Absolute Measurement of the Abundance of Atmospheric Carbon Monoxide C

Home , Equivalent concentration

Absolute measurement of the abundance of atmospheric carbon monoxide C. Brenninkmeijer, C. Koeppel, T. Röckmann, D. Scharffe, Maya Bräunlich, Valerie Gros

To cite this version:

C. Brenninkmeijer, C. Koeppel, T. Röckmann, D. Scharffe, Maya Bräunlich, et al.. Absolute measurement of the abundance of atmospheric carbon monoxide. Journal of Geophysical Research: At- mospheres, American Geophysical Union, 2001, 106 (D9), pp.10003-10010. 10.1029/2000JD900342. hal-03117189

HAL Id: hal-03117189 https://hal.archives-ouvertes.fr/hal-03117189 Submitted on 8 Feb 2021

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 106, NO. D9, PAGES 10,003-10,010, MAY 16, 2001

Absolute measurement of the abundance of atmospheric carbon monoxide

C. A.M. Brenninkmeijer,C. Koeppel,T. R6ckmann,D. S. Scharffe, Maya Br•iunlich,and Valerie Gros

AtmosphericChemistry Division, Max PlanckInstitute for Chemistry,Mainz, Germany

Abstract.The main aspects of anabsolute method for measurementof the mixing ratio of atmos- phericcarbon monoxide (CO) are presented. The method is based on cryogenic extraction of CO fromair afteroxidation to CO2followed by accuratevolumetric determination. Gravimetric measurementis usedto determinethe quantity of sampleair processed.In routineoperation the overall errorcan be kept below 1%. Furthermore, the results of a laboratoryintercomparison are analyzed. It is shownhow offsets in thecommonly applied analytical methods can occur and how these can seriouslyaffect results obtained at the low concentrationend (<100 nmole/mole).

1. Introduction Chartography(SCIAMACHY)). Validation of dataobtained by suchremote sensing instruments is important. Theneed to betterknow the budgetof CO is reflectedby the Relevantis alsothe question whether there have been gradual scaleof internationalefforts in measuringand modeling the distri- changesin OH [Crutzenand Zimmermann,1991; Krol et al., butionand interactionsof this ubiquitouschemically active trace 1998; Prinn et al., 1995]. For establishingthe causes,knowledge gas[Novelli, 1999; Sanhueza et al., 1997;Moxley and Srnit,1998; aboutchanges in CO and also CH4 of similarprecision is neces- Reichle et al., 1999; Connors et al., 1999; Bakwin et al., 1994; sary.The same applies to lnCO,for whichit maybe particularly Novelli et al., 1998b, and the special issue of Chemosphere: importantto have accurateobservations because of its uniquepo- GlobalChange Science, September1999, 28 furtherpapers]. Be- tential as a long term diagnosticfor changesthe oxidativecapacity ing themain reaction partner of hydroxyl(OH), changesin CO af- of the atmosphere.Although one may never be able to model at- fect OH and vice versa. Because of the central role of OH in the mosphericCO (or laCO for that matter) with such a highdegree of chemistryof the atmosphere,CO is an importanttrace gas. confidencethat changesof the order of 1% would matter,there is Closelyrelated to the cycleof CO is thatof lnCO,which no doubt that it would be unwise not to have accurate information sharesthe same sinks, namely OH and soils. Given that the major atpresent. Even though it seemsthat an accuracy of 1%is feasi- originofthe inc in laCO is cosmogenic, andtherefore independent ble,there have been serious problems. ofthe sources ofCO itself, lnCO is of interest, even though it pre- Mostof thetechniques formeasuring COare described by sentsonly a verysmall fraction of troposphericCO rangingbe- Novelli[1999]. The analysis of COis almost exclusively based on tween10'•l and10 -12. This is equivalent toonly about 5 to 20 relativemethods, thatis the comparison withstandard mixtures of moleculespercm 3air STP. Nonetheless, lnCOis a uniqueuseful CO, and the procuring and maintaining oftraceable standards is tracerfor diagnosing OH distribution, large-scale hemispheric cir- difficultat times. Since the discovery of large differences between culation,and fluxes from the stratosphere into the troposphere CO calibration scales used by different laboratories [Novelli et al., [Volzet al., 1981;Brenninkrneijer, 1993; Brenninkrneijer and 1991],good progress in standardization hasbeen made [Novelli et ROckmann,1998; Brenninkrneijer etal., 1992;Mak and Southon, al., 1994;1998a]. Primary gravimetric standard mixtures have 1998;JOckel et al.;2000, Quay et al., 2000]. Because determina- been prepared and standards were prepared for laboratoryinter- tionof the abundance of•4CO isbased onthat of CO, the ability to comparison.[Novelli et al., 1994].Two intercomparions have makeaccurate and precise observation ofCO is of direct relevance been organized by NationalOceanic and AtmosphericAdmini- forlnCO applications. stration/ClimateMonitoring and DiagnosticsLaboratory Of importanceishow frequently, how precisely, and how ac- (NOAA/CMDL),in 1994and 1999. These "ring tests" enable di- curatelyatmospheric CO shouldbe measured.The question of rectcomparison and allow the various laboratories to systemati- frequencyand precision islinked to the variability ofCO which is callyfollow and diagnose drifts in COstandards. This paper dis- dominatedby its lifetime and influencedby localized sources. cusses an absolute method for the determination of CO. Its results Generallythe requirementsare not as high as for CO2, CH4, and differ significantlyfrom those of severalother laboratoriesin the N20, and a 1% precisionis consideredadequate. The accuracy 50 to 100 nmole/molerange, and the possiblereasons will be dis- with whichone hasto knowthe mixingratio of CO is ratherfun- cussed. damentalbecause of the long term atmosphericchanges. Besides the analysisof samplesof air on site or in the laboratory,remote sensingviasatellite borne optical detectors israpidly gaining im- 2. ExperimentalMethods portance(e.g. Measurementof Air Pollutionfrom Space(MAPS), Measurementof Pollution in theTroposphere (MOPITT), and 2.1. ExtractionSystem ScanningImaging Absorption Spectrometerfor Atmospheric The techniqueof Stevens[Stevens and Krout, 1972] for the Copyright2001by the American Geophysical Union. isotopicanalysis ofCO is used. In essence, CO2 is removed from air, after which the CO contentis oxidizedand subsequentlycol- Papernumber 2000JD900342. lectedas CO2. The mixing ratio of CO is obtainedby determining 0148-0227/01/2000JD900342509.00 the mole fraction of CO2. The system(Figure 1) is an improved

10,003 10,004 BRENNINKMEIJER ET AL.: ABSOLUTE CO MEASUREMENT

Lab air PMST

PIR

DF MAN SB

CAG RDT1 RDT2

VENT DP TV

Figure 1. The CO extractionsystem. The following abbreviationswere used:ZAG, zero gas generator;CAG, calibrationgas; PR, pressureregulator; MFC, massflow controller;RDT, Russiandoll trap; SR, Schatzereactor; COT, collectiontrap; H, heater;DP, diaphragmpump; PMST, purge molecularsieve trap; DF, drying finger; MAN, manometer;PT, pressuretransducer; SB, samplebottle; PIR, Piranivacuum gauge; HV, high vacuumpump stand;BV, buffer volume;TV, throttlevalve; BGM, bellowsgas meter.

versionof a predecessor[Brenninkmeo'er, 1993]. The main prop- min afterwhich the air is shuntedthrough the Schiitzereactor. The erties of the CO extractionprocedure are as follows: (1) It is an systemis flushedfor 10 min and the collectiontrap cooled.After absolutemethod in which the CO is quantitativelyextracted; (2) it 1 min, the glassfiber thimbleshave reachedliquid nitrogentem- is an integratingmethod, which implies that by increasingthe perature,and sample collection of CO2 commences.The integrator samplesize throughprocessing more air, the signal-to-noiseratio of the flow controlleris initialized,and the gasmeter readings are improves;and (3) the methodhas a low detectionlimit in the sub- recorded.Subsequently, the flow is increasedto 5 L min'• upon nmole/molerange. whichthe pressureincreases to 180 hPa.This is the highestpossi- Theflow of sampleair (5 L min-• STP)is regulatedwith a ble pressurefor optimizingthe residencetime of CO moleculesin thermal massflow controller(Hastings type HFC-202F). Con- the Schiitzereactor without freezing out oxygenin the traps. densablecompounds are removedby two ultra efficientRussian After processing350 to 400 1 of air the flow is throttledto 1.5 Doll trapssubmerged in liquid nitrogen[Brenninkmeijer, 1991]. L min4 andterminated after 1 min.The bypass isopened, and the Suchtraps consist of stainlesssteel cylinders incorporating three Schiitzereactor is isolated.The inlet valve of the collectiontrap is concentricborosilicate glass fiber thimbles.Cooling at the outlet closed.The processpump valve is closedat 5 hPa, and the valve of thesetraps is preventedby sheathedthermocouples heater ele- to high vacuumpumping stage (Pfeiffer molecular drag pump) is ments[Brenninkmeijer and Hemmingsen,1988]. After removalof openedto evacuatethe collectiontrap. Subsequently,the U tubeis all CO2, N20, and hydrocarbons(C3 and higher),CO is oxidized submergedin liquid nitrogen,its outlet valve is closed,and the to CO2in a reactor(kept at 35øC)filled with 0.8 I of Schatze's largedewar surrounding the collectiontrap is removed.During 5 reagent[Smiley, 1965], which consistsof acidifiedI2Os on sili- min the sample CO2 is distilled into the U tube. After further cagel. The reagentdoes not induceisotopic exchange and is very evacuation,the inlet valve of the U tubeis closed,and the sample efficientin oxidizingCO. CO2 is distilledinto a smallfinger containingP205 for removing The smallquantity of CO2 from the oxidationof CO is trapped tracesof water.Next the CO2 is transferredinto a manometerby at liquid nitrogen temperature in a glass Russian Doll trap cryogenicdistillation. After measuringpressure (typically 20 to 60 [Brenninkmeo'erand ROckmann, 1996]. Measurementof the hPa)and the temperature, the sampleCO2 is distilledinto a sample quantityof CO2 producedis volumetric.The respectivemanome- bottlefor transferto the massspectrometer. The manualprocedure ter consistsof a volume isolatedwith a 5 mm bore O-ring glass can be automated by using methods developed before valve [Brenninkmeo'er,1981] and is fitted with a piezo-resistive [Brenninkmeo'er,1983]. absolutepressure transducer (Institute of Geologicaland Nuclear Sciences(IGNS), model 9401). The volume being only 0.816 cm 3 increasesthe pressureobtained and thereforethe resolution. 2.3. Determinationof the Quantity of Air Calculationof the mixing ratio is basedon the volumeof CO- 2.2. Extraction Procedure derivedCO2 and the quantityof dry air processed.The quantityof air is determinedusing the electronicmass flow controller.This When not in use, the systemis back-flushedwith laboratory device is calibratedusing a gas burette.Furthermore, a bellows air cleanedusing a 10 L reactorwith 13X syntheticzeolite. The type gas meter integratesthe amount of air at the exhaustof the Schatzereagent is kept sealedoff. Air samplesare processedas system.A buffer volume betweenthe diaphragmpump and the follows. Back-flushingis terminated,and the two RussianDoll meter reducespressure pulses and improvesthe accuracyof the trapsare submergedone afterthe otherin liquidnitrogen. Sample reading. air is admitted via the mass flow controller,and pumpedaway To ensureprecise and accuratemeasurement of the amountof usingthe diaphragmpump (Vacubrand, type MZ4, O-ring setfor air, the cylinder containingthe air to be processedis at times improvedsealing was used). At aninitial flow rate of 1.5L min't placedon an electronicbalance. With a resolutionand stabilityof the systempressure reaches 50 hPa. The systemis flushedfor 10 0.05 gram,the amountof air processed,which typically is 350 L, BRENNINKMEIJER ET AL.' ABSOLUTE CO MEASUREMENT 10,005

canbe measuredto betterthan 0.1%. Thereare threedevices me- accuratelymeasured using a micrometer.By doing tests as de- teringthe amount of air processed(the correction for theamount scribedabove using two or three needlesinserted into the calibra- of air CO2extracted by theRussian Doll cleaningtraps is too tion volume (thus reducingits volume), a value of 815.1 gL was smallto be significant),namely the integratingmass flow con- obtainedfor the manometer.(In this determinationthe volumeof troller, the bellowsgas meter, and the electronicbalance. The the needles,provided they are identical,does not play a role.) Fi- mostaccurate and precise measurement is by the electronicbal- nally, by also usingthe calculatedvolume of the needles,a value ance. of 816.8 gL was obtained.The averagedvalue is 816.0 +1 gL. 2.4. VolumetricDeterminations of CO2 2.5. Verification by Calibration Runs The determinationof the smallamount of CO2recovered is critical.The volume of themanometer, the pressure, and the tem- Regular calibrationruns are performedinjecting isotopically peraturehave to be knownaccurately. The electronicpressure definedCO into a flow of air devoidof CO (zero air). A main rea- transducerhas in the range0 to 200 hPaa resolutionof 0.01 hPa. sonfor calibrationis to establishthe effect of the oxidationstep on The zeroreadings of thisgauge over a periodof I yearranged the oxygen isotopiccomposition of the CO2 produced.As the without adjustmentaround 0.1 hPa, with a standarddeviation of carefulwork by Stevenshas shown,this oxidationsolely addsan lessthan 0.1 hPa.The dayto dayvariability is typically0.02 hPa, oxygen atom to the CO molecule, and there is no isotopic ex- andbefore each measurement a zero reading is madefor correc- change with the CO2 produced [Brenninkrneijer, 1993; tion.The atmosphericpressure reading of the gaugeis compared Brenninkrneijerand ROckmann,1997]. In additionto the isotopic everyday with that of a preciseatmospheric pressure gauge (Paro- information,the calibrationruns give a diagnosticof how well the scientific,Digiquartz 740). This instrumentis extremelystable, systemproduces quantitative conversion and recoveryof CO as is hasa resolutionof 0.001hPa, and is calibratedagainst the read- discussed below. ingsfrom the weatherservice. Furthermore, the linearityof the A certainquantity of CO is injectedand exactly this shouldbe pressuresensor is betterthan 0.1%. Note that the linearityin the recoveredas CO2. The quantityof air processedduring a calibra- actualvolumetric measurement also assumes a constancyof the tion run is not of primary importanceprovided the zero air is de- volume.To ensurereproducibility and minimal volume variations, void of CO. The flow ratesof both the zero air and CO injected a 5 mm Vacutap[Brenninkmeijer and Louwers,1985] is used. can be varied thus providingdifferent concentrations of CO. The Testsperformed by repeatedsealing of the manometervolume zero air is producedusing heated reactors that containI L of Hop- show that the error made is less than 0.1%. calite, or a Pt on aluminum-oxidecatalyst (Merck). Temperatures Thevolume of themanometer is establishedusing a calibrated of 100to 200øC are applied. After oxidation of theCO content, the volume.This consistsof a valve sealingoff a shortsection of air is passedthrough a bed of 13X syntheticzeolite. This stripsout stainlesssteel tubing cut froma longersection. By severaltimes water, CO2, and other impurities.With the zero air generatorsair fillingthis longer section with water and accurately weighing, the withoutCO can be produced(less then 0.5 nmole/mole)at flow volumeof the shortsection (0.2442 cm 3) wasdetermined with ratesof upto 20 L rain4. It is assumedthat at 5 L min4 thezero highaccuracy. By fillingthis calibration volume with dry CO2 at a air contains less than 0.1 nmole/mole CO. Tests in which the mo- well-definedpressure (mostly near 1000 hPa) and temperature, lecular sieve of the zero air generatoris replaced by Drierite andtransferring this CO2 into the manometeraccurate calibrations (Hammond,Ohio), which doesnot removeCO2, do not showany aremade. To excludemeasurement errors or mistakes,several ap- differencein yield or isotopiccomposition. This is proof of the proacheswere used, which lead to the values in Table 1. In the extreme efficiency of the two Russian Doll traps (combined firstseries of teststhe calibrationvolume was filled with CO2at >99.99997% CO2 removal).The small quantityof condensables nearatmospheric pressure, cross-checked with the Digiquartzin- (CO2 and SO2)that is recoveredin a zero run is subtractedfrom strument.This CO2 was distilledinto the manometervolume, and the quantityobtained for a sample. a pressurereduction of 0.3001occurred, leading to the manometer Experienceshows that the zero yield of the systemincreases volumegiven in the table.This valueis subsequentlycorrected when not in use. After one run the zero yield returnsto a low usingthe van der Waalsequation. The endresult is 816.1 gL. In level between0.1 and 0.2 hPa. This is equivalentto approxi- the next series of tests the calibration volume was filled to lower mately 0.25 to 0.5 nmole/mole for a 350 L air sample.A correc- startingpressures. In this instancethe van der Waals correction tion for this is applied,and the remaininguncertainty is lessthan becomesnegligible; however, the starting pressure is notknown as 0.1 nmole/mole. Thus, for air samplescontaining 50 to 200 well asin thepreceding case (no directcomparison at nearatmos- nmole/moleCO, the uncertaintyin the blank correctionintroduces phericpressure is possible).The endresult is 815.9gL. Thisalso an error of 0.2 to 0.05%. confirmsthe high degreeof linearityof the pressuretransducers For calibration,CO is not injectedas pure CO gas, but as a used. mixtureof 269 ñ 3 [tmole/moleCO in nitrogengas. This mixture The problemintrinsic to determiningthe manometervolume is meteredusing a thermalmass flow controller(Hastings HFC- accuratelyis that internal volume measurementis difficult. Inde- 202A) which is occasionallycalibrated with a gas burette.Ac- pendentconfirmation of the volume calibrationwas thereforecar- cording to the supplier of the mixture (NZIG, Lower Hutt), it riedout using external volume determinations. Three identical cy- contained271 ppm by volume CO; accordingto a bulk analysis lindricalstainless steel needles, with a diameterof 1.5 mm, were carriedout in the laboratoryin New Zealandit contained269 ñ 3

Table 1. Determination of the Volume of the Manometer a • " Stan'dard-- Lø•ve 'i•ressure ' 2.:3-needles' 2Needles AbS01u•e • Ratio 0.3001+0.0001 0.29934-0.0001 n.a. c 0.167954-0.0002 Volume,gL 813.734-0.3 815.94-0.3 815.14-0.6 813.64-1 Van der Waals corrected 816.14-0.3 815.94-0.3 815.14-0.6 816.84-1 aThe last row lists the final value after correction forthe non-ideal gasbehavior of CO•. bIn thisdetermination the volume of the needlesis irrelevant. cHere not applicable. 10,006 BRENNINKMEIJER ET AL.' ABSOLUTE CO MEASUREMENT

18- cidentallybe compensatedby a concurrentincrease in trappingef- ficiencyof the collectiontrap, is unlikely. No changein isotopic compositionwas observed.Independently, the effectof lower lev- 14 els of liquid nitrogenaround the samplecollection Russian doll trap hasbeen investigated earlier [Brenninkmeijerand ROckmann, 1996]. Altogetherthese tests corroboratethat the values of the m 10 main operatingparameters are not in a criticalrange.

m 8 Despite the a priori linearity of the system,evidence for a

o small systematicdeviation was obtainedby analysisof data for '- 6 duplicates.For a total of 22 cylinderswith air samplescollected at E Spitsbergen[R6ckmann and Brenninkmeijer,1997], two sub- ::3 4 Z samplesof nominally 400 and 700 L were processed.It turnsout 2 that the mixing ratios establishedfor the 700 L samplesare sys- tematicallylower by 1.3%. This is a puzzlingeffect becausethere 0 97.5 98.0 98.5 99.0 99.5 100.0 100.5 101.0 101.5 is no obvious mechanismby which less CO2 would result for Yield [%] largersamples, or an error in the amountof air processedwould be made.The causeof the effect is tracedback to the efficiencyof Figure 2. Frequencydistribution for 67 calibrationruns performed the Russian doll collection trap. Apparently, when the trap is over 1 year. cooledfor a longerperiod of time, its efficiencydrops. Because nearly all samplesand calibrationsare over a period of lessthan I•mole/mole. On the basis of calibrationsusing this mixture, the 80 min, during which 350 to 400 L is processed,this effect had yield of the extractionsystem is 99.8%, with a standarddeviation escapedobservation during normal operatingconditions. To de- for an individual run of 0.7%. These numbers are based on 67 de- termine the effect, normal calibrationswere followed by delayed terminationsfor September1998 to October 1999. No calibration calibrationsduring which no calibrationmixture was injectedinto result has been omitted. Two personshave operatedthe system the zero-air flow during the first hour. Under these conditions without this causingsystematic differences. The frequencydistri- therewas a drop of 2 + 0.5% in efficiency.These tests confirm the bution obtainedis symmetrical(Figure 2). The near 100% yield existenceof a lossof efficiencywhen samplesare processedover and 0.7% standarddeviation prove independentlythat the CO can a longerperiod of time. The value of 2 _+0.5%will not be usedto be determinedwith a precisionand accuracyof about 1%. This correctfor the lossof efficiency.The 22 samplepairs of 400 and overall test involvesconversion, trapping, releasing, and transfer- 700 L give the more accuratevalue of 1.3%. ring. The causefor the abstruseloss in efficiencywhen sampleair is processedover longer periods is not fully understood.The 2.6. Tests of the Overall Efficiency amountof CO2 seemstoo small to affectthe propertiesof the large surfacearea of glass available for trapping it. Analysis of the In total efficiency the conversionof CO to CO2 is also critical. compositionof the air passingthrough the systemdid not show If this stepis not quantitative,inserting a secondSchiitze reactor unexpecteddelays in one of the main components.A recent im- into the systemand thus doubling the reactiontime for CO will in- provementto the procedureis to heat the collectiontrap to about creasethe yield. If, for instance,the yield is only 90%, the addi- 50øCduring back flushing. After correcting the results for the 700 tion of a second reactor boosts it to 99%. A less cumbersome and L samplesfor loss in efficiency,the duplicatescan be usedto as- effectivealternative is the reductionof the pressurein the Schiitze sessthe reproducibilityof the system.A standarddeviation for a reactor during sample processing.Accordingly, by using two singledetermination of only 0.5% is obtained.Thus, for routine processingpumps in parallel the systempressure was reduced operation,a randomerror of 0.5% can be guaranteed.This error is from the standard 180 hPa to 50 hPa. This reduces the residence even smaller than that obtained for the calibration runs. time of a CO molecule in the reactor from about 2 s to 0.5 s. It can be shownthat when the yield during the normal 180 hPa condi- 2.7. Linearity Tests tionsis 99%, the yield shoulddrop substantially to a mere70%. In contrast,during these experiments a dropin efficiencyof lessthan Deviations in linearity are investigatedusing two different 1% was observedfrom which it is calculatedthat the efficiencyis linearity tests.Test one is basedon injectingthe calibrationgas over 99.9999%. During thesetests the pressurein the entire sys- mixture at different rates. With the range of the massflow con- tem was lowered, from which it is inferred that also the trapping trollerbeing 0 to 10 mL min-• thereis a practicallower limit of of CO2 in the collectiontrap is nearly 100 % efficient.That the about1 mL min'•. Belowthis the uncertainty in the injected quan- yield would drop in the Schiitzereactor, and that this would coin- tity becomestoo large.The resultsare shownin Table 2. For each

Table2,. Linearity,,.Test1 Runa InjectionFlow, mL Mixing'Ratiob 'inj'e•i•h Quantity, Recov&red Quantity,' Yield, min-• •tL pL % 1600 1.0510 58 23.19 23.32 100.56 1601 1.0510 58 23.27 23.34 100.30 1602 2.0441 113 45.16 45.35 100.42 1609 2.0448 113 45.20 45.54 100.75 1603 3.0409 169 67.99 68.12 100.19 1608 3.0280 169 66.81 66.75 99.91 1599 3.9974 223 88.41 88.52 100.12 1604 4.0205 223 89.95 89.63 99.64 1605 6.3471 349 139.71 139.92 100.15 1606 8.0916 446 178.42 178.05 99.79 1607 •8738 544 217.58 215.76 99.16 aOctober 1998. bNominal mixing ratios in nmole/mole related to 400 L sampleair. BRENNINKMEIJER ET AL.' ABSOLUTE CO MEASUREMENT 10,007 settingof thecalibration gas mass flow controllerit wascalibrated BecauseRussian doll traps are efficient to a degreethat can againsta gasburette with a pistonsealed by a mercuryring. At a cause a gas chromatographicdelay for CO passing through flow rateof 5 L min'l STPzero air, anda totalof 400 L air, the [Brenninkmeo'erand ROckmann,1996], and becausethe 1 L equivalentmixing ratio varied from about 60 to 550nmole/mole. Schatze reactor itself in effect is a gas chromatographycolumn, The yield obtainedwas 100.1% with a standarddeviation for an the speedwith which a pulseof CO passesthrough the systemwas individualmeasurement over the entirerange of 0.45%. The four measured.CO at 8 cm3 min 'l wasinjected only during the very last duplicatedeterminations agree well within 0.4%. A linear least 5 min of a run. The correspondingyield was 99% which assures squaresfit to the data give for the recoveredquantity gives y = that there is no significantdelay or retentionof CO. The testscon- 0.9927x+0.52 nmole/mole (r• = 0.99996).The datasuggest a sistentlyconfirm the high linearity, quantitativeyields, and high smalldrop in efficiency.At 544 nmole/molethe yield is 1% below reproducibility.One remainingquestion is whetherit is pure CO2 that obtainedfor the lower concentrations.Furthermore the yield that is collected.Mass spectrarecorded at high amplificationcon- is higherthan that obtainedfor the calibrationruns (99.8%). The firm purity of the samplecollected. calibrationruns are performedat nominally200 nmole/mole,for A sourceof interferencecould be the oxidationof trace gases whichthe linearregression gives 99.5%. Thereare no significant other than CO to CO2 in the Schatzereactor. This was testedby discrepanciesbetween the results. Stevensand Krout [1972] and is a negligiblesource of interference Linearitytest 2 is basedsolely on the durationof the injection for normal air samples.Generally for the lower CO mixing ratios periodat a constantsetting of 2.0 mL min-l. Thistest checks the in remoteair, concurrentnonmethane hydrocarbon mixing ratios linearity in the low nmole/molerange up to 300 nmole/mole. are also lower. Testsbased on the injectionof C2H4into calibra- Calibrationgas was passedthrough a three-way valve into the tion or zero air has no effect. Injection of 100 gmole/moleCH4 zero-airflow allowinga rapidswitching between waste and injec- does not changethe amount of CO• recovered,or its isotopic tionposition. Injections ata flowof 2.0 cm 3 min -I wereperformed composition. Higher hydrocarbons(>C3) are effectively trapped for 5 min (with incrementsof 10 min) to a maximumperiod of by the Russiandoll traps and are presentat very low concentra- 120 min. The minimum equivalenteffective concentrationwas tionsonly. thereforeonly approximately 6.5 nmole/molefor the equivalentof 400 L of air processed.Figure 3 showsa goodlinearity for up to 2.8. Accuracy and Precision about60 hPa, and a very smallrandom error. Applying a linear regression,excluding the uppertwo datapoints gives y = 1.087lx Estimated and measured random errors for the 50 nmole/mole + 0.2214hPa. The standarddeviation for a singledetermination is level and processing350 L of air are as follows:blank correction 0.4%. Usingthis formula,the deviationsfor the two highestcon- 0.2%, pressurereading 0.3%, temperatureof CO: samplein the centrationsare 0.4 and 1.4% respectively.The test suggestsa manometer0.3%, andquantity of air 0.1%. smalldrop in efficiencywhen air is processedover long periods of The combinedrandom error therefore is approximately0.5%. This time. This is equivalentto the effectnoted for the processingof canbe comparedwith the independentlydetermined value for the the 400 and700 L subsamplesof the air cylindersand for the ex- measuredstandard deviation of 0.5% which is based on dupli- tended calibration run. The effect is not of relevance for the rou- cates. The precision obtained therefore is 0.5%. For 200 tine conditions.Test number2 is not the exact equivalentof the nmole/molethis valueimproves to about0.4%. injectionof, for instance,7 nmole/moleover the full periodof Estimatedvalues for the maximum systematicerror are as sampleprocessing. It doeshowever establish the linearityof the follows:Quantity of air 0.2% and quantityof sampleCO: 0.3%. systemin the importantlow concentrationrange. The maximumdeviation is thus0.5%. This canbe comparedwith the independentlydetermined value using the calibrationruns of 0.2 +1% deviation.

160 3. Laboratory Intercomparison Tests 140 / 3.1. Analysis and Evaluation of the 1993-1995 Laboratory Intercomparison 120 / - Over morethan 10 yearsthorough work hasbeen carried out

100 to continuouslyimprove the qualityof CO measurements[Novelli et al., 1994].Ring testswhich are laboratoryintercomparisons in / whichlaboratories analyze subsamples of air from the samesuite 80 of cylindershelp to comparedata [Novelli et al., 1998a].It is im- / - portantnot only to comparescales but alsoto understandthe dif- 60 ferences.A predecessorof the presentsystem [Brenninkmeijer, / 1993]was used in the 1994-1995ring test [Novelli et al., 1998a]. 40 Fourcylinders with nominally50, 100, 150 and200 nmole/mole wereanalyzed at theNational Institute of Waterand Atmospheric / Research(NIWA), New Zealandin 1994.Because it was consid- 2O eredthat by withdrawingas muchas 350 L of air fromthe cylin- dersthe CO concentrationmight be somehowaffected, thus possi- / bly spoilingthe resultsof other laboratorieslater in the test se- 0 20 40 60 80 1 O0 120 140 quence,it wasdecided to usethe NIWA gaschromatograph (GC) Injection time of calibration gas [rain] systemequipped with a reducinggas analyzer. Therefore the re- sultscould not be as preciseas they wouldhave been with direct Figure 3. Linearitytest 2 showingthe pressurerecorded in the analysisby extraction. manometeragainst the duration of thecalibration gas injection at a The GC wascalibrated using the extractionsystem and a setof flowrate of 2 cm3 min'l. Thenominal equivalent concentration 10 cylinderswith CO valuesfrom 50 to over200 nmole/mole.The rangeis approximately7 to 260 nmole/mole. responseof the GC systemwith reducinggas analyzerwas not 10,008 BRENN1NKMEIJER ET AL.: ABSOLUTE CO MEASUREMENT

300 - What seems to cast some doubt on the NIWA data is that AL usingthe gasfilter correlationtechnique (GFC) and NationalIn- stituteof Standardsand Technology(NIST) standards,seems to 250 confirm the trend. Indeed, if the AL values are basedon an inde- pendentscale and technique, the trend is confirmedindependently, and the NIWA values must be in error. However, the UMD re- 200 sults,which also used GFC and NIST standardscontradict the existenceof thetrend. Actually, both NASA andUMD do havevery similarresults, despite the useof differenttechniques and different 150 standards.Admittedly their agreementat the 50 nmole/molelevel is not as close as for the other three values, but with the GFC techniquelarger errorsoccur in the lower reaches,and theseare lOO alsogiven. However, because UMD andNASA use independent scalesand independenttechniques, their resultsgive supportfor theNIWA datashowing an opposingtrend. 50 Fromthe availableinformation it is not possibleto retrospec- tively assesswhich valuesare correct.The resultsof the absolute methoddo not necessarilyappear more correct.For betterunder- o 0 50 100 150 200 250 standingthe causesfor the observeddeviations a model basedon

absolute (nmolelmole) the appliedmetrology is used.The NIWA data are usedas reference;this is a working hypothesis.Consider the GC and GFC Figure 4. The calibrationwith quadraticleast squaresfit of the techniques.The GC techniquesuffers from two deficiencies.One NIWA CO gaschromatograph in 1994 using 10 cylindersassayed is that the detectorresponse is not alwayslinear. This effect will by the absolutemethod. not be consideredhere. The secondproblem is that the detection limit is not zero. When air containingonly a few nmole/moleis linear.Furthermore, an air samplewith a CO mixing ratio of 3.7 injected,there may be no chromatographypeak that can be inte- nmole/moledid not give a chromatographypeak that could be in- gratedfor determininga signal.The gas filter correlation(GFC) tegrated.The responseof the GC systemswith reducinggas ana- techniqueis intrinsicallylinear, but suffersfrom a worsesignal- lyzersis not alwayslinear and may vary dependingon the operat- to- noiseratio. Severallaboratories have improvedthese instru- ing conditions.The resultsare shownin Figure4. The scatterof ments,yet eventhen, air containinga few nmole/moleCO cannot the individualdata points relative to the quadraticleast squares fit be assayedaccurately. Thus both techniquescannot directly de- is less than 1%. terminewhether an air sampleis free from CO at the nmole/mole Table 3 showsthe ring testresults [Novelli et al., 1998a].The level. The otherproblem is the familiar one that both techniques mixing ratiosrelative to the NOAA-CMDL scalediffer by 3% at areindirect, and depend on the comparisonof the systemresponse the high endto 8% at the 50 nmole/molelevel. In Figure5 all re- with that for standards. sultsexcept those from INPE andHKP whichshowed rather large The errorsintroduced by offset and calibrationerrors in sys- deviations(for acronyms,see Novelli et al. [1998a]) are plotted tems based on standards can be calculated. It is assumed for the againstthe NIWA dataas percentagedeviation. presentdiscussion that the assumedor measuredmixing ratio "a" Severalinferences that may improvethe understandingof the is related to true or real value "r" and the offset of the detector "z" differenceswill be made. CMDL, the Common Wealth Scientific via a = cr + z. The relativeerror is A = a/r-1 = (c - 1) + z/r. In and IndustrialResearch Organization (CSIRO), and NASA used the next case the effect of errors in standards is addressed. Stan- the GC techniqueand the CMDL scale.NASA agreesfairly well dardscan be producedwith greataccuracy, based on gravimetric with the NIWA values, but CMDL and CSIRO do not. The latter procedures.In these,uncertainties in the degreeof dilutioncan be two are mutually consistent,and their resultssuggest that the minimal.Another option is dynamicdilution, where the dilution NIWA valueshave an increasinglypositive bias with decreasing errorsmay be larger,but often still acceptable.For reachingthe concentration,for brevity referredto here as "the trend".IFU and low mixing ratiosdilution with zero air or gas is applied.The CSIR, both usinga mutual scale,and the GC technique,closely pointis thata largeamount of air is addedto lowerthe mixingra- confirm the trend. Therefore one almost cannot escapebut one tio. Thereforethe CO contentof the dilutiongas must be in the conclusion.The NIWA valuesmust be wrong,and this explains sub-nmole/molerange. If a highdegree of dilutionis needed,a re- the trend. But is this correct?Not necessarily,because this also sidualCO level of 0.5 nmole/molewill leadto approximatelythe dependson whetherthe IFU scaleused was independentof the same bias of the diluted concentrationproduced. At the 50 CMDL scale. Both scales have a history of intercomparisons. nmole/mole level this would induce an error of 1%. It is assumed Consequently,there may be no independentproof that the NIWA for the presentdiscussion that a standardis availablewith an ac- valuesare wrong. tual mixing ratio "s". The assumedmixing ratio is, however,cs

Table3. Resultsfor Two Laboratop, Intercomparisons a 1993-1995Rin• Test 1998-1999 RinlgTest NIWA gas½hromato- NOAA-CMDL MPI-C absolute MPI-C gas NOAA-CMDL graphy/absolute chromatolgrapy 1999b 56.1 51.5 51.8 52.1 (0.5)c 45.0 (0.9)c 102.7 97.6 98.8 98.9 (1) 91.6 (0.9) 154.1 147.7 151.4 149.5 (1.5) 147.1 (1.4) 204.5 198.0 183.4 182.3 (2) 183.6 (1.8) 350.7 351.4 (7) 355.7 (6) Units are nmole/mole. bp. •Novelli (personal communication, 1999). Uncertainty. BRENNINKMEIJER ET AL.' ABSOLUTE CO MEASUREMENT 10,009

UMD modelJ

2 NASA

NIWA

CSIR

IFU CMDL

CSIRO -10

JMA

-12 i ! ! i 0 50 100 150 200 250

nmole/mole

Figure 5. The resultsof the 1994-1995ring test, plotted as the percentdeviation relative to the NIWA data.The curveshown is basedon c = 0.989 andz = - 2.2 nmole/mole(see text). For the laboratoryidentifiers, see Novelli et al., 1998a.

culatingthe averagevalue, the secondvalue submittedby CSIRO nmole/mole.Further, it is assumedthat the air for dilutionis as- hasbeen used, otherwise the samecriteria apply as before.Using sumedto have 0 nmole/molealthough it actuallycontains "z" the equationsderived above givesz = - 2.2 nmole/moleand c = nmole/mole.This leads,for the standardmade by dilution of a 0.989 as offsetand calibrationerror, respectively, for the average factor'f ", to the assumptiona = fcs. The real mixingratio is, comparedto NIWA. however,r =j• + (1-J)z.The resulting error is A = a/r-1 =(c - 1) Without further information no further conclusions can be + z/r. This error in the standardscauses the sameerror for sample drawn. On the basis of the previoussections on the absolute values.A thirdpossibility for introducingsystematic errors is that method,it would seemthat a plausiblescenario is that NIWA had CO is graduallyproduced in a setof standardsstored in cylinders. an error in the 200 nmole/molerange of about2%. This for in- This is a commonproblem, and in the simplestcase CO would stancecould be the resultof a wrongcalibration of the gasmeter. growat constantand identical rates in all cylindersirrespective of If this applies,it still doesnot explainthe highervalues in the 50 the CO concentration.This leadsto a similarexpression as for the nmole/molerange. In view of the low detectionlimit of the abso- two cases considered. lute method,and the high linearity,a conclusionwould be thatthe Whatbecomes apparent from inspectionof Figure5 is that average of the other laboratoriesis too low by a several nearlyall curvesexhibit the hyperboliccharacteristic of the ex- nmole/mole.For a direct comparisonbetween NIWA and the pressionderived above. The differences between the laboratories NOAA scale,this would mean that in the 50 nmole/molerange, canbe expressed in terms of offsetsand calibration differences. At the NOAA value was 3.5 nmole/mole too low in 1995. low values the offset becomesa dominant source of error. These deviationsbetween the laboratoriescan alsobe gleanedfrom the 3.2. The 1998-1999 Laboratory Intercomparison tabulationby Novelliet al., [1998a]based on quadraticequations comparingall institutesto theNOAA scale.In Figure5, NASA The secondring test involvedtwo circuitsof laboratoriesand andUMD havecurves that showan increasingpositive deviation. two sets of cylinders.One set of cylinderswith nominal values CMDL, CSIRO, AL, JMA, andIFU togetherwith CSIR showan from 50 to 350 nmole/mole was made available to the Max Planck increasingnegative relative deviation towards the low concentra- Institutefor Chemistry.This time, in contrastto the first compari- tions.The curvesmost probably represent either the effect of bias son in New Zealand, the mixing ratio for each cylinderwas di- in calibration and the existence of an offset, or the use of a rectly determinedby processing400 L of air from each cylinder nonzerodilution gas and a deviatingvalue for thegas that was di- with the systemdescribed. This volume of air was lessthan 10% luted,or thegrowth of CO in cylinderswith standard mixtures. of the cylinder's content. The resultsfrom NIWA will now be comparedwith the aver- It was also decided to additionally perform gas chroma- agevalues of theother laboratories. At thenominal value of 200 tographicdeterminations with a reducinggas analyzer (type RGA nmole/molethe deviationbetween NIWA and the averageis 4.4 3). The gaschromatograph was calibratedindependently from the nmole/mole,or 2.2%. For calculatingthe averages,the valuesin extractionsystem results by usinga singlecylinder that was cali- Table3 [Novelliet al., 1998a]are used without weighting the in- bratedby NOAA-CMDL in 1996 to have 172 + 3 nmole/mole, dividualerrors. Only the last CMDL valueand one of the values and applyingdynamic dilution downward using zero air. During submittedby CSIRO were used.HKP and INPE resultswere this calibrationby dynamicdilution, two furthercylinders with air omitted.Because some scalesare linked, this averageis biased. were fixed in their mixing ratio relative to the dilution line. The Despitethis, the most probableconclusion would be that the threecylinders were then usedto assaythe five unknownNOAA- NIWA dataat nominally200 nmole/molewere 2.2% too high.At CMDL test standards.Extrapolation was usedfor the highercon- the lowestmixing ratio near 50 nmole/mole,the averagevalue is centrations.The resultslisted in Table 3 show good agreement. 53.3 nmole/mole,which is 5.0% below the NIWA value. For cal- There is a discrepancywith the valuesof NOAA/CMDL obtained 10,010 BRENNINKMEIJER ET AL.: ABSOLUTE CO MEASUREMENT in 1999, which indicatesa problem with the standards.Such Brenninkmeijer,C.A.M., and T. R0ckmann,Using isotopeanalysis to im- problemswere acknowledged(P. Novelli personalcommunica- proveatmospheric CO budgetcalculations, inlnternational Symposium tion, 1999) and are treatedin a paperin press(K.A. Masarieet al., on IsotopeTechniques in the Studyof Past and CurrentEnvironmental Changesin the Hydrosphereand the Atmosphere,edited by P. Murphy, The NOAA/CSIRO Flask Air IntercomparisonExperiment: A pp. 69-77, Int. AtomicEnergy Agency, Vienna, Austria, 1998. strategyfor directly assessingconsistency among atmospheric Brenninkmeijer,C.A.M., M.R. Manning, D.C. Lowe, R.J. Sparks, G. measurementsderived from independentlaboratories, submitted to Wallace, and A. Volz-Thomas,Interhemispheric asymmetry in OH Journal of GeophysicalResearch, 2000). abundanceinferred from measurements of atmospheric 14CO, Nature, 356, 50-54, 1992. 4. Conclusions Connors,V.S., B.B. Gormsen,S. Nolf, and H.G. Reichle,Spaceborne ob- 1. The extractionsystem has well definedcharacteristics. The servationsof the global distributionof carbonmonoxide in the middle linearity (better than 1%), the low detectionlimit (< 0.2 troposphereduring April and October 1994,at. Geophys.Res., 104, 21,455-21,470, 1999. nmole/mole),and the quantitativeconversion and trapping of the Crutzen,P.J., and P.H. Zimmermann,The changingphotochemistry in the CO2 formed, even in the smallestquantities equivalent to 7 troposphere,Tellus, Ser. AB, 43,136-151,1991. nmole/mole,support the contention thatCO mixing ratios can be J0ckel,P.,C.A.M. Brenninkmeijer, andM.G. Lawrence, Atmospheric re- determinedwith an error of lessthan 1%. sponsetime of cosmogeniclsCO to changesin solaractivity, d. Geo- 2. The gaschromatography results provide independent evi- phys.Res., 105, 6737-6744,2000. dencefor conclusion1, assumingthat the standardonce assayed Krol,M., P.J.van Leeuwen, and J. Lelieveld,Global OH trendsinferred byNOAA as having 172 nmole/mole indeed had this value at the frommethylchloroform measurements, d.Geophys. Res.,103, 10,697- momentofthe test. This assumption isnot unreasonable, butthe Mak,10,711, J.E.,1998. andJ.R. Southon, Assessment of tropicalOH seasonalityusing agreementispossibly coincidental because the standard isa few atmospheric14COmeasurements fromBarbados, Geophys. Res.Lett., yearsold and may have drifted. 25,2801-2814, 1998. 3. The intercomparisonresults for the 1993-1995ring test Moxley,J.M., and K.A. Smit, Factors affecting utilisation of atmospheric systematicallyreflect the effect of offsetsat low mixing ratios. COby soils, Soil Biol. Blochem., 30, 65-79,1998. The resultsfrom the different laboratoriesdo not allow a conclu- Novel!i,P.C., CO in the atmosphere:measurements techniques and related sionto be madeas to whichvalues were correct. There is no inde- issues,Chemosphere Global Change Sc., 1 (1-3),115-126, 1999.

pendentevidence that the absolute method asused in 1994pro- Novelli, tion ofP.C., a gravimetric J.W. Elkins, reference and L.P. scale Steele, for measurementsThe development of atmosphericandevalua- ducedbetter results. However, theresults ofthe 1998-1999 ring carbonmonoxide, d.Geophys. Res,96, 13,109-13,121, 1991. testshow an even stronger trend effect. With the refinement inthe Novelli,P.C., J.E. Collins Jr., R.C. Myers, G.W. Sachse, andH.E. Scheel, absolutemethod, the test for linearity, and the confirmationby Reevaluationof the NOAA/CMDL carbonmonoxide reference scala means of gas chromatography,this strongly suggeststhat the andcomparisons with CO referencegases at NASALangley and the lowerend of theNOAA scalehas dropped. FraunhoferInstitut, d. Geophys.Res, 99, 12,833-12,839,1994. 4. The resultsfrom the absolutemethod in 1994 and 1999 to- Novelli,P.C. et al., An internallyconsistent set of globallydistributed at- gether suggestthat the lowest range NOAA standardsactually mosphericcarbon monoxide mixing ratiosdeveloped using results from an intercomparisonof measurements,d. Geophys.Res., 103, 19,258- were alreadylow in 1994 by about5%. 19,293, 1998a. 5. It is recommendedthat in publicationsof laboratoryinter- Novelli, P.C., K.A. Masarie,and P.M. Lang, Distributionand recenttrends comparisonsof CO standardizationthe laboratoriesinvolved sup- in carbonmonoxide in the lower troposphere,J Geophys.Res., 103, ply informationabout zero gas, offsets,standards, and linearity. 19,015-19,033, 1998b. Prinn,R.G., R.F. Weiss,B.R. Miller, J. Huang,F.N. Alyea,D.M. Cunnold, Acknowledgments.The authorsthank WolfgangHanewacker for P.J.Fraser, D.E. Hartley,and P.G. Simmonds,Atmospheric trends and carefulprocessing and PeterPohl and Inn Hemmingsenfor IGNS, New Zealand for developingthe pressuregauges. The GC developmentat lifetimeof CH3CCI3 andglobal OH concentrations,Science, 269, 187- NIWA (New Zealand)were carriedout by D.C. Lowe and G. Brailsford. 192, 1995. We thankEckehart Scheel for discussions.We particularlythank Paul Quay,P., S. King,D. White,M. Brockington,B. Plotkin,R. Gammon,S. Novellifor organizingthe extensivering tests, and for criticalremarks and Gerst,and J. Stutsman,Atmospheric (CO)-C-14: A tracerof OH con- discussions.Mark Perrichecked the grammar. centrationand mixing rates, at. Geophys. Res. 105, 15,147-15,166,2000. Reichle,H.G. et al., Spaceshuttle based global CO measurementsduring References April andOctober 1994, MAPS instrument,data reduction, and data validation,d. Geophys.Res., 104, 21,443-21,454, 1999. Bakwin,P.S., P.P. Tans, and P.C. Novelli, Carbon monoxide budget in the R0ckmann,T., andC.A.M. Brenninkmeijer, COand CO2 isotopic compo- NorthernHemisphere, Geophys. Res. Lett., 21, 433-436, 1994. sitionin Spitsbergenduring the 1995ARCTOC campaign,Tellus Ser. B, Brenninkmeijer,C.A.M., A pneumaticallyoperated high vacuumglass tap, 49, 455-465, 1997. Int. d. Appl.Radiat. Isotop.,32, 679-681, 1981. Sanhueza,E., Y. Dong,D. Scharffe,J.M. Lobert,and P.J. Crutzen, Carbon Brenninkmeijer,C.A.M., Deuterium,carbon-13 and oxygen-18 in tree and monoxideuptake by temperateforest soils: The effectsof leavesand peat depositsin relation to climate, Ph.D. thesis,Univ.of Groningen, humuslayers, Tellus, Ser. B, 50, 51-58, !997. Groningen,the Netherlands,146 pp., 1983. Smiley,W.G., Note on a reagentfor oxidationof carbonmonoxide,Nucl. Brenninkmeijer,C.A.M., Robust,high efficiency,high-capacity cryogenic Sci. A bstr., 3, 391, 1965. trap,Anal. Chem.,63, 1182-1184, 1991. Stevens,C.M., andL. Krout,Method for the determinationof the concen- Brenninkmeijer,C.A.M., Measurementof the abundanceof l4co in the trationand of the carbonand oxygenisotopic composition of atmos- atmosphereand the 13C/12C and 180/160 ratio of atmosphericCO, with phericcarbon monoxide, lnt. d. MassSpectrom. Ion Phys.,8, 265-275, application in New-Zealand and Antarctica,d. Geophys. Res., 98, 1972. 10,595-10,614, 1993. Volz, A., D.H. Ehhalt,and R.G. Derwent,Seasonal and latitudinalvaria- Brenninkmeijer,C.A.M., and I. Hemmingsen,Sheathed thermocouples tionof 14CO,and the tropospheric concentration of OH radicals,JGeo- usedas heaterelements,,/. Phys. E, Sci. Instrum.,21,502-503, 1988. phys.Res., 86, 5163-5171, 1981. Brenninkmeijer,C.A.M., and M.C. Louwers, Vacuum actuated high- vacuumglass valve, Anal. Chem.,57, 960-962, 1985. M. Brltunlich,C.A.M. Brenninkmeijer,V. Gros, C. Koeppel, T. Brenninkmeijer,C.A.M., and T. R0ckmann,Russian doll type cryogenic R0ckmann,and D.S. Scharffe,Atmospheric Chemistry Division, Max Planck traps:Improved design and isotopeseparation effects,Anal. Chem., 68, 3050-3053, 1996. Institute for Chemistry, D-55020 Mainz, Germany. (carlb•mpch- Brenninkmeijer,C.A.M.,and T. R0ckmann, Principal factors determining mainz.mpg.de) thelgo/160 ratio of atmosphericCO asderived from observations in the southernhemispheric troposphere and lowermoststratosphere, a[ Geo- (ReceivedFebruary 4, 2000; revisedMay 9, 2000; phys.Res., 102, 25,477-25,485, 1997. acceptedMay 30, 2000.)