the oxygen exchange efficiency in CuO oven was calculated to be > 99.5 %.We conclude that the oxygen exchange reaction with CuO/Ni wires at 900 • C is complete.

Error analysis
In order to investigate the long-term stability of our CF-IRMS system and the measurement uncertainty, 540 aliquots www.atmos-meas-tech.net/9/5607/2016/Atmos.Meas.Tech., 9, 5607-5620, 2016 of reference air were injected continuously to the CF-IRMS system for 50 h, comprising 270 individual measurements of the complete isotopic composition of CO 2 .The standard deviation of 17 O(CO 2 ) over all 270 measurements was 1.22 ‰.This is the error that we assign to a single measurement with the new analytical system.In our previously published method (Mrozek et al., 2015), the 17 O(CO 2 ) standard deviation in such a long-term stability test was 1.68 ‰.The improvement compared to the system described in (Mrozek et al., 2015) is due to the replacement of the isotope exchange medium from powdered CeO 2 to CuO wires and through abandonment of a liquid nitrogen trap for re-focusing of the isotopically equilibrated CO 2 .For δ 18 O(CO 2 ) and δ 13 C(CO 2 ) the standard deviation over all 270 measurements was 0.06 and 0.07 ‰, respectively, vs. 0.16 and 0.09 ‰ in (Mrozek et al., 2015).
To investigate how the measurement error is reduced statistically with multiple measurements on an air sample, we divided the 270 measurements from the stability test into different packages of n numbers of measurements (two packages of 135 measurements each, three packages of 90 measurements each, etc.).The experimental standard error (SE) for each set of packages was then compared to the theoretically expected error.Calculations for 17 O(CO 2 ) are shown in Table 1; the results for 17 O(CO 2 ), δ 18 O(CO 2 ) and δ 13 C(CO 2 ) are displayed in Fig. 8. Since the SE is defined as σ/n 0.5 the expected slope on this double logarithmic plot is −0.50.The error reduction follows the theoretically expected relation quite well.When the two groups with only two and three members each are neglected, the linear fit to the data has a slope of −0.46 ± 0.02 for 17 O, −0.42 ± 0.03 for δ 18 O and −0.45 ± 0.03 for δ 13 C. Within the error, this is close to the theoretical slope of −0.50.
We conclude that repeated measurements on one air sample (up to 54 repetition that corresponds to 2.6 µmol CO 2 ) can reduce the uncertainty in 17 O(CO 2 ) to 0.2 ‰.More than 54 repetitions on one air sample improves the 17 O(CO 2 ) uncertainty only marginally.This uncertainty is in the same range as previously reported techniques for large samples, most of which did not allow the measurement of very small samples.(Bhattacharya and Thiemens, 1989) reported an uncertainty of 0.1 ‰ using a BrF 5 -based technique, (Brenninkmeijer and Röckmann, 1998) obtained 0.2 ‰ with a two-step fluorination method, (Assonov and Brenninkmeijer, 2001) reported 0.33 ‰ with a CeO 2 exchange method and (Mahata et al., 2012) improved this method to an uncertainty of 0.12 ‰.The (Kawagucci et al., 2005) method is the only technique that also targeted very small sample sizes (like our system) and they reported an uncertainty of 0.35 ‰.For five repeated measurements on one stratospheric air sample stored in the SAS (an example of a CF-IRMS measurement sequence is given in Fig. 5) we therefore expect an uncertainty of 0.57 ‰ for 17 O(CO 2 ) and 0.03 ‰ for both δ 18 O(CO 2 ) and δ 13 C(CO 2 ).This will be compared to the reproducibility of the actual SAS measurements in Sect.4.2. 4 AirCore flight on 5 November 2014 The AirCore payload was launched about 100 km upwind from Sodankylä, using a meteorological balloon (67.35 • N, 26.93 • E).The AirCore was released from the balloon at 34.5 km altitude (at 4.2 hPa) and the coil filled with air during the balloon descent.Time of launch was 10:04 UTC, while the payload landed at 12:40 UTC; thus the total flight duration was 2 h and 36 min.During the balloon ascent the payload travelled south-east towards Sodankylä due to winds in the stratosphere and troposphere.The payload landed with a parachute 53.8 km east from the analysis laboratory (67.24 • N, 27.84 • E).After landing the AirCore was transported to FMI.The analysis of trace gases started 2 h and 15 min after the landing of the payload.In the laboratory the mole fractions of CO 2 , CH 4 and CO were measured and the Atmos.Meas. Tech., 9, 5607-5620, 2016 www.atmos-meas-tech.net/9/5607/2016/(Foucher et al., 2011).The distribution of the air sample in the SAS segments is proportional to the pressure change during the AirCore descent: here 16 mbar per SAS segment.
stratospheric air was transferred into the SAS.The SAS was sent to the laboratory in Utrecht for CO 2 isotope analysis.

Assigning the trace gas data to the SAS segments
We assigned the trace gas mixing ratios measured with the Picarro instrument to the individual SAS segments based on the flow rate of the carrier gas and the time required to fill the SAS.As the SAS has 10 segments, the air in each segment represents 38.6 s of trace gas measurements with the Picarro instrument.The second piece of information that is necessary for correct assignment is the starting time for the sub-sampling.This starting time was established by taking into account the transfer time of air from the outlet of the Picarro analyser to the inlet of the SAS and the time required to flush out the initial portion of fill-gas from the AirCore.
To ensure that no fill-gas was transferred into the SAS, the top fraction of the stratospheric air (which immediately follows the fill-gas) also had to be discarded.This way we lost the stratospheric air from above 25 km.The timing of subsampling has in the meantime been improved to discard less stratospheric air in future samplings.
Figure 9 shows which part of the stratospheric AirCore air is stored in which segment of the SAS, both as function of altitude and pressure.Note that the SAS segments are equally spaced in pressure change (here 16 mbar/segment) during the descent rather than altitude.For example, the SAS segment 1 contains air from 24.5 to 21.4 km, while the SAS segment 2 contains air from 21.4 to 19.2 km.An analysis of the trace gas profiles from the AirCore flight will be published separately.

4.2
17 O(CO 2 ) analysis of the AirCore air stored in the SAS The stratospheric part of the AirCore air from 5 November 2014 flight was measured in Utrecht between 25 November and 1 December 2014.The results showed that segments 10 and 8 were contaminated with CO 2 from outside air.The precise origin of this contamination could not be determined, but it could have happened when connecting SAS segments to the injection sub-unit of the CF-IRMS system.Due to an accidental instability in the helium carrier gas flow, the air sample from the SAS segment 6 was lost.
As shown in Fig. 5, we were able to perform five repeated CF-IRMS measurements on the other seven stratospheric air samples stored in SAS.The average (1σ ) standard deviation of the raw molecular mass ratios (45/44 and 46/44) of an air sample was 0.04 and 0.02 ‰ for nonequilibrated CO 2 and 0.07 and 0.11 ‰ for equilibrated CO 2 respectively.The main contribution to the measurement error of 17 O is the uncertainty in the isotope ratio 45/44 (Brenninkmeijer and Röckmann, 1998).Based on the reproducibility of m/z 45 of equilibrated and non-equilibrated CO 2 ( (0.04) 2 + (0.07) 2 × 15), the uncertainty in δ 17 O for a single measurement was 1.25 ‰ and for a package of five measurements it was 0.56 ‰ (1.25 ‰/ √ 5).The error of 0.56 ‰ determined for the real AirCore air measurements agrees very well with the uncertainty of 0.57 ‰ expected from the long-term stability test (see Sect. 3.3).The standard deviations for all seven successfully measured SAS segments were 0.08 ‰ for δ 13 C and 0.05 ‰ for δ 18 O.This results in the standard error of 0.03 ‰ for δ 13 C and 0.02 ‰ for δ 18 O for a www.atmos-meas-tech.net/9/5607/2016/Atmos.Meas.Tech., 9, 5607-5620, 2016 package of five repeated analyses from an SAS segment.This is again very similar to the expected error from the long-term stability test as presented in Fig. 8.

17 O(CO 2 )-N 2 O correlation
The mole fraction of N 2 O is a good tracer for the photochemical processing of long-lived trace gases in the stratosphere (Park et al., 2004;Kaiser et al., 2006) In Fig. 10 we show the N 2 O mole fractions deduced from the CH 4 profile vs. the 17 O(CO 2 ) data of the AirCore air samples analysed from the SAS.The linear negative correlation between N 2 O and 17 O(CO 2 ) is well known from previous studies and was discussed before (Boering et al., 2004;Kawagucci et al., 2008;Wiegel et al., 2013).Here, we use this correlation as an additional, independent check on the system calibration.The AirCore air samples agree with the already published data, suggesting that our measurements are well calibrated.The scatter of our data is similar to the variability of the data presented by Kawagucci et al. (2008).The data presented by Wiegel et al. (2013) show smaller variability, but our measurements were obtained from samples of 25 mL of air at ambient pressure only; this is up to 80 times smaller sample size than used in the measurements of Wiegel et al. (2013).The variability of our data around the linear fit is within the analytical uncertainty of 0.56 ‰ (1σ ) per SAS segment.The uncertainty of the N 2 O mole fraction in Fig. 10 reflects the standard deviation of the N 2 O profile stored in an individual SAS segment.It is up to 20 ppb (1σ ) for the high-altitude AirCore air samples where N 2 O showed a strong trend and 2 ppb (1σ ) for lower altitude samples where N 2 O is more constant.The complete CH 4 , CO and CO 2 mole fraction dataset and in-depth analysis on CO 2 isotope measurements from the 5 November 2014 AirCore flight over Sodankylä, Finland, will be published separately.
We conclude that air sampling with an AirCore, followed by sub-sampling of stratospheric air in 10 segments of the SAS and analysis with the analytical system described here provides a relatively cost-effective technique for obtaining vertical profiles of the complete isotopic composition of CO 2 in the stratosphere.

Conclusions
This article describes the concept of the SAS, a device constructed to recover and store the stratospheric part of air obtained with an AirCore sampler.The SAS described here is an assembly of 20 m, 1/4 inch o.d., stainless steel tubing divided into 10 segments.Each SAS segment contains 25 mL of air sample at ≈ 980 mbar.During the procedure of subsampling air into the SAS, the air from an AirCore is measured for CO 2 , CH 4 and CO mole fractions with a Picarro analyzer.This leads to some mixing of air and partial loss of the vertical information, but once stored in the SAS the stratospheric profile of the AirCore air after online analysis is preserved and the individual sub-samples can be supplied to relatively slow analytical instrumentation.The SAS is easy to construct and simple to use, also in the field.
To illustrate the scientific possibilities, the SAS was coupled to a continuous flow analytical system for measurement of the 17 O excess of CO 2 .The air from the SAS was supplied via a new sample introduction system to allow isotope analysis on very small samples.The sample injection occurs at ambient pressure and we use the reference air as the carrier gas.The determination of 17 O of CO 2 is performed by measuring CO 2 before and after complete oxygen isotope exchange with a large oxygen reservoir provided by CuO / Ni.The standard error for a 25 mL air sample at stratospheric CO 2 mole fraction is 0.56 ‰ (1σ ) for 17 O and ‰ (1σ ) for both δ 18 O and δ 13 C.The analytical system operates free of liquid nitrogen because the isotopically equilibrated CO 2 does not require a focusing step, unlike the (Mrozek et al., 2015) method.
The concept of SAS and its coupling to the analytical system for measurement of 17 O was validated through measurements on stratospheric air samples obtained during an AirCore flight over Sodankylä, Finland, in November 2014.
17 O shows the expected negative linear correlation with N 2 O, which provides an independent check on the system calibration.In the future, we plan routine measurements of stratospheric profiles of 17 O together with the mole fractions of CO 2 , CH 4 and CO from AirCore flights on a regular basis, which could expedite the use of the isotope signatures for studying stratospheric circulation patterns and reaction mechanisms.The concept of the SAS will further broaden the scientific questions that can be addressed by AirCore soundings (e.g.Paul et al., 2016).

Figure 1 .
Figure 1.A schematic diagram showing the overall procedure described in this work, from AirCore sampling on the site to IRMS analysis in laboratory.

Figure 2 .
Figure 2. Schematic diagram of the analytical system for trace gas analyses with a Picarro instrument and for transferring the air from the AirCore coil into the Stratospheric Air Sub-Sampler (SAS).The three-way valves in the SAS are in "open to the right" position when the AirCore air is transferred into the SAS and closed after sub-sampling.The crossed circles are conventional valves.The open circles at the inlet of the fill-gas and the calibration-gas cylinder represent cylinder valves and pressure regulators.The black circle represents the shut-off valve at the end of the AirCore coil.The drying tube is filled with magnesium perchlorate (Mg(ClO 4 ) 2 ) from Sigma-Aldrich.

Figure 3 .
Figure 3. Schematic diagram of the CF-IRMS analytical system for complete isotope analysis (including 17 O) of stratospheric CO 2 .The system is divided into four units: (a) sample injection, (b) gas chromatographic column (GC) to separate CO 2 from air and N 2 O, (c) optional oxygen isotope exchange unit and (d) isotope detection unit including open split interface and isotope ratio mass spectrometer (IRMS).MFC is mass flow controller, V1-V3 are Valco valves, RefCO 2 is Utrecht working reference CO 2 , the crossed circles are conventional valves and the blue T's are the three-way valves.The SAS three-way valves are normally closed but here they are shown in the sample admission position.

Each
Valco valve has two possible positions: LOAD and INJECT.The initial Valco valve configuration is shown in Fig. 3: V1 and V3 are in position LOAD and V2 is in position INJECT.After an SAS segment has been connected and the connecting lines flushed as described above, sample admission starts.The MFC is set to a flow rate of 1 mL min −1 Atmos.Meas.Tech., 9, 5607-5620, 2016 www.atmos-meas-tech.net/9/5607/2016/

Figure 4 .
Figure 4. Example of a typical IRMS chromatogram of a single 17 O(CO 2 ) analysis with the CF-IRMS system described here.The eight square peaks are the working reference CO 2 (RefCO 2 ) peaks injected directly via the open split interface.The non-equilibrated CO 2 (PreCO 2 ) peak and the equilibrated CO 2 (PostCO 2 ) peak are the two sample gas peaks.
Figure 4 presents an example of an IRMS chromatogram.A single measurement including two injections takes 650 s.The CO 2 peak from the first injection (PreCO 2 ) is detected between 250 and 280 s, and the equilibrated CO 2 peak from the second injection (PostCO 2 ) between 560 and 640 s.The eight working reference CO 2 peaks (RefCO 2 ) are injected to the IRMS directly via the open split interface: four before detection of the PreCO 2 (0-200 s) and four before detection of the PreCO 2 (300-500 s).The two sample peaks (PreCO 2 and PostCO 2 ) have a different shape because the second one has passed the exchange unit and the additional volume.

Figure 6 .
Figure 6.IRMS chromatogram demonstrating the separation of N 2 O from CO 2 on the GC column and the absence of an N 2 O peak after the hot CuO / Ni isotope exchange unit.The artificially prepared mixture of N 2 O and CO 2 was injected to the analytical system following the procedure described in Sect.2.4.2.In order to observe the N 2 O peak the fifth working RefCO 2 peak was omitted in the chromatogram and the measurement time was extended to 750 s.

Figure 7 .
Figure 7. Efficiency of oxygen isotope equilibration in the CuO oven.The δ 18 O(CO 2 ) values of RefAir and a mixture of RefCO 2 in synthetic air are shown before and after isotope equilibration.The last point of the RefAir measurement sequence after exchange is missing because the peak of the last run was accidentally not registered.

Figure 8 .
Figure 8. Correlation of ln(SE) vs. ln(n) for 17 O(CO 2 ) (based on data from Table 1), δ 18 O(CO 2 ) and δ 13 C(CO 2 ).SE is the standard error of the mean, and n is the number of measurements in a package.The experimental SE of 17 O, δ 18 O and δ 13 C are shown as circles, triangles and black diamonds respectively.The dashed/dotted lines are the theoretically expected slopes for the linear correlation between ln(SE) and ln(n).For a packages of n ≤ 54, the experimentally derived slopes are −0.46 ± 0.02 for 17 O, −0.42 ± 0.03 for δ 18 O and −0.45 ± 0.03 for δ 13 C.The values are close to the theoretical slope of −0.50 (dotted/dashed lines).Based on this experiment we expect an uncertainty of ±0.57‰ for 17 O of an AirCore air sample stored in an individual SAS segment.

Figure 9 .
Figure9.Stratospheric CO 2 mole fraction profile for an AirCore descent near Sodankylä, Finland (67 • N), in November 2014, as a function of altitude (km) and pressure (mbar).The numbers and the various line styles indicate the sub-sampler segments into which the respective part of the air from the AirCore coil was transferred after the trace gas analysis.The general trend of the profile agrees with the already published data(Foucher et al., 2011).The distribution of the air sample in the SAS segments is proportional to the pressure change during the AirCore descent: here 16 mbar per SAS segment.

Figure 10 .
Figure 10.Comparison of the linear negative correlation between 17 O(CO 2 ) and N 2 O from measurements of the AirCore/SAS samples (red rhombuses) presented here with literature data from (Kawagucci et al., 2008) (blue crosses) and (Wiegel et al., 2013) (black triangles).For the AirCore dataset, N 2 O mole fractions were deduced from CH 4 mole fractions as described in the text.The horizontal error bars are a combination of the range of CH 4 mole fractions that is combined within a single SAS segment and the error of the CH 4 -N 2 O transfer function.The vertical error bars represent 17 O(CO 2 ) uncertainty of 0.56 ‰ (1σ ).

Table 1 .
Statistical improvement of 17 O analytical error by combining multiple (n) measurements of the same gas into packages and taking the standard error of these packages.
(Engel et al., 2016)not measured in the AirCore air, but CH 4 was measured and we use the relationship between stratospheric CH 4 and N 2 O mole fractions at high latitudes to translate the CH 4 mole fractions to N 2 O mole fractions.For this purpose we used CH 4 -N 2 O data from two cryosampler flights in the Arctic in 2009 and 2011(Engel et al., 2016).