Modification of air standard composition by diffusive and surface processes

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110,, doi:10.109/004jd00548, 005 Modification of air standard comosition by diffusive and surface rocesses R. L. Langenfelds, 1 M. V. van der Schoot, R. J. Francey, and L. P. Steele Atmosheric Research, Commonwealth Scientific and Industrial Research Organisation, Asendale, Victoria, Australia M. Schmidt Laboratoire des Sciences du Climat et de l Environnement, Gif-sur-Yvette, France H. Mukai National Institute for Environmental Studies, Ibaraki, Jaan Received 30 Setember 004; revised 8 March 005; acceted 9 March 005; ublished 15 July 005. [1] We resent exerimental evidence of modification of O /N, Ar/N, 9 N / 8 N, 34 O / 3 O ratios and CO mole fraction in dry air standards resulting from gas handling. Correlated variations among high recision measurements of multile molecular air ratios show diffusive fractionation to be the main modifying rocess. This mechanism can account for much of the reviously unexlained CO instability commonly observed in air standards used by the atmosheric CO measurement community. Identification of the effects of diffusive fractionation hels to isolate effects of other rocesses such as surface adsortion, which is also imlicated as a cause of CO modification in some alications. These findings have direct imlications for the rearation and maintenance of O /N /Ar and CO calibration standards and are relevant to other atmosheric measurement activities that involve gas handling. Citation: Langenfelds, R. L., M. V. van der Schoot, R. J. Francey, L. P. Steele, M. Schmidt, and H. Mukai (005), Modification of air standard comosition by diffusive and surface rocesses, J. Geohys. Res., 110,, doi:10.109/004jd Introduction [] Many atmosheric trace gas monitoring rograms rely on air standards for calibration of measurements. A desirable requirement is that the comosition of such standards remains stable within accetable limits, referably over an extended eriod of time. However, there have been extensive observations of instability in a wide range of trace gases and isotoes (see below). In some cases, ensuing research has identified the causes with subsequent imrovements in selection of materials, gas handling rocedures, and erformance of standards, while in others the causes of instability remain obscure. In the field of carbon cycle research, drifting standard comosition has long been recognized as a major roblem for high recision monitoring of atmosheric CO [e.g., Thoning et al., 1987]. This continues to be one imediment to achieving the long-standing World Meteorological Organization (WMO) targets for intercomarability among CO measuring laboratories of ±0.1 and ±0.05 mmol mol 1 in the Northern and Southern Hemisheres, resectively [WMO, 1981; Toru and Kazuto, 003]. The otential for artefacts resulting from drifting 1 Also at Institute of Antarctic and Southern Ocean Studies, University of Tasmania, Hobart, Tasmania, Australia. Coyright 005 by the American Geohysical Union /05/004JD00548 standards is also of major concern for measurements of the O /N ratio, another key tracer of carbon cycle rocesses [Bender et al., 1994; Keeling et al., 1998; Langenfelds, 00]. [3] Observed instability in air standards has in some cases been attributed to surface interactions (e.g., for some halocarbon secies [Yokohata et al., 1985; Fraser et al., 1991; Thomson et al., 1994]). While mass-deendent fractionation has long been recognized as an imortant modifying rocess for studies involving isotoomer ratios [e.g., Craig et al., 1988], its imact on measurements of trace gas concentrations such as CO in air has received little attention. This is due to historically oorer recision of measurement of trace gas concentrations as comared to that achieved for isotoomer ratios, and the assumtion that the recision is limited by other factors such as chemical rocesses that differentiate between the trace gas of interest and the gas matrix (e.g., air) in which it resides. The develoment of techniques to measure O /N ratio to a recision of better than 1 in 10 5 that is necessary to constrain the global carbon budget has brought new focus to the imortance of diffusive fractionation of air samles. Evidence of O /N modification by diffusive and surface rocesses has been reorted, and some reventative measures have been imlemented in exerimental rograms [Bender et al., 1994; Keeling et al., 1998; Manning, 001; Langenfelds, 00]. Associated changes in the Ar/N ratio have been observed and used to investigate the nature of the 1of11

2 modifying rocesses [Langenfelds, 00; Keeling et al., 004]. Such effects have also been observed in nature, for examle in the diffusive columns of air in both sand dunes [Severinghaus et al., 1996] and firn [Severinghaus et al., 001]. As we will show here, the knowledge gained from these rograms has relevance to measurements of other trace gases and isotoes. [4] The matters raised here in relation to drifting comosition of air standards may also be relevant to the rearation of standards, such as those reared gravimetrically to be accurate in absolute terms. There is growing interest in imroving the calibration of atmosheric greenhouse gas measurements, for examle among nations working to meet their obligations under the Kyoto Protocol. Work is in rogress in various national measurement institutes to establish the caability for roducing suitable absolute standards with trace gas concentrations linked to fundamental constants. It has generally been assumed that diffusive and surface rocesses have negligible influence on the comosition of such mixtures. Information that is now becoming available from measurements of O /N /Ar and associated secies allows us to consider the validity of this assumtion. [5] In this aer, we resent evidence of drifts in CO, O /N, and other secies in air standards and discuss the likely causes. Reorted measurements come from various oerational rograms and exeriments that involve transfer of air for analytical or diagnostic uroses. They include tracking the change in comosition of air standards in high-ressure cylinders and measuring the difference in comosition of decanted subsamles. Our aroach to understanding the observed changes focuses on highrecision measurements of multile trace gas and isotoomer ratios and using the signature of changes in these ratios to hel identify the modifying rocesses.. Theoretical Background.1. Diffusion [6] Fractionations associated with kinetic rocesses reflect varying rates of diffusion among different secies, resulting from gradients in ressure, concentration or temerature, and where diffusivities are deendent on mass, molecular geometry, and intermolecular forces. A summary of these rocesses and their otential imact on measurements of atmosheric comosition, esecially in relation to high-recision measurement of O /N ratios, is rovided by Keeling et al. [1998]. The general equation of diffusion for a binary mixture is given by Chaman and Cowling [1970] as C 1 C ¼ ðx 1 x Þ 1 D 1 rx 1 þ n 1n ðm m 1 Þ r ln nr r 1r ð r F 1 F Þþk T r ln T where subscrits denote secies 1 and, C is mean velocity, x is mole fraction, D 1 is the diffusion coefficient, n is number density, m is mass, is ressure, r is mass density, F reresents external forces, k T is the thermal diffusion ratio, and T is temerature. The four terms in the bracket at the right-hand side reresent comonents due to gradients in concentration, ressure, external forces, and temerature, ð1þ resectively. If the molecules are assumed to behave as rigid elastic sheres, the diffusion coefficient is aroximated by: D 1 ¼ 3 1 ktðm 1 þ m Þ 8ns ðþ 1 m 1 m where s 1 is the mean diameter of secies 1 and, and k is the Boltzmann constant. Mass is the key arameter, esecially for isotoic (geometrically and chemically like) airs where relative fractionation among multile molecular ratios can be reasonably aroximated from consideration of relative mass alone. However, for ratios involving different gases, such as O,N, and Ar, the geometric roerties of the molecules may also become significant. For examle, some exerimental evidence attributes a higher diffusivity to O than N in some circumstances [e.g., Severinghaus et al., 1996], desite O having the greater mass. [7] Three tyes of diffusion are considered here. [8] 1. The first tye is ressure diffusion. In the resence of a ressure gradient, diffusion causes heavier molecules to referentially accumulate in the region of higher ressure. One examle is in the gravitational settling of heavier molecules towards the base of air columns where gas transort is dominated by diffusion, such as observed in firn and in sand dunes. Of more relevance to the results reorted here is the behavior in nonequilibrium, flowrelated ressure gradients where fractionation induced by the ressure gradient is attenuated by turbulent mixing and advection of the air samle. Relative fractionation among two airs of secies (Table 1) can be calculated from equations (1) and () as a function of mass and molecular diameter (Aendix A). [9]. The second tye is thermal diffusion. Under a thermal gradient, heavier molecules generally (but not always) referentially accumulate in the colder region, for examle as observed in olar firn air [Severinghaus et al., 001]. Fractionation effects can be calculated in terms of a thermal diffusion factor [Grew and Ibbs, 195; Chaman and Cowling, 1970]. Ratios in Table 1 are calculated from values of this factor, determined exerimentally for temeratures of 93 K, as summarized for most of the molecular airs considered here by Severinghaus et al. [1996], and taken from Chaman and Cowling [1970,. 78] for Ar/N (direct measurement) and CO in air (estimated from CO / N,CO /O and CO /Ar). Keeling et al. [004] reorted exerimental results demonstrating this effect for Ar/N and O /N, with evidence that the d( 40 Ar/ 8 N )/d( 3 O / 8 N ) ratio characteristic of fractionation by thermal diffusion is also deendent on samle ressure, with values of 3.77 and.17 observed at ressures of 0.1 and 1 MPa, resectively. [10] 3. The third tye is effusion. Gas molecules escaing from a ressurized vessel through a tiny orifice are subject to molecular effusion (also known as Knudsen diffusion). This rocess occurs when the size of the orifice is small comared to the mean free ath between molecular collisions so that most collisions are with the walls of the orifice. The rate of effusion is inversely roortional to the square root of the mass as described by Graham s Law for a air of gases a and b: effusion rate of a effusion rate of b ¼ m 1 b ð3þ m a of11

3 Table 1. Signatures of Fractionation for Different Molecular Pair Ratios and Different Diffusive Processes Predicted From Theory and Observed in Data Presented Here Molecular Ratios a Pressure Diffusion b Predicted Thermal Diffusion Effusion Data Reorted in This Study Drift in Standard ID c Decanted Subsamles c d( 40 Ar/ 8 N )/d( 3 O / 8 N ) ± ± 0.3 d( 40 Ar/ 8 N )/d( 9 N / 8 N ) ±.4 68 ± 146 d( 40 Ar/ 8 N )/d( 34 O / 3 O ) ± ± 15 d( 3 O / 8 N )/d( 9 N / 8 N ) ± 0.8 ± 31 d( 3 O / 8 N )/d( 34 O / 3 O ) ± ± 3.9 d( 34 O / 3 O )/d( 9 N / 8 N ) ± ± 0.6 d( 40 Ar/ 8 N )/d( 44 CO / 9 air) ± ± 1.1 a The d notation refers to the relative deviation in a ratio of two secies a and b such that d(a/b) = [(a/b) samle /(a/b) reference 1], and is exressed elsewhere in this aer in units of er meg, obtained by multilying this term by a factor of b Calculated using molecular diameter values of Chaman and Cowling [1970]. c Sloes determined by orthogonal distance regression from data in Figures 3, 4, 6; uncertainties are 1s. The derivation of this relationshi as given by Moore [196] does not allow for any deendence on molecular geometry, yet it is ossible that geometry could have significant influence on effusion rates. The rocess of effusion can also be described by equations (1) and (), where one collision artner is considered to have infinite mass and an effective diameter that is large relative to a gas molecule and deendent on the geometry of the orifice through which the gas is escaing. For cases where the diameter of the orifice aroaches that of the gas molecules, the rate of effusion may also become deendent on the shae of the molecules, rather than just their mean diameters. For examle, Battle et al. [1996] observed fractionation of O /N in South Pole firn air that they attributed to the ability of the more rolate O molecules to referentially escae through channels searating air bubbles from oen firn during the bubble close-off rocess. Thus Graham s Law might be more robust for fractionation ratios involving airs of isotoomers of the same secies rather than airs of different gases. The molecular air ratio values for effusion in Table 1 are calculated from equation (3)... Molecular Pair Tracers [11] A useful aroach for identifying (or aortioning) effects of fractionating rocesses is to measure multile molecular and isotoic ratios in the same air samles. The isotoic ratios 9 N / 8 N and 34 O / 3 O comare chemically and geometrically like molecules and are not subject to any significant biogeochemically forced variations in the contemorary free atmoshere. They thus serve as useful tracers of diffusive fractionation associated with exerimental techniques and some natural rocesses. The ratio 40 Ar/ 8 N is articularly useful as a diagnostic of diffusive fractionation as it can be measured very recisely (to better than 1 in 10 5 ) by mass sectrometry, it is strongly sensitive to mass-deendent fractionation rocesses owing to the comaratively large 1-unit mass difference between Ar and N, and it comares two inert gases with only secondorder variations in the free atmoshere resulting from seasonal variations in air-sea flux due to temerature deendence of their solubility in seawater [Battle et al., 003]. The 3 O / 8 N ratio is another useful diagnostic as it rovides a signal-to-noise ratio (relative to recision of measurement) similar to that for Ar/N, yet higher by close to an order of magnitude than that for the other molecular airs considered here. However, its usefulness as a diagnostic is limited for some atmosheric measurement alications by biogeochemical signals resent in the atmoshere..3. Surface Interactions [1] A further otentially imortant rocess is the adsortion of gases on solid surfaces. Of interest here are interactions with wetted materials used in the exerimental equiment, in articular the metal bodies of cylinders, valves and regulators (stainless steel, aluminum, brass, nickel) and elastomers used in seals (e.g., Viton and olytetrafluoroethylene [PTFE]). Adsortion may involve either hysisortion, which results from weak intermolecular forces and is generally reversible, or chemisortion which involves stronger bonding between molecules and is generally irreversible [Moore, 196]. The equilibrium artitioning of gases between gaseous and adsorbed hases is sensitive to ressure and differentiates between chemically different secies. Effects of adsortion cannot be easily characterized in terms of multisecies signatures because they deend heavily on the combination of gas and surface material, and data characterizing such interactions are not readily available. What can be said is that because adsortion of a gas molecule to a surface is likely to be rimarily a function of the chemical roerties of the molecule rather than its mass or diameter, one would exect resulting fractionations among the measured trace gases to be decouled from those of the isotoic tracers d( 9 N / 8 N ) and d( 34 O / 3 O ). Modification of O /N and CO by surface adsortion was identified by Keeling et al. [1998] in tests involving exosure of air samles to a large number of Viton O-rings. Two rocesses were identified, (1) referential hysisortion of O and CO relative to N, characterized by deletion of the O /N ratio and CO mole fraction in samle air, and () chemisortion characterized by declining O /N and increasing CO and attributed to oxidation of the Aiezon grease used to lubricate the Viton O-rings. 3. Exeriment 3.1. Drifting Comosition of High-Pressure Air Standards [13] All measurements reorted here were erformed at the Commonwealth Scientific and Industrial Research 3of11

4 Figure 1. The change in CO in dry, natural air standards contained in 9.5 L aluminum cylinders, lotted as a function of final cylinder ressure. Data are from standards used between 1997 and 003 on a conventional Siemens Ultramat 5E non-disersive infrared (NDIR) analyser for in situ monitoring of CO at Cae Grim. This analyzer is distinct from the LoFlo CO analyzers [Da Costa and Steele, 1999] oerated at Cae Grim in arallel with the conventional system since 000. Initial cylinder ressures were MPa. The line of best fit was obtained by linear regression and gives a sloe of 0.07 ± 0.03 (1s) umol mol 1 MPa 1. Organisation s Global Atmosheric Samling Laboratory (CSIRO GASLAB). Details of analytical techniques have been rovided elsewhere. CO was measured by gas chromatograhy [Francey et al., 003] and O /N /Ar by mass sectrometry [Langenfelds, 00]. All measurements are made on samles of dry natural air, which in some cases were modified by minor addition or removal of CO to roduce air standards suitable for calibration of atmosheric CO measurements. All high-ressure standards were contained in 9.5 L, aluminum cylinders manufactured by Luxfer (Riverside, California, USA) and urchased from Scott Marrin Inc. (Riverside, California, USA). The cylinders are fitted with brass valves manufactured by Ceodeux (Lintgen, Luxembourg) using all-metal seats and Kel-F or nickel stems. This tye of cylinder is used to store CSIRO s rimary CO -in-air and O /N /Ar calibration standards. It is also used by the WMO-designated Central CO Calibration Laboratory (CCCL) at the National Oceanic and Atmosheric Administration s Climate Monitoring and Diagnostics Laboratory (NOAA-CMDL) for its CO calibration activities and is distributed by the CCCL to other CO measurement laboratories for calibration uroses. [14] Exerience with these cylinders at CSIRO indicates that CO stability of contained gas mixtures is comarable to or better than other cylinder tyes tested, roviding the gas is well dried (H O ] mmol mol 1 ). Nevertheless, many such standards exhibit significant drift and the cause(s) have not yet been identified. The most common occurrence of CO drift is as a function of ressure (or gas consumtion) rather than time, as exemlified by the working standards used for in situ CO monitoring at the Cae Grim Baseline Air Pollution Station. These standards are tyically in service for a short ( 4 month) eriod, during which most of their initial contents (10 15 MPa) are consumed. They have been oerated using either two-stage, nickel-lated, brass regulators (series 1000), manufactured by Alhagaz (Walnut Creek, California, USA), or singlestage, stainless steel regulators manufactured by Tescom Cor. (Elk River, Minnesota, USA). Measurements erformed at CSIRO before and after deloyment at Cae Grim show systematic enhancement of CO that increases with falling cylinder ressure (Figure 1). Significant changes of about 0.1 mmol mol 1 emerge below ressures of 0.8 MPa. [15] Evidence of changes in comosition across a range of secies is rovided by the measurement history of one CSIRO dry air standard (samle ID 99191), fitted with an Alhagaz regulator. It was drained from 14 to 0.07 MPa in years, mainly through short bursts at flow rates of 3 4Lmin 1 for filling test flasks (low-ressure samles in 0.5 L glass flasks used for quality control uroses). Because of its test flask role for several CSIRO measurement rograms, this standard was regularly analyzed for O / N /Ar ratios, isotoes of O and N, mole fraction of CO and other secies. Measurement histories for selected secies are shown in Figure as a function of cylinder ressure. The data show enrichment of d(ar/n ), d(o / N ), d( 9 N / 8 N ), d( 34 O / 3 O ) and CO with falling cylinder ressure, and acceleration in the drift at very low ressures, at least for d(ar/n ) and d(o /N ) where the magnitude of the drift is large comared with measurement uncertainty. [16] The fact that all ratios, including the urely isotoic tracers, show rogressive enrichment in the heavier secies oints to the signature of a mass-deendent rocess. This can be exlored further by assuming that air drawn from the cylinder was fractionated with resect to its bulk contents by a constant amount over the life of the standard. It is by no means certain that this assumtion is accurate, as fractionation factors may have varied with cylinder ressure (and/or other arameters such as flow rate etc.), but nevertheless it rovides a useful benchmark against which to assess the observed multisecies drifts. One can exloit the fact that the roduct of cylinder ressure P and the ratio R of any two secies is conserved, analogous to the case for amount of CO and its isotoic labeling (e.g., [CO ] d 13 C [Tans et al., 1993]). Thus for a change in cylinder ressure of dp from P 0 to P 1 that is accomanied by a fractionation causing R to change in the cylinder by an amount dr from R 0 to R 1 : R 0 P 0 ¼ R out dp þ R 1 P 1 If the two-secies ratio in the gas removed from the cylinder (R out ) is modified by a fractionation factor a with resect to the ratio in the cylinder such that R out ¼ ar 0 ð4þ ð5þ 4of11

5 Figure. Multisecies measurements from one cylinder standard (ID 99191) as a function of cylinder ressure or fractional usage (uer horizontal axis). Data are lotted as deviations from the mean of measurements at highest cylinder ressure ( MPa for O /N /Ar and MPa for CO ). Error bars reresent tyical uncertainties for daily mean values and are within the symbol size for d(ar/n ) and d(o /N ). For d(ar/n ), the solid curve aroximates the change with ressure in air drawn from the cylinder, assuming a constant fractionation relative to bulk cylinder contents (dashed curve; see text). Curves for other secies are redicted from d(ar/n ), assuming fractionation exactly roortional to mass difference, and fixed to ass through the mean of data oints at highest cylinder ressure. and ignoring the second order term dr.dp, Equations (4) and (5) give the relationshi dr dp ¼ Rða 1Þ P and a solution for R as a function of ressure ð6þ is drawn from the cylinder and is analyzed. The curve is constrained to ass through the mean of data airs at maximum and minimum cylinder ressure. The other (dashed) curve is offset by a constant amount, in this case 65 er meg, to reresent d(ar/n ) of air remaining in the cylinder. The difference between the two curves shows the imortance of the fractionation effect. Although the modeled curves slightly underestimate acceleration of d(ar/n ) drift at low cylinder ressure, they nevertheless suggest that the assumtion of a constant fractionation factor over the life of this standard is reasonable. [17] Similar curves are lotted for other secies in Figure but where the magnitude of drift is redicted from d(ar/ N ) data. We assume for this urose that fractionation is exactly roortional to mass difference, but note that this is only an aroximation as different factors of roortionality aly for different diffusive rocesses (Table 1). This would imly fractionation of er meg ( 65 4/1) for d(o / N ), 5 and 11 er meg for d( 9 N / 8 N ) and d( 34 O / 3 O ), resectively, and 0.03 mmol mol 1 for CO with resect to air. All solid curves are constrained to fit the mean of oints near maximum cylinder ressure only. [18] The modeled curves rovide a good fit to data for all secies excet CO. Significant drift was observed in d(o / N ) and in the isotoic tracers d( 9 N / 8 N ) and d( 34 O / 3 O ), roviding comelling evidence of diffusive fractionation. Furthermore, the multisecies signature defined by the molecular air ratios lotted in Figures 3 and 4 and listed in Table 1 is consistent with much or all of the O / N /Ar variations being due to diffusive fractionation. The most recisely defined ratio among these secies exists for d(ar/n )/d(o /N ). The observed value of 3.5 ± 0.1 lies between the redicted values for ressure and thermal diffusion, suggesting that one or both rocesses are likely major influences. [19] The CO observations cannot be as easily reconciled with diffusive fractionation alone, even after allowance for the range of redicted values for different rocesses shown in Table 1. The measurements show a drift that is of the same sign and with magnitude near the uer end of the range observed in the Cae Grim standards (Figure 1) but larger than what is redicted for this standard from d(ar/n ) assuming urely diffusive fractionation. At minimum cylinder ressure the observed CO enrichment is double that redicted from d(ar/n ). This discreancy is at least artly due to a shift of about +0.1 mmol mol 1 CO between 10 and 8 MPa that conflicts with the shae of the modeled curves. The absence of similar features in the other secies R ¼ e ½ða 1 Þln Pþk Š ð7þ where k is a constant. Equation (7) is a reresentation of the Rayleigh fractionation model, which is commonly exressed in terms of f, the fraction of the initial contents remaining in the reservoir such that f = P/P o : R R o ¼ f a 1 ð Þ ð8þ Two curves of this form are shown for the d(ar/n ) data in Figure. One (solid line) reresents d(ar/n ) of the air that Figure 3. The relationshi of concurrently measured d(ar/ N ) and d(o /N ) in standard ID 99191, with the line of best fit determined by orthogonal distance regression. 5of11

6 Figure 4. The relationshis of multile secies with d 9 N concurrently measured in standard ID 99191, with lines of best fit determined by orthogonal distance regression. indicates that it cannot have resulted from diffusive fractionation. The exact cause of this shift is not clear but is likely due to surface rocesses, involving either the cylinder or the attached regulator used to analyze the cylinder contents. Exchange of CO with the internal cylinder surfaces would imly that the shift is a genuine comonent of drift in this standard, which occurred in tandem with the drift due to diffusive fractionation. On the other hand, it is lausible that the shift may have resulted from regulator-induced artefacts during the early set of measurements (cylinder ressure >8 MPa). While we have no auxiliary evidence to indicate that measurements of this standard were subject to such artefacts, we have observed comarable CO anomalies from time to time in measurements of other standards, where the anomalies were linked to leaks or insufficient equilibration of regulator surfaces with the air being analyzed at cylinder ressure. If the CO data in Figure 1 are adjusted by 0.1 mmol mol 1 to allow for a single ste change at 8 MPa ressure, then the data closely follow the modeled fractionation curve redicted from d(ar/n ). 3.. Decanting [0] Another exeriment that demonstrates changes in air comosition due to gas handling and hels to elucidate the causes, involved rearation of subsamles decanted from high-ressure, dry, natural air standards. This exeriment was erformed at CSIRO in suort of intercomarison activities conducted by the Terrestrial and Atmosheric Carbon Observing System (TACOS) and National Institute for Environmental Studies (NIES). [1] A total of 3 subsamles were reared from a suite of eight individual high-ressure standards by decanting into either of two cylinder tyes: (1) the same 9.5 L, aluminum cylinders described above and () 34 L, electroolished stainless steel cylinders (model 80C ) manufactured by Essex Cryogenics (St. Louis, Missouri, USA) and fitted with all metal Parker valves (recently develoed in collaboration with Professor Ray Weiss of Scris Institution of Oceanograhy and DOT rated to hold air at ressures u to 6 MPa). The high-ressure standards used in this exeriment were also contained in 9.5 L, aluminum cylinders. Before the start of this exeriment, all subsamle cylinders were re-conditioned for several weeks with clean, dry, natural air at aroximately 0.5 MPa. A single Alhagaz regulator was used to decant air from all arent cylinders and for all subsamles. It was attached directly to the arent cylinders using a brass nut and nile. The regulator delivery ressure was set at a constant value of 0.55 MPa (gauge ressure) for all subsamles. The length and internal diameter of stainless steel caillary tubing connecting the regulator to subsamle cylinders was selected to achieve the desired decanting flow rate, which was varied among subsamles in the range of L min 1. Prior to decanting, subsamle cylinders were vented, evacuated to Ma then filled to 0.1 MPa with air from the arent cylinder ( cycles) and again evacuated. Finally, they were filled to 0.5 MPa. There was no temerature control of any comonents. [] The selection of regulators and gas handling rocedures for analysis of both arent and subsamle air are critical issues for this exeriment. Our aim is to investigate fractionation associated with the decanting rocess, but the analysis rocedure itself involves decanting of air, which may induce fractionation of the air being analyzed. There are several lines of evidence to show that any fractionation associated with the analytical rocedures was minor comared to that incurred during transfer of air from highressure cylinders to generate the subsamles. First, any deendence of subsamle measurements on the flow rate used to decant from the high-ressure cylinders cannot be due to analysis-related artefacts because analysis techniques and samle ressures were effectively identical for all subsamles. Second, there was consistency between changes in the comosition of subsamles and concomitant changes in arent air (discussed in more detail below). Third, data selection took account of the sensitivity of O / N /Ar measurements to regulators and flow rates indicated by auxiliary tests. A variety of regulators were used for analysis of both arent and subsamle air. They included two-stage Alhagaz, and single and two-stage Tescom regulators. The O /N /Ar analysis rocedures used at CSIRO [Langenfelds, 00] allow for cylinder air to be introduced into the mass sectrometer inlet system under variable flow rates. Many of the regulators were tested for flow rate deendence on O /N /Ar in the ml min 1 range. Significant variation was observed in the magnitude of flow rate deendence among individual samles (and/or individual cylinders or regulators), but there was no obvious difference among the three regulator tyes. No flow rate deendence was observed for some samles (e.g., d(ar/n ) varied by <10 er meg). Where significant variations were observed, they always showed falling O /N and Ar/N with increasing flow rate and a d(ar/n )/d(o /N ) relationshi consistent with diffusive fractionation. We thus assume that the sense of any fractionation is to always favor removal of lighter molecules from the air samles being analyzed and we reject any data obtained using flow rates above 5 ml min 1 that showed significant, relative deletion in O /N or Ar/N. In those cases where samles were analyzed with different tyes of regulators at the same cylinder ressure, 6of11

7 Figure 5. Measured difference in subsamles relative to arent cylinder air as a function of decanting flow rate for stainless steel (crosses) and aluminum (diamonds) cylinder data, with linear regressions shown as solid and dashed lines resectively. there was good agreement among measurements obtained using a flow rate of 5 ml min 1. The key oint is that while there may be some interference from analysis-related artifacts, we can be confident that this is small comared to the magnitude of changes attributable to the decanting rocess. [3] Figure 5 shows differences in the comosition of subsamles relative to that of their arent cylinder air. In calculating the differences, allowance was made for changes in comosition of the arent air, established from measurements before and after decanting events. Both d(o /N ) and d(ar/n ) show increasing deletion with flow rate and a relationshi [d(ar/n )/d(o /N ).4; Figure 6] that is consistent with diffusive fractionation. The molecular air ratio is significantly lower than that observed for standard in Figure 3 and more in keeing with redicted values for effusion or ressure diffusion than for thermal diffusion (Table 1). The factors of roortionality for other molecular air ratios are, at best, marginally significant relative to exerimental recision. Data in Figure 5 are lotted searately for subsamles in aluminum and stainless steel cylinders. There are no significant differences (at the 1s level) between the cylinder tyes in the sloes of the lines of best fit for any of the measured secies, indicating that the nature of the reciient cylinder has no influence on fractionation of the air being transferred. If the extent of fractionation is assumed to be exactly roortional to mass difference, the d(ar/n ) data can be used to estimate the effect on the other measured secies. Differences of about 10 and 0 er meg would be exected for d( 9 N / 8 N ) and d( 34 O / 3 O ) resectively at a flow rate of 4 L min 1, but neither data set is sufficiently recise to confirm or dismiss the link. A difference of 0.05 mmol mol 1 that would be exected for CO is also too small to be resolved with our gas chromatograhic recision. However, CO data for the two cylinder tyes do show a significant offset of about 0.08 mmol mol 1 that is indeendent of decanting flow rate. The absence of a corresonding signal in the other measured secies eliminates diffusive fractionation as the cause. We attribute the offset to ressure-deendent surface adsortion of CO on the internal surfaces of the aluminum cylinders. This would require CO to be referentially adsorbed (relative to O, N and Ar) onto the cylinder surfaces as samle air is introduced into the evacuated cylinders. Because any changes in CO mole fraction of samle air due to this mechanism are rimarily driven by ressure changes and cylinder materials, they would aear in Figure 5 as a constant, nonzero offset. We have frequently observed such effects in other exerimental activities involving contact of air with olymer surfaces (e.g., in regulators and other tests conducted at CSIRO) subject to changes in ressure, but the results resented here rovide our first firm indication of ressure-deendent surface effects on CO occurring inside high-ressure cylinders. [4] Without confirmation from the isotoic tracers d( 9 N / 8 N ) and d( 34 O / 3 O ), there is greater uncertainty in attributing the d(o /N ) and d(ar/n ) variations in Figure 5 to diffusive fractionation. However, there is suorting evidence to show that this is indeed the dominant modifying rocess. The method of removing air from the arent cylinders was similar to that used for standard 99191, so it seems likely that both exeriments would be subject to the same fractionating rocesses. There is comlementary information in measurements of the changing comosition of the arent cylinder air used in this exeriment. Where several subsamles were taken from the same arent cylinder at high flow rates, there was measurable enrichment of d(o /N ) and d(ar/n ) in the arent air. The magnitude of the changes was consistent with diffusive fractionation, as determined by mass balance calculations based on observed subsamle-arent differences. Figure 7 shows measurements from the arent cylinder that was most heavily deleted in ressure and which showed the largest enrichment in the measured secies. Observed changes in both d(o /N ) and d(ar/n ) over the full ressure range are within 0% of those redicted from subsamle-arent differences, consistent with diffusive fractionation being the main modifying rocess. The 0% discreancy might Figure 6. The relationshi of d(ar/n ) and d(o /N ) deviations in subsamles in stainless steel (crosses) and aluminum (diamonds) cylinders relative to their arent air, with the line of best fit determined by orthogonal distance regression. Data are for all flow rates as lotted in Figure 5. 7of11

8 Figure 7. Direct measurements from one of the arent cylinders used for decanting into subsamles, showing enrichment of measured secies with falling ressure. The solid line shows assumed values of the arent air for the urose of calculating subsamle-arent differences. Changes in the arent air are redicted (dashed lines) by mass balance calculations based on measured subsamlearent differences. An additional (dotted) line for CO shows the redicted change after allowance for an offset of 0.08 mmol mol 1 in aluminum cylinder subsamles due to nondiffusive rocesses. reflect other (e.g., surface) effects on O /N /Ar or errors in calculating subsamle-arent differences due to fractionation effects incurred when analyzing arent or subsamle air. The d(o /N ) and d(ar/n ) data in Figure 7 also show some divergence between observed and redicted curves in the 3 7 MPa ressure range, which is significant at least for the two measurements made at 3.4 and 4.7 MPa. This robably reflects fractionation-related artifacts in these measurements of the arent air. A similar calculation for CO roduces a larger discreancy between observed and redicted curves, but the discreancy can be removed by assuming the 0.08 mmol mol 1 offset in aluminum cylinders is due to non-diffusive rocesses and is indeendent of changes in Ar/N. Linear regression of the d( 9 N / 8 N ) and d( 34 O / 3 O ) data for this cylinder (not shown) indicated enrichment of 10 ± 6 and 9 ± 1 (1s uncertainties) er meg, resectively, again consistent with exectation for diffusive fractionation based on the d(o /N ) and d(ar/n ) observations. 4. Discussion [5] The results resented here describe modification of the comosition of air drawn from high-ressure cylinders and of the air that remains in the cylinders. For some secies, these effects can be large enough to seriously comromise atmosheric measurement rograms if they are either not recognized or not suitably managed within suorting calibration rocedures. A better understanding of the modifying rocesses can be used to imrove exerimental techniques, with immediate relevance to carbon cycle studies based on high-recision measurements of CO and O /N. [6] The signature of correlated variations among multile gas and isotoomer air ratios measured in our exeriments oints to diffusive fractionation as the main mechanism resonsible for the observed changes. Of these ratios, d( 40 Ar/ 8 N )/d( 3 O / 8 N ) has the highest signal-tonoise characteristics and returned values within the range exected for different diffusive rocesses (Table 1). This gas air ratio is well suited as a diagnostic of such rocesses. The relative abundance of 8 N, 3 O, and 40 Ar can be measured simultaneously by mass sectrometry, and with high recision owing to their high abundances in air. Substantial mass differences romote fractionation signals that are larger than for corresonding changes in the isotoomers of these gases. In the absence of suorting measurements, the main limitation of this gas air ratio is the ossibility that chemically fractionating (e.g., surface) rocesses might bear a similar signature to that of diffusive rocesses. The nature of any relevant chemical effects is not known at this time but could be better constrained with further testing. The ability to also measure d( 9 N / 8 N ) and/ or d( 34 O / 3 O ) rovides a means of distinguishing effects of diffusive and chemical fractionation, if the fractionation is large enough to roduce detectable changes in these quantities. Significant changes in d( 9 N / 8 N ) and d( 34 O / 3 O ) were observed in some of our exeriments, confirming diffusive fractionation as the dominant modifying rocess, though not excluding secondary effects on O /N /Ar or larger effects on other trace gases due to other rocesses. [7] The sense of the observed changes imlies that lighter molecules are referentially removed from cylinders during decanting of air and that this effect is exacerbated by higher flow rates. The exact cause of the fractionation, in terms of the different diffusive rocesses listed in Table 1 and the oint(s) at which it occurs in the flow ath of air drawn from a cylinder, has not been established. Observed multisecies signatures are not sufficiently recise to distinguish between the different diffusive rocesses, and this situation is further comlicated by exerimental evidence of significant ressure-deendence in thermal diffusion signatures [Keeling et al., 004]. The d(ar/n )/d(o /N ) ratio offers the best oortunity to make such a distinction. At face value, the results for standard favor thermal diffusion and those for the decanted subsamles favor ressure diffusion or effusion. However, the similarity between gas handling rocedures in both exeriments suggests that the modifying rocess(es) should also be similar. The differing d(ar/n )/d(o /N ) ratios for the two data sets may reflect a secondary effect on O /N /Ar due to surface adsortion or variations in fractionation factors, such as those imlied by the results of Keeling et al. [004] for thermal diffusion at different ressures. What can be said is that the diffusive fractionation is likely related to the assage of air either through the cylinder valve or the regulator. Both environments are characterized, at least art of the time, by narrow athways and gradients in ressure and temerature (e.g., due to adiabatic cooling of exanding air), all of which are factors conducive to fractionation. An interesting sideline to this discussion is the observation by Keeling et al. [004] of Ar/N drifts in high-ressure cylinder standards that were also a function of usage but 8of11

9 which were of the oosite sign to those reorted here, i.e., in the sense of rogressive deletion of Ar/N with declining cylinder ressure. The drifts were attributed to thermal diffusion on the basis of observed d(ar/n )/d(o /N ) ratios. This suggests that the magnitude and even sign of diffusive fractionation in such alications is critically deendent on the exact gas handling rocedures emloyed. [8] Another fractionating mechanism to consider here is gravitational settling of air inside a cylinder, whereby heavier molecules referentially accumulate toward the base of the air column. Its relevance to air standards stored in cylinders was noted by Keeling et al. [1998] who calculated that at barometric equilibrium, d(o /N ) would be enriched at the to of a 1 m column by 17 er meg. However, it would take 1 year to achieve this equilibrium and the gradient would likely be strongly diminished by mixing due to other influences such as convection due to differential heating or cooling of different arts of the cylinder. The high-ressure cylinders used in this study are 1. m high, have valves located at the to of the cylinders and have always been used and stored in a vertical osition. Thus gravitational settling may have contributed to some of the changes in comosition reorted here, although it can be eliminated as a major influence on the grounds that (1) even the maximum ossible d(o /N ) gradient of 0 er meg is too small to account for all of our observations, () there is no exectation of significant flow rate deendence during decanting, and (3) other decanting exeriments using horizontally ositioned cylinders and testing a lower flow rate regime [Langenfelds, 00] also exhibited flow rate deendent fractionation. [9] Our observations suggest that while diffusive fractionation can account for much of the CO drift observed in high-ressure cylinder standards as a function of usage (Figures 1 and ), there must also be some influence due to other rocesses in the aluminum cylinders. The observed CO drift in standard was about double that exected from diffusive fractionation alone, based on the other measured molecular air ratios, and the loss of 0.08 mmol mol 1 CO in subsamles decanted into the aluminum cylinders was indeendent of changes in the other tracers. We suggest at least the latter observation might be exlained by ressure-deendent surface adsortion of CO onto the wetted materials of evacuated cylinders of this tye. If so, this might involve the aluminum surfaces of the body of the cylinder, the brass, Kel-F or nickel surfaces of the valve, or the PTFE thread sealant used in attaching the valve to the cylinder. In our exerience, soft olymer materials such as PTFE have far greater roensity for adsortion of CO than do clean metal surfaces of aluminum or stainless steel. However, in terms of the effect on the stored air, this is mitigated by the small surface area of exosed PTFE as comared with the larger metal surface area of the cylinder body and valve. It is also our exerience that artifacts involving surface adsortion of CO are exacerbated by the resence of moisture. This resumably reflects a roerty of CO that imacts on its interaction with surfaces and which is related to its high solubility by comarison with the other gases considered in this study. [30] The magnitude of changes observed in our exeriments is large comared with recision and calibration stability requirements of atmosheric CO and O /N measurement rograms. Drifts in CO mole fraction of as much as 0.3 mmol mol 1 over the lifetime of a cylinder standard are significantly higher than the target levels for CO intercomarability among laboratories. If such drifts were to go unrecognized, they could lead to significant errors in carbon budget calculations. One area of work that would be articularly sensitive to such errors is in the inversion of atmosheric CO observations to establish regional surface fluxes using data from different sites and/or laboratories, if such data sets are not adequately intercalibrated. Awareness of the otential for such consequences arising from CO instability in air standards has led the CO measurement community to imlement elaborate calibration strategies. They involve regular calibration of a laboratory s rimary standards by the CCCL, which maintains the WMO CO scale using a manometric technique [Zhao et al., 1997; Tans et al., 003]. This technique is accurate in absolute terms to better than ±0.1 mmol mol 1. Laboratories are also required to maintain a hierarchy of calibration standards that enables adequate roagation of CO assignments from rimary through to working standards. These can be resource intensive activities, esecially where analytical techniques demand a high rate of consumtion of air standards, as is the case for the NDIR technique most commonly used by CO measurement laboratories. For examle, working standards used for the historical in situ CO monitoring rogram at Cae Grim (Figure 1) have an average lifetime of only 3 months. Understanding the rocesses resonsible for CO drifts in air standards has the otential to imrove the efficiency of calibration rocedures, and thereby reduce caital, logistical, instrument, and oerator time costs associated with maintaining CO measurement rograms. [31] Carbon cycle studies based on O /N measurements are also suscetible to calibration errors from drifting air standards. A key motivation for measuring O /N is to use its atmosheric trend to constrain the artitioning of global utake of fossil CO between terrestrial and oceanic reservoirs. For such measurements to be useful, the trend in d(o /N ) must be known to better than er meg yr 1, yet drifts as large as 80 er meg were found here in one cylinder standard. At this time there is no established absolute scale used within the O /N measurement community, although efforts are underway to establish such a scale [Toru and Kazuto, 003, Annex A4]. Individual laboratories have instead based their calibration on relative stability among a suite of regularly analyzed air standards. This strategy is otentially more suscetible to calibration errors arising from unrecognized drifts in standards than the system in lace for CO measurement rograms where measurements are ultimately linked to an absolute scale. Assessment of the relative stability of standards has revealed significant uncertainty of ±1.5 er meg yr 1 in the stability of CSIRO s O /N scale [Langenfelds, 00] but stability of better than ±0.5 er meg yr 1 in the scale maintained by the Scris Institution of Oceanograhy [Keeling et al., 1998]. [3] Actions can be taken to reduce the fractionation of air standards and the imact on calibration of measurements. Standards can be retired from service before falling to low ressures where drift rates accelerate. While our exeriments have not identified all of the imortant arameters, they do show that fractionation increases with flow rate, at least within the range of flow rates tested. Therefore 9of11

10 oerating at low flow rates should hel to limit the extent of fractionation and resultant drift in standard comosition. Further reventative measures are ossible and will deend on the nature of the fractionating rocess. For examle, thermal diffusion could be limited by suitable insulation or temerature control of affected comonents (e.g., cylinder valves or regulators). Maintaining a caability for highrecision measurement of O /N /Ar and isotoomers of O and N can be a valuable diagnostic for (1) monitoring fractionation of air standards and thus constraining drift in secies such as CO and () develoing techniques to reduce modification of air standards. Further testing should hel to identify those materials resonsible for the susected CO surface effects in aluminum cylinders. Such effects might be avoided or reduced by better selection of materials, methods of cylinder construction, surface assivation treatments, and/or exclusion of moisture. [33] These issues are also directly relevant to the rearation of absolute standards. Prearation of CO -in-air standards in the same tye of aluminum cylinders used here may be subject to significant modification by both surface and diffusive effects. To be useful for this urose, the absolute accuracy and recision of the CO mole fraction assignment must be at least comarable to that currently achieved by the CCCL using their manometric technique. The usefulness of standards would be seriously comromised if they were subject to the same 0.08 mmol mol 1 change in CO mole fraction that we observed in our decanting exeriment. Errors could conceivably be even larger when cylinders are filled to their maximum ressure of about 15 MPa, as comared with the fill ressure of just 0.5 MPa used for our exeriments, and after allowance for diffusive effects. Our exeriments were restricted to limited ressure and flow rate ranges and thus may not be directly alicable to the gas handling rocedures or equiment used in the rearation of high-ressure cylinder standards. Nevertheless, there is otential for significant errors, and due attention should be aid to limiting or quantifying any such effects. [34] This study has focused on a small number of atmosheric secies and on one asect of maintaining an atmosheric measurement rogram, namely calibration using air standards. However, the fractionating rocesses discussed here are otentially relevant to other secies and other gas handling alications. Other secies measured in CSIRO GASLAB, such as the isotoomers of CO, mainly the 13 C/ 1 C ratio, and mole fraction measurements of N O, CH 4, and H, could be significantly affected in some circumstances, although none are affected nearly as much as O /N /Ar or CO in terms of routine exerimental activities. This reflects either lower recision of measurement, lower rate of consumtion of air standards, and/or lower ratio of the magnitude of inferred fractionation effects to biogeochemical signals of interest. We have seen numerous other examles of diffusive fractionation occurring during the collection, storage and analysis of various air samles analyzed at CSIRO [Langenfelds, 00]. For examle, mass deendent changes in comosition of a subset of Cae Grim Air Archive samles were attributed to effusion of air through tiny leaks in the corroded walls of the stainless steel cylinders. Flask air samles collected from aircraft-based samling above Cae Grim were in some cases subject to altitude-deendent fractionation, with d(ar/n ) variations of as much as 00 er meg observed over an altitude range of 7 km. Systematic differences were also observed between ground and aircraft-based samles collected at the same altitude at Cae Grim (mean Dd(Ar/ N ) = 40 er meg), suggesting diffusive fractionation of air in one or both intake systems. As exerimental techniques imrove and scientific demands for higher recision and accuracy continue to increase, fractionation is likely to become a matter of concern for a wider range of atmosheric measurement alications. It is hoed that our exeriences will hel to address these roblems. Aendix A [35] The derivation of molecular air ratio values for ressure diffusion listed in Table 1 is given here. The ratios were calculated from equations (1) and () and take the form d(a/b)/d(c/d) for four secies a, b, c, and d. If amount of substance N for any one secies undergoing diffusive transort is roortional to mean molecular velocity C, then it follows that the ratio of secies a and b will be modified such that and Na 0 Nb 0 ¼ C a C b N a N b a = b Þ ¼ C a 1 C b ða1þ ðaþ The molecular air ratio is thus given by the ratio of the differences in mean velocity of each gas air as a roortion of the mean velocity of secies b and d, resectively, d a Ca Cb ð= b Þ c = d Þ ¼ C b C c C d This relationshi can alternatively be exressed as C d a = b Þ c = d Þ ¼ C d C a C b C b C c C d ða3þ ða4þ To define molecular air ratios for ressure diffusion in air, the mean velocity difference of each secies is first calculated relative to air. Assuming no gradients in mole fraction or temerature and no external forces, equations (1) and () give the relationshi for secies a 1 = 3 ktðm a þ m air Þ C a C air ¼ 8n s a þ s air m a m air n an air ðm air m a Þ r ln nr ða5þ Substituting x a = n a /n and x air = n air /n and rearranging C a C air ¼ 3 ffiffiffiffiffi kt 8r ffiffiffiffiffi r ln m a þ m 1 = air ðm air m a Þ m a m ða6þ air s aþs air The difference in mean velocity for two secies a and b can then be calculated from the exression C a C b ¼ ðc a C air Þ ðc b C air Þ ða7þ 10 of 11