An analytical system for stable isotope analysis on carbon monoxide using continuous-flow isotope-ratio mass spectrometry

A fully automated system for the determination of δC and δO in atmospheric CO has been developed. CO is extracted from an air sample and converted into carbon dioxide (CO2) using the Schütze reagent. The isotopic composition is determined with an isotope-ratio mass spectrometer (IRMS) technique. The entire system is continuously flushed with high-purity helium (He), the carrier gas. The blank signal of the Schütze reagent is ∼ 4 nmol mol, or 1– 3 % of the typical sample size. The repeatability is 0.1 ‰ for δC and 0.2 ‰ for δO. The peak area allows for simultaneous determination of the mole fraction with an analytical repeatability of ∼ 0.7 nmol mol for 100 mL of ambient air (185.4 nmol mol of CO). An automated single measurement is performed in only 18 min, and the achieved time efficiency (and small volume of sample air) allows for repetitive measurements practically.


Introduction
Carbon monoxide (CO) has an average mole fraction of only ∼ 100 nmol mol −1 (parts per billion or ppb) in the atmosphere, but it has a large yearly turnover of about 2700 Tg (Brenninkmeijer et al., 1999) because of its fast reaction rate with the hydroxyl radical (OH q ).It is produced by numerous sources at the earth's surface and in the atmosphere.About one-third of the atmospheric CO originates from methane oxidation while fossil fuel combustion, biomass burning, and oxidation of non-methane hydrocarbons (NMHC) are other important sources (Brenninkmeijer et al., 1999).The strong latitudinal gradient is a result of the main sources being in the Northern Hemisphere, and the seasonal cycle of CO is largely driven by the seasonality of the OH q (Röckmann et al., 2002).The reaction CO + OH q is the main sink not only for CO but also for OH q , with CO consuming approximately 60 % of OH q in the atmosphere (Crutzen and Zimmermann, 1991).An increase in the CO mole fraction will therefore cause a decrease in the oxidation efficiency of the atmosphere, resulting in a build-up of other gases, such as the long-lived greenhouse gas methane, which are primarily removed by OH q .Consequently, CO is established as an important indirect greenhouse gas in the recent Intergovernmental Panel on Climate Change (IPCC) assessment report (Hartmann et al., 2014).In addition, under high-NO x (mononitrogen oxides) conditions the oxidation of CO leads to the production of ozone, contributing to the build-up of photochemical smog (Westberg et al., 1971).
The isotope ratio, R, is 13 C / 12 C in the case of carbon (δ 13 C) and 18 O / 16 O for oxygen (δ 18 O).For CO, δ values for 13 C and 18 O are usually reported against the international reference materials V-PDB (Vienna Pee Dee Belemnite) and V-SMOW (Vienna Standard Mean Ocean Water), respectively.Since stable isotope variations in nature are small, these δ values are expressed in per mill (‰).
Published by Copernicus Publications on behalf of the European Geosciences Union.

S. L. Pathirana et al.: An analytical system for stable isotope analysis on carbon monoxide
Precise measurements of CO mole fraction and isotopic composition are useful in constraining individual source and sink processes.The combination of the δ 13 C and δ 18 O values gives a distinct isotopic signature for each individual CO source.CO from methane (CH 4 ) oxidation is the most 13 C-depleted source, with δ 13 C values around −50 ‰ (Brenninkmeijer and Röckmann, 1997).δ 13 C values for CO from fossil fuel combustion, biomass burning, and NMHC oxidation range between −27 and −32 ‰ (Manning et al., 1997;Stevens and Wagner, 1989).For CO sources that have a range overlap in δ 13 C values (vehicle emissions range ∼ −36 to −20 ‰, biomass burning range ∼ −25 to −21 ‰, and NMHC oxidation range ∼ −37 to −27 ‰), δ 18 O proves to be a better tracer.Carbon monoxide from vehicle emissions has the highest δ 18 O values, 24 ‰ (Popa et al., 2014;Tsunogai et al., 2003) compared to 7-10 ‰ for biomass burning (Röckmann et al., 1998;Tarasova et al., 2007).For NMHC oxidation no direct measurements are available and a δ 18 O value of 0 ‰ was indirectly derived from isotope budget considerations (Brenninkmeijer and Röckmann, 1997).
Originally, CO isotope analysis was carried out with offline extraction systems, which require large amounts of air (Brenninkmeijer, 1993;Stevens and Krout, 1972), but in recent years, continuous-flow techniques have been developed to accommodate smaller sample volumes (Tsunogai et al., 2002;Wang and Mak, 2010).The method of Brenninkmeijer (1993) required sample sizes of the order of 100 L, whereas the method of Wang and Mak (2010) was optimized for a volume of 0.1 L. There are two conceptually different techniques allowing for isotopic analysis of δ 13 C and δ 18 O of carbon monoxide.One uses the principle of conversion to carbon dioxide (CO 2 ) and subsequent isotope analysis of CO 2 (Brenninkmeijer, 1993;Stevens and Krout, 1972;Wang and Mak, 2010), and the other uses isotope measurement on CO directly (Tsunogai et al., 2002).The technique of converting CO into CO 2 has the advantage as high-precision mass spectrometry is often based on CO 2 , which allows for the possibility of using standardized techniques and isotope calibration scales (Brenninkmeijer et al., 1999).CO is converted to CO 2 using an oxidizing agent, and if the isotopic composition of this oxidizing agent is constant, its effect on the isotopic composition of the CO 2 product can be taken into account.The need to correct for the additional O atom is the weakness of this method.
In the direct method, CO + ion currents at masses 28, 29, and 30 are monitored simultaneously (Tsunogai et al., 2002).Since CO is not converted to CO 2 , there is no introduction of an additional oxygen atom that needs to be calibrated.
Current continuous-flow isotope-ratio mass spectrometry (CF-IRMS) techniques move towards faster methods, smaller sample sizes, and most importantly more precise and reproducible results.Rapid methods allow for multiple measurements that can be combined to improve the error of the mean of measurements for one sample.A method that requires a smaller sample volume for a single run provides the opportunity not only to measure samples multiple times but also to measure small air samples, e.g. from ice cores (Wang and Mak, 2010) or firn air (Petrenko et al., 2013;Wang et al., 2012), expanding the range of possible applications.
This paper presents a method to measure the mole fractions, δ 13 C and δ 18 O of CO, in an air sample in less than 20 min.The method uses 100 mL STP (standard temperature and pressure) sample gas.The minimum pressure required in a 1 L glass flask, to perform a single run with a stable flow rate of 20 mL min −1 , is 1550 mbar abs.The blank from the oxidant (Schütze reagent) that is used to convert CO to CO 2 is strongly reduced by continuously flushing the reagent with high-purity He, which leads to highly reproducible results.
Conceptually, the key difference to the system by Wang et al. (2010) is that our extraction system always operates under a flow of He; it is thus a conceptually different realization of the same idea of extraction and preparation of CO.The system is particularly well-suited for routine operation and automated analysis of many samples.In terms of precision, the overall results are similar to the system described by Wang et al. (2010), and the main improvement is the strong reduction of the system blank.

Method and instrumentation
A diagram of the analytical system is shown in Fig. 1.The system consists of an automated multiple sample inlet system, the CO extraction and conversion set-up, a gas chromatograph (GC) for purification of the CO 2 , an open-split system, and an IRMS.The system is at all times flushed with ultra-high-purity helium (He) with BIP ® technology (BIP is Built In Purification; specification number: He-26507; assay: 99.997 %), provided by Air Products.Fused silica capillaries are used for connecting components unless specified differently.There are two membrane vacuum pumps (P 1 and P 2 ) attached to the set-up: one to evacuate the multi-sample inlet line (P 1 ) and the other one at the exhaust of the extraction (P 2 ), conversion, and collection unit.A single, automated, measurement is performed in 18 min.

Multi-sample inlet system
Sample flasks are connected to the automated multi-sampling unit for analysis.This unit allows for automated measurements of the reference gas and up to eight sample flasks.Sample flasks are connected to a 8-port 16-position dead-end flow path selector (V 1 : VICI; product number: SD8MWE).The 3-port switching valve (V 2 : VICI; product number: 3UWE), after the sample multi-port, provides the option to select either the reference gas cylinder or one of the samples.The air to be analysed is then directed via a mass flow controller (MKS; model 1179, 100 sccm) to a 6-port 2-position valve (V 3 : VICI; product number: 6UWM) from where the A diagram of the continuous-flow isotope-ratio mass spectrometry system for measuring 13 C and 18 O of CO. S 1 to S 8 represent the sample flasks; the connections to the flask can be changed to accommodate other types of cylinders or cans.The multi-sample inlet system can be evacuated by a pump (P 1 ).V 1 , V 2 , V 3 , V 4 , and V 5 are Valco valves.The sample is admitted to the analytical system via a mass flow controller (MFC) and a 6-port valve (V 3 ) and either sample gas or He carrier gas are processed through the system by a vacuum pump (P 2 ) at the outlet of the extraction system.T 1 , T 2 , T 3 , and T 4 are the cryogenic trap filled with glass beads, the glass tube with Schütze regent, the collection trap to extract the CO 2 from CO, and the focus trap, respectively.Final separation of the CO-derived CO 2 is achieved on a gas chromatographic column (GC: Poraplot-Q, 25 m × 0.25 m).
gas can be either injected to the extraction system (position 1) or evacuated (position 2).A 1/16 inch Restek Silcosteel ® coated stainless steel tubing is used for connecting the individual components (Tsunogai et al., 2000).This multisampling unit is controlled by LabView software, which specifically controls V 1 , V 2 , the flow rate of the mass flow controller, the sample injection time, the sample flush time, and the number of times each sample is measured.The sample is injected into the system at a flow rate of 20 mL min −1 for 5 min.A higher flow rate and longer injection time can be used for measuring samples with lower mole fractions.The LabView program also records the values from the pressure sensor before the mass flow controller and gives a start signal to the ISODAT program to start its acquisition.The interface that is essential for the communication between the valves and the PC is National Instruments USB NI-6008 unit.
When starting an automated measurement sequence, first the eight samples are connected to the inlet system (V 1 ) and V 2 is set in the direction of the samples.Then V 3 is set to evacuation position and the membrane pump valve is opened, allowing the air from the point of the sample connection to V 3 to be evacuated.Following this procedure the lines connected to all the flasks are evacuated by switching V 1 and tested for leaks.After the leak test, the sample bottle/can/cylinder valves are opened.From this point onwards the method is fully automated.The final pressure in the sample admission part of the system prior to the introduction of the sample is ∼ 1 mbar.
To avoid contamination with remaining air when switching between samples via the multi-sample inlet system, V 1 is first set to the close position between two sample ports and the system is evacuated for 60 s.Afterwards the multi-sample inlet system is flushed with the new sample air for 55 s at a flow rate of 20 mL min −1 before it is injected via V 3 .

Extraction and conversion set-up
By switching the injection valve (V 3 ), the sample is injected into the extraction system and directed through a chemical trap containing Ascarite ™ (CO 2 absorbent, 8-20 mesh, Aldrich) and magnesium perchlorate (Sigma-Aldrich), removing CO 2 and H 2 O, respectively.The subsequent cryogenic trap (T 1 −3 mm ID, 6 mm OD, 62 cm length, glass), containing glass beads (US mesh 40-60), removes CO 2 , N 2 O, and other condensable gases at liquid Nitrogen temperature (−196 • C).
CO is then selectively oxidized to CO 2 using the Schütze reagent (Schütze, 1949;Smiley, 1965) in T 2 .The oxidation reactor (T 2 ) consists of a 10 cm length 6 mm OD glass tube and it is filled with 3 g of Schütze reagent.In order to reduce the Schütze blank, the oxidation tube is located in the loop position of a 6-port 2-position valve (V 4 : VICI; product number: C6UWM) and it is continuously flushed with He (flow rate ∼ 8 mL min −1 ) when not in use.The flow is directed through the Schütze oxidant only during the sample injection and flushing period.

Synthesis of Schütze reagent
In 12.5 mL of water (Sigma-Aldrich, product number 270733: CHROMASOLV ® for High Performance Liquid Chromatography (HPLC) graded water, filtered through a 0.2 µm filter) 2.5 g of diiodine pentoxide (I 2 O 5 , 99.9 %, Aldrich) was dissolved to obtain a solution of iodic acid (colourless); 20 g of silica gel (Grade 40, 6-12 mesh, Sigma-Aldrich) was added to this solution.The mixture, covered with a watch glass, was dried in the oven for 1.5 h at 145 • C. Immediately out of the oven, 5 mL of concentrated sulfuric acid (H 2 SO 4 ) was added to the mixture and the covered beaker was left overnight in the laboratory hood.This allows the H 2 SO 4 to coat the mixture well, dehydrating it.The mixture was then placed in the Schütze reagent reactor (Fig. 2).

Nitrogen
Figure 2. A diagram of the apparatus used for preparing the Schütze reagent.It is mounted in an oven and flushed with a slow continuous flow of nitrogen gas in order to dehydrate the iodine pentoxidecoated silica gel efficiently, while the temperature distribution is homogeneous.
The Schütze reagent reactor was continuously flushed by a slow nitrogen gas stream and heated at 220 • C for 6 h or until the mixture was dry.The air exiting the reactor was passed through a molecular sieve and a large beaker of water.This was done to ensure that the H 2 SO 4 vapour carried out with the N 2 was removed before the N 2 was released to the laboratory.The N 2 flow rate was adjusted so that there was a slow release of bubbles visible in the beaker.The active chemical (I 2 O 5 ) was formed only when all the water was removed.The Schütze reagent is white, but sometimes it may have a bright yellow tint when iodosyl salts are formed (Schmeisser and Brändle, 1963).When the reagent turns brown (Formation of iodine: 5CO + I 2 O 5 → 5CO 2 + I 2 ) with use, it must be replaced with a new batch.When the Schütze reagent is replaced, not only the capillaries but also the reagent is exposed to the atmosphere causing a build-up of CO 2 .Once the connections are properly tested for leaks, the reagent needs to be flushed well with helium, for 3 days to a week, until the CO 2 blank is back to normal low levels (see Sect. 3.1).

Collection, focus, and separation
The CO-derived CO 2 is trapped in the collection trap, T 3 (1/16 inch stainless steel tubing), using liquid nitrogen while the other gases are removed via the vacuum pump.The CO 2 sample is then transferred to a focus trap (T 4 ), when the 6port 2-position valve (V 5 : VICI; product number: 6UWM) is switched to the inject position.In T 4 , the 320/430 µm fused silica capillary, used throughout the system, continues through 1/16 inch stainless steel tubing (tubing is used only to protect the capillary).The liquid nitrogen level of the cold traps is controlled by a liquid nitrogen refiller (NORHOF 900 series LN2 micro-dosing system) to improve the reproducibility of the peak areas.When T 1 , T 3 , and T 4 are in the down position the traps acquire the temperature of liquid nitrogen (−196 • C), and when the traps are in the up position they warm up to room temperature (25 • C).
The sample is separated from other residual components, on a Poraplot-Q (25 m × 0.25 mm) gas chromatography column (at 50 • C).It is then dried via a Nafion dryer.Finally, the sample is transferred into a Thermo-Finnigan Delta V Plus IRMS through a custom-made (Röckmann et al., 2003) opensplit interface.The ISODAT program controls the components from the cryogenic trap to the open split.All the valves and the traps are air actuated and controlled by solenoids linked to the interface with IRMS.
In routine operation, the entire system is flushed for 425 s between runs, the Schütze reagent is introduced into the main gas stream 425 s before injection of sample air, the sample processing takes 300 s, followed by another 300 s of flushing before the sample is transferred from trap T 3 to T 4 .The cryogenic trap that removes the remaining traces of CO 2 and N 2 O is warmed to room temperature for 302 s in between runs to remove the eluted gases and is cooled again for 123 s before the next sample is admitted.

Data processing and calibration
In order to monitor the performance of the CO isotope system, we prepared a reference air (Ref) sample with a known mole fraction and isotopic composition (see Sect. 2.2.2).This reference air is dry ambient air in a 30 L aluminum cylinder (Luxfer with Rotarex Ceodeux brass valve, used with a Scott Specialty Gases type 51-14C pressure regulator) pressurized up to 130 bar at the Max Plank Institute for Biogeochemistry (MPI-BGC) in Jena, Germany, in 2009.
Ref is run multiple times and an evaluation of these runs helps determine the reproducibility and accuracy of the system.Ref is analysed before and after every sample run to enable calibration and to detect variations in system sensitivity.
CO 2 derived from CO in a sample is analysed by the mass spectrometer.CO is quantitatively converted to CO 2 using Schütze reagent (Brenninkmeijer, 1993).Therefore the quantity (in moles) of CO is equal to the quantity of CO 2 derived from the CO in the air sample.
Carbon monoxide mole fractions are calculated using a 1point calibration, according to where c S is the mole fraction of the sample, area all S is the area of the sample peak, area all R is the area of the Ref peak, f R f S is the ratio of the reference flow rate and sample flow rate, t R t S is the ratio of the reference injection time and sample injection time, and c R is the mole fraction of the reference air cylinder.For typical ambient air samples f R f S = t R t S = 1.The ISODAT software reports the δ values of each peak in the chromatogram (both sample and reference air) vs. the laboratory working gas, δ S vs. WG , and δ R vs. WG .In our data reduction procedure, we first use these values to calculate the isotopic composition of the sample vs. the reference, δ S vs. R , according to For δ R vs. WG we use the average of the reference δ values before and after the sample run.Then, the δ value of the sample is converted to the international reference scales via δ R vs. V is the δ value of the reference air cylinder vs. the international standard, V-PDB or V-SMOW.
In δ 18 O data evaluation, δ S vs. V is the δ value of the sample vs. the international standard, V-SMOW for CO 2 .CO 2 is generated when the CO from the sample is oxidized by the Schütze reagent.
Therefore, a correction has to be made to get the, δ S vs. V for CO (Brenninkmeijer, 1993).
In Eq. ( 5), δ 18 O S vs. V : CO is the δ value of the sample vs. the international standard for CO, δ 18 O S vs. V : CO 2 is the δ value of the sample vs. the international standard for CO 2 and δ 18 O Schütze Reagent is the O from the Schütze reagent, which is derived using the δ value of the reference vs. the international standard for CO 2 (δ 18 O R vs. V : CO 2 ) and the δ value of the reference vs. the international standard for CO (δ 18 O R vs. V : CO ).

Mole fraction calibration
The mole fraction in Ref air cylinder was determined to be 185.4nmol mol −1 at MPI-BGC, by using a Trace Analytical reduction gas analyser and it is linked to the WMO X2004 calibration scale.The Ref air cylinder has been regularly measured (since March 2013) against other gas cylinders, for isotope calibration (see Sect. 2.2.2) and for checking the system stability.No significant drift in CO mole fraction relative to other gases has been observed since the measurements described here have begun.
An independent calibration of the Cal cylinder was published in 1997 (Brenninkmeijer and Röckmann, 1997), which confirmed the originally assigned values after a long period of storage.The estimated maximum uncertainty for 13 C Cal vs. VPDB was given as −44.30± 0.2 ‰.The uncertainty for 18 O Cal vs. VPDB-CO 2 was given as 11.43 ± 0.3 ‰.
The Cal gas was diluted to a suitable mole fraction (130 nmol mol −1 ) with CO-free zero air (checked with a Peak Performer 1 reduction gas analyser).This diluted calibration gas is referred to as DiCal and it is assumed that Di-Cal has the same isotopic composition as Cal.Then, Ref and DiCal were measured 10 times vs. the lab CO 2 working gas using the new measurement system, and the averages were used for calibration of the isotopic composition of CO in Ref relative to the isotopic composition of CO in DiCal.
For δ 18 O, the correction for the oxygen atom from the Schütze reagent is done in the same manner as Eq. ( 5).
The δ 18 O of the Schütze reagent oxygen is derived by δ 18 O Schütze Reagent = 2δ 18 O DiCal vs. V:CO 2 − δ 18 O Cal vs. V:CO , where δ 18 O DiCal vs. V:CO 2 is the measured δ 18 O of CO 2 from DiCal CO and δ 18 O Cal vs. V:CO is the known δ 18 O of Cal CO.Following this calibration, the values of the reference gas (Ref) against the international standards were determined as δ 13 C (Ref, V-PDB) = −29.61± 0.1 ‰ and δ 18 O (Ref, V-SMOW) = 8.45 ± 0.2 ‰.In the absence of international standards for the isotopic composition of CO, we note that there may be additional systematic errors (e.g.temporal changes of the primary calibration cylinder or dilution artefacts), which may introduce an additional unspecified uncertainty to these values.

Blanks
A blank run is performed using the same method as a sample run but without the injection of reference or sample gas.The continuous He flow collects the background of the system for the usual injection time of 5 min.The peak area of this system blank is ∼ 0.1 Vs, which is 2.2 % of the average reference gas (Ref) peak area (the system blank ranges between 8-23 pmol in a 0.7 nmol Ref sample).The majority of the blank signal originates from CO 2 that is released by the Schütze reagent.It is an accumulation of CO 2 formed by the system CO blank or CO 2 from the reagent itself, which are released in later measurements.
When a blank run is done excluding the Schütze reagent trap, the peak has an area of ∼ 0.019 Vs.The blank including the background CO 2 released from the Schütze reagent is used as the system blank.
When the system was first built, the system blank was 10 % of the sample peak.The simple modification of adding a 6-port Valco valve to continuously flush the Schütze reagent with He at a flow rate of 8 mL min −1 reduced the blank to 1-3 %.This blank affects both the sample and the reference air in a typical measurement sequence and is not considered when calculating the mole fractions and δ values of a sample.

Removal efficiency of CO 2 and N 2 O
When the air sample is injected into the extraction system, CO 2 and nitrous oxide (N 2 O) must be completely removed.CO 2 in the sample must be removed as CO is converted to CO 2 for isotope analysis.N 2 O shares the same molecular mass as CO 2 and interferes with the CO 2 peak derived from CO on the chromatogram.CO 2 is largely and efficiently removed by the Ascarite trap, and remaining traces are together with N 2 O condensed in the cryogenic trap (T 1 ) with glass beads.The glass beads increase the surface area for condensation.T 1 is warmed and evacuated at the end of each run.Periodically, checks are done to confirm that CO 2 and N 2 O traps (Ascarite trap and T 1 trap respectively) work efficiently.This is done by bypassing the Schütze reagent so the CO is not converted to CO 2 .The result of such runs should be the same as a blank run without the Schütze reagent trap, such as a blank run that does not show a CO 2 peak in the chromatogram.If the result shows a CO 2 peak on the chromatogram then the Ascarite trap needs to be changed.N 2 O has the same nominal isotopocule masses as CO 2 , but with much smaller molecular isotope ratios 45 R and 46 R. Therefore, a small amount of N 2 O seeping through the cryogenic trap can be detected in the resulting δ values.Table 1 shows a comparison of average values of 10 runs from the reference gas (Ref) and 10 runs from a can with approximately 2000 nmol mol −1 N 2 O (15 µL of N 2 O was added to a 2.5 L steel can and was filled with reference gas, which already contained atmospheric levels of N 2 O).The results show that there is no evidence of N 2 O leaking from the cryogenic trap (T 1 ) even at a high mole fraction.Figure 3 shows a case where the N 2 O peak appears on a chromatograph because trap T 1 is not used to remove N 2 O.The retention time of N 2 O is ∼ 355 s, about 25 s longer than for CO 2 at ∼ 330 s.Due to the different isotope ratios of N 2 O (see above), the isotope ratios show an inverted peak; thus, an N 2 O interference is easy to recognize.

Repeatability
The peak area of 100 mL aliquots of Ref is ∼ 4.6 Vs with a standard deviation of ∼ 0.03 Vs, which corresponds to a relative repeatability of 0.7 % for the mole fraction on a single sample.δ 13 C has a repeatability of 0.1 ‰. δ 18 O has a repeatability of 0.2 ‰.System reproducibility is tested on a daily basis and often with overnight runs.When the system stays idle, at least five runs should be performed to regain its normal repeatability.Ref gas is measured often (several times per day) and all the samples are measured relative to the Ref; it is the repeatability on a short term (hours to days) that is the most important.

Linearity
Ideally, the δ value of a sample measured on an isotope instrument vs. a certain reference should be independent of the amount of sample that was injected into the instrument.In reality, isotope systems often show a dependence of the isotope results on the total amount of sample injected, which is commonly called a non-linearity.The non-linearity of our system was calibrated by injecting varying amounts of sample (for sample linearity) and He (for blank linearity) into the system.

Blank linearity
A blank linearity test was performed to characterize the effect of the CO 2 released by the Schütze reagent.No sample was injected in these blank experiments.Injection time on the x axis of Fig. 4 depicts the period, in seconds (s), for which the He flow was directed through the complete system (including the Schütze reagent).These injection times are 100, 300, 600, 900, and 1200 s.Each injection time test

Sample linearity
The amount of sample was varied by changing the mole fraction of a sample with initial high mole fraction, by dilution with CO-free air.For the dilution test, 8 mL of high mole fraction (∼ 269 µmol mol −1 ) CO sample was injected into a 1 L glass flask, which was then filled with zero air to 1.8 bar.A sequence of 59 measurements was made while the flask air pressure kept constant at 1.8 bar by refilling with zero air after every run, which results in an extended dilution series.Figure 5 shows the δ 13 C and δ 18 O values as a function of area all (Vs).The δ 13 C values are constant down to about 4 Vs and then start deviating systematically for the low peak areas (i.e. they become non-linear).The δ 18 O values are relatively constant (with a small trend) for areas above 1.5 Vs (60 nmol mol −1 ).The average δ 13 C and δ 18 O values in the peak area range between 4 and 15 Vs where samples usually measured are −44.2± 0.1 ‰ and 54.1 ± 0.2 ‰, respectively.The small trend at higher peak areas visible in Fig. 5 in particular for δ 18 O is not further investigated.Measurements over many months indicate that the area below, where nonlinearity is observed, depends on the state of the filament in the IRMS.Therefore, the non-linearity is checked regularly.

Application example: CO emissions from vehicles
As an application example, Fig. 6 shows δ 13 C and δ 18 O values of air samples collected at the Islisberg highway tunnel in Switzerland that were presented and discussed in detail in Popa et al. (2014).Air samples from the entrance and the exit of the tunnel were collected in 1 L glass flasks under ∼ 1.8 bar pressure and analysed on the analytical system described here for δ 13 C and δ 18 O.The different colour markers in Fig. 6 represent the samples from the entrance and the exit of the tunnel.The exit samples contain air that has accumulated emissions of vehicles passing through the tunnel; these samples have very high (2-10 ppm) CO mole fractions, and their isotopic values represent the isotopic signature for CO emitted by vehicles.The isotopic composition of the vehicle emissions based on these samples was δ 13 C = −25.6 ± 0.2 ‰ and δ 18 O = 24.1 ± 0.2 ‰, respectively (Popa et al., 2014).The air collected near the entrance is much closer to background air, but since the collection was actually in the tunnel, it is also influenced in varying proportions by the emissions of vehicles on the highway.These entrance data thus fall in between the CO isotopic signature of fossil fuel combustion and the isotopic composition of background atmospheric CO.

Conclusions
A new continuous-flow isotope-ratio mass spectrometry system was presented allowing quick and precise measurement of CO mole fractions, δ 13 C and δ 18 O.The technique is based on the conversion of CO in the sample to CO 2 using the Schütze reagent.Prior to the conversion of CO, CO 2 and N 2 O are quantitatively removed from the sample by chemical and cryogenic traps.Helium is used not only as a carrier gas to transfer the sample from one component to another but also for flushing the system.The repeatability for the mole fraction measurement is 0.7 nmol mol −1 for a reference air cylinder with a CO mole fraction of 185.4 nmol mol −1 .This corresponds to a relative error of 0.4 %.The CO 2 blank that originates from the Schütze reagent is minimized by continuously flushing the reagent with helium.The peak area of the system blank is ∼ 4 nmol mol −1 , which is 2.2 % of the peak area obtained with the reference air.The δ 13 C and δ 18 O repeatability are 0.1 and 0.2 ‰, respectively.The amount of air that is used for a single analysis is only 100 mL, but in the present configuration this has to be supplied from a sample container at a minimum pressure of 1550 mbar abs, so that the actual amount of sample air required is larger.A single measurement is completed in 18 min, and the system is fully automated and well-suited for measurements on air samples from flasks that are routinely employed in atmospheric monitoring networks.
Figure1.A diagram of the continuous-flow isotope-ratio mass spectrometry system for measuring 13 C and 18 O of CO. S 1 to S 8 represent the sample flasks; the connections to the flask can be changed to accommodate other types of cylinders or cans.The multi-sample inlet system can be evacuated by a pump (P 1 ).V 1 , V 2 , V 3 , V 4 , and V 5 are Valco valves.The sample is admitted to the analytical system via a mass flow controller (MFC) and a 6-port valve (V 3 ) and either sample gas or He carrier gas are processed through the system by a vacuum pump (P 2 ) at the outlet of the extraction system.T 1 , T 2 , T 3 , and T 4 are the cryogenic trap filled with glass beads, the glass tube with Schütze regent, the collection trap to extract the CO 2 from CO, and the focus trap, respectively.Final separation of the CO-derived CO 2 is achieved on a gas chromatographic column (GC: Poraplot-Q, 25 m × 0.25 m).

Figure 3 .
Figure 3.A visual comparison of the system blank peak (top), the sample peak (middle), and the N 2 O peak (bottom).During a normal measurement, N 2 O is removed from the sample to avoid isobaric interference with the CO-derived CO 2 masses.As a further precaution N 2 O and CO 2 are separated on the gas chromatograph where N 2 O peaks at ∼ 355 s compared to CO 2 peaking at ∼ 330 s.The mass ratio 45/44 and 46/44 traces show inverted compared to the CO 2 and are easily recognizable.The absence of a N 2 O signal shows that N 2 O is quantitatively trapped in the cryogenic trap and does not reach the IRMS.The mass traces (44, 45 and 46) shown are direct output from the ISODAT software.For the ratios 45/44 and 46/44, a value of 100 mV was arbitrarily added to the signals in order to avoid artificial noise arising from the ratio of two small numbers.

Figure 4 .
Figure 4. Dependence of the peak area all (a) and δ 13 C and δ 18 O (b) on the injection time for blank runs, i.e. when the sample inlet valve was actually not opened.The peak area of the blank increases approximately linearly with injection time, and the dependence of the isotope values on injection time is relatively small.Note that the scatter in the isotopic composition measurements is so large because of the very small peak areas of these blank experiments.

Figure 5 .
Figure 5. Dilution test: δ 13 C V-PDB (black) and δ 18 O V-SMOW (red) plotted against peak area (area all in Vs).The δ values show a strong non-linearity at peak areas below 4 Vs (shaded region).Values between 4 and 15 Vs are used to calculate the repeatability of the system.

Figure 6 .
Figure 6.δ 13 C V-PDB (‰) vs. δ 18 O V-SMOW (‰) plot of Islisberg highway tunnel samples.Data from Popa et al. (2014).The δ values of the samples collected at the entrance and exit are depicted by blue and red markers, respectively.
. Both 13 C 16 O 16 O and 12 C 17 O 16 O contribute to the ion signal at m/z = 45.This means that, when assuming MDF, the contribution of 17 O to the ion beam at mass 45 is underestimated, leading to an overestimation in www.atmos-meas-tech.net/8/5315/2015/Atmos.Meas.Tech., 8, 5315-5324, 2015

Table 1 .
Peak area and isotopic composition of the reference gas, and an aliquot of the reference gas that was spiked with 2000 nmol mol −1 N 2 O.