GLOBAL BIOGEOCHE•C• CYCLES, VOL. 14, NO. 1, PAGES 373-387, MARCH 2000 In situ evaluationof air-seagasexchangeparameterizations usingnovel conservativeand volatile tracers PhilipD. Nightingale, • GillMalin, 2CliffS.Law,• Andrew J.Watson, 2PeterS.Liss2 Malcolm I. Liddicoat, l Jacqueline Boutin, 3andRobert C.Upstill-Goddard 4 Abstract. Measurements of air-seagasexchangeratesarereportedfrom two deliberate tracerexperiments in the southernNorth SeaduringFebruary1992 and 1993. A conservative tracer,sporesof the bacteriumBacillusglobigiivar. Niger, wasusedfor the first time in an in situair-seagasexchangeexperiment.This nonvolatiletraceris usedto correctfor dispersive dilutionof the volatiletracersand allowsthreeestimationsof the transfervelocityfor the sametimeperiod. The firstestimationof the powerdependence of gastransferon molecular diffusivityin the marineenvironment is reported.This allowsthe impactof bubbleson estimatesof the transfervelocityderivedfrom changesin the helium/sulphurhexafluorideratio to be assessed.Data from earlierdual tracerexperimentsare reinterpreted,and findingssuggestthatresultsfrom all dualtracerexperiments are mutuallyconsistent.The completedata setis usedto testpublishedparameterizations of gastransferwith wind speed. A gasexchangerelationshipthatshowsa dependence on wind speedintermediate betweenthoseof Lissand Merlivat [1986] and Wanninkhof[1992] is foundto be optimal. The dualtracer dataare shownto be reasonablyconsistentwith globalestimatesof gasexchangebasedon the uptakeof naturalandbomb-derived radiocarbon.The degreeof scatterin the datawhen plottedagainstwind speedsuggests thatparameters not scalingwith wind speedare alsoinfluencinggasexchangerates. levels[Hahnand Crutzen,1982;Solomon et al., 1994;Wofsy 1. Introduction et al., 1975], and climate [Charlsonet al., 1987]. Addition- One of the majoruncertainties in studiesof globalbiogeo- ally,thecalibration of oceanicglobalcirculation modelsusing chemicalcycleslies in estimatingthe fluxesof gasesbetween marine tracers is sensitive to uncertainties in air-sea fluxes the atmosphereand the oceansand vice versa. An accurate [e.g.,Englandet al., 1994]. Currently,no directandaccurate quantification of this processis necessary in orderto under- measurementof the flux of a gas acrossthe air-seainterface stand the role of the oceansin atmosphericchemistryand hasbeenmadeat sea,mostlydueto technological limitations. globalclimate. It is well knownthatthe oceansareresponsi- Fluxesare thereforecommonly derivedfrom the productof ble for theuptakeof a significantproportionof anthropogenic the concentration difference between the surface ocean and carbondioxide (CO2), but the magnitudeof this oceanicsink atmosphere,which drives the flux, and a kinetic factor known is not well established[Houghtonet al., 1995]. Considerably asthe gastransfervelocity(k), [seeLiss, 1983]. less is known about the size of the marine source of other bioGastransferratesarethoughtto be regulatedby turbulence genicgasessuchas dimethylsulphide(DMS), nitrousoxide, at the air-waterinterface[Jahneet al., 1987b]. Laboratory low molecularweighthalocarbons, and hydrocarbons.These experiments andtheorysuggest that ratesare influencedby compounds can influenceatmospheric acidity[Charlsonand friction velocity,wave type, breakingwaves and bubbles, Rodhe, 1982], troposphericand perhapsstratospheric ozone temperatureand/or humidity gradients, and surface films [Broecker et al., 1978; Broecker and Siems, 1984; Memery and Merlivat, 1985; Monahah and Spillane, 1984; Phillips, 1991]. Wind speedis linked to mostof theseparametersand •Plymouth MarineLaboratory, Centre forCoastal andMarine Sciences,Plymouth,United Kingdom. has been shownto have a considerableinfluenceon gas ex2School of Environmental Sciences, University of EastAnglia(UEA), changein laboratory,lake, and estuarineexperiments,[e.g., Norwich,UnitedKingdom. Liss, 1983; Wanninkhofet al., 1985; Clark et al., 1994]. The 3Laboratoire d'Oc•anographie Dynamique etdeClimatologie, UMR apparentease with which wind speedcan be measured,and 7617: CNRS / ORSTOM / Universitfi Pierre et Marie Curie, Paris, the availabilityof globaldata setshasthereforemeantthat k is France. 4Department ofMarine Sciences andCoastal Management, Universityusuallyexpressedas a functionof this parameter[Liss and Merlivat, 1986; Smethie et al., 1985; Tans et al., 1990; Wanof Newcastle uponTyne,Newcastle, UnitedKingdom. ninkhof, 1992], seeFigure 1. The Liss and Merlivat [1986] relationship(hereafterreCopyright2000 by theAmericanGeophysical Union. ferredto asLM86) is basedon dataobtainedfrom a lake study Papernumber1999GB900091. usingthe purposefullyaddedtracersulphurhexafluoride(SFr) 0886-6236/00/1999GB900091512.00 [Wanninkhofet al., 1985] and a laboratoryinvestigationfor 373 374 NIGHTINGALE ET AL.: IN SITU MEASUREMENT OF AIR-SEA GAS EXCHANGE technique has some well-documented shortcomings[Liss, 1983;Roether andKromer, 1984].Although the•4Cmethod 70 , 60 A :/ 5O /' shouldallow a reasonableestimateof the globally averaged CO2 uptakeby the oceans,it yieldslittle informationon how k variesin time and spaceor indeedhow to calculatek for other gases. There is a great need for such informationto enable better estimatesof the air/sea flux of gasesof environmental significancein both regionalbudgetstudiesand globalmodeling exercises. The use of different global wind regimesto calculatethe yearly global flux of CO2 leads to discrepanciesof 40-85% / "' /0ø between LM86 andrelationships basedon •4C[Boutin and Etcheto, 1995; Erickson, 1989], the range dependingon the wind regime selected. Various explanationshave been proposedto reduce,or even resolve, these discrepancies.These includebubbles/breaking waves [WooIf, 1993; 1997], the skin temperature effect [Robertson and Watson, 1992], and changesin the power dependenceof k on the moleculardiffusivity of the gas with increasingwind speed [Wanninkhofet al., 1993]. However, lessprogresshas been made on obtaining direct measurementsof gas exchangein order to test whethertheseor otherparameterizations are indeedapplicable at sea. Exciting new approachesto the indirect determination of gas exchangeratesinclude massbudgetingof atmospheric / ,: o 40 // o: "• 30 10 0 -"'"/•, , , , , , • ' 0 4 8 12 16 DMS [Gabric et al., 1995] and seasonal variations in the at- 20 Windspeed (ms mosphericoxygen/nitrogenratio [Keeling et al., 1998]. The former studysupportedLM86 while the latterfavoredW92. We have previouslydescribedhow a dual tracertechnique utilizing SF6and3-Helium (3He)canbeusedto obtain direct estimatesof k at sea [Watson et al., 1991]. These results from Figure l. Publishedparametedzations of k with wind speed the southernNorth Sea suggestedthat gas exchangerates in andpreviousdual tracerdata. Solid line represents Lixxand the oceanmay indeedbe similarto thosepredictedby LM86. Merlivat [1986], the sho• dashedline representsWa,,i,khof Additional studiesusing the radon deficiency techniquein [1992], and the longerdashedline representsSmethieet al. [1985]. Solid circlesshowthe dual tracerdatafrom the Noah sucha way as to overcomesomeof the previousdrawbacks Sea [Watxo, et al., 1991], solid squaresshowthe dual tracer also tend to supportLM86 [Emersonet al., 1991]. In condatafrom the GeorgesBank (as reworkedby Asherand Wan- trast, resultsfrom later applicationsof the dual tracer tech,i,khof [1998]) and the solid triangle showsthe dual tracer niqueon the GeorgesBank [Wanninkhofet al., 1993] andthe datum from the Florida Shelf [Wanninkhofet al., 1997]. The Florida Shelf [Wanninkhofet al., 1997] showeda strongerde- opensquareis thebomb-derived radiocarbon estimate of Bro- pendence on wind speedandgavebetteragreement with W92. ecker et al. [ 1985]. Unfortunately, there is an inherent drawback to the dual tracer technique as previously employed, in that it actually windspeeds above13.5m s-I [Broecker andSiems,1984]. It purposefullyadded gases. Some simple assumptions,dis- was an extrapolationof the then availabledata, as measurementsof gasexchangein the oceans(obtainedby the Rn deficiencytechnique[Peng et al., 1979]) had failed to show any obviouscorrelationwith wind speed. The relationshipproposedby Wanninkhof[1992] (hereafterreferredto as W92) is simply a quadraticcurvefitted suchthat when averagedover global wind speedsit is in agreementwith the global mean transfervelocitydeterminedfrom the oceanicuptakeof bomb- cussed in more detail later, have to be made in order to esti- measures the difference in the transfer velocities of the two matek foreither3Heor SF6or indeed foranyothergas.Ideally, one of the tracersdeployedshould be nonvolatileto allow dilution and dispersioncorrectionsto be applied to the volatile compoundso that unequivocaldirect estimatesof k can be made. Until now, a suitable marine tracer has not been identified. Here we investigatethe use of microbialtracersin this context. Previouslythey have been used to track sewage derived radiocarbon (•4C)[Broecker et al., 1985]. Similarly, dispersionin coastalwaters[Pike et al., 1969] and to tracethe Tanset al. [1990] adjustedthe radon-derivedparameterization movementof contaminatingmicrobesin groundwatersystems of Smethie et al. [1985] (hereafter referred to as S85) to fit [Keswicket al., 1982]. Bacterial sporeshave been employed through the •4Cvalue. Therearedrawbacks to theseap- in such studiesbecausethey persistin theseenvironmentsin proaches.Oceanicgasexchangeratesmay well be different their metabolically inactive state and have better detection to those observed in tanks and lakes where the wave field is limits than commonlyusedchemicaltracers. They are considrestrictedand bubblesize and spectraare likely to be different ered to be innocuousin use, incapableof growthin the envito those in seawater. The radon data on which S85 is based ronment, and unlikely to interfere with the biota [Gameson, showeda large amountof scatterwith wind speed,and the 1986]. These characteristicssuggestedto us that bacterial NIGHTINGALE ET AL.: IN SITU MEASUREMENT sporescouldbe an ideal conservativetracerfor use in gasexchangestudiesin openseawater. In this paperwe reportnew measurements of gasexchange using the dual tracer techniquein the southernNorth Sea in February1992 (NS92) and February1993 (NS93). A careful reexaminationis made of the original data from our earlier dual tracer experimentsin the North Sea during March 1989 (NS89a) and October 1989 (NS89b) as describedby Watson et al. [1991]. These data are combinedgiving a comprehensive data set (NStotal) spanninga wide rangeof wind speeds that is usedto test publishedparameterizations of k. We also describethe use of a conservativetracer,sporesof the bacterium Bacillusglobigii vat. Niger (BG), in the NS93 studyand present estimatesof gas exchange using three tracer pairs 54øN 375 North Sea 53øN + 52øN Netherlands - Belgium 51øN -111111111111• - _ IIIIIIIIll•ll•illlllllllllll- (3He/SFr, 3He/BG, SF6/BG).Thisallowsthefirstestimation of the power dependenceof gas transferon diffusivity to be reportedin the marine environment. As far as we are aware, theseresultsrepresentthe first deploymentof bacterialspores in open seawaterand the first use of conservativetracersin estimatingin situ the air-seaexchangeof volatile compounds. OF AIR-SEA GAS EXCHANGE OøE 1øE 2øE 3øE 4øE 5øE 6øE Figure2. The southern NorthSeashowingthetracerrelease sitefor all four experiments (filled square)andthe locations of the three Royal NetherlandsMeteorological Institute (KNMI) operated meteorological platforms (plussigns). 2. Methods tial uncertaintyof water depth from the calculationof the transfervelocity. Unfortunately,this area of the North Sea is Therelease of SF6and3Heto seawater is complicated by heavily used and allowance had to be made for possible theirlow solubilities (around3.6 x 10-4and3.5 x 10-4mol movementof the patch toward either shippinglanes or the dm-3,respectively, at thetemperatures andsalinities encoun- many gas productionplatformsin the vicinity. The release teredin this study). Our preferredmethodof deploymentis to sitein NS92 was movedslightlycomparedto the earlierstudy 2.1. Tracer Release dissolve the tracers in a known volume of water in a steel tank of Watson et al. [1991] to - 55 km from the coast, and the (- 1000dm3)andthenpumpthiswaterintotheseato create NS93 the tracer-enrichedpatch. This approachis time-consuming, and the massof tracer deployedis limited by the size of the tank. However,it avoidsthe lossof gasto the atmospherevia direct bubblingof seawaterand ensuresthat the ratio of the tracersis initially constantthroughoutthe patch. In both NS92 and NS93, the tracer-taggedwater was preparedpriorto sailingin orderto preventany inadvertentcontaminationof the shipwhich mighthaveled to difficultiesin deploymentoccurred in shallower waters (20 m comparedto 30 m) at a distanceof- 35 km from the coast. Seawatersampleswere collectedprior to tracerdeployment from various depthsin the water column to determinebackground levels of the tracers,including BG. CTD profiling verified that the water column was well mixed. A concern in NS93 was the possibleeffect of storageunder low-oxygen conditionson BG viability. The declineof dissolvedoxygen levels during the saturationof the tank water with SF6 had obtainingaccuratemeasurements duringthe later phasesof beenclearlyseenby TCD-GC. In orderto verify the BG denthe experiment. In NS93 only, two units of BiotracerM(Inter- sity, a sampleof tracer-enrichedwater was takenfrom the tank national Biochemicals (UK) Ltd., Berkshire, United King- immediately prior to tracer release. Considerablecare was dom),eachcontaining 10TM spores of BGper2 dm3of suspen-taken not to contaminateeither equipmentor personnelwhile sion, were addedto the tank. In both experiments,the tank collecting this sample. The sample was stored in a sterile, wasfilledwith- 1000dm3of freshwater adjusted toa salinity polystyrene,disposablecentrifugetube kept underwaterin a of- 35 by additionof sodiumchloride. The tank was then further sealed vessel and remained in the dark at 5øC until sealedwith respectto the atmosphereand saturatedwith SF6 analysison return to the laboratoryat UEA. Resultsshowed over a periodof- 24 hoursusingthe methoddescribedby thatthe viabilityof BG had not beenaffectedby storageunder Upstill-Goddard etal. [1991].A known volume of 3Hewas low-oxygenconditionsin the sealedtank. The tracer was deployedas describedby Upstill-Goddard subsequently addedto the headspace in the tank, and this was recirculatedthroughthe water phase. Initial concentrations in et al. [1991]. In brief, seawaterwas pumpedinto the tank via the tank, as determinedby gas chromatography with thermal a header,minimizingpossiblelossof tracerto the air and subconductivity detection (TCD-GC), were8.6 x 10-smoldm-3 sequentmodificationof tracerratios,asthe tank contentswere and3.1 x 10-4moldm-3in NS92,and5.0 x 10-5moldm-3and releasedat a depth of- 10 m. The flow rate was increased 2.3x 10-4moldm3inNS93for3HeandSF6respectively. The throughoutthe deploymentto compensatefor the dilution of tank wasthensealedfor up to 11 daysuntil deploymentat the the tracer remainingin the tank. Drogued surfacedrifting tracerreleasesite (Figure2). buoyswere releasedat regularintervalsduring the tracer deThe releasesite was specificallychosenbecausethe water ploymentand subsequently trackedusing the Argos satellite columnwas fully mixed to the seafloor,eliminatingthe need monitoringnetwork and direction-locatingradio transmitters. to budgetfor tracerlossesacrossthe thermocline.Addition- Thesewere initially employedto give guidanceas to the locaally, the seafloorwas relativelyuniform, removingthe poten- tion of the patchand were subsequently usedto correctthe 376 NIGHTINGALEET AL.: IN SITU MEASUREMENTOF AIR-SEAGASEXCHANGE data for the influenceof the strongtidal oscillationsthat are sion was typically 15%. No colonieswith this pigmentation prevalentin this region. On completionof the deployment, werefound in seawatersamplestakenprior to the deployment. the tracer tank was moored on the seafloor some fifteen miles from the releasesitein orderto preventany contamination of the vesselduringthe remainderof the experiment. 2.2. Analytical An automatedon-line analyticalsystemcapableof a SF6 analysisevery3 min was usedto obtainnear real-timerepresentationof the patchwhile the shipwasunderway.The output from this systemwasinterfacedto the ship'selectromagneticnavigational log, andthe resultantinformationwasused as a first approximation to reducethe effectsof the tidal oscillation. These continuouslyupdatedplots were used as a furtheraid in navigating the shipwithinandwithoutthepatch. The tracer-enriched water body was regularlyrelocatedusing thesetechniqueseven thoughthe vesselsteamedup to five hundredmilesfrom the deploymentareain orderto satisfythe requirements of otherscientists duringthe 1992experiment. Discreteseawatersampleswere collectedfrom the center The sporedensityin the tank immediatelyprior to the tracer deploymentwas determinedin the first batchof BG analyses and again after 8 weeks storagewhen the BG analyseshad beencompleted. Bothof theobserved densities (1.6x 10• spores dm-3)werethesameasoriginally determined in the Blotraceunits, after allowing for dilution in the tank. The ratio of BG to SF6 in the tank was the same as measuredat the first CTD station12 hoursafter the release,indicatingthat the deploymentof the BG had been successful. Wind data were collected from a certified anemometer and wind vane mountedon the mast of the Royal ResearchShip Challenger. Wave height and spectraldata were obtained from a 0.7 m moored wave-riderbuoy during NS93. Wind and wave data were also routinelyrecordedon three platforms (MeetpostsNoordwijk, Europlatform, and Goeree) situated close to the releasesite (see Figure 2) and operatedby the Royal Netherlands Meteorological Institute (KNMI). Archived meteorologicaldata from these platformsfor the two earlier dual tracerexperiments(NS89a and NS89b) have also ofthepatch using 10dm-3stainless steelsprung Niskin bottles been fromtypicallythree,but occasionally up to five, depthsin the made available to us. watercolumn.Samples for3Heanalysis weredrawnfirstand sealedin coppertubesfor laterdetermination by isotopicmass spectrometry with an analyticalprecisionof 3% (D. Martell, personalcommunication, 1993). Analysesfor SF6 werecompletedon-boardusinga novelvacuumpurgeandtrap system with an experimentalprecisionof 2 and 1% for NS92 and NS93, respectively [Law et al., 1994]. Eachdetermination of SF6 represents the meanof at least5 replicates,unlessthe standarddeviationof 3 or 4 analyseswas better than 1%, in whichcasefurtherreplicateswerediscarded. Duplicateseawatersampleswere collectedin sterile1.3 dm-3 polystyrene rollerbottles(BibbySterilinLtd., Staffs, 3. Theory The use of the dual tracer techniquein estimatingk has been discussedpreviously[Wanninkhofet al., 1993; Watson et al., 1991], so the following sectionis only a brief review, which focuseson how we have interpretedresultsfrom the method. If 3HeandSF6areusedasthetracer pair,thenit is the difference between the two transfer velocities (i.e. k3He ksF6)that is actually determinedfrom the equationshownbelow: k3He- kSF6----In (r• / r2) h / (t2 - ti), (1) United Kingdom) for analysisof BG concentrations.The r, is3He/SF6 attimeti (aftersubtraction of background bottleswerestoredin a constanttemperature roomin the dark where and h is the meandepthof the watercolumn. and at ambientseawatertemperature(5øC) both at seaand at concentrations) Clearly, it is necessaryto know how ksv6 and k3m are rethe UEA laboratory.Analysesfollowedthe methodoutlined by Pike et al. [1969]. Replicatesamples(between4 and 11 latedin order to obtain an estimateof k for either gas. From replicates of 100- 500cm3depending onthevolume of water micrometeorologicaltheory, transfer velocities are generally relatedto a power law dependence(n) on the Schmidtnumber requiredand/oravailableper analysis)were dispensedinto glassbottles.Thesewerethenheatedin a waterbathat 63øC (Sc) shown below: for 30 min to inducesporegerminationand reducethe numksF6]k3H e= (SCsF 6/ SC3He) n, (2) bersof backgroundbacterialflora. The heatedsamplewas then filtered throughdisposable,sterile,analyticaltest filter whereSc is the kinematic viscosityof seawater! molecular funnels,containing 47 mm diameter,0.45 gm poresize,grid- diffusivity of the gas. ded, cellulose nitrate filters (Nalgenem). The bottles were Modeling work [Ledwell, 1984], laboratorystudies[Jahne rinsedwithautoclaved seawater (- 25 cm-3)to ensure com- et al., 1987a], and lake experiments[Watsonet al., 1991] all pleteremovalof spores, andthisfluid alsopassed throughthe filter. Filters were then incubated on Petri-Pads (Millipore show that the expected value of n is close to -0.5 at wind speeds greater than3.6 m s-l. In calculating ourresults, we UK Ltd., Herts,UnitedKingdom)soakedin recoverymedium. have assumedthat this dependenceis correctfor wind speeds The mediumcomprisedan autoclavedsolutioncontaining2 g above 3.6 m s-• and have correctedall values of k derived tracerpairto a Scof 600 (k600), theScof tryprone and0.5gNaC1 in90cm3distilled water (pH6.8)and fromthe3He/SF6 at 20øC. If bubbles/breaking wavesareim10cm3 of a filter-sterilized solution of glucose andmannitol CO2in freshwater (each 10% weight/volume)addedimmediatelybefore use. portantin gas exchangeat intermediateto high wind speeds Standard aseptic microbiologicaltechniques were used then our interpretationwill be too simple only in situations throughout.All bottles,measuringcylinders,pipettes,and where any enhancementforcesthe value of n to differ from filter funnels were sterile and used only once to preventcross- -0.5. The modelingwork of WooIf[1997] indicatesthat the Sc of 3HeandSF6oughtto remain closeto -0.5in contamination.Followingincubationin the darkat 30 øCfor dependency - 48 hours, coloniesexhibitingorange/brownpigmentation the presenceof bubbles. This is discussedin more detail in characteristicof BG were countedby eye. Analytical preci- section 5.4. NIGHTINGALE ET AL.' IN SITU MEASUREMENT OF AIR-SEA GAS EXCHANGE 1000 2000 I 3000 SFs (fmol dm 'a) I 10 15 ..... I 20 25 , , '.... 30 35 Temp. 377 In NS93 we took a novel approach by employing a nonvolatile tracer, BG, and used it to correct for dilution and dispersionof the patch. The transfervelocity of the volatile , . , 8•!ini• gascan then be calculateddirectlyfrom the changein the ratio of the volatile and nonvolatile tracers. In this situation an ac- curatevalue for n is not requiredand either ksv6or k3.ecan be calculateddirectly;that is, kvo,ati,e tracer = In (r./r2) h / (t2 - t.), 1o (3) where r• is the volatile tracer/BG concentrationsat time ti. The use of BG allows our assumptionthat any bubble con- tribution tothegasexchange of 3HeandSF6doesnotforcea departurefrom n = -0.5 to be tested, as we can comparethe absolute estimates of k3Heor ksv6with those derived from the 3He/SF6 tracerpair. Further,by employing a conservative tracer with two volatile tracers,it shouldbe possibleto calculate n independently. As far as we are aware, no previousestimatesof this parameterhave beenobtainedat sea. 25 4. Results 3o , , 4.1. Mass Budgeting ,, An alternativeway of obtaininga direct estimateof ksF6is A typicaldepthprofile for SF6 obtainedon February10, 1992,9 hoursaftercompletion of thetracerre- to producea massbudgetof the gas in the patch and observe lease. The SF6concentrations are represented by the filled its decline over time as it degassesto the atmosphere. This squares, and the errorbarsindicatevariabilityin replicate approachwas attemptedusing data from the releasein OctoFigure 3. measurements.Where no error bars are apparent,the vari- ber 1989 during which time wind speedswere typically 10 m abilityis smaller thanthesymbol.Alsoshownarethetem- s-•. Owingto strong diurnal tides,thewatermoved a considperature (filleddiamonds) andthesalinity (filledcircles) from erable distanceduring the 24 hoursrequiredfor each survey. the bottle cast. 150.00 52.3," e 120,00 90.00 52.3( 70.00 .... • 50.00 *'•X.'-:'• 30,00 52.2• *• 15.00 i)•6,00 52.2( I..., 3.30 , ,.......... t....i , t ,. , 3.35 , , .... l... I..I 3.40 , , t , I_,•.1 3.45 3.50 , , 1 3.55 Figure 4. Reconstructionof the SF6patchin the October 1989 experiment4 days after the tracer release. The griddedoutputfrom the interpolateddata was usedto obtaina massbudgetfor SF6. A Lagrangiancorrectionwas applied using estimatesof tidal drift derived from the low-profile surface drifting buoys. The filled circlesrepresentthe samplingpoints. 378 52.65 NIGHTINGALE ET AL.: IN SITU MEASUREMENT ......................... OF AIR-SEA GAS EXCHANGE cantly alter the mass budget. We therefore suspectthat the samplecoverageof the patch was insufficientfor this purpose and that a considerablyfaster analysiswould be requiredbefore the massbudgetapproachcould be usedto obtain meaningful estimatesof gasexchange. 4.2. Wind Data Thermal stabilityeffectswere removedfrom the wind data by correctionto an equivalentwind speedat 10 m heightunder neutralair boundaryconditions(Ui0,. This was achieved by estimatingthe bulk stabilityparameterZ/L (whereZ is the measurementheight, and L is the Monin-Obukhov length) from air and seawatertemperaturesas describedby Large and Pond [ 1981, section3c] using the Stantonnumbersfor unstable and stable stratification proposed by Large and Pond [1982]. U•0,was then determinediteratively following the approachof Large and Pond [1981, equation16]. The functional dependenceof the neutraldrag coefficient(Ct•) on U•0n was taken from Oost et al. [1998] who derived the relation$2,25 3.3 3.4 3,6 3.6 3.7 3.8 3.9 Figure 5. A comparisonof the movementof one of the droguedlow-profile Industrial DevelopmentBoard (IDB), surfacedrifting buoysand the SF6 tracerpatch during the March 1989tracerelease.The drift trackof thebuoyis representedby filled circles, and its positionat 0000 UT on consecutivedays in March is given by the date next to the track. Shown below the buoy track are wind vectors (1 cm is equivalent to 5 m s-•). The maximum extentof the tracer patchwithoutanyLagrangiancorrection duringthreetime pehods of- 24 hoursis indicatedby the threeboxes(March 20, (box A), March 22 (box B), March 24 (box C)). ship from data setscollectedon one of the KNMI platforms (MeetpostNoordwijk) usedin this presentstudy. Wind speed data from all four tracer experimentsare shownin Figure 6. In general,there is excellentinternalconsistencybetweenthe data from the three platforms. Average wind speedsfor MeetpostNoordwijk and Europlatformagree to - 1% for all four experimentalperiods. Data from Meetpost Goereeare in similar agree•nentduring NS92 and NS93 releasesbut are significantly lower (12 and 17%) for the cruisesin NS89a and NS89b, respectively. A comparisonof ship and platform-derived wind speeds is also shown for NS89a and NS89b (Figures6a and 6b). The ship-borneanemometer overestimatedthe wind speed by up to 20% during strongwinds. The shipboardwind instrumentusedin NS89a and NS89b showedno systematicoffset when checkedagainst a calibrated anemometer and was later lost at sea. However, In orderto estimatethe massof SF6in the patch,it was necessary to reconstructthe patch relative to a fixed point in the water and integrate acrossit, assumingthe tracer was well mixed throughoutthe water column. Evidence from depth profiles suggeststhat this was generally true (Figure 3). A typical reconstructionof the patch is shownin Figure 4. The total massof tracer was determinedby interpolatingthe SF6 data onto a 100 x 100 grid and summingit after subtracting the backgroundSF6concentration.Althoughestimatesof the changein the total massof SF6 over time derived from three different interpolationmethodswere in good agreement,it was not possibleto calculate reliable estimatesof k6(x)using the massbudgetapproach. Estimatesof the massof tracer in the patch were 0.093, 0.109, and 0.097 mol SF6 on October 12, 14, and 17, 1989, respectively.One difficulty is causedby the low-profile drifting buoys used to make the Lagrangian correction. These clearly did not track the water patch accurately (Figure 5), and hence data from them do not allow for there are severalother possiblereasonswhy the platform data are likely to be more accurate. An allowancehasto be madefor the movementof the ship relativeto the wind, leadingto potentialerrorsgiventhe relatively rapid maneuveringof the vesselwhile surveyingthe patch. The anemometermay also have beenpumpedby lateral rolling of the ship, leading to a likely overestimateof wind speedof typically 1-5% [Taylor et al., 1995]. Finally, and perhapsmostimportantly,flow effectscausedby the bulk of a vesselcommonlylead to errorsof at least 10% in mean wind speeds[Taylor et al., 1995]. Compensationfor errors due to air-flow disturbanceis complex althoughthe use of threedimensionalcomputationfluid dynamicsappearsto offer a solution [Yelland et al., 1998]. Previous studiesindicate that anemometers sitedon the main mastsof variousships,as an accurate correction shipsin recentdecadeshas been ascribedto the introduction of shipboardanemometers[Peterson and Hasse, 1987; Ramage, 1987]. We thereforebelieve that the platform data are of consid- of the movement of water while sam- pling. However, attemptsto remove the effects of the tidal oscillationby applying latitudinal and longitudinalcorrections basedon the size of the tidal ellipsecalculatedfrom the buoys when little wind-driven drift was apparentfailed to signifi- in thiswork,arelikely to overestimate truewind speedsby 5 20% [Kahmaand Lepparanta, 1981; Taylor et al., 1995]. Indeedthe apparentincreasein mean wind speedsreportedby erably better quality than can be currentlyobtainedfrom a NIGHTINGALE ET AL.' IN SITU MEASUREMENT OF AIR-SEA GAS EXCHANGE (9--IS/OHt) Ul o o& a• i '1.o I • I I I I I • • I i -- • • i i i i cq cq (s/m) poodspu!A,k • (9-1S/OHO Ul • 379 380 NIGHTINGALE ET AL.' IN SITU MEASUREMENT OF AIR-SEA GAS EXCHANGE 6O is onethatshouldbeurgentlyaddressed. The majorityof field studieswhereparameterizations of k are likely to be usedoften occurin areaswherehigh quality wind data are not available. Even in the presentratherfavorablecase,there could $0 / # 4O well havebeendifferences in thewindoverthepatchandthat measured at the platforms(between10 and50 km awayde- # pendingon the tide and drift of the tracer). We assumethat localshort-term variabilityin windswouldaverageoutdueto o the length of time over which the measurementswere made. Numerical tests show that the calculation of Urn,is not particularlysensitiveto the approximations and constantsusedin the estimationof Z/L, nor the relationshipof Co,with U•0, The main errorwouldappearto be in the precisionof the data, / • 30 which arerecorded bytheKMNIinintervals of0.5ms-•. 2O 4.3. in k Timedependent determinations of the3He/SF6 ratiowere ////..• 10 Errors eachcalculatedfrom the mean of three independentestimates of this ratio measured at the surface, middle and bottom of the water columnfor the releasesin NS92 and NS93 (seeFigure 6). Theaverage standard deviation in theratioof 3He/SF6 in ß• ' ' • 2 •, ,I I 4 I I 6 I, ,, 1, , I 8 I , 10 I I 12 14 Wind speed (ms'•) these samplesover the depth of the water column was 7%, greaterthan that calculatedfrom the individualprecisionof replicates(< 4% for the ratio). This suggestssomevariability in the3He/SF6 ratiothrough thewatercolumn, although we Figure 7. Estimatesof kr{•}derivedfrom the February1992 experiment. The solid line representsLiss and Merlivat [1986], the short dashedline representsWanninkhof[1992], and the longer dashedline representsSmethieet al. [1985]. Solid circles show the new dual tracer data. ship. Wind datafromMeetpostNoordwijkhavebeenusedas theplatformwasusuallytheclosestto thetracerpatch;it has beenusedin severalmicrometeorological studies(e.g.,Humidity Exchangeover the Sea (HEXOS), Air-Sea Gas Exchange Experiment (ASGASEX) and ASGASEX Marine AerosolGasExchange (ASGAMAGE),anddataarealwaysin goodagreementwith at leastone otherplatform. However, believethat the water columnwas usuallyrapidly mixed (see Figure3). Owing to the influenceof strongtidesand winds,it was not easy to remain stationarywith respectto the water while on station,and it is thereforedifficult to separatevertical from horizontalvariability. Significantmovementthrough the patch while samplingwith depth was, however, easy to detectusing the continuous-flowSF6 system. Samplestaken fromtheedgeofthepatch appeared tocontain lower3Heconcentrationscomparedto SF6,presumablybecauseof edge effects, and were not used. Given the higher than expected variability in the3He/SF6 ratiothrough thewatercolumn, it did not seemappropriateto calculateerrorsbasedon analytical precision,and uncertaintiesin k have been estimatedfrom the field data. Typically, errorsof 7% (seeabove)in the ratios theproblemof obtaining accurate windspeeddatafromships of two timedependent determinations of the 3He/SF6 ratio Table 1. Estimates of k6(x)Derived FromtheFourTracerExperiments in the SouthernNorthSea Day/Month/Year Depth Water Air U•{),, sd Temp" Temp" 20.92 22.4613.09 14.14- 22.46 24.89 14.14 15.35 March 1989 March 1989 October 1989 October 1989 10.88 14.3315.9417.34 18.61 - 14.67 February1992 17.90 February1992 17.34 February1993 18.77 February1993 20.68 February1993 Sc (3He b) (SF6 •) 30 30 30 30 7.5 7.5 15.5 15.5 5.9 6.8 13.8 12.2 10.0 14.7 10.1 10.6 3.1 2.3 2.1 2.7 246 246 171 171 30.7 30.7 20.0 20.0 20.0 6.3 6.3 6.0 6.0 6.0 7.5 3.4 6.0 6.8 6.3 8.2 11.4 5.9 7.6 12.5 2.8 3.6 1.8 2.6 2.7 260 260 264 264 264 Here sdis standarddeviationandtempis temperature. "Meantemperatures overtheperiodof measurement busing relationship of Wanninkhof etal. [1993] •Usingrelationship of KingandSaltzman[1995] dDerived assuming thatn= -0.5. Sc 1959 1959 1253 1253 2099 2099 2136 2136 2136 k3H½ kr(x) d error 37.7 94.0 28.7 30.1 21.9 62.6 10.8 32.5 24.8 61.9 18.9 19.8 14.4 41.7 7.17 21.6 26.5 3.4 10.3 6.7 4.5 0.9 0.65 0.17 3.4 1.4 40.0 381 NIGHTINGALE ETAL.:IN SITUMEASUREMENT OFAIR-SEAGASEXCHANGE speed value lends support toobservations ofenhanced gasexchange under stormy conditions [e.g.,Watson etal.,1991].It wasnotpossible to obtainan accurate measurement of k6{•} 6O fromthelow windspeedperiodat theendof theexperiment 5O as the changein the ratio is low comparedto the variabilityin 3He/SF6. 4.5. February 1993 Experiment (NS93) 4O Two periodsof very heavyweatherwere encounteredduring this cruise,and when the ship was hove to, it was impossible to obtain discretewater samplesfrom bottle casts. Although replicate sampleswere routinely collectedfrom the over-sidepump at these times, most of these sampleswere o 30 foundto contain excess 4Hesuggesting aircontamination of the samples.This was presumablydue to the presenceof very 2O fine bubbles that were observed in the surface water. Conse- quently, these data were not subsequentlyused. Measure- mentsof excess 3He/SF6 derivedfromverticalprofiling are plottedwith wind speedin Figure 6d. Three estimatesof 10 werederived fromlinearregressions through In 3He/SF6 and 0 2 4 6 8 10 12 8O 14 Windspeed(ms4) 7O Figure 8. Estimates ofk6(•)derived from theFebruary 1993 experiment. Solid linerepresents LissandMerlivat [1986], 6O theshort dashed linerepresents Wanninkhof [1992],andthe longer dashed linerepresents Smethie etal.[1985].Solid cir- clesshowthe new dual tracerdata. '-' ,." 5O // t' o 40 / leadto a totaluncertainty of- 30% in k, depending strongly onthemagnitude of thechange in theratio,i.e.,onthelength ß.• 3O of time over which the measurementswere made. 4.4. February 1992 Experiment(NS92) // // ." T 2O A wide rangeof wind speedswere encountered during 10 NS92 (Figure6c). Our sampling strategy duringthiscruise wasin partdictated by theneedsof othergroups ontheship. Whilesurveying the patchwe endeavored to obtainvertical profiles of thetracerat leasttwiceoverperiods of 12hours or 2 4 6 8 10 12 14 16 18 lessin orderto try to obtainestimates of gasexchangethat Windspeed(ms'1} weretightlyconstrained with respect to windspeed.Occasionally, adjacent areasof tracer-tagged waterweresampled Figure9. Thereanalysis of the Watson et al. [1991]data. withinasshorta timeaspossible in a furtherattemptto esti- Solidsquares showtheoriginaldataof Watson et al. [1991], matetheprecision of thetechnique. Thechange in theratioof andtheopensquares showtheoriginalestimates of k6(•} with the two tracerswith time is shown in Figure 6c. Unfortu- thewindspeed datafromtheKNMI platforms andcorrected of nately,it wasnotpossible to obtainaccurate estimates of gas to neutralstability(U•0n). The opencirclesare estimates k6{•) derivedusingthesametemperature dependence for Scof transferover timescalesof < 14 hoursas the changein the raet al. [1993]in their tio of excess 3He/SF6 waslow compared to theprecision of SF6 and3Heas usedby Wanninkhof of dualtracerdata. The filled circlesrepresent our theanalyses. However, thisintensive sampling of thepatch analysis revisedestimates of theoriginaldataof Watsonet al. [1991] allowedextremelyaccurate estimates of k6•)to be obtained derivedusingU•on,the measurements of Kingand Saltzman overlongerperiods whiletheshipwasawayfromthepatch [1995] to calculateSc of SF6,andthe relationship usedby (Figure7 andTable1). Thesewerederivedfromlinearre- Wanninkhof etal. [1993]forScof3He.Thesolidlinerepre- gressions through In3He/SF6 versus timeasshown inFigure sentsLiss and Merlivat [1986], the short dashedline repre- 6c. The lowerwind speedestimateof k6(x} is in goodagree- sentsWanninkhof [1992], and the longerdashedline reprementwith the NS89a and NS89b resultswhile the higherwind sents$methie et al. [1985]. 382 NIGHTINGALEET AL.' IN SITUMEASUREMENT OFAIR-SEAGASEXCHANGE 4.7. Bacillus Globigii vat. Niger At the time of the NS93 deployment,we were unable to carryout BG analysesonboardthe ship. Sampleswere stored S0 in thedarkat 5øCfor upto 7 weekspriorto analysis, although //0,,ø 40 the majorityof sampleswere analyzedwithin 4 weeks. We were concernedthat samplestoragemight have affectedBG sporeviability either directlyor due to a declinein culturing efficiencydue to outgrowthby otherthermoduricbacteriaon the filters. These possibleconcernspromptedus to reanalyse somesamplesseveraltimesduringthe analysisperiod. Eighteen repeatanalyseswere made,of which 13 showedevidence of a decline in recovery. However, none were statistically significantat the 5% level (two-tailedt test). As a precaution, only resultsfrom samplesthat had been storedfor similar pehods prior to analysishave been used. Long-termstorageof the sampletakenfrom the tank, which containedtap water and sodium chloride, immediately before tracer deployment showedno declinein BG viability even after 2 years. Althoughit wasnot feasibleto samplefor BG asfrequently /" ,, ..... /,.' •.Eo30 ; ,' •0 0 .....-'*** '*'; 0 2 4 asfor thevolatiletracers,thisdid notaffectthenumberof gas 6 8 •0 •2 •4 exchangevalues we have been able to calculate. The relatively low precisionof the BG analysisforcedthe calculation •6 Windspeed(ms'•) of k overlongertimeperiods thanfor thetraditional 3He/SF6 Figure ]0. F.stimat½s of • derivedfromtwo differenttracer pairsfor two differentwind events.F.stimat½s calculatedfrom the3He/SF6 and3He/BG data arerepresented byfilled circles andfilledtriangles, respectively. Thesolidlinerepresents LissandMerlivat[1986],theshortdashed linerepresents Wanninkhof [1992],andthe longerdashed line represents 45 40 / / Smethieet al. [ 1985]. / 35 -' ,* / ß ß / ! / 00ø / time(Figure 6d)andagainthesespana rangeof windspeeds 30 / (Figure8, Table1). Oneestimate is at a relativelylow aver- • agewindspeedof 5.9 m s-• whereit is difficultto obtaina a= 25 E precisevaluefor k6(•}as the changein the ratio is so small compared to likelyanalytical errors.Interestingly, the other • 20 1I ' two estimatesof k6(x} have been determinedfor rather different windevents butaresimilarin magnitude suggesting thatvariablesnotintimately linkedto windspeed arealsoinfluencing 15 II - / 10 - 4.6. Reevaluation of the Watsonet al. [1991] Data wind speedsusedfor the earlier estimatesof k6(•}in the experimentsin 1989 [Watsonet al., 1991] were undoubtedly overestimates, particularlyat higherwind speeds.Addition- ally,theassumption of a neutrallystableatmosphere hadbeen madein orderto correctthedatato equivalent windspeedsat 10 m height. The revisedwind speeddataareshownin Table 1. TheScvalues forSF6and3Heused byWatson etal. [1991] to derivek6(•)from k3He -- kSF6 havebeenrecalculated using more recent estimatesof diffusivities, and a small error in the calculation (the difference between seawater and fresh water diffusivities)has been corrected(see Table 1). The effect of / .' // air-seagasexchange rates. As discussed in moredetailin section4.2, the ship-borne .: / //,-* 5 ./." 0 2 4 6 8 10 12 14 Windspeed(ms4) Figure11. Estimates of k6(•)dehved fromallthreetracerpairs for the same wind event. Estimates calculated from the 3He/SF6, 3He/BG, and SF•BG data arerepresented byanopen circle,filledtriangle, andopensquare, respectively. Thee•or bar represents the standarde•or for k6(•)dehvedfrom the F•BGtracer pair,thee•orsassociated withthe•He/SF6 and He/BGtracerpairsbeingsmaller thanthesymbols. Thesolid our reinterpretation of the datais to increasesignificantlythe linerepresents LissandMerlivat [ 1986],the sho• dashedline dependence of k6(x} on windspeed,particularly for the highest represents Wanninkhof [1992],andthelongerdashed linerepwind speedpoint (seeFigure9). resentsSmethieet al. [ 1985]. NIGHTINGALE ET AL.: IN SITU MEASUREMENT pair. We haveusedthe datato obtainestimatesof k3H e during two different wind events (see Table 2). These values have been scaledto kr(•),assumingthat n is -0.5, and comparedto krl)o derivedfromthechange in the3He/SF6 ratiooverthe same time interval (Figure 10). It should be noted that the OF AIR-SEA GAS EXCHANGE 383 8ø t , 3He/SF6 derived estimates of kr(•)are notadditional datapoints to those describedin Figure 8. The valuesof kr(•)calculated fromthe3He/BG ratios arenotsignificantly different, evenat the 90% level (two tailed t test), from estimates determined fromthe3He/SF6 tracer pair. In order to obtain an estimateof k6(•)from the SF6/BG tracer pair, it is necessaryto averagedata over an even longer time period as the molecular diffusivity of SF6 is considerably lowerthanfor 3He(i.e.,thechange in theSF6/BGratiois smaller). Only one value for ksv6was determined (Table 2). The value has been reducedto kr(•),again assumingn is -0.5, and plottedagainstk6(•)obtained usingthe 3He/BGand 3He/SF6 tracerpairs(Figure11). Theestimates arein close agreement. Again we shouldcautionthat the last two values are not new estimatesof kr(•)but previouslydiscusseddata averagedover longertime periodsfor comparativepurposes. The direct estimatesof k3He and ksF6 derived from the use of BG as a tracerallow a comparisonwith k3H e- ksF6 measureddirectly by the volatile tracerpair. The differencebetweenk3H e 0 2 4 6 8 10 12 14 16 18 20 Windspeed(ms'•) andksv 6 wasfoundto be 20.3cmhr-1,whichis in excellent agreement withthevalueof 21.3cmhr-1obtained fromthe Figure 12. All of the dual tracerdataplottedagainstwind 3He/SF6 ratios.Thesemeasurements aretherefore mutually speed(U•0n). Solid circlesrepresentdata from all four of the supportive. Using the values for k3H e and ksF6given in Table 2, it is simple to calculatea value for n of-0.51. No previousestimateof n hasever beenobtainedat sea. The uncertaintyassociated with this value is estimated to be + 0.14, -0.19. This value is in agreementwith estimatesof 0.505 and 0.515 determined by Watson et al. [1991] in a lake study, and with lab/modelingstudiesof gas transferacrossunbrokensurfaces [Jahne et al., 1987a; Ledwell, 1984]. tracerexperiments in the North Sea. The solidsquaresarethe GeorgesBank dual tracerdata of Asher and Wanninkhof [1998]. Errors in k were derived in the same manner as for our North Seadata. The solidtriangleis the estimatefrom the Florida Shelf [Wanninkhofet al., 1997]. All data were derived assuming n = -0.5. The solid line represents Liss and Merlivat [1986], the shortdashedline represents Wanninkhof [1992], and the longerdashedline represents $methieet al. [1985]. The dark filled line represents a quadraticdeconvolvedfit to the North Seadata (seetext). 5. Discussion 5.1. Total North Sea Data Set Estimates of kr(•), obtained from all four tracer releasesin the North Sea (NStotal), have been plotted againstthe platform-derivedwind data (Figure 12). There is a good correlation with wind speed,but there is considerablescatterin the data suggestingthat wind speedis not the only variableinfluencinggasexchangerates. The bestfit to the data was found that in order it is necessary to applythe to have correction. someidea Theofmost howcommon k varieswith relationU•o, shipsarebasedon linearor quadraticequations(W92 is of the forma = 2, b=0; whileLM86 canbe closelyapproximated by a = 0.166andb = 0.133wherek60o = a (U•0,)2+ b (Ulon) and a zero interceptis assumed.Modeledfits were madeby applyingthe aboveequationto the Ul(),dataand summingthe resultantkr(•)acrossthe period of time over which each k6(x) to bek60 o: 0.25(W10n) 2 (R2=0.79). Clearly, datascatter be- datapoint wasexperimentallydetermined.The modelfit was tweenboth the LM86 and the W92 relationships. However, as shownin Figure6, the wind speedwas highly variable during the periods over which k6(• was determined. In order to properlycomparek6(• with wind speed(and with parameterizations of k6(•)),it is necessaryto allow for the effect of wind speedvarianceas Figure 12 suggeststhat the relationshipbetweenk and wind speedis nonlinear. As originally discussedby Wanninkhof[1992], but rarely considered by otherworkers,if the relationshipbetweenk6tx)andU10 n has a positivecurvature,thenkr(• derivedover a periodof variable windswill be increasedcomparedto a periodof steadywinds. Both LM86 and W92 were developed for instantaneous winds. The difficulty in correctingthe data for this effect is adjustedby minimizingthe leastsquaresbetweenthe modeled valuesof kr(•)andthe actualdatapoints. The deconvolvedfit wask6o0 = 0.23(U•o,)2+ 0.1(U•0,)(seeFigure 12),andthis equationexplains81% of the total varianceof the k6o0 data. It hasso far not provedpossibleto identifywhichotheren- vironmentalvariablesare responsible for the remainingvariabilityin the k6oo data. Two obviouscandidates are breaking waves/bubbles and surface films. As the wind direction also showedconsiderable variabilityduringtheseexperiments, it occurredto us that this could be a factor. However, there is no evidencein the data of reducedgas exchangewhen the windwasfromcontinental Europeandthereforepossiblylimitedby fetch. A plot of k6oo versuswaveheightdoesnot give 384 NIGHTINGALE ET AL.: 1N SITU MEASUREMENT OF AIR-SEA GAS EXCHANGE Table 2. Estimates of theTransferVelocityDerivedUsingMultipleTracers Dayin Windspeed March 1993 U10n sd 15.94- 18.77 16.65 - 18.77 6.8 7.5 18.61 - 20.68 18.77 - 20.68 16.65 - 20.68 SFdBG ksF6 k6oo* 3He/BG se 3He/SF6 k3He k6oo* se k3He- k6oo* se 2.4 2.2 19.0 20.6 12.6 13.7 3.1 4.0 15.2 18.0 15.5 18.4 1.8 2.6 12.5 12.6 2.7 2.8 54.8 51.1 36.4 33.9 3.6 5.2 26.7 24.8 27.3 25.5 2.1 2.9 9.9 3.6 30.9 20.5 2.3 21.3 21.8 0.7 10.6 20.0 5.5 Here sd is standard deviation and se is standard error *Derived assuming thatn = -0.5,Sc(SF6)- 2136[KingandSaltzman, 1995]andSc(3He)=264[Wanninkhofet al., 1993]. of organicmatterfrom a better correlation thanwithU10 . (R267%,Figure13). There sourcesor perhapsdueto resuspension is also no evidenceof increasedgas exchangeduring periods the sedimentafter periodsof high winds. We note that the during the high wind of rapid changesin wind directionwhen incidentsof breaking relatively low value for k6{x)obtained waves and bubble formation might be thought to be more event (11/2 to 14/2) in NS93 was coincidentwith a high suspendedloadobservedin thewatercolumn. likely. Surface films have been shown to greatly inhibit gas exchangein laboratoryexperimentseven at high air phasestir- 5.2. Comparison with Other Dual Tracer Data ring rates[Frew, 1997]. We haveno measurements of surface The only other studiesusingthe dual tracer methodat sea films from our experiments,but biological activity is low in have been madeon GeorgesBank (GB90) [Wanninkhofet al., this region during the month of February [Howarth et al., 1993] and on west Florida shelf (FS96) [Wanninkhofet al., 1994]. However,NS93 (Figure 8) doesseemto showconsid1997] (see Figure 1). Estimatesof k6(•)from theseexperierable variability in k6(x)forwhich one explanationcould be ments showeda considerablystrongerdependenceon wind the presence of surface films either from anthropogenic speedthan the resultsof Watsonet al. [1991]. Both of these otherstudiesconcludedthat the data supportedW92 and were consistent withtheInCglobal measurements [Broecker etal., 90 1985]. However, in the original interpretationof the GB90 data [Wanninkhofet al., 1993], the laboratorystudyof Asher et al. [1992] was usedto includea bubble/breakingwave enhancement to gastransferby simplyvaryingthe power dependence of ksF 6 andk3.eon the diffusivityof the gas(i.e., varyingn in (2)) from -0.56 to -0.35. This hasthe effect of increasingk6lx) 80 7O calculatedfrom k3He-ksF6 measured on the GeorgesBankby up 60 œ to 26%. The authorscautionedthat kco2could not therefore be estimatedfrom the data set, and hencetheir conclusionthat 5o thedatawereconsistent withthe •4Cglobalmeasurements wasinappropriate.In their recentreanalysisof the GB90 data, Asherand Wanninkhof[1998] arguethat their originalinter- ,o 30 pretationoverpredictedthe effect of bubbleson k6lx).The favoredinterpretationof Asherand Wanninkhof[ 1998] is to use valuesof n that varyfrom 0.50 to 0.44 andhenceincreasek6(• 20 calculatedfromthek3He-ksF 6 measured in GB90 by up to 7%. For comparativepurposes, we ignorethis interpretation of the dual tracer results and derive k6(• for GB90 in the same mannerasN Stotal,i.e., n = -0.5 andthe samerelationships for 10 Scof 3HeandSF6(seeFigure 12). Wefoundnosignificant difference between our estimates of GB90 and those described 0 5 Significant Waveheight (m) by Asher and Wanninkhof[1998]. An analysisof variance between the NStotal data set, the GB90 + FS96 data set and Figure 13. Estimatesof k6(•)versussignificantwave height. Filled circles are estimatesfrom all four experimentsin the North Sea and the filled squaresare estimatesfrom the Geor- the total data set (NStotal + GB90 + FS96) showsthat thereis' no significant difference between the NStotal data and the GB90 + FS96 data set(F2.95% level). There is thereforenow agreementbetweenthe differentdatasets.Incorporation of the gesBank (R. Wanninkhofpersonal communication1992). revised GB90 data and the value for FS96 with the NStotal NIGHTINGALE ET AL.: IN SITU MEASUREMENT OF AIR-SEA GAS EXCHANGE 385 contributionto gas exchange[e.g., Merlivat and Memery, data extendsthe range of wind speedsfor which data are availabledown to 3.5 m s-• A bestfit to the total dual tracer 1983; Woolfand Thorpe, 1991]. Severalmodelingapproaches have beenadoptedin an efdataset (NStotal + GB90 + FS96) givesa relationshipof k6{xl fort to calculatethe size of the likely bubble enhancement. = 0.222U•0,2 + 0.333U•0,(R2= 0.80). The mostcommonapproachis to add a term kb,dependenton both diffusivity and solubility, to the nonenhancedtransfer 5.3. Comparisonof Dual Tracer Data With Other velocityk,,that is assumedto be dependentonly on diffusivity Estimates of Gas Exchange to the power-0.5 [e.g., Keeling, 1993; Memery and Merlivat, In orderto comparethe resultsof dual tracerexperiments withthe•4Cderived globalaverage valueforkco 2andthees- 1985;WooIf,1993]. Unfortunately,thereis no simpleway to calculatekbfor a givenwind speedor for a givengas,andthe timatesof ko2recentlycalculatedby Keelinget al. [1998], it is dependenceof kbon wind speedvariesconsiderablybetween necessary to know how to calculatek for differentgases. If the different studies. Most laboratorystudieshave shownthat we assumethat bubbles/breaking wavesdo not force a departurefromn - -0.5for3He,SF6,CO2,and02, thenit is simple kco2 is enhancedto a considerablylower extent by bubbles/breakingwaves than are k3Heor ksF6be:ause of its relato derivek for 02, CO2 or indeedany other poorly solublegas. tively high solubility. The modelingstudy of WooIf [1997] Wetherefore combine thebestfit (k6•x} = 0.222U•0,2 + 0.333 hasindicatedthat breakingwaves/bubbles do not force an apUi0,) to the total data set (NStotal + GB90 + FS96) with a preciabledeparturefrom n = -0.5 (at leastat the wind speeds . Rayleighdistributionbasedarounda meanglobalwindspeed weencountered) for3HeorSF6. Thisisinagreement withthe of 7.4m s-• [Wanninkhof, 1992]anddetermine a globalmean resultsfrom the triple tracer experimentin this study. Howk6{•of 18cmhr-•. Estimates of k61•} fromnatural andbombderivedradiocarbonare 21.2 +/- 5 cm hr-• and21.9 +/- 3.3 cm hr-I, respectively [Broecker et al., 1986]. Theuncertainty in the bomb-derivedvaluemay be greatergiventhat thereare in- consistencies in theglobal•4Cbudget[Hesshaimer et al., 1994]. ever, WooIf [1997] also proposedthat kco2will be overestimatedby simply scalingthe dual tracerdatafrom Sc6• and assumingn = -0.5. Our relationshipfor k6{x} may thereforelead to overestimationof k½o 2, increasingthe gap betweenthe dual tracer andthe•4Cvaluebyan,asyet,unknown amount. Alternatively,Asher and Wanninkhof[1998] have very recently developeda parameterizationof k that includesan enwascloseto theaverage of 7.10m s-I fortheperiod1992- hancementdue to breakingwavesby incorporatinga dependence on both white cap coverageand solubility. Effectively, 1995. The details of the calculationare describedin detail by this usesa variablen approachto includebreakingwave and Boutinand Etcheto[1995]. The globalmeank60ocalculated bubbleenhancements but n also variesfor each pair of gases from the dual tracerdata but ignoringany Sc temperature debeing compared. The use of the Asher and Wanninkhof pendence is 17cmhr-•. A morethorough comparison is to [1998] equationswith the total dual tracerdata set presented calculatethe meanCO2 exchangecoefficient,K, after allowhere gives valuesfor n that range from -0.43 to-0.50 for the We have also made a comparisonusing 1 year of ERS-1 -1 datafrom 1993 when the mean 10 m wind speedof 7.13 m s ing for the effectof variabilityin seasurfacetemperature on bothCO2solubilityandSchmidtnumber. The averagevalue of K using theERS-1datais5.0x 10-2molm-2yr-• I.latml, 18%lessthantheestimates of 6.1x 10-2molm-2yr-• •atm-! deriveddirectlyfrom the radiocarbon techniques [Broeckeret al., 1986]. This, togetherwith theobviousscatterin the dual tracer data,implies thatthefieldbased and14C techniques are close to agreement. Possiblechemicaland biologicalenhancements to the exchangeof CO2will actto narrowthe difference. There is a largerdifferencebetweenthe total dual tracer 3He/SF6 tracer pairand0.51to0.80forthe3He/CO2 pair.Estimatesof k½o 2 calculatedusing this relationshipare lower than valuesof k6{•lshownin Figure 12 by between1 and 17% for the lowest and highestwind speeds,respectively. However, the exact magnitude of kco2calculated from the dual tracerdata usingthis approachis subjectto someuncertainty given that the relationshipof Asher and Wanninkhof[ 1998] is sensitiveto the whitecap dependenceon wind speed used. Use of the relationshipof Monahan [1993] leadsto valuesof k½o2 that are up to a maximumof 11% lower than a simplen = -0.5 interpretation. This approachis in agreementwith the conclusionsof WooIf [1997] in that kco2ought to be lower datasetand the estimatesof k02recentlyreportedby Keeling et al. [1998]. The O2/N2derivedestimatesare higherthanthose than k6(•}derived from the dual tracer data and n = -0.5 in the calculatedfrom the dual tracer data by about 50%, considerapresenceof breakingwaves. bly greaterthanthe uncertaintyin ko2(25%). 5.4. Effect of Bubbles and Breaking Waves Laboratoryand field studieshave shown that there is an enhancementin gas exchangewith breakingwaves and the subsequent formationof bubbles[e.g., Farmer et al., 1993; Merlivat and Memery, 1983]. One possibleflaw in the calculations above is that we have assumed that an n - -0.5 rela- 6. Conclusions New measurements of the transfervelocity obtainedusing thetraditional 3He/SF6 tracer pairhavebeendetermined overa wide range of wind speecls.A clear relationshipwith wind speedwas observed. The original resultsof Watsonet al. [1991] have been reinterpreted,and good agreementis now tionship isapplicable toSF6,3He,CO2,and02,thatis,thereis found no effect of solubilityon gas exchange. While this is to be expectedfor an unbrokenwavy surface,variousmodelingand laboratorystudieshaveshownthat thereis likely to be a solubility effect if breaking waves/bubblesmake a significant agreementextendsto the dual tracerdataof Wanninkhofet al. [ 1993] and Wanninkhofet al. [ 1997] providedthe data are interpretedin the samemanner. A wind speeddeconvolution hasbeenattemptedwith the North Sea data in order derive a between the total data set from the North Sea. This 386 NIGHTINGALE ET AL.: IN SITU MEASUREMENTOF AIR-SEA GAS EXCHANGE parameterizationof k6(x)with instantaneouswinds. The best References fit wasfoundto bek6(x)-0.23(Umn) 2 + 0.1Umn, explaining Asher, W.E., P.J. Farley, R. Wanninkhof, E.C. Monahan, and T.S. Bates,Laboratoryand field measurements concerningthe correlahavebeencombinedwith the data of Wanninkhofet al. [1997] tion of fractionalareafoam coveragewith air/seagastransport,in and Asher and Wanninkhof[1998] in order to cover a wider PrecipitationScavengingand Atmosphere-Surface Exchange,edited by S.E. Schwartz, and W.G.N. Slinn, pp. 815-828, Hemirange of wind speeds. The best fit to the total data set was sphere,Washington,D.C., 1992. foundtobe0.222(S10n 2q-0.333Urn,explaining 80%of the Asher,W.L., and R. Wanninkhof,The effect of bubble-mediated gas total variancein the data set. The remainingvariabilitysugtransferon purposefuldual-gaseoustracer experiments,J. Geogestseither that parametersnot scalingwith wind speedare phys.Res., 103, 10555-10560, 1998. alsoinfluencinggasexchangeratesor that 10 m wind speeds Bates, N.R., A.H. Knapp and A.F. Michaels, Contributionof hurricanesto local and global estimatesof air-sea exchangeof CO2, are not intimately linked to turbulenceratesat the seasurface. 81% of the total variance in the data set. The North Sea data Nature, 395, 58-61, 1998. It is reassuringthat resultsfrom the dual tracerexperiments Boutin, J., and J. Etcheto,Estimatingthe chemicalenhancementefare now in good agreementfrom three different areasof the fect on the air-sea CO2 exchangeusing the ERS-1 scatterometer marineenvironment. However, there is still a paucityof gas wind speeds,in Air-Water Gas Transfer,editedby B. Jahne,and E.C. Monahan,pp. 827-841, AEON Verlag and Studio, Hanau, exchangedata at very high wind speeds. Thesehave recently 1995. been highlightedas potentiallyprovidinga disproportionately Broecker,H.C., and W. Siems,The role of bubblesfor gas transfer largecontributionto the air-seaexchangeof CO2 [Bateset al., from water to air at higherwind speeds.Experimentsin the wind1998]. We should also caution that no dual tracer data have yet been reportedfor the open ocean where it might be expectedthat air-sea gas exchangerates may well be different. Regionsof enhancedgasexchangemightbe expectedto occur as a responseto productionof breaking waves/bubblesby a more fully developedwave field. Equally, regionsof high marineproductivitymightbe expectedto reducegasexchange ratesvia the presenceof surfactantfilms as hasbeen reported at high stirringratesin laboratoryexperiments[Frew, 1997]. We believe that the resultsfrom NS93 clearly demonstrate the considerablevalue of deployinga nonvolatiletracerin airseagasexchangestudies. Its utility is presentlylimited to pehodsof 48 hoursor greateralthoughimprovements in analytical technique(e.g., on board analysis)might reducethis averaging time. The data obtainedallow direct in situ derivation of ksF6 and k3.e and calculation of the first estimate of the Schmidt number dependence'n' at sea. The value of-0.51 obtained is in excellent agreement with results from lake studies [Watson et al., 1991]. This result leads us to the con- wave facility in Hamburg,in Gas Tran•fer at Water Surfaces,edited by W. Brutsaertand G.H. Jirka, pp. 229-236, D. Reidel, Norwell, Mass., 1984. Broecker, H.C., J. Petermann, and W. Siems, The influence of wind on CO2 exchangein a wind wave tunnel,includingthe effectsof monolayers,J. Mar. Res.,36, 595 - 610, 1978. Broecker,W.S., T.H. Peng,G. Ostlund,andM. Stuiver,The distribution of bomb radiocarbonin the ocean, J. Geophys.Res., 90, 6953-6970, 1985. Broecker, W.S., J.R. Ledwell, T. Takahashi, R. Weiss, L. Merlivat, L. Memery,T.H. Peng,B. Jahne,and K.O. Munnich,Isotopicversus micrometeorologic oceanCO2 fluxes: A seriousconflict, J. Geophys. Res.,91, 10517-10527, 1986. Charlson,R.J., and H. Rodhe,Factorscontrollingthe acidityof natural rainwater, Nature, 295, 683-685, 1982. Charlson, R.J., J.E. Lovelock, M.O. Andreae, and S.G. Warren, Oce- anic phytoplankton,atmosphericsulfur,cloud albedoand climate, Nature, 326,655-661, 1987. Clark, J.F., R. Wanninkhof, P. Schlosser,and H.J. Simpson,Gasexchangeratesin the tidal HudsonRiver usinga dual tracertechnique,Tellus,Ser. B, 46, 274-285, 1994. Emerson,S., P. Quay, C. Stump,D.O. Wilbur, and M. Knox, 02, Ar, 222 N2, and Rn in surface waters of the subarctic ocean: Net bio- logical 02 production, Global Biogeochem.Cycles, 5, 49-70, clusionthat bubbles/breakingwavesdo not force a significant departure froma Schmidt number of -0.5 for the SF6/3He 1991. England,M.H., V. Garcon,and J.F. Minster, Chlorofluorocarbon upcombinationat the wind speedswe encounteredin this study. take in a world ocean model, 1, Sensitivityto the surfacegas If this is also true for CO2, then the dual tracer data are reaforcing,J. GeophysRes., 99, 25215-25233, 1994. sonablyconsistentwith other estimatesof gas exchangede- Erickson,D.J., Variationsin the global air-seatransfervelocityfield of CO2, Global Biogeochem.Cycles,3, 37-42, 1989. rived from radiocarbon.There is a discrepancywith estimates of ko2derived from atmosphericO2/N2 ratios. The roles of bubbles/breaking wavesand surfaceactivematerialin gasexchangeare onesthat clearlyrequireurgentinvestigation. Farmer, D.M., C.L. McNeil, and B.D. Johnson, Evidence for the im- portanceof bubblesin increasingair-sea gas flux, Nature, 361, 620-623, 1993. Frew, N.M., The role of organicfilms in air-seagasexchange,in The Sea Surface and Global Change, edited by P.S. Liss and R.A. Duce,pp. 121- 172, CambridgeUniv. Press,New York, 1997. Acknowledgments.The authorsare gratefulto Wiebe Oostand Gabric, A.J., G.P. Ayers, and G.C. Sander,Independentmarineand the KNMI for allowing us the useof their data and Rik Wanninkhof atmospheric modelestimatesof the sea-airflux of dimethylsulfide forproviding uswithhisoriginal SF6and3Hedata.WiebeOostand in the SouthernOcean,Geophys. Res.Lett., 22, 3521-3524, 1995. MargaretYellandbothhelpedto providea preliminaryeducationin Gameson,A.H.L., Tracersfor the WaterIndustry,Water Res.Cent., meteorologicalmatters. Many other scientistscontributedtoward the Marlow, U.K., 1986. fieldworkphaseof this work includingPhil Goy, Alison George, Hahn,J., andP.J. Crutzen,The role of fixed nitrogenin atmospheric SarahJones,SimonDavies,andAndyThompson.We wouldlike to photochemistry, Philos. Trans.R. Soc. London,Ser. B, 296, 521- thankthegroupof KeithO'Nionsforundertaking the3Heanalyses. 541. 1982. NickOwenssuggested thepossibility of usingbacterialspores asmarinetracers,andGarethLeehelpedwith someof theanalyses.Commentsby Rik Wanninkhofandan anonymous reviewerimprovedthe Hesshaimer, V.,M. Heimann, andI. Levin, Radiocarbon evidence for manuscript. Finally, we would like to thank the staff of the Research VesselServices,in particularDave Teare, and the crew and officers Houghton,J.T., L.G.M. Filho, B.A. Callender,N. Harris,A. Katten- of RRS Challengerfor theirassistance.Fundingfor thiswork was providedby the EC Environmentcontracts(EV5V-CT92-0124) and (ENV4-CT95-0132)andPML corefundingfromtheUK NaturalEnvironment Research Council. a smalleroceaniccarbon-dioxide sinkthanpreviously believed, Nature, 370, 201-203, 1994. berg,andK. Maskel,Climatechange1995.'Thescience of climatechange,CambridgeUniv. Press,New York, 1995. Howarth, M.J.,et al.. Seasonal cyclesandtheirspatialvariablity, in Understanding the North Sea system,editedby H. Charnocket al., pp. 5-26, ChapmanandHall, New York, 1994. NIGHTINGALE ET AL.: IN SITU MEASUREMENT Jahne, B., G. Heinz, and W. Dietrich, Measurement of the diffusion coefficientsof sparinglysolublegasesin water, J. Geophys.Res., 92, 10767-10776, 1987a. Jahne,B., K.O. Munnich, R. Bosinger,A. Dutzi, W. Huber, and P. Libher, On the parametersinfluencingair-water gas exchange,J. Geophys.Res.,92, 1937-1949, 1987b. Kahma, K.K., and M. Lepparanta,On errorsin wind speedobservationson R/V Aranda,Geophysica,17, 155-165, 1981. Keeling, R.F., On the role of large bubblesin air-seagas exchange andsupersaturation in the ocean,J. Mar. Res.,51,237-271, 1993. Keeling, R.F., B.B. Stephens,R.G. Najjar, S.C. Doney, D. Archer, and M. Heimann, Seasonalvariationsin the atmosphericO2/N2 ratio in relation to the kineticsof air-sea gas exchange,Global Biogeochem.Cycles,12, 141- 163, 1998. Keswick,B.H., D.S. Wang, and C.P. Gerba,The useof microorganismsas groundwater tracers,Ground Water, 20, 142-149, 1982. King, D.B., and E.S. Saltzman,Measurementof the diffusion coefficientof sulfurhexafluoridein water, J. Geophys.Res., I00, 70837088, 1995. Large, W.G., andS. Pond,Openoceanmomentumflux measurements in moderateto strongwinds, J. Phys. Oceanogra., II, 324-336, 1981. OF AIR-SEA GAS EXCHANGE 387 oceanandits implications for CO2uptake,Nature,358, 738-740, 1992. Roether,W., andB. Kromer,Optimalapplication of theradondeficit methodto obtainair-seagasexchange rates,in Gas Transferat Water Surfaces,editedby W. Brutsaertand G.H. Jirka, D. Reidel, Norwell, Mass., 1984. Smethie, W.M., T. Takahashi, andD.W. Chipman, Gasexchange and CO2 flux in the tropicalAtlantic Oceandeterminedfrom Rn-222 andpCO2measurements, J. Geophys. Res.,90, 7005-7022,1985. Solomon,S., R.R. Garcia,andA.R. Ravishankara, On the role of iodinein ozonedepletion, J. Geophys. Res.,99, 20491-20499,1994. Tans,P.P., I.Y. Fung,andT. Takahashi,Observationalconstraints on the global atmospheric CO2 budget,Science,247, 1431-1438, 1990. Taylor,P. K., E. C. Kent,M. J. YellandandB. I. Moat,Theaccuracy of windobservations fromships,COADSWindsWorkshop, pp. 132-155, Environ.Res.Labs.,Nat. Oceanicand Atmos.Admin., Boulder, Color., 1995. Upstill-Goddard, R.C., A.J. Watson,J. Wood, and M.I. Liddicoat, Sulfurhexafluoride andHe-3 as seawater tracers- Deployment techniquesand continuousunderway analysis for sulfurhexafluoride,Anal. Chim.Acta, 249, 555-562, 1991. Large, W.G., andS. Pond,Sensibleandlatentheatflux measurements over the ocean,J. Phys. Oceanogra., 12,464-482, 1982. Wanninkhof, R., Relationship between windspeed andgasexchange Law, C.S., A.J. Watson, and M.I. Wanninkhof, R., J.R. Ledwell,andW.S. Broecker, Gasexchange Liddicoat, Automated vacuum analysisof sulfurhexafluoridein seawater- Derivationof the atmospherictrend (1970-1993) and potential as a transienttracer, Mar. Chem., 48, 57-69, 1994. Ledwell, J.R., The variationof the gas transfercoefficientwith molecular diffusivity, in Gas Transferat Water Surfaces,editedby W. Brutsaertand G.H. Jirka, pp. 293-302, D. Reidel, Norwell, Mass., 1984. Liss, P.S., Gas transfer:experimentsandgeochemicalimplications,in Air-Sea Exchangeof Gasesand Particles, editedby P.S. Liss and W.G.N. Slinn, pp. 241 - 298, D. Reidel, Norwell, Mass., 1983. Liss, P.S., and L. Merlivat, Air-sea gas exchangerates: Introduction and synthesis,in The Role of Air-Sea Gas Exchange in Geochemical Cycling, edited by P. Buat-Menard, pp. 113-127, D. Reidel, Norwell, Mass., 1986. Memery,L., andL. Merlivat, Modelingof gasflux throughbubblesat the air-water interface, Tellus, Ser. B, 37, 272-285, 1985. Merlivat, L., and L. Memery, Gas exchangeacrossan air-waterinterface: Experimentalresultsand modelingof bubblecontributionto transfer,J. Geophys.Res.,88, 707-724, 1983. Monahan,E.C., Occurenceandevolutionof acousticallyrelevantsubsurfacebubbleplumesand their associated,remotelymonitorable, surface whitecaps, in Natural Physical Sources of Underwater Sound,editedby B.R. Kerman,pp. 513-517, Kluwer Acad., Norwell, Mass., 1993. Monahan,E.C., and M.C. Spillane, The role of whitecapsin air-sea gas exchange,in Gas Transfer at Water Surfaces,edited by W. Brutsaert,andG.H. Jirka, pp. 495-504, D. Reidel, Norwell, Mass., 1984. Oost, W.A., The KNMI HEXMAX stressdata- A reanalysis,Boundary Layer Meteorol., 86, 447-468, 1998. Peng, T.H., W.S. Broecker,G.G. Mathieu, Y.H. Li, and E.A. Bainbridge,Radonevasionratesin the Atlantic and Pacific Oceansas determinedduringthe GEOSECS program,J. Geophys.Res., 84, 2471-2486, 1979. Peterson, E.W., and L. Hasse, Did the Beaufort scale or the wind cli- matechange?,Am. Meteorol. Soc.,17, 1071-1074, 1987. Phillips, L.F., CO2 transportat the air-waterinterface:Effectsof coupling of heat and matter fluxes, Geophys.Res. Lett., 18, 12211224, 1991. Pike, E.B., A.W.J. Bufton, and D.J. Gould, The use of Serratia mar- cesensand bacillus subtilisvar niger sporesfor tracing sewage dispersionin the sea,J. Appl. Bacteriol.,32,206-216, 1969. Ramage, C.S.,Secular change in reported surface windspeeds over overtheocean,J. Geophys.Res.,97, 7373-7382, 1992. wind speedrelationmeasured with sulfurhexafluorideon a lake, Science, 227, 1224-1226, 1985. Wanninkhof, R., W. Asher,R. Weppernig, H. Chen,P. Schlosser, C. Langdon, andR. Sambrotto, Gastransfer experiment on Georges Bankusingtwo volatiledeliberate tracers,J. Geophys. Res.,98, 20237-20248, 1993. Wanninkhof, R., et al.,Gasexchange, dispersion andbiological productivityon the westFloridashelf:Resultsfrom a lagrangian tracerstudy,Geopt•ys.Res.Lett., 24, 1767-1770, 1997. Watson,A.J.,R.C. Upstill-Goddard, andP.S.Liss,Air seagasexchangein roughandstormyseasmeasured by a dualtracertechnique,Nature, 349, 145-147, 1991. Wofsy,S.C.,M.B. McElroy,andY.K. Yung,The chemistry of atmospheric bremine,Geophys.Res.Lett.,2,215 - 218, 1975o Woolf,D.K., Bubbles andtheair-seatransfer velocityof gases, Atmos.Ocean, 31, 517-540, 1993. Woolf,D., Bubbles andtheirrolein air-seagasexchange, in TheSea Surfaceand Global Change,editedby P.S. Liss andR.A. Duce, pp. 173-205,Cambridge Univ.Press, Cambridge, 1997. Woolf,D.K., andS.A.Thorpe,Bubbles andtheair-seaexchange of gasesin near-saturation conditions,J. Mar. Res., 49, 435-466, 1991. Yelland, M.J.,B.I. Moat,P.K.Taylor,R.W.Pascal, J. Hutchings, and V.C. Cornell,Windstress measurements fromtheopenoceancorrectedfor airflowdistortion by theship,J. Phys.Oceanogr., 28, 1511-1526, 1998. J. Boutin,d'Ocfianographie Dynamique et deClimatologie, UMR 7617:CNRS/ ORSTOM/ Universitfi Pierreet MarieCurie,Tour 15, 2•meEtage,4, PlaceJussieu, 75252ParisCedex05, France.(e-mail: Jacqueline. Boutin@1odyc.j ussieu.fr) C.S. Law, M.I. Liddicoat,P.D. Nightingale, CCMS-Plymouth Marine Laboratory,ProspectPlace,West Hoe, Plymouth,Devon, PL1 3DH, UnitedKingdom.(e-mail:csl@ccms.ac.uk; mil@ccms.ac. uk; pdn@ccms.ac.uk) P.S. Liss, G. Malin, A.J. Watson, School of Environmental Sci- ences,University of EastAnglia,Norwich,NR4 7TJ, UnitedKing- dom.(e-mail:P.Liss@uea.ac.uk; G.Malin@uea.ac.uk; A.Watson@ uea.ac.uk) R.C. Upstill-Goddard, Department of MarineSciences andCoastal Management, University of Newcastle uponTyne,Newcastle, United Kingdom.(e-mail:R.Goddard@newcastle.ac.uk) theocean,Am. Meteorol.Soc.,26, 525-528, 1987. (Received October27, 1998;revised April 13, 1999; Robertson, J.E.,andA.J.Watson,Thermalskineffectof thesurface acceptedApril 28, 1999.)