PUBLICATIONS Journal of Geophysical Research: Biogeosciences Fluvial carbon export from a lowland Amazonian RESEARCH ARTICLE 10.1002/2016JG003464 rainforest in relation to atmospheric fluxes KeyPoints: LeenaE.Vihermaa1,SusanWaldron1,TomasDomingues2,JohnGrace3,EricG.Cosio4, • FsmoraelsltC-toso-autrmceo,s0p.h7eMregflCuhxai(cid:3)nd1iycra(cid:3)t1eda FabianLimonchi4,ChrisHopkinson5,HumbertoRibeirodaRocha6,andEmanuelGloor7 • Forest-to-riverand water-to-atmospherefluxeswere0.3 1SchoolofGeographicalandEarthSciences,UniversityofGlasgow,Glasgow,UK,2FaculdadedeFilosofia,CiênciaseLetras and0.09-0.14MgCha(cid:3)1yr(cid:3)1, deRibeirãoPreto,UniversidadedeSãoPaulo,SãoPaulo,Brazil,3SchoolofGeoSciences,UniversityofEdinburgh,Edinburgh, respectively UK,4SecciónQuimica,PontificiaUniversidadCatólicadelPerú,Lima,Peru,5DepartmentofGeography,Universityof • FluvialClosswashighestduringwet Lethbridge,Lethbridge,Alberta,Canada,6InstitutodeAstronomia,GeofísicaeCiênciasAtmosféricas,UniversidadedeSão season,especiallyinthecaseofDOC Paulo,SãoPaulo,Brazil,7SchoolofGeography,UniversityofLeeds,Leeds,UK SupportingInformation: • SupportingInformationS1 AbstractWeconstructedawholecarbonbudgetforacatchmentintheWesternAmazonBasin,combining drainagewateranalyseswitheddycovariance(EC)measuredterrestrialCO fluxes.AsfluvialCexportcan 2 Correspondenceto: representpermanentCexportitmustbeincludedinassessmentsofwholesiteCbalance,butitisrarelydone.The L.E.Vihermaa, footprintareaofthefluxtowerisdrainedbytwosmallstreams(~5–7km2)fromwhichwemeasuredthedissolved [email protected] inorganiccarbon(DIC),dissolvedorganiccarbon(DOC),particulateorganiccarbon(POC)export,andCO efflux. 2 TheECmeasurementsshowedthesiteCbalancetobe+0.7(cid:1)9.7MgCha(cid:3)1yr(cid:3)1(asourcetothe Citation: atmosphere)andfluvialexportwas0.3(cid:1)0.04MgCha(cid:3)1yr(cid:3)1.Ofthetotalfluvialloss34%wasDIC,37%DOC, Vihermaa,L.E.,S.Waldron, T.Domingues,J.Grace,E.G.Cosio, and29%POC.ThewetseasonwasmostimportantforfluvialCexport.Therewasalargeuncertainty F.Limonchi,C.Hopkinson,H.R.da associatedwiththeECresultsandwithpreviousbiomassplotstudies((cid:3)0.5(cid:1)4.1MgCha(cid:3)1yr(cid:3)1);hence,it Rocha,andE.Gloor(2016),Fluvialcar- bonexportfromalowlandAmazonian cannotbeconcludedwithcertaintywhetherthesiteisCsinkorsource.Thefluvialexportcorrespondsto rainforestinrelationtoatmospheric only3–7%oftheuncertaintyrelatedtothesiteCbalance;thus,otherfactorsneedtobeconsideredtoreduce fluxes,J.Geophys.Res.Biogeosci.,121, theuncertaintyandrefinetheestimatedCbalance.However,streamCexportissignificant,especiallyfor 3001–3018,doi:10.1002/2016JG003464. almostneutralsiteswherefluviallossmaydeterminethedirectionofthesiteCbalance.ThefateofC Received1MAY2016 downstreamthendictatestheoverallclimateimpactoffluvialexport. Accepted25OCT2016 Acceptedarticleonline27OCT2016 Publishedonline14DEC2016 1. Introduction Tropicalforestscoveravastarea,variouslyestimatedatbetween17and25×106km2[Achardetal.,2002; Graceetal.,2014].Theyexchangeenergy,mass,andmomentumwiththeatmosphere,andsohavethecapa- citytoinfluencetheclimatesystem,locallyandregionally[CostaandFoley,2000;LeanandWarrilow,1989; Shuklaetal.,1990;WerthandAvissar,2002].OfparticularinterestistheexchangeofCO withtheatmosphere, 2 andthecapacityoftheforesttoactasastoreofcarbonandacarbon“sink,”thusprovidingavaluableglobal environmentalservicebyabsorbinganthropogenicCO emissions[Gibbsetal.,2007].Ourunderstandingof 2 howthemetabolismoftropicalforestsdeterminescarbonfluxes,andhowunderlyingprocessesareinflu- encedbyclimatologicalphenomenasuchasdrought,hasdevelopedrapidlyasaresultof(i)measurements andexperimentationontreesatthescaleof1haplots[CostaandFoley,2000;Lewisetal.,2009;LoladaCosta etal.,2010;Nepstadetal.,2007;Phillipsetal.,1998,2010],(ii)micrometeorologicalobservationsattheecosys- temscaleusingeddycovariance[Araujoetal.,2002;Carswelletal.,2002;Kruijtetal.,2004;Saleskaetal.,2009], and(iii)observationsatthelandscapescalefromaircraft-basedmeasurementsofatmosphericconcentra- tionsofCO [Chouetal.,2002;Gattietal.,2010,2014;Lloydetal.,2007]orlidar-basedassessmentofstanding 2 biomass[Marvinetal.,2014]. ManyoftheecosystemstudiesuseeddycovariancetomeasuretheverticalexchangeofCO andapportion 2 thistophotosyntheticandrespiratoryprocesses.Yetthereislateraltransportofdissolvedandparticulate carbon away from the forest in the drainage water, and hence, a strong likelihood of overestimating the ©2016.TheAuthors. terrestrialcarbonsinkiftheCleakageindrainagewaterisnotaccountedfor[Richeyetal.,2002;Waterloo Thisisanopenaccessarticleunderthe etal.,2006].Foragivensiteanycarbontransporteddownstreamindrainagewaterconstitutesapermanent termsoftheCreativeCommons lossofcarbon. AttributionLicense,whichpermitsuse, distributionandreproductioninany Thedrainagefluxescomprisedissolvedinorganiccarbon(DIC),dissolvedorganiccarbon(DOC),particulate medium,providedtheoriginalworkis properlycited. organiccarbon(POC),andalsoCO2effluxfromtheriversurface.IntheAmazonBasin,despitethelargeEC VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3001 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 network,onlyatasinglesitehasorganicCexportbeenrelatedtoECresults[Monteiroetal.,2014;Waterloo etal.,2006].ThefluvialCO efflux,iftakingplacewithinatowerfootprint,willbecapturedwithintheECmea- 2 surements,butthisCO poolrepresentsadifferentpathwayofcarbonmovement,distinctfromecosystem 2 respiration,andisrequiredtounderstandandmodelthesystembehaviorindetail.Forunderstandingwider ecosystemfunctioning,thetypeandprocessingoftheexportedCisalsoimportant.DOCmaybedecom- posedtoCO andlosttotheatmospherealongwithCO derivedfromsoilrespiration.POCmayundergo 2 2 decomposition, be deposited on floodplains, or ultimately, any fraction that has escaped decomposition maybe buried in ocean sediments. Deep-sea burial of ecosystem-derived POCwill contribute to removal ofatmosphericCO [Galyetal.,2015,Hilton,2016]. 2 Understandingthenatureofthislateralfluxindependentlyofthenetcarbonbalanceataforestedsiteis important. There is a long history of measuring Amazonian aquatic efflux [Bartlett et al., 1990; Devol et al., 1987; Richey et al., 2002], but these measurements have not been combined with eddy covariance measurements.MostrecentestimateoftheCO emissionsfromalltheriversandstreams(takenasrivers 2 <100mwide)oftheAmazonBasinupstreamofÓbidossuggestsasmuchas0.8PgCyr(cid:3)1isreleasedtothe atmosphere[Raseraetal.,2013],farmorethanisdischargedtotheoceanatthemouthoftheriver.Afluxof thismagnitudeislargerthanthelossesfromdeforestationandlandusechangeinSouthAmericaofabout 0.5PgCyr(cid:3)1 [Gloor et al., 2012] and comparable to fossil fuel burning in the tropics 0.74PgCyr(cid:3)1 [Grace etal.,2014]andthereforeneedstobeconsideredinanyconstructionofthetropicalcarbonbudgetthataims toseparateexplicitlythedifferentpathwaysofCmovement.Theaquaticeffluxmayalsoincludeacontribu- tionofagedcarbon,i.e.,notrecentlyfixedfromtheatmospherebytheecosystem[Clarketal.,2013;Mayorga etal.,2005;Vihermaaetal.,2014].Ifitisknownthatthisoccurs,acorrectionforthefractionoffossilcarbon inputsmustbemadefor,asotherwisethebudgetwouldoverestimatethelossofrecentlyfixedC. The fluxes from forest to drainage water, hereafter called the fluvial export, are likely to vary seasonally accordingtothehydrologicalcontrols.DOCandPOCconcentrations,[DOC]and[POC],aretypicallyhigher in the wet season and during rain events that leach carbon from the forest canopies and soils [Johnson etal.,2006;Monteiroetal.,2014;Salimonetal.,2013;Townsend-Smalletal.,2008].Heavyraineventsusually resultinapeak[DOC]butthetimesincepreviousraineventandtheintensityofraininfluencethisrelation- ship [Monteiro et al., 2014]. DIC concentration, [DIC], has been found to be maximal at low water levels [Salimonetal.,2013;Sousaetal.,2008]withthetimingofmaximumexportfluxdependingonthebalance betweentheincreaseddischargeandthedilutioneffect.AstheCconcentrationsandfluxesvaryseasonally, accordingtohydrologicalcontrols,quantifyinganannualexportbudgetrequiresdatacaptureoverthefull hydrologicalrange. EddycovariancemeasurementshavenotpreviouslybeencarriedoutinthePeruvianAmazonwherethesoils areyoungerandmorefertileingeneralthanthesitesstudiedpreviouslyinBrazil.Herewereportcatchment- scalemeasurementsfromalowlandrainforestinPeruinthismorefertilearea,whereforest-to-atmosphere andadditionallyforest-to-riverandriver-to-atmospherefluxesweremeasuredfor1yearduringOctober2011 toSeptember2012. 2. Methods 2.1. StudySite ThestudysitewaslocatedintheTambopataRivercatchment(latitude12°49′54.30″S,longitude69°16′52.37″W), near Puerto Maldonado, Madre de Dios, Peru (Figure 1). It is within the Reserva Nacional de Tambopata, notedforitshighbiodiversity.Forexample,a1haplotatthesitewasfoundtocontain556treesin115taxa [Lopez-Gonzalezetal.,2012].Theclimateiswarmandhumid.Anaverageannualtemperatureof25.4°Cand precipitationof2377mmwasrecentlyreportedtobethe112yearaverage[Weatherbase,2015].Thewind directionisquitevariable,withNWasprevailingdirection.Elevationgradientsofthesurfacearoundtheflux toweraresmall(Figure1),amaximumof10mincreaseinelevationwithin0.5kmradiusinalldirections.The towerfootprinthasanaveragedaytimeareaofabout2km2andiscoveredby30mtallprimaryevergreen forestonsoilsthatvaryfromwelldrainedtoswampy,classedasHaplicCambisolsorInceptsolsintheU.S. Soil Taxonomy [Quesada et al., 2011]. The Tambopata site is part of the more fertile areas within the AmazonBasin,withthesoilsamplesclassedinthetop50percentileofthedataintermsofnutrientcontent inanAmazonBasin-widesoilstudy[Quesadaetal.,2010].Theabovegroundbiomassinthemonitoringplot VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3002 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 Figure1.LocationoftheRamiroChacon-SAGESeddycovariancefluxtower,thesurroundingtopography,thetwostudy streams,andtheforestbiomasssamplingplots(TAM5,TAM6,andTAM9)attheTambopatasite.NewColpitaistheper- enniallyactiveStream1,andMainTrailistheephemeralStream2.Thestreamcatchmentboundariesaremarkedwith dashedlineandthesamplingpointsineachsystemwithadot(SP1andSP2).Inset:locationofthestudysiteinthewestern partoftheAmazonBasininMadredeDiosregion,Peru. knownasTAM09was81MgCha(cid:3)1and70MgCha(cid:3)1inthewiderforestareasurroundingtheplot[Marvin etal.,2014].TheselectedsiteisidealforconstructingadetailedCbalanceasthelocalforestareaisdrained onlybyheadwaterstreamsandupstreamCinputsdonotneedtobeconsidered. 2.2. EddyCovarianceMeasurements Aneddycovariancefluxtower,named“TheRamiroChacon-SAGESTower,”wasconstructedin2010–2011 and flux measurements commenced in September 2011. The tower is a free-standing structure made of 42mtallsteelgirders.Theinstrumentsforeddycovariance,chosenfortheirlowpowerconsumption,were for CO and H O concentration, LI-7200 with a short-path sampling tube and high-flux pump LI-7550 2 2 (LI-COR, Lincoln, Nebraska, USA); for CH concentration, LI-7700 (LI-COR, Lincoln, Nebraska, USA); and for 4 windspeedanddirectionasonicanemometer,CSAT3(CampbellScientific,Utah,USA).Meteorologicalinstru- mentsweresolarandthermalradiation,NR01(CampbellScientific,Utah,USA);directanddiffuseradiation, BF3 (Delta-T Devices, Burwell, UK); photosynthetically active radiation downwelling and upwelling, LI-190 (LI-COR, Lincoln, Nebraska, USA); temperature and relative humidity, HMP45C (Campbell Scientific, Utah, USA);windspeedanddirection,VectorA100PandW200P(CampbellScientific,Utah,USA);wetnesssensor, SKLW1900(SkyeInstruments,Llandrindod,UK);barometricpressure,CS100(CampbellScientific,Utah,USA); andatippingbucketraingauge,TB4(CampbellScientific,Utah.USA).Airwasdrawnfromabovethetowerat 15Lmin(cid:3)1via5.3mminternaldiameterSynflextubeanddirectedtotheCO analyzer.Dataforeddycovar- 2 ianceanalysiswerecollectedat10HzonaCampbelldatalogger(CR3000,CampbellScientific,Utah.USA);for the meteorological data the sensors were scanned every minute. The entire array of instruments runs at around 12V and 5A. Power for the instruments was provided by six solar panels: 67cm×148cm, 12V VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3003 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 nominal,andratedat42Weach;storagewas560Ahinanarrayofleadacidbatteriesmountedatthetopof thetower. Datawerefilteredtoremovecaseswheretheenergybalanceclosurewasinadequateduetolowwindspeed andwhenthefrictionvelocity(u*)valuewaslessthan0.17ms(cid:3)1(basedonaplotofnightfluxversusu*). GapswerefilledafterfittingcurvestotherelationshipbetweenCO fluxandphotosyntheticallyactiveradia- 2 tion,followingtheprocedureofGilmanovetal.[2007].GPP(grossprimaryproductivity)wasestimatedby subtractinganestimateofdarkrespirationfromthenetfluxduringthedaylighthours.Darkrespirationin the day was estimated from dark respiration at night by correcting for the warmer day temperatures [Aubrechtetal.,2016;Stoyetal.,2006].Finally,themonthlyandannualtotalfluxeswerederivedbyaccumu- latingthehalf-hourvalues. From28Mayto16June2012therewasacontinuousperiodof20daysofmissingdatainthefluxtowerrain- fallrecord.Thiswasgap-filledusingthedatafromthreenearestSENAMHI[ServicioNacionaldeMeteorologíae Hidrología del Perú, 2014] met stations (Tambopata, Limbani, and Puerto Maldonado; Figure S1 in the supporting information). The estimated missing rainfall (37.8mm; standard deviation 10.4) was added to therainfallrecordedbythetowerinthelatterhalfofJune(69.4mm),yieldingatotalof107.2mm. 2.3. FluvialCarbonSampling Twosmallstreamsdrainthetowerfootprintarea(Figure1):oneisperenniallyactive(Stream1,NewColpita), whiletheother(Stream2,MainTrail)isactiveonlyduringthewetseason.Surfacewatersamplesformeasure- ment of [DIC], [DOC], and [POC] were collected during three field campaigns: February–April 2011, September–December2011,andMarch–May2012.Stream2driesupduringthedryseason;therefore,no samplescouldbecollectedduringSeptember–November2011.Differentflowconditionsweretargetedin samplingtounderstandthehydrologicalcontrolsonthecarbonconcentrationsandfluxes.DICsampleswere collectedinpre-acidified(150μLofconcentratedphosphoricacid)evacuated12mlExetainersandthehead- spaceanalyzed[Waldronetal.,2014]onThermo-Fisher-ScientificGasBench/DeltaVPlusfor[DIC].DOCsam- pleswerefilteredthroughpre-furnaced(8hat450°C)0.7μmglassfiberfilterpaperonthedayofsampling and stored refrigerated. Samples treated this way show little change in composition over 3months [Gulliver et al., 2010]. Prior to the measurement of [DOC] by combustion (Thermalox TOC 2020, Analytical Sciences)thesampleswereacidifiedtopH3.9anddegassedtoremoveanyDIC.[POC]wasmeasuredbyloss onignition(LOI)onthefilterpapers.Inthismethodovendry(3hat105°C)weightwascomparedtothe weight after furnacing (16h at 375°C). The OC content of the mass loss was assumed to be 50% [Atjay et al., 1977; Pregitzer and Euskirchen, 2004], which corresponds well with the measured range C content (45–53%)ofdifferentlitterfallfractionsfromthesouthernAmazonforestbutwashigherthanthefraction (43%)foundincoarsePOM[Selvaetal.,2007].ToassessCinputsinrainfall,[DIC],[DOC],and[POC]werealso measuredinalimitednumberofrainwatersamplescollectedonthe2012campaign.Thedeliveryperland area was calculated using rainfall volume-weighted mean concentration of C fractions and the annual totalrainfall. DirectmeasurementsofCO effluxfromwatersurfaceswerecarriedoutusingafloatingchamberconnected 2 toaCO analyzer(Li-840A,LI-COR,Lincoln,Nebraska,USA).CO accumulationinthefloatingchamberhead 2 2 space was measured every second for 4min, the measurement repeated three times, and the flux rates calculatedaccordingtoFrankignoulle[1988].Additionalwaterflowvelocitymeasurements(Handheldflow meter, Geopacks, Hatherleigh, UK) were taken in the exact location where the chamber was deployed. A limitednumber(n=4)ofCO effluxmeasurementswerecarriedoutinthefloodedforest. 2 2.4. StreamCatchmentAnalysis The streams were mapped by walking along them while recording the route with a hand-held GPS. The stream catchment areas were analyzed in ArcGIS 10 (ESRI, USA) using a digital elevation model (DEM) constructed fromlidar data[Boyd etal.,2013;Hill etal.,2011]. LidarDEM-basedstream locationdatawas comparedto thefieldmeasurements,andgoodagreementwasfound. Thelidardataallowedanalysisof thelargerareaofthecatchmentthanhadbeenpossibletomapinthefieldduetodifficultyincuttingroutes through the dense vegetation. Upstream of the sampling point, Stream 1, New Colpita, drains an area of 7.2km2,andtheseasonalStream2,MainTrail,drains4.9km2.AtthesamplingpointStream1was4.5–7.5m andStream2was3.5–5mwidedependingonwaterlevel. VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3004 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 2.5. HydrologicalandWaterChemistryMeasurements Waterchemistry(pH,conductivity,dissolvedoxygen,andtemperature;Troll9500,In-SituInc.,USA)andstage height(RuggedTroll,In-SituInc.,USA)werecontinuouslyloggedevery15mininthesestreamsfromFebruary 2011toOctober2012.Thewaterchemistryisimportantasthecontinuoustimeseriescouldbeusedtopre- dictCconcentrations,andstageheightisneededtocalculateexportbudgets.Flowvelocity(Flowlink2150, ISCOInc.,USA)wasmonitoredineitherstreaminturnascampaignsandstreamcrosssectionsmeasuredper- iodically.AsflowvelocityisthekeycontrolforCO outgassing[Longetal.,2015]wefocusedonderivinga 2 flowvelocity timeseriesfrom thestage tovelocityrelationship (FigureS2)andthen calculateddischarge fromvelocityandtheactivecrosssectionatgivenstageheight. During the study period the stage height ranged between 24 and 717cm and between 0 and 90cm in Streams1and2,respectively.ThestagevelocityregressionobtainedforStream1wouldyieldhigh-velocity estimatesathighstageheightvalues,whereasbasedonfieldobservations,thestreamvelocitywouldstartto decreasewhenstageheightexceeded100cmandwouldhaveceasedcompletelyby180cmstageheight. Therewerealimitednumberofvelocitymeasurementsfromfloodedconditionsastheflowloggerunitcould notbedeployedduringthoseperiodsduetotheriskofdamagingorlosingtheunit.Alineardecreaseinflow velocitywasassumedbetween100cmto180cmstageheights. ThestagetovelocityrelationshipinStream1wascomplicatedasthefastest-flowingsectionwasattimes divertedbydebristrappedbetweentherocksintheriverbed;thus,similarstageheightscouldbeassociated withwiderangeofflowvelocities(FigureS2).However,anexponentialincreaseinvelocitywasnotconsid- ered realistic, so the linear relationship was fitted but with the high predictions during flooding adjusted basedonfieldobservationsasdescribedabove.Thestagetoflowvelocityconversionappearstooverpredict- inginthelowervelocityrange.However,thisvelocityrangehasalimitedcontributiontothetotaldischarge withtheflow<0.2m/scontributingjust5%tothetotaldischargeduringthestudyperiod.Ofmoreconcern wouldbeuncertaintiesinthepredictedhigher-velocityrange.Thecalculatedflowvelocities>0.7m/scontri- buteapproximately30%oftheannualdischarge.However,quantifyingtheerrorinthatextrapolatedrange, orintherangewhenvelocitymeasurementsdiverge(>0.4ms(cid:3)1),isdifficult.Toyieldanestimateonthemag- nitudeoftheerrorinthevelocitymeasurement,thepercentresidualsinthemidrangeofpredictedvelocity (0.25–0.35ms(cid:3)1;n=485)wereinvestigated(FigureS3).Themedianpercentresidualinthisrangewasapproxi- mately2%withthefirstandthirdquartilesat(cid:3)25%and23%,respectively.Themeanofthesequartiles(24%) was taken to describe the uncertainty associated with flow velocity estimate. To propagate the error to discharge,anestimated10%errorinthestreamcross-sectionmeasurementwasassumed.Toinvestigate theeffectofpotentialdischargeoverestimationinthevelocityrangewheredatawaslacking(>0.7ms(cid:3)1),a sensitivityanalysiswascarriedoutbyreducingthedischargeinthatrangeby37.5%andrecalculatingthe annualtotaldischargeandtheresultingCexport. InthecaseofStream2thestageheighttoflowvelocityrelationshipwasbetterconstrained(FigureS2).The medianpercentresidualacrossthewholemeasurementrangewasapproximately3%withthefirstandthird quartileat(cid:3)6%and10%.Theresultinguncertaintyintheflowvelocitymeasurementwasestimatedtobe themeanofthequartiles(8%),whichwasthenpropagatedinthedischargedataalongwiththeestimated 10%uncertaintyinthecross-sectionmeasurement. The closure of water balance was investigated comparing the flux tower measurements of rainfall and evapotranspirationandtheresultingcalculatedwateravailabilityto theobservedstreamflow. Stream1 is perenniallyactive asa result of a groundwater inputs, and hence,thetotal streamflow was split to event andbaseflowusingtheEcoHydRologypackageinRsoftware,version3.1.0.TheBaseflowSeparationcode appliesadigitalfilter[LyneandHollick,1979]tothestreamflowdata;threepassesofthefilterwereused. The seasonal Stream 2 does not have a groundwater component, and the total discharge was compared tothecatchmentwateravailability. 2.6. DerivingContinuousFluvialCarbonExportTimeSeries InordertocomparefluvialexportwiththeECestimates,detailedandcontinuoustimeseriesarenecessary. ContinuousmeasurementoffluvialexportisnotpossibleassensorsforallrelevantCpoolsdonotexist.Thus, the continuously logged hydrological and water chemistry data was explored to assess if these measure- ments formed strong relationships with directly measured values and so could be used to generate VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3005 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 semicontinuous time series (models) of [DIC], [DOC], and [POC]. This approach has worked successfully elsewhere[e.g.,Waldronetal.,2007;Monteiroetal.,2014;Zanchietal.,2015].Fromsuchconcentrationtime series,usingcontinuousdischargetimeseriesfluvialexportcanbecalculated.Similarly,tounderstandthe fractionofCO thatevadesfromaquaticsystems,ratherthanfromplantandsoilrespiration,amodelfor 2 continuousCO effluxtimeserieswasrequired. 2 Generalizedlinearmodels(glm)werefittedtothewaterchemistryandhydrologicaldatausingtheRversion3.1.0 [RCoreTeam,2014]toconstructmodelstoestimatethecarbonconcentration(DIC,DOC,andPOC)andCO 2 effluxtimeseries.Thesetimeserieswerethencombinedwithdischargedatatocalculateannualcarbonexport. Thisfullmodelwassimplifiedbysequentiallyexcludingtheleastsignificantexplanatoryvariablestoderivethe bestfitmodel.Aglmmodelhasauser-definederrorstructureandalinkfunctionthatcanbeselectedtoensure thatthefittedvaluesstaywithinarealisticrange[Crawley,2003].ThegeneralizedlinearmodelsdonotyieldR2 values,andtherefore,theconcordancecorrelationcoefficients(ccc),whichconsidertheinfluenceofbothloca- tionandscaleshiftintherelationshipbetweentwovariables[Lin,1989],wereusedtoassessthequalityoffit. For[DIC],[DOC],and[POC],anidentitylinkwasused.FortheCO effluxaloglinkwasselectedtopreclude 2 negativepredictionsofCO efflux;asthiswasneverobservedinfieldmeasurementandallpCO valuescal- 2 2 culatedfrommeasured[DIC],pHandtemperaturedata[Rebsdorfetal.,1991]wereabovetheatmospheric equilibrium(Stream 1: 2399–10,712ppm; Stream 2: 1256–4205ppm (details on stream pCO are found in 2 Longetal.[2015]).TobuildtheCO effluxmodel,themeasurementofwaterflowvelocityatthechamber’s 2 locationwasused.InStream1,thewaterflowvelocitymeasuredatchamberdeploymentspotsrangedfrom 0.05 to 0.9ms(cid:3)1, with one extreme value of 1.58ms(cid:3)1 which was excluded from the modeling data set. Stream1alsoflooded,duringwhichtheflowstoppedasthehighwaterlevelintheTambopataRiver,into whichthesmallstreamsdrained,backedupandblockedthestreamwateroutflow.Duringfloodedperiods themodelwasnotapplied;instead,aneffluxrateof0.33μmolm(cid:3)2s(cid:3)1(n=1)measuredduringfloodingin thefieldwasused.Thechambermeasurementsonagivendaywerecarriedoutinmultipleplacesinthe streamandmaynotbecoincidentwiththemodelprojectionasthesespotmeasurementswereoverarange ofvelocities,whereastheflowvelocityloggerwasdeployedatafixedpoint.InStream1,analternativeCO 2 effluxmodelbasedonflowvelocityalonewasderived;thiswasappliedwhendissolvedoxygenandconduc- tivitywereoutsidetherangeobservedduringfieldmeasurementsasthefullmodelproducedextremepre- dictionsintheseconditions. ToderiveannualaquaticCO effluxcontributioninunitsthatarecomparabletotheothermeasurements,the 2 modeledeffluxrate(μmolCm(cid:3)2s(cid:3)1;relatingtowatersurface)wasconvertedtoMgCyr(cid:3)1usingthestream surfaceareacalculatedfrommappedstreamlengthsandstreamwidthestimatesfromdifferentseasons.This totalCO effluxfromthestreamsurfacewasthenrelatedtothecatchmentlandarea(asinBillettetal.[2004]). 2 PriortoproducingtheCtimeseriesfromthewaterchemistrydata,agapintheStream1pHdatadueto probemalfunctioning(1Julyto7September2011)wasfilledbyusingpooledcoefficientsderivedfrommulti- ple fits (100) of a model based on conductivity and stage height (MICE package in R [van Buuren and Groothuis-Oudshoorn,2011]).InStream2waterchemistrytimeseriesthereweremuchlargerdatagapsin 2012duetoproblemswiththeinstruments.Thesewerebeyondthescopeofreliablegapfilling,andthere- fore,amorerestrictedCconcentrationtimeserieswasproduced. ThebestenergybalanceclosureintheECdatawasfoundwhenthewindcamefromtheNE.Then,thetower wasmeasuringgasexchangeinafluxfootprintcontainingStream1catchment.Hence,datafromthisstream weremoresuitableforcalculatingthefluvialcarbonexport,andthesiteCbalancefocusesonthesedata. Whencontinuouscarbontimeserieswereusedtocalculatemonthlyexportandannualtotalfortheperiod fromOctober2011toSeptember2012,anymissingvalueswerefilledwiththemonthlymeanvalue.Results fromStream2wereincludedtoillustratethedifferencesfoundintwoadjacentstreamcatchmentsdraining thesameforestarea.Uncertaintyintheestimatesisexpressedasstandarddeviationunlessstatedotherwise. 3. Results 3.1. TerrestrialFluxesofCO 2 TheECmeasurementsindicatedaphotosyntheticuptakeof27.2MgCha(cid:3)1yr(cid:3)1andanestimatedrespira- torylossof27.9MgCha(cid:3)1yr(cid:3)1(Table1).Uncertaintiesintheseestimatescannotbecalculatedfromthedata VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3006 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 a Table1.MonthlyBreakdownoftheEddyCovarianceCarbonSinkEstimatesandFluvialCarbonExportattheTambopataStudySite,MadredeDios,PeruFromOctober2011toSeptember2012 11Oct11Nov11Dec12Jan12Feb12Mar12Apr12May12Jun12Jul12Aug12SepTotal (cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:3)(cid:1)FluxtowerGPP2,3372,2112,1862,4041,9862,3282,3042,1142,0162,3732,5072,41427,1816795estimate(cid:1)Ecosystem2,4251,8442,1422,5652,1682,1582,6772,1182,2612,4282,6102,49227,8886972respiration(cid:3)(cid:3)(cid:3)(cid:1)87.6366.644.1161.4181.6169.9372.84.4244.954.6103.377.67089736FluxtowerNEEestimatebb(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)DICexport4.21.54.41.69.43.310.73.69.93.512.54.18.22.612.54.112.94.413.14.59.03.26.32.311311.8bb(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)1.401.012229.8DOCexport1.20.71.51.010.87.021.512.923.717.128.817.111.06.47.64.25.93.16.53.41.62.3bb(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)(cid:1)POCexport2.22.62.32.77.99.212.614.812.614.816.319.09.210.78.910.47.89.28.49.84.85.63.33.89636.7bb(cid:1)fl11.033148.7Totaluvialcarbon7.78.228.244.946.257.628.428.926.728.015.5exportTotalClossfromthesite1039fl(NEE+uvial) aflSignconvention:negativemeansuptakeofcarbonfromtheatmosphere.TheaquaticcarbonexportvaluesarebasedonStream1thatisdirectlydrainingtheuxtowerfootprintarea.Theunits(cid:3)(cid:3)(cid:3)(cid:3)1111flmonthforalltheothercolumnsexceptthetotalwhichiskgChayr.TheuncertaintiesintheuvialexporthavebeenderivedpropagatingtheuncertaintiesfromtheCareinkgChaconcentrationmodelanddischargecalculation.OnannualleveltheuncertaintyrelatedtotheECmeasurementswastakentobe25%accordingtoKruijtetal.[2004].bThesevaluescomewithalargeruncertaintyasthepressuresensordatatoderivestageheightanddischargedidnotappeartorecordafairlylargerainevent,andsoitsfunctioningmaybelimited. swat(eliwar3ThSfcpwwt(A2tahaa9dfit(iccFcctiondetf6uttasnn6FTeohahhhhhiheoooouvvnnsohtxr7..efpo.aierian78tahhaiagh7rri2rrrueaeseeeeipaparlvnsg0rdendanetenteoeeoentsstlboie8rhot.ayssnMMaiieaauclo5hdudfuncea,rsllrcruyeertcl)ameaaksawiracrSha1(cid:1)Seewrttrreliem.halantgg(cid:1)saiaidtotlabebbitstpeviltt2alattllialtolriineeotieyltarmitcostcWwtorotC1C8huiiseptnttfloiseo7nameillShcdot2geoaeouhheaoii)n8ialeriuceflaottdghh0ls3dre.ednndadtentnvnloiumnyy1essio(opdamxccosmohso3)eaainnaTenmepdtiTd,dex.dn.5ceoonptsfewett(cid:3)(cid:3)nebhhcssrnamieceri(mulcgae%hsmnntttaehcmtIc1oree7tealS117eboritdhhnahCelaineeiehirtdrdfiaaemco5srvn)tar.veayylreayiseeusrp0,hcaatreeteeiidrubrfludde7anlerrentfnptheheeteltdwetlaer(cid:3)(cid:3)fiiMtoceevcyeousshled2m(cid:1)faiessesrwedhsrosimeatctttp.aalnndcw11ca(tfserihatom)myyrighaieuotTciga.vmil.aaaocaavsocirmet1Tnseollrtitgeetieaccoifea.husgrTusCtiewasnsWttainhnhn8ooftwwseoehrbtheseiibnarhtrax.oet2oCbyendre9itetmhintI0eedrinaimotilenitoetiraniehOooinnostesmslyoesamho.gmo.mlwsinhbyds7ratal(ailturvobtans(cid:3)fasmtesi1afehmbphdeflyinaaujvnae.ieasMmStuctnenoonn5mr1i3esys.reloiinredlmstatntnaplTteotarstcyiardufueged)ayhetWgir)nhtwchdwichtdial.ngSrhseahtecaslrlinKmehd,teeCehtlterlntyswCes(cid:3)tuchacwhahheamoarririeDiicsbromagnAgeganudooncmt1fsqeerweihoc(flhlsaoserocctisauehpge1iohgfuSnofdnaohuseaghjamnnociaetffogrltaenpde3etasenrpmtcoealam(cid:3)iaeteasdlmnnwprAlnehpcrsrttus91cttchhhasctrstreevsneenrhh1.ehiutkHasiteulmeepnw9m,tloesoeotaeeqaoytcelnpoeedloytnice1aUrcd,l,lcfyrnntarram(cid:1)hnuaombictaoorettooltihdosflbeeagcsnasinras(cid:3)bhtaoeeesunztanntnd1rnboasuawSrudleetaieel3coflhiooees.1rnrsnltoythdd1snardltuoxbdsnrdernrap6tarucehpiienngr1itnteca6eteitpreri[etetdrnrneseiutl4syecnva2tehaaoytentcmocyitttlw1dsid,oenaaduaeolcirhat,t0i,remp,ptawwva2flnaidt9mmrttoilawdrrieolwy0evdaaditetttttelniioii5onan5fucftmnriinsonbehhhhhdmit4alicrttatoroomieestt%aiwaasslsiirfCdhngnhhngoit2eeeeee]kyyyylcsssrrr-------ff)lll:,. VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3007 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 1) ingandapproachingfloodedconditions;estimatingrain- flEfuxCO2(cid:3)(cid:3)2molms (cid:1)16.9(15.5)12.9–0.379.651 (cid:1)4.5(2.5)4.6–0.412.042 fHeasoltlwimceaavnteerb,wetaheserrrewoarastpoernroabnbaelleatanoncoed[pein.ogdt.ie,cnMattieeaksloetnhrrnaoetrnsthienettdahlie.s,ch2hi0ga1rhg5ee]r. μ( flow range had limited impact on the total export esti- mates and does not change the unambiguous finding cmber2012 POC(cid:3)1(mgCL) (cid:1)5.3(6.8)3.5–1.054.182 (cid:1)2.7(1.6)2.3–1.210.053 ents. tTThhaametbtflhooepoadritivanegRrsivoaebrres,eawrvhneeedrteCintehSxeptrsoetrartem.am1swdarasinc,abulsoecdkibngy tthhee e m y2011toSept DOC(cid:3)1(mgCL) (cid:1)3.6(2.3)3.1–0.712.382 (cid:1)7.0(1.3)6.58–4.711.452 ctionmeasure mwpwoaaixittnieentrrgaeosfwufeTtifltcahotmswtbihn.oeTpthhsateetraemflawomeoaadtsweiunrargtecemoer.upeTlnidshtoesbdtebeauskflleokcnwoouaifnltdsgtahcmienauwpsslaaeemrsdbpuaalsicnnekddg- Dios,Peru,Februar DIC(cid:3)1(mgCL) (cid:1)4.5(2.2)4.3–1.09.3172 (cid:1)1.2(0.3)1.2–0.61.9104 edstreamcross-se drHTtaoaietnmminogcbnoeeodso,,pebtaalhsnDteeadIrCvRmteaihvdonoedddsreeuDwlrsOtianaCdtgkeeehrnrnsaiov.dnedHbdfluoeorwiesonhnedgoveteuafldrklod,ceosondinnniiondcnteigntibootahennreeflsao(DfwFofIeidiCgtcheutmdeirnedcoSotdhb4nee)y-l. e,Madrede Discharge(cid:3)a31(ms) (cid:1)0.47(0.83)0.26–05.1b9,053 (cid:1)0.2(0.2)0.15–0.011.5b6,468 ndthedetail itcwniovacuitltuleyddraaecsrhdiesecxemopdnliausdnteruaytctotofriTvroyaimtmvya,bwaroinhapdibcalthethaste,bheDearOcrCokCwrcsmoainntoecdtreheenelftfpereaHxctptiaooonnnrdttSbcitmourendeadgsmueectr1s-- heTambopataSit StageHeighta(cm) am1(cid:1)58(66)40–2471756,610am2(cid:1)32(16)30–109024,482 (cid:3)1velocity(ms)a istiiennitvasccigrrtweeyeeaaharsseeeneidcigcnoohpntuthHslwtde,rasubbesceeatvyecdaodcrunoi.aeDdmbtulpo1erai8snTn0wagicemenmdrobewrosmoipttbahaagstlaeasetrdwrvhreeoaeadptimeg(irFnhfliatobgnwuoadtrt,ehiintnciScmflo5ruen)e.eadsTnsuheaccidnes- att Stre Stre owam. themodeledcarbonconcentrations. However,since dis- theStreamsSampled genTemperaturen)(°C) (cid:1)23.7(1.3)23.9–18.827.550,765 (cid:1)23.9(1.0)24.0–16.926.516,164 flenstageheightandwasanephemeralstre lidcbanheataicsotcrkhngrweiseba.eFwtcedoaarrlascePbuftOfaoleakCvcteeiteonsxfndopasorso,Dnr0taIoCtndthduaienrniflmnidtugeDeidstnOhicaCenen.sotcehtoepnseccuarebironbjedtorcsant,teteihdoxenptowporoattsicesausnlsucteieuads-l Table2.WaterChemical,Hydrological,andFluvialCarbonValuesof ConductivityDissolvedOxy(cid:3)1μSamplingLocationpH(Scm)(%Saturatio (cid:1)(cid:1)(cid:1)(cid:1)Mean(SD)6.3(0.3)33(12)72(12)Median6.33275–––Range5.07.1976090n50,78450,78440,265 (cid:1)(cid:1)(cid:1)(cid:1)Mean(SD)5.1(0.1)7(2)75(5)Median5.1775–––Range4.56.25284093n15,01213,57915,241 aflStream2datarelatetotheperiodsthatthestreamwasowing.bDischargewascalculatedfromtherelationshipestablishedbetwecStream1,NewColpita,wasaperennial,andStream2,MainTrail, TT3a2s(wTstId2w1ma(n23htonhhh0..n.i92raa.f.3een961deeeefdssSa.e(cid:1)ms1a(cid:1)(cid:1)td[scrwmESmr)Deugdar,0ehfe12isinIrrerefl.fladaCCi3aeby..wgta23rmovue]1sdoLnnhew,xmmosr(cid:3)nsmaiee1nfie1([1nggddnap–Dicgca,asgsi5uoCCatsIrflCoChehf4eornntPeLLni]eshivLnnctmwgr(cid:3)(cid:3)apla(cid:3)eeesigy(htn11awel)gsnn1tu.;)esacetteCepiSprtmrhnwtrSlt,iatrho<vLgnetrov[tpiSaawre(cid:3)edDarisapeotlo0nnalv1eerlIeyaionuCu.m,deyndr0.rmrsSenw]a1iItnntAmidddiw<)1iorrntdeaurqwSS1nhayia02wurtuStnasirmrm.,trnta0etsasehhirom6enegt0aseeaaef.riag2m0a2amacdi4Ssnmnh1rc(cid:1)go.eCk(5;1io16grplneton11no(cid:1)Cta.hhdtst9hw.[utes=oh7eeDvl5re2ymnheienwasm8bO.(cid:1)dat2rlrc6boeuridCgwbewecame1twtm]ofndneiCar:.eaoos3gen–ggttentirnLsrergDcimeCaedn(cid:3)asr4ben(ena.TtnLig1.nosdinc2isfia,(cid:3)ieiDotteorCnnbca[rh1ya[nmuDngaPanlLsessrtaaanwOOo(cid:3)dbtiielnnnoht2nl1CCoeealyddinge))ystrt-f]]... VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3008 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 Table3. FluxTowerEstimatedWaterAvailabilityComparedtotheDischargeintheTwoStudyStreams 11Oct 11Nov 11Dec 12Jan 12Feb 12Mar 12Apr 12May 12Jun 12Jul 12Aug 12Sep Total Rainfall(mm) 143 104 199 345 243 154 148 33 107a 7 32 115 1630(cid:1)408 Evapotranspiration(mm) 79 82 70 69 57 87 63 72 50 72 89 84 873(cid:1)218 Availablewater 64 22 129 276 186 67 85 (cid:3)39 57 (cid:3)65 (cid:3)57 31 757(cid:1)189 Stream1totalflow(mm) 63 66 224 359 348 456 260 252 219 230 135 93 2705(cid:1)703 Stream1baseflow(mm) 34 33 67 143 61 190 197 139 122 130 109 83 1306(cid:1)340 Stream1eventflow(mm) 28 33 158 216 287 266 64 113 97 100 27 10 1399(cid:1)364 Stream2flow(mm) 0 0 35 146 213 159 104 21 NA 0 0 0 678(cid:1)88 aGapfilledasdetailedabove. December2011thatactivatedtheflowofStream2afterthedryperiod.TheStream1maximumconcentra- tion(54mgCL(cid:3)1)wasmorethantwiceashighasthenexthighestconcentration(22.2mgCL(cid:3)1).Themed- ian[POC]inStream1was3.5(cid:1)6.8mgCL(cid:3)1andinStream22.3(cid:1)1.6mgCL(cid:3)1.Themaximum[DOC]inboth streamswasalsoobservedduringthoseheavyrainsattheendofthedryseason2011,butintraeventmax- imum[DOC]waslaterthanthe[POC].InStreams1and2maximum[DOC]wereobservedinsamplescol- lected14hand21hafterthemaximum[POC].AquaticCO effluxrangedwidely:0.3–79.6μmolm(cid:3)2s(cid:3)1in 2 Stream1and0.4–12μmolm(cid:3)2s(cid:3)1inStream2.Effluxfromthefloodedforestwas2.4(cid:1)0.93μmolm(cid:3)2s(cid:3)1. OurPOCsamplingexcludedthecoarsefractionofcarbonlostfromthesystemwhichwouldincludeleaves, twigs and branches transported by the streams. However, such coarse particulate organic carbon (CPOC; collectedusingasamplerwitha2mmmeshsizenet)hasbeenfoundtobe1.5kgCha(cid:3)1yr1elsewherein the Amazon Basin [Selva et al., 2007] which is only 1–1.6% of our estimated POC export. Further, in a cool-temperate deciduous system where a larger fraction of trees shed their leaves annually, the [CPOC] (total of size classes 1–10mm and >10mm) was 0.34mgL(cid:3)1 [Shibata et al., 2001] which is equivalent to 10–15%ofourmedian[POC].Hence,itappearsthatthecoarseparticulatefractionisaminorcomponent ofthecarbonexport,anditsomissionwillnotleadtolargeunderestimationofexportinthisstudy. 3.4. ModelstoPredictCConcentrations Thecccvaluesofthebestfitmodelsforthetwostreamsrangedfrom0.65to0.91,withthepoorestvaluefound inStream2(Table4andFigureS6).InStream1,conductivitywasincludedasanexplanatoryvariableinallthe models.Conductivityreflectsthecontributionsofsurfacerunoffandgroundwater,withdryseasoncharacter- izedbyhigherconductivityduetoproportionallygreatergroundwatercontribution.[DOC]peaksduringevent flowconditionsandstageheightwasincludedasexplanatoryvariableinthemodelforStream1.Nosignificant explanatoryvariableswerefoundtopredict[DOC]inStream2.[POC]wasnotsuccessfullypredictedbyany modelineitherstream.Whereamodelcouldnotbederived,themedianconcentrationswereusedtoestimate exportforthestudyyear.Stream2hadalowpH,andhence,mostoftheDICpoolwasinformoffreeCO .The 2 [DIC]modelforthisstreamincludeddissolvedoxygensaturationwhichislikelytoberelatedtorespiration levelsinthestream.DissolvedoxygensaturationwasincludedintheCO effluxmodelsforbothstreams. 2 Theeffluxratewasstronglydependentonthewaterflowvelocity,withtheslopeoftherelationshipreflecting thedifferencesinpCO (FigureS7).Themodel(FiguresS8andS9)showstheprojectedCO effluxfromtheposi- 2 2 tionthatthevelocitysensorwasemplaced.ThefullCO effluxmodel(FigureS8a)producedveryhighCO 2 2 effluxestimateswhenthemodelparameters(oxygensaturationorconductivity)wereoutsidetherangeunder whichthemodelwasconstructed.Duringtheseconditionstheeffluxestimateswereimprovedusingtheflow- basedmodel(FigureS8b);however,somepeakeffluxratesabovethemeasuredmaximum79.6μmolm(cid:3)2s(cid:3)1 werestillobserved,butthesehadlimitedinfluenceontheannualfluxescalculated. 3.5. FluvialCarbonTimeSeriesandExportBudgets Fortheeddycovariancemeasurementperiodfairlycontinuoustimeseriesofthecarbonspeciescouldbe producedforStream1(Figure2).ThenumberofmissingvaluesinStream1eachmonth,rangedfrom0to 496datapoints(equalto0–5daysofmeasurements),exceptforOctober2011where15daysofdatawere missing.Hence,theexportvalueforOctoberhasagreateruncertainty.Duetodatagapsinthecontinuously loggedexplanatoryvariables,onlyamuchshortercontinuousDICtimeseries(FigureS10)couldbederived fortheephemeralStream2,andtherefore,themedian[DIC](Table2)wasusedtocalculatetheannualexport VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3009 Journal of Geophysical Research: Biogeosciences 10.1002/2016JG003464 Table4. StreamModelsUsedtoDerivethe15minResolutionCarbonTimeSeriesandtheConcordanceCorrelation CoefficientfortheAgreementBetweenFittedandMeasuredValuesandtheNumber(n)ofCarbonSamplesUsedto BuildtheModel n ccc PredictiveEquation DIC Stream1 172 0.91 0.74+0.13*conductivity Stream2 104 0.65 (cid:3)4.65+1.22*pH+0.014*stageheight(cid:3)0.012*dissolvedoxygen DOC Stream1 82 0.78 23.48(cid:3)2.95*pH(cid:3)0.066*conductivity+0.0083*stageheight Stream2 52 - Nosignificantexplanatoryvariables;median6.58mgCL(cid:3)1(SD1.3) POC Stream1 82 - Nosatisfactorymodelfound;median3.52mgCL(cid:3)1(SD4.02a) Stream2 53 - Nosignificantexplanatoryvariables;median2.25mgCL(cid:3)1(SD1.6) CO2Efflux Stream1 46 0.87 Exp(9.47+3.30*flowvelocity(cid:3)0.04*conductivity(cid:3)0.08*dissolvedoxygen) b 46 0.84 Exp(1.92+3.00*flowvelocity) Stream2 41 0.84 Exp(3.96+4.14*flowvelocity(cid:3)0.05*dissolvedoxygen) CoanCdaurcbtoivnitysp(μecSiecsm(cid:3)un1)it,sp:HD,ICd,issDoOlvCe,dPoOxCyg(emng(%CLs(cid:3)a1tu),raatniodn)C,Oan2desftflaugxe(hμemigohltm((cid:3)cm2)s.(cid:3)T1h).eEhxipglhaensattovraylueva5ri4a.b1lemgunCit/sL: wasexcludedfromthis. bThemodelbasedonflowvelocitywasusedwhenoxygendataweremissingduetoprobemalfunction(17Julyto4 September2012)andwhenwaterchemistryvariableswerebelowtherangeobservedduringeffluxmeasurements (oxygensaturation<64%;conductivity<19μScm(cid:3)1). budget.Asnopredictiverelationshipwasobservedfor[DOC]and[POC],theseexportbudgets(FigureS11) werealsocalculatedusingthemedianconcentrationsanddischargedata(Table2). Thewetseasonwasfarmoreimportanttomassofcarbonexported(inrelationtothecatchmentlandarea) byStream1(Figure3),with71%oftheannualCexporttakingplaceDecembertoMay.Theeffectofwet seasonwasmostpronouncedforDOCwith61%ofannualexporttakingplaceinJanuarytoMarch.[DIC]was higherduringthedryseason(June–November),andhence,differencesinC exportbetweendryand wet seasonsarelessmarked,withtheDecember–Mayperiodconstituting56%ofthetotalDICexport.Theannual totalexportof331(cid:1)48.7kgCha(cid:3)1yr(cid:3)1inStream1(Table1)consistedof34%(113(cid:1)12kgCha(cid:3)1yr(cid:3)1)DIC, 37%(122(cid:1)30kgCha(cid:3)1yr(cid:3)1)DOC,and29%(96(cid:1)37kgCha(cid:3)1yr(cid:3)1)POC.Naturally,intheseasonalStream 2alltheCexporttookplaceduringthewetseason(FigureS11).Theamountexported,66(cid:1)3.5kgCha(cid:3)1, wasalsomuchlessthanforStream1(234(cid:1)45kgCha(cid:3)1)forthecorrespondingDecember–Mayperiod. CorrectionsfortheinputofCfromrainandexportofagedcarbonarerequiredtoavoiderroneouslyincorpor- ating ecosystem C loss from sources that are external or not derived from recently fixed C. The volume- weighted carbon concentrations in rain water were found to be 0.16(cid:1)0.06mgCL(cid:3)1 of DIC (n=5), 2.3(cid:1)0.23mgCL(cid:3)1ofDOC(n=7),and1.3(cid:1)0.22mgCL(cid:3)1ofPOC(n=7).Duringtheyearofcomparisonthe rainfallwas1630mmwhichwouldyieldtotaldepositionofapproximately0.06(cid:1)0.005metrictonCha(cid:3)1yr(cid:3)1. 3.6. FluvialCarbonExportinRelationtoTerrestrialCO Flux 2 Incomparisontotheindicatedsmallnetsourceof0.7MgCha(cid:3)1yr(cid:3)1(EC)orsink0.5MgCha(cid:3)1yr(cid:3)1(amea- surementfromanearbybiomassplot[Malhietal.,2014])thefluvialexportisequivalentto43–60%ofthese smallnetCfluxes.TheCO effluxedfromdrainagewasaminorcomponent,approximately0.3–0.7%,ofthe 2 plantandsoilrespiration(Figure4b).ThesiteClossindrainage0.3MgCha(cid:3)1yr(cid:3)1wassmallcomparedtothe uncertainty of the terrestrial C balance: 3% in relation to EC measurement and 7% to biomass plot study results.Thefractionderivedfromfossilcarbonatesources(estimatedfromVihermaaetal.[2014])is likely tobelessthan1%ofthefluvialexportbutcouldbeamaximumof7%(Figure4a).Rainfallcontributionswere equaltoapproximately20%ofthefluvialexport. 4. Discussion 4.1. CO Fluxes:ComparisonsofThisSiteWithOtherSitesintheAmazonBasin 2 Thepresenteddycovarianceresultsuggeststhattheforestisneitherastrongsourcenorastrongsinkbutis closetobeinginequilibriumshowingacarbongainof+0.7Mgha(cid:3)1yr(cid:3)1whichcorrespondsto~1%ofthe VIHERMAAETAL. COMPLETERAINFORESTSITECBALANCE 3010