Biogeosciences 17 515ndash527 2020httpsdoiorg105194bg-17-515-2020copy Author(s) 2020 This work is distributed underthe Creative Commons Attribution 40 License

The carbon footprint of a Malaysian tropical reservoir measuredversus modelled estimates highlight the underestimated key role ofdownstream processesCynthia Soued and Yves T PrairieGroupe de Recherche Interuniversitaire en Limnologie et en Environnement Aquatique (GRIL)Deacutepartement des Sciences Biologiques Universiteacute du Queacutebec agrave Montreacuteal Montreacuteal H2X 3X8 Canada

Correspondence Cynthia Soued (cynthiasouedgmailcom)

Received 24 September 2019 ndash Discussion started 14 October 2019Revised 9 December 2019 ndash Accepted 7 January 2020 ndash Published 31 January 2020

Abstract Reservoirs are important sources of greenhousegases (GHGs) to the atmosphere and their number is rapidlyincreasing especially in tropical regions Accurately predict-ing their current and future emissions is essential but hin-dered by fragmented data on the subject which often fail toinclude all emission pathways (surface diffusion ebullitiondegassing and downstream emissions) and the high spatialand temporal flux variability Here we conducted a com-prehensive sampling of Batang Ai reservoir (Malaysia) andcompared field-based versus modelled estimates of its annualcarbon footprint for each emission pathway Carbon dioxide(CO2) and methane (CH4) surface diffusion were higher inupstream reaches Reducing spatial and temporal samplingresolution resulted in up to a 64 and 33 change in theflux estimate respectively Most GHGs present in dischargedwater were degassed at the turbines and the remainder weregradually emitted along the outflow river leaving time forCH4 to be partly oxidized to CO2 Overall the reservoiremitted 2475 gCO2 eqmminus2 yrminus1 with 89 occurring down-stream of the dam mostly in the form of CH4 These emis-sions largely underestimated by predictions are mitigatedby CH4 oxidation upstream and downstream of the dam butcould have been drastically reduced by slightly raising thewater intake elevation depth CO2 surface diffusion and CH4ebullition were lower than predicted whereas modelled CH4surface diffusion was accurate Investigating latter discrepan-cies we conclude that exploring morphometry soil type andstratification patterns as predictors can improve modelling ofreservoir GHG emissions at local and global scales

1 Introduction

Reservoirs provide a variety of services to humans (wa-ter supply navigation flood control hydropower) andcover an estimated area exceeding 03 millionkm2 globally(Lehner et al 2011) This area is increasing with an ex-pected rapid growth of the hydroelectric sector in the nexttwo decades (International Hydropower Association (IHA)2015) mainly in tropical and subtropical regions (Zarfl et al2015) The flooding of terrestrial landscapes can transformthem into significant greenhouse gas (GHG) sources to theatmosphere (Prairie et al 2018 Rudd et al 1993 Teodoru etal 2012) While part of reservoir GHG emissions would oc-cur naturally (not legitimately attributable to damming) theremainder results from newly created environments favour-ing carbon (C) mineralization particularly methane (CH4)production (flooded organic-rich anoxic soils) (Prairie et al2018) Field studies have revealed a wide range in measuredfluxes with spatial and temporal variability sometime span-ning several orders of magnitude within a single reservoir(Paranaiacuteba et al 2018 Sherman and Ford 2011) Moreoverreservoirs can emit GHG through several pathways throughdiffusion of gas at the airndashwater interface (surface diffu-sion) through the release of gas bubbles formed in the sed-iments (ebullition) for some reservoirs (mostly hydroelec-tric) through gas release following pressure drop upon waterdischarge (degassing) and through evasion of the remainingexcess gas in the outflow river (downstream emissions) Therelative contribution of these flux pathways to total emissionsis extremely variable While surface diffusion is the most fre-quently sampled it is often not the main emission pathway

Published by Copernicus Publications on behalf of the European Geosciences Union

516 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

(Demarty and Bastien 2011) Indeed measured ebullitiondegassing and downstream emissions range from negligi-ble to several orders of magnitude higher than surface dif-fusion in different reservoirs (Bastien and Demarty 2013DelSontro et al 2010 Galy-Lacaux et al 1997 Keller andStallard 1994 Kemenes et al 2007 Teodoru et al 2012Venkiteswaran et al 2013) making it a challenge to modeltotal reservoirs GHG emissions

Literature syntheses over the past 20 years have yieldedhighly variable global estimates of reservoirs GHG footprintranging from 05 to 23 PgCO2 eqyrminus1 (Barros et al 2011Bastviken et al 2011 Deemer et al 2016 St Louis et al2000) These estimates are based on global extrapolations ofaverages of sampled systems representing an uneven spa-tial distribution biased toward North America and Europeas well as an uneven mixture of emission pathways Re-cent studies have highlighted the lack of spatial and tempo-ral resolution as well as the frequent absence of some fluxpathways (especially degassing downstream and N2O emis-sions) in most reservoir GHG assessments (Beaulieu et al2016 Deemer et al 2016) More recently studies have fo-cused on identifying drivers of reservoir GHG flux variabil-ity Using global empirical data Barros et al (2011) pro-posed the first quantitative models for reservoir carbon diox-ide (CO2) and CH4 surface diffusion as a negative functionof reservoir age latitude and mean depth (for CO2 only) anda positive function of dissolved organic carbon (DOC) in-puts (Barros et al 2011) An online tool (G-res) for predict-ing reservoir CO2 and CH4 emissions was later developed onthe basis of a similar empirical modelling approach of mea-sured reservoir fluxes with globally available environmentaldata (UNESCOIHA 2017) Modelling frameworks to pre-dict GHG emissions from existing and future reservoirs areessential tools for reservoir management However their ac-curacy is directly related to available information and inher-ently affected by gaps and biases of the published literatureFor example while the G-res model predicts reservoir CO2and CH4 surface diffusion as well as CH4 ebullition and de-gassing on the basis of temperature age percent of littoralzone and soil organic C it does not consider N2O emissionsCO2 degassing and downstream emissions due to scarcity ofdata Overall the paucity of comprehensive empirical studieslimits our knowledge of reservoir GHG dynamics at a localscale introducing uncertainties in large-scale estimates andhindering model development

The research reported here focuses on building a compre-hensive assessment of GHG fluxes of Batang Ai a tropicalreservoir in Southeast Asia (Malaysia) over four samplingcampaigns spanning 2 years with an extensive spatial cover-age The main objective of this study is to provide a com-prehensive account of CO2 CH4 and N2O fluxes from sur-face diffusion ebullition degassing and downstream emis-sions (accounting for riverine CH4 oxidation) to better under-stand what shapes their relative contributions and their poten-tial mitigation The second objective is to compare our mea-

sured values with modelled estimates from each pathway andgas species to locate where the largest discrepancies are andthereby identify research avenues for improving the currentmodelling framework

2 Materials and methods

21 Study site and sampling campaigns

Batang Ai is a hydroelectric reservoir located on the islandof Borneo in the Sarawak province of Malaysia (latitude116 and longitude 1119) The regional climate is tropicalequatorial with a relatively constant temperature throughoutthe year on average 23 C in the morning to 32 C duringthe day Annual rainfall varies between 3300 and 4600 mmwith two monsoon seasons November to February (north-east monsoon) and June to October (southwest) (SarawakGovernment 2019) Batang Ai reservoir was impounded in1985 with no prior clearing of the vegetation and has a damwall of 85 m in height a mean depth of 34 m and a total areaof 684 km2 The reservoir catchment consists of 1149 km2

of mostly forested land where human activities are limitedto a few traditional habitations and associated croplands aswell as localized aquaculture sites within the reservoir mainbasin The reservoir has two major inflows the Batang Aiand Engkari rivers which flow into two reservoir branchesmerging upstream of the main reservoir basin (Fig 1) Foursampling campaigns were conducted (1) 14 November to5 December 2016 (NovemberndashDecember 2016) (2) 19 Aprilto 3 May 2017 (AprilndashMay 2017) (3) 28 February to13 March 2018 (FebruaryndashMarch 2018) and (4) 12 to 29 Au-gust 2018 (August 2018)

22 Water chemistry

Samples for DOC total phosphorus (TP) total nitrogen(TN) and chlorophyll a (Chl a) analyses were collected fromthe water surface (lt 05 m) at all surface diffusion samplingsites shown in Fig 1 and during each campaign For TPand TN we collected non-filtered water in acid-washed glassvials stored at 4 C until analysis TP was measured by spec-trophotometry using the standard molybdenum blue methodafter persulfate digestion at 121 C for 20 min and a calibra-tion with standard solutions from 10 to 100 microg Lminus1 with a5 precision (Wetzel and Likens 2000) TN analyses wereperformed by alkaline persulfate digestion to NO3 subse-quently measured on a flow Alpkem analyser (OI AnalyticalFlow Solution 3100) calibrated with standard solutions from005 to 2 mgLminus1 with a 5 precision (Patton and Kryskalla2003) Water filtered at 045 microm was used for DOC analy-sis with a total organic carbon analyser 1010-OI followingsodium persulfate digestion and calibrated with standard so-lutions from 1 to 20 mgLminus1 with a 5 precision (detectionlimit of 01 mgLminus1) Chl a was analysed through spectropho-tometry following filtration on Whatman (GFF) filters and

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 517

Figure 1 Map of Batang Ai showing the location of sampled sites and reservoir sections Represents the reservoir inflow sites

extraction by hot 90 ethanol solution (Sartory and Grobbe-laar 1984)

23 Surface diffusion

Surface diffusion is the flux of gas between the water sur-face and the air driven by a gradient in gas partial pressureSurface diffusion of CO2 and CH4 to the atmosphere wasmeasured at 36 sites in the reservoir and 3 sites in the inflowrivers (Fig 1) and sampling of the same sites was repeatedin each campaign (with a few exceptions) Fluxes were mea-sured using a static airtight floating chamber connected ina closed loop to an ultraportable gas analyser (UGGA fromLGR) Surface diffusion rates (Fgas) were derived from thelinear change in CO2 and CH4 partial pressures (continu-ously monitored at 1 Hz for a minimum of 5 min) throughtime inside the chamber using the following Eq (1)

Fgas =sV

mA (1)

where s is the gas accumulation rate in the chamber V =25 L the chamber volume A= 0184 m2 the chamber sur-face area andm the gas molar volume at current atmosphericpressure

N2O surface diffusion was estimated at seven of the sam-pled sites (Fig 1) using the following Eq (2) (Lide 2005)

Fgas = kgas(CgasminusCeq

) (2)

where kgas is the gas exchange coefficient Cgas is the gasconcentration in the water and Ceq is the theoretical gas

concentration at equilibrium given measured water temper-ature atmospheric pressure and ambient gas concentrationCN2O was measured using the headspace technique with a112 L sealed glass serum bottle containing surface water anda 012 L headspace of ambient air After shaking the bottlefor 2 min to achieve airndashwater equilibrium the headspace gaswas extracted from the bottle with an airtight syringe andinjected in the previously evacuated 9 mL glass vial cappedwith an airtight butyl stopper and aluminium seal Three ana-lytical replicates and a local sample of ambient air were takenat each site and analysed by gas chromatography using a Shi-madzu GC-2040 with a Poropaq Q column to separate gasesand an electron capture detector (ECD) calibrated with 031 and 3 ppm of N2O certified standard gas After analysis theoriginal N2O concentration of the water was back-calculatedbased on the water temperature before and after shaking (forgas solubility) the ambient atmospheric pressure the ratioof water to air in the sampling bottle and the headspace N2Oconcentration before shaking kN2O was derived from mea-sured kCH4 values obtained by rearranging Eq (2) for CH4with known values of Fgas Cgas and Ceq The kCH4 to kN2Otransformation was done using the following Eq (3) (Coleand Caraco 1998 Ledwell 1984)

kN2O =

(ScN2O

ScCH4

)minus067

kCH4 (3)

where Sc is the gas Schmidt number (Wanninkhof 1992)CH4 and CO2 concentrations in the water were mea-

sured using the headspace technique Surface water was col-lected in a 60 mL gas-tight plastic syringe in which a 30 mL

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

518 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

Figure 2 Boxplots of measured CH4 (on a log axis) and CO2fluxes grouped according to spatial position Boxes are bounded bythe 25th and 75th percentile and show medians (solid lines) andwhiskers show 10th and 90th percentiles Grey circles show singledata points

headspace was created (using either ambient air or carbon-free air) The syringe was shaken for 2 min to achieve gasequilibrium between air and water The gas phase was theninjected in a 12 mL airtight pre-evacuated vial and subse-quently analysed through manual injection on a ShimadzuGC-8A gas chromatograph with a flame ionization detectorfollowing a calibration curve with certified gas standards (0ndash10 000 ppm for CO2 and 0ndash50 000 ppm for CH4) The sam-ples were also analysed for isotopic δ13CO2 and δ13CH4 sig-natures by manually injecting 18 mL of gas in a cavity ring-down spectrometer (CRDS) equipped with a small sampleisotopic module (SSIM A0314 Picarro G2201-i analyser)set in a non-continuous mode with a three-point calibrationcurve based on certified gas standards (minus40 permil minus39 permil and253 permil for δ13CO2 and minus665 permil minus383 permil and minus239 permilfor δ13CH4)

Figure 3 Concentrations (black symbols and solid line) and δ13C(grey symbols and dotted lines) of CO2 and CH4 from right up-stream of the dam (grey band) to 19 km downstream in the out-flow river Circles squares diamonds and triangles represent val-ues from NovemberndashDecember 2016 AprilndashMay 2017 FebruaryndashMar 2018 and August 2018 respectively

24 Ebullition

Ebullition is the process through which gas bubbles formedin the sediments rise through the water column and are re-leased to the atmosphere Sediment gas ebullition was mea-sured at four sites in the reservoir and two sites in the inflows(Fig 1) by deploying 0785 m2 underwater inverted funneltraps at 2 to 3 m deep for approximately 20 d in the reservoirand 1 h in the inflows The top part of a closed plastic sy-ringe was fixed to the narrow end of the funnel trap wherethe emerging bubbles accumulated Upon recovery bubblegas volume was measured collected from the syringe and in-jected in 12 mL pre-evacuated airtight vials for CO2 and CH4concentration analyses (using the aforementioned method)Ebullition rate was calculated assuming the original bubblecomposition was similar to bubbles collected almost right af-ter ascent in the inflows sites which was 100 CH4 Hencewe considered CO2 and N2O ebullition to be null

In order to estimate the potential for sediment accumula-tion fuelling ebullition in the littoral zone we calculated themud energy boundary depth (EBD in metres below whichfine-grained sediment accumulation occurs) using the reser-voir surface area (E in km2) as the exposure parameter in thefollowing Eq (4) (Rowan et al 1992)

EBD= 2685E0305 (4)

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 519

Table 1 CO2 and CH4 dynamics downstream of the dam gas export rate from upstream to downstream of the dam degassing result ofCH4 oxidation (CO2 production and CH4 consumption) downstream emissions and total emissions to the atmosphere below the damUncertainties based on variation coefficients are reported in parentheses Units are in mmolmminus2 dminus1 of reservoir surface area

GHG downstream of the dam (mmolmminus2 dminus1)

Exported Degassed Gainloss by oxidation Downstream emissions Total emissions

CO2

NovndashDec 2016 4062 (plusmn227) 1526 (plusmn085) 090 (plusmn013) 1267 (plusmn071) 2793 (plusmn156)AprndashMay 2017 3780 (plusmn211) 1491 (plusmn083) 059 (plusmn008) 983 (plusmn055) 2470 (plusmn138)FebndashMar 2018 3798 (plusmn212) 958 (plusmn054) 180 (plusmn026) 970 (plusmn054) 1930 (plusmn108)Aug 2018 3807 (plusmn213) 2167 (plusmn121) 038 (plusmn005) 831 (plusmn046) 3000 (plusmn168)

CH4

NovndashDec 2016 1484 (plusmn210) 1156 (plusmn164) 090 (plusmn013) 219 (plusmn031) 1376 (plusmn195)AprndashMay 2017 732 (plusmn104) 400 (plusmn057) 059 (plusmn008) 190 (plusmn027) 590 (plusmn084)FebndashMar 2018 1247 (plusmn177) 492 (plusmn070) 180 (plusmn026) 399 (plusmn057) 891 (plusmn126)Aug 2018 1071 (plusmn152) 954 (plusmn135) 038 (plusmn005) 051 (plusmn007) 1005 (plusmn142)

25 Degassing downstream emissions and CH4oxidation

Degassing of CO2 and CH4 right after water discharge(Fdeg) and downstream emissions of the remaining reservoir-derived GHGs in the outflow river (Fdwn) were calculatedusing the following Eqs (5) and (6)

Fdeg =Q(CupminusC0

) (5)

Fdwn =Q(C0minusC19+Cox) (6)

where Q is the water discharge and Cup C0 and C19 themeasured gas concentrations upstream of the dam at the wa-ter withdrawal depth at the powerhouse right after water dis-charge and in the outflow 19 km downstream of the dam re-spectively Cox is the net change in gas concentration due tooxidation (loss for CH4 and gain for CO2) For downstreamemissions we considered that after a river stretch of 19 kmall excess gas originating from the reservoir was evaded andgas concentration was representative of the outflow riverbaseline This assumption potentially underestimates actualdownstream emissions (in the case of remaining excess gasafter 19 km) However given the observed exponential de-crease in gas concentration along the outflow (Fig 3) emis-sions after 19 km are expected to be small compared to thosein the 0 to 19 km river stretch consistent with observations inother reservoirs (Gueacuterin et al 2006 Kemenes et al 2007)

Gas concentrations upstream and downstream of the damwere obtained by measuring in each campaign CO2 andCH4 concentrations in a vertical profile right upstream ofthe dam at a 1 to 3 m interval from 0 to 32 m and at fourlocations in the outflow at 0 (power house) 06 27 and19 km downstream of the dam (Fig 1) Sampling was doneusing a multi-parameter probe equipped with depth oxygenand temperature sensors (Yellow Spring Instruments YSImodel 600XLM-M) attached to a 12 V submersible Tornado

pump (Proactive Environmental Products) for water collec-tion Gas concentration and δ13C were measured as describedin Sect 23 Water withdrawal depth ranged from 20 to 23 mand was estimated based on known values of elevations ofwater intake and water level compared to sea level Gas con-centration in the water exiting the reservoir was defined asthe average measured gas concentrations in the plusmn1 m rangeof the withdrawal depth

Estimates of downstream CH4 oxidation were obtainedfor each sampling campaign by calculating the fraction ofCH4 oxidized (Fox) using the following Eq (7)

Fox =

minus(ln

(δ13CH4resid+1000)minusln(δ13CH4source+1000

))middot

(1minus [CH4]resid

[CH4]source

)(1minus 1

α

)ln

([CH4]resid

[CH4]source

) (7)

Equation (7) is based on a non-steady-state isotopic modeldeveloped considering evasion (emission to the atmosphere)and oxidation as the two loss processes for CH4 in the out-flow river assuming negligible isotopic fractionation for eva-sion (Knox et al 1992) and a fractionation of α = 102 foroxidation (Coleman et al 1981) (see derivation in Supple-ment) [CH4]source [CH4]resid δ13CH4source and δ13CH4residare the concentrations of CH4 and their corresponding iso-topic signatures at the beginning of the outflow (km 0)and 19 km downstream representing the source and residualpools of CH4 respectively The amount of CH4 oxidized toCO2 along the 19 km of river stretch for each sampling cam-paign was calculated as the product of Fox and [CH4]sourceThe resulting loss of CH4 and gain of CO2 in the outflowwere accounted for in downstream emissions (Cox in Eq 6)Note that downstream N2O emissions were considered nullsince N2O concentrations measured in the deep reservoirlayer were lower than concentrations in the outflow

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

520 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

26 Ecosystem-scale C footprint

Batang Ai annual C footprint was calculated as the sumof surface diffusion ebullition degassing and downstreamemissions of CO2 CH4 and N2O considering a greenhousewarming potential of 1 34 and 298 respectively over a 100-year lifetime period (Myhre et al 2013) For each flux path-way annual flux was estimated as the average of the sam-pling campaigns Ecosystem-scale estimate of surface diffu-sion was calculated for each campaign as the average of mea-sured flux rates applied to the reservoir area for N2O and forCO2 and CH4 it was obtained by spatial interpolation of mea-sured fluxes over the reservoir area based on inverse distanceweighting with a power of two (a power of one yields similaraverages CV lt 11 ) using package gstat version 11ndash6 inthe R version 341 software (Pebesma 2004 R Core Team2017) Ebullition at the reservoir scale was calculated as theaverage of measured reservoir ebullition rates applied to thelittoral area (lt 3 m deep)

The estimated GHG emissions of Batang Ai based on mea-sured data was compared to values derived from the G-resmodel (UNESCOIHA 2017) and the model presented inBarros et al (2011) Both models predict surface CO2 andCH4 diffusion as a function of age and account for the effectof temperature using different proxies the G-res uses effec-tive temperature while the Barros et al model uses latitude(an indirect proxy that integrates other spatial differences)In terms of CO2 surface diffusion the G-res uses reservoirarea soil C content and TP to quantify the effect of C in-puts fuelling CO2 production while the Barros et al modeluses DOC inputs directly (based on in situ DOC concentra-tion) For CH4 surface diffusion both models account formorphometry using the fraction of littoral area (G-res) or themean depth (Barros et al model) Overall both models pre-dict surface diffusion based on the same conceptual frame-work but use different proxies CH4 ebullition and degassingare modelled only by the G-res being the sole model avail-able to this date Details on model equations and input vari-ables are presented in the Supplement (Tables S2 and S3)

3 Results and discussion

31 Water chemistry

The reservoir is stratified throughout the year with a ther-mocline at a depth around 13 m and mostly anoxic condi-tions in the hypolimnion of the main basin (Fig S1 in theSupplement) The system is oligotrophic with very low con-centrations of DOC TP TN and Chl a averaging 09 (SDplusmn02) mgLminus1 59 (SDplusmn24) microgLminus1 011 (SDplusmn004) mgLminus1and 13 (SDplusmn 07) microgLminus1 respectively (Table S1) and highwater transparency (Secchi depth gt 5 m) In the reservoirinflows concentrations of measured chemical species areslightly higher but still in the oligotrophic range (Table S1)

however the transparency is much lower due to turbidity(Secchi lt 05 m) The oligotrophic status of the reservoirlikely results from nutrient-poor soils (Wasli et al 2011)and a largely undisturbed forested catchment in the protectedBatang Ai National Park The reservoirrsquos low Chl a concen-trations are comparable to the neighbouring Bakun reservoir(Ling et al 2017) and its DOC concentrations are on the lowend of the wide range of measured values in nearby rivers(Martin et al 2018)

32 Surface diffusion

Measured CO2 diffusion in the reservoir averaged 77 (SDplusmn182) mmolmminus2 dminus1 (Table S1) which is on the low endcompared to other reservoirs (Deemer et al 2016) and evento natural lakes (Sobek et al 2005) but similar to CO2 fluxesmeasured in two reservoirs in Laos (Chanudet et al 2011)CO2 diffusion across all sites ranged from substantial up-take to high emissions (from minus308 to 5939 mmolmminus2 dminus1Table S1) reflecting a large spatial and temporal variabil-ity Spatially CO2 fluxes measured in the main basin andbranches had similar averages of 79 and 73 mmolmminus2 dminus1

respectively (overall SDplusmn182) contrasting with higher andmore variable values in the inflows with a mean of 1373(SDplusmn 1924) mmolmminus2 dminus1 (Fig 2) Within the reservoirCO2 fluxes varied (SDplusmn182 mmolmminus2 dminus1) but did not fol-low a consistent pattern and might reflect pre-flooding land-scape heterogeneity (Teodoru et al 2011) Temporally high-est average reservoir CO2 fluxes were measured in AprilndashMay 2017 when no CO2 uptake was observed contraryto other campaigns especially FebruaryndashMarch and Au-gust 2018 when CO2 uptake was common (Fig S2) andaverage Chl a concentrations were the highest This reflectsthe important role of metabolism (namely CO2 consumptionby primary production) in modulating surface CO2 fluxes inBatang Ai

All CH4 surface diffusion measurements were positive andranged from 003 to 1134 mmolmminus2 dminus1 (Table S1) Spa-tially CH4 fluxes were progressively higher moving furtherupstream (Figs 2 and S3) with decreasing water depth andincreasing connection to the littoral This gradient in mor-phometry induces an increasingly greater contact of the waterwith bottom and littoral sediments where CH4 is producedexplaining the spatial pattern of CH4 fluxes CH4 surface dif-fusion also varied temporally but to a lesser extent than CO2being on average highest in August 2018 in the reservoir andin NovemberndashDecember 2016 in the inflows

Reservoir N2O surface diffusion (measured with a limitedspatial resolution) averaged minus02 (SDplusmn 21) nmolmminus2 dminus1

(Table S1) The negative value indicates that the systemacts as a slight net sink of N2O absorbing an estimated21 gCO2 eqmminus2 yrminus1 (Table 2) Atmospheric N2O uptakehas previously been reported in aquatic systems and linkedto low oxygen and nitrogen content conducive to completedenitrification which consumes N2O (Soued et al 2016

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

516 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

(Demarty and Bastien 2011) Indeed measured ebullitiondegassing and downstream emissions range from negligi-ble to several orders of magnitude higher than surface dif-fusion in different reservoirs (Bastien and Demarty 2013DelSontro et al 2010 Galy-Lacaux et al 1997 Keller andStallard 1994 Kemenes et al 2007 Teodoru et al 2012Venkiteswaran et al 2013) making it a challenge to modeltotal reservoirs GHG emissions

Literature syntheses over the past 20 years have yieldedhighly variable global estimates of reservoirs GHG footprintranging from 05 to 23 PgCO2 eqyrminus1 (Barros et al 2011Bastviken et al 2011 Deemer et al 2016 St Louis et al2000) These estimates are based on global extrapolations ofaverages of sampled systems representing an uneven spa-tial distribution biased toward North America and Europeas well as an uneven mixture of emission pathways Re-cent studies have highlighted the lack of spatial and tempo-ral resolution as well as the frequent absence of some fluxpathways (especially degassing downstream and N2O emis-sions) in most reservoir GHG assessments (Beaulieu et al2016 Deemer et al 2016) More recently studies have fo-cused on identifying drivers of reservoir GHG flux variabil-ity Using global empirical data Barros et al (2011) pro-posed the first quantitative models for reservoir carbon diox-ide (CO2) and CH4 surface diffusion as a negative functionof reservoir age latitude and mean depth (for CO2 only) anda positive function of dissolved organic carbon (DOC) in-puts (Barros et al 2011) An online tool (G-res) for predict-ing reservoir CO2 and CH4 emissions was later developed onthe basis of a similar empirical modelling approach of mea-sured reservoir fluxes with globally available environmentaldata (UNESCOIHA 2017) Modelling frameworks to pre-dict GHG emissions from existing and future reservoirs areessential tools for reservoir management However their ac-curacy is directly related to available information and inher-ently affected by gaps and biases of the published literatureFor example while the G-res model predicts reservoir CO2and CH4 surface diffusion as well as CH4 ebullition and de-gassing on the basis of temperature age percent of littoralzone and soil organic C it does not consider N2O emissionsCO2 degassing and downstream emissions due to scarcity ofdata Overall the paucity of comprehensive empirical studieslimits our knowledge of reservoir GHG dynamics at a localscale introducing uncertainties in large-scale estimates andhindering model development

The research reported here focuses on building a compre-hensive assessment of GHG fluxes of Batang Ai a tropicalreservoir in Southeast Asia (Malaysia) over four samplingcampaigns spanning 2 years with an extensive spatial cover-age The main objective of this study is to provide a com-prehensive account of CO2 CH4 and N2O fluxes from sur-face diffusion ebullition degassing and downstream emis-sions (accounting for riverine CH4 oxidation) to better under-stand what shapes their relative contributions and their poten-tial mitigation The second objective is to compare our mea-

sured values with modelled estimates from each pathway andgas species to locate where the largest discrepancies are andthereby identify research avenues for improving the currentmodelling framework

2 Materials and methods

21 Study site and sampling campaigns

Batang Ai is a hydroelectric reservoir located on the islandof Borneo in the Sarawak province of Malaysia (latitude116 and longitude 1119) The regional climate is tropicalequatorial with a relatively constant temperature throughoutthe year on average 23 C in the morning to 32 C duringthe day Annual rainfall varies between 3300 and 4600 mmwith two monsoon seasons November to February (north-east monsoon) and June to October (southwest) (SarawakGovernment 2019) Batang Ai reservoir was impounded in1985 with no prior clearing of the vegetation and has a damwall of 85 m in height a mean depth of 34 m and a total areaof 684 km2 The reservoir catchment consists of 1149 km2

of mostly forested land where human activities are limitedto a few traditional habitations and associated croplands aswell as localized aquaculture sites within the reservoir mainbasin The reservoir has two major inflows the Batang Aiand Engkari rivers which flow into two reservoir branchesmerging upstream of the main reservoir basin (Fig 1) Foursampling campaigns were conducted (1) 14 November to5 December 2016 (NovemberndashDecember 2016) (2) 19 Aprilto 3 May 2017 (AprilndashMay 2017) (3) 28 February to13 March 2018 (FebruaryndashMarch 2018) and (4) 12 to 29 Au-gust 2018 (August 2018)

22 Water chemistry

Samples for DOC total phosphorus (TP) total nitrogen(TN) and chlorophyll a (Chl a) analyses were collected fromthe water surface (lt 05 m) at all surface diffusion samplingsites shown in Fig 1 and during each campaign For TPand TN we collected non-filtered water in acid-washed glassvials stored at 4 C until analysis TP was measured by spec-trophotometry using the standard molybdenum blue methodafter persulfate digestion at 121 C for 20 min and a calibra-tion with standard solutions from 10 to 100 microg Lminus1 with a5 precision (Wetzel and Likens 2000) TN analyses wereperformed by alkaline persulfate digestion to NO3 subse-quently measured on a flow Alpkem analyser (OI AnalyticalFlow Solution 3100) calibrated with standard solutions from005 to 2 mgLminus1 with a 5 precision (Patton and Kryskalla2003) Water filtered at 045 microm was used for DOC analy-sis with a total organic carbon analyser 1010-OI followingsodium persulfate digestion and calibrated with standard so-lutions from 1 to 20 mgLminus1 with a 5 precision (detectionlimit of 01 mgLminus1) Chl a was analysed through spectropho-tometry following filtration on Whatman (GFF) filters and

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 517

Figure 1 Map of Batang Ai showing the location of sampled sites and reservoir sections Represents the reservoir inflow sites

extraction by hot 90 ethanol solution (Sartory and Grobbe-laar 1984)

23 Surface diffusion

Surface diffusion is the flux of gas between the water sur-face and the air driven by a gradient in gas partial pressureSurface diffusion of CO2 and CH4 to the atmosphere wasmeasured at 36 sites in the reservoir and 3 sites in the inflowrivers (Fig 1) and sampling of the same sites was repeatedin each campaign (with a few exceptions) Fluxes were mea-sured using a static airtight floating chamber connected ina closed loop to an ultraportable gas analyser (UGGA fromLGR) Surface diffusion rates (Fgas) were derived from thelinear change in CO2 and CH4 partial pressures (continu-ously monitored at 1 Hz for a minimum of 5 min) throughtime inside the chamber using the following Eq (1)

Fgas =sV

mA (1)

where s is the gas accumulation rate in the chamber V =25 L the chamber volume A= 0184 m2 the chamber sur-face area andm the gas molar volume at current atmosphericpressure

N2O surface diffusion was estimated at seven of the sam-pled sites (Fig 1) using the following Eq (2) (Lide 2005)

Fgas = kgas(CgasminusCeq

) (2)

where kgas is the gas exchange coefficient Cgas is the gasconcentration in the water and Ceq is the theoretical gas

concentration at equilibrium given measured water temper-ature atmospheric pressure and ambient gas concentrationCN2O was measured using the headspace technique with a112 L sealed glass serum bottle containing surface water anda 012 L headspace of ambient air After shaking the bottlefor 2 min to achieve airndashwater equilibrium the headspace gaswas extracted from the bottle with an airtight syringe andinjected in the previously evacuated 9 mL glass vial cappedwith an airtight butyl stopper and aluminium seal Three ana-lytical replicates and a local sample of ambient air were takenat each site and analysed by gas chromatography using a Shi-madzu GC-2040 with a Poropaq Q column to separate gasesand an electron capture detector (ECD) calibrated with 031 and 3 ppm of N2O certified standard gas After analysis theoriginal N2O concentration of the water was back-calculatedbased on the water temperature before and after shaking (forgas solubility) the ambient atmospheric pressure the ratioof water to air in the sampling bottle and the headspace N2Oconcentration before shaking kN2O was derived from mea-sured kCH4 values obtained by rearranging Eq (2) for CH4with known values of Fgas Cgas and Ceq The kCH4 to kN2Otransformation was done using the following Eq (3) (Coleand Caraco 1998 Ledwell 1984)

kN2O =

(ScN2O

ScCH4

)minus067

kCH4 (3)

where Sc is the gas Schmidt number (Wanninkhof 1992)CH4 and CO2 concentrations in the water were mea-

sured using the headspace technique Surface water was col-lected in a 60 mL gas-tight plastic syringe in which a 30 mL

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

518 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

Figure 2 Boxplots of measured CH4 (on a log axis) and CO2fluxes grouped according to spatial position Boxes are bounded bythe 25th and 75th percentile and show medians (solid lines) andwhiskers show 10th and 90th percentiles Grey circles show singledata points

headspace was created (using either ambient air or carbon-free air) The syringe was shaken for 2 min to achieve gasequilibrium between air and water The gas phase was theninjected in a 12 mL airtight pre-evacuated vial and subse-quently analysed through manual injection on a ShimadzuGC-8A gas chromatograph with a flame ionization detectorfollowing a calibration curve with certified gas standards (0ndash10 000 ppm for CO2 and 0ndash50 000 ppm for CH4) The sam-ples were also analysed for isotopic δ13CO2 and δ13CH4 sig-natures by manually injecting 18 mL of gas in a cavity ring-down spectrometer (CRDS) equipped with a small sampleisotopic module (SSIM A0314 Picarro G2201-i analyser)set in a non-continuous mode with a three-point calibrationcurve based on certified gas standards (minus40 permil minus39 permil and253 permil for δ13CO2 and minus665 permil minus383 permil and minus239 permilfor δ13CH4)

Figure 3 Concentrations (black symbols and solid line) and δ13C(grey symbols and dotted lines) of CO2 and CH4 from right up-stream of the dam (grey band) to 19 km downstream in the out-flow river Circles squares diamonds and triangles represent val-ues from NovemberndashDecember 2016 AprilndashMay 2017 FebruaryndashMar 2018 and August 2018 respectively

24 Ebullition

Ebullition is the process through which gas bubbles formedin the sediments rise through the water column and are re-leased to the atmosphere Sediment gas ebullition was mea-sured at four sites in the reservoir and two sites in the inflows(Fig 1) by deploying 0785 m2 underwater inverted funneltraps at 2 to 3 m deep for approximately 20 d in the reservoirand 1 h in the inflows The top part of a closed plastic sy-ringe was fixed to the narrow end of the funnel trap wherethe emerging bubbles accumulated Upon recovery bubblegas volume was measured collected from the syringe and in-jected in 12 mL pre-evacuated airtight vials for CO2 and CH4concentration analyses (using the aforementioned method)Ebullition rate was calculated assuming the original bubblecomposition was similar to bubbles collected almost right af-ter ascent in the inflows sites which was 100 CH4 Hencewe considered CO2 and N2O ebullition to be null

In order to estimate the potential for sediment accumula-tion fuelling ebullition in the littoral zone we calculated themud energy boundary depth (EBD in metres below whichfine-grained sediment accumulation occurs) using the reser-voir surface area (E in km2) as the exposure parameter in thefollowing Eq (4) (Rowan et al 1992)

EBD= 2685E0305 (4)

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 519

Table 1 CO2 and CH4 dynamics downstream of the dam gas export rate from upstream to downstream of the dam degassing result ofCH4 oxidation (CO2 production and CH4 consumption) downstream emissions and total emissions to the atmosphere below the damUncertainties based on variation coefficients are reported in parentheses Units are in mmolmminus2 dminus1 of reservoir surface area

GHG downstream of the dam (mmolmminus2 dminus1)

Exported Degassed Gainloss by oxidation Downstream emissions Total emissions

CO2

NovndashDec 2016 4062 (plusmn227) 1526 (plusmn085) 090 (plusmn013) 1267 (plusmn071) 2793 (plusmn156)AprndashMay 2017 3780 (plusmn211) 1491 (plusmn083) 059 (plusmn008) 983 (plusmn055) 2470 (plusmn138)FebndashMar 2018 3798 (plusmn212) 958 (plusmn054) 180 (plusmn026) 970 (plusmn054) 1930 (plusmn108)Aug 2018 3807 (plusmn213) 2167 (plusmn121) 038 (plusmn005) 831 (plusmn046) 3000 (plusmn168)

CH4

NovndashDec 2016 1484 (plusmn210) 1156 (plusmn164) 090 (plusmn013) 219 (plusmn031) 1376 (plusmn195)AprndashMay 2017 732 (plusmn104) 400 (plusmn057) 059 (plusmn008) 190 (plusmn027) 590 (plusmn084)FebndashMar 2018 1247 (plusmn177) 492 (plusmn070) 180 (plusmn026) 399 (plusmn057) 891 (plusmn126)Aug 2018 1071 (plusmn152) 954 (plusmn135) 038 (plusmn005) 051 (plusmn007) 1005 (plusmn142)

25 Degassing downstream emissions and CH4oxidation

Degassing of CO2 and CH4 right after water discharge(Fdeg) and downstream emissions of the remaining reservoir-derived GHGs in the outflow river (Fdwn) were calculatedusing the following Eqs (5) and (6)

Fdeg =Q(CupminusC0

) (5)

Fdwn =Q(C0minusC19+Cox) (6)

where Q is the water discharge and Cup C0 and C19 themeasured gas concentrations upstream of the dam at the wa-ter withdrawal depth at the powerhouse right after water dis-charge and in the outflow 19 km downstream of the dam re-spectively Cox is the net change in gas concentration due tooxidation (loss for CH4 and gain for CO2) For downstreamemissions we considered that after a river stretch of 19 kmall excess gas originating from the reservoir was evaded andgas concentration was representative of the outflow riverbaseline This assumption potentially underestimates actualdownstream emissions (in the case of remaining excess gasafter 19 km) However given the observed exponential de-crease in gas concentration along the outflow (Fig 3) emis-sions after 19 km are expected to be small compared to thosein the 0 to 19 km river stretch consistent with observations inother reservoirs (Gueacuterin et al 2006 Kemenes et al 2007)

Gas concentrations upstream and downstream of the damwere obtained by measuring in each campaign CO2 andCH4 concentrations in a vertical profile right upstream ofthe dam at a 1 to 3 m interval from 0 to 32 m and at fourlocations in the outflow at 0 (power house) 06 27 and19 km downstream of the dam (Fig 1) Sampling was doneusing a multi-parameter probe equipped with depth oxygenand temperature sensors (Yellow Spring Instruments YSImodel 600XLM-M) attached to a 12 V submersible Tornado

pump (Proactive Environmental Products) for water collec-tion Gas concentration and δ13C were measured as describedin Sect 23 Water withdrawal depth ranged from 20 to 23 mand was estimated based on known values of elevations ofwater intake and water level compared to sea level Gas con-centration in the water exiting the reservoir was defined asthe average measured gas concentrations in the plusmn1 m rangeof the withdrawal depth

Estimates of downstream CH4 oxidation were obtainedfor each sampling campaign by calculating the fraction ofCH4 oxidized (Fox) using the following Eq (7)

Fox =

minus(ln

(δ13CH4resid+1000)minusln(δ13CH4source+1000

))middot

(1minus [CH4]resid

[CH4]source

)(1minus 1

α

)ln

([CH4]resid

[CH4]source

) (7)

Equation (7) is based on a non-steady-state isotopic modeldeveloped considering evasion (emission to the atmosphere)and oxidation as the two loss processes for CH4 in the out-flow river assuming negligible isotopic fractionation for eva-sion (Knox et al 1992) and a fractionation of α = 102 foroxidation (Coleman et al 1981) (see derivation in Supple-ment) [CH4]source [CH4]resid δ13CH4source and δ13CH4residare the concentrations of CH4 and their corresponding iso-topic signatures at the beginning of the outflow (km 0)and 19 km downstream representing the source and residualpools of CH4 respectively The amount of CH4 oxidized toCO2 along the 19 km of river stretch for each sampling cam-paign was calculated as the product of Fox and [CH4]sourceThe resulting loss of CH4 and gain of CO2 in the outflowwere accounted for in downstream emissions (Cox in Eq 6)Note that downstream N2O emissions were considered nullsince N2O concentrations measured in the deep reservoirlayer were lower than concentrations in the outflow

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

520 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

26 Ecosystem-scale C footprint

Batang Ai annual C footprint was calculated as the sumof surface diffusion ebullition degassing and downstreamemissions of CO2 CH4 and N2O considering a greenhousewarming potential of 1 34 and 298 respectively over a 100-year lifetime period (Myhre et al 2013) For each flux path-way annual flux was estimated as the average of the sam-pling campaigns Ecosystem-scale estimate of surface diffu-sion was calculated for each campaign as the average of mea-sured flux rates applied to the reservoir area for N2O and forCO2 and CH4 it was obtained by spatial interpolation of mea-sured fluxes over the reservoir area based on inverse distanceweighting with a power of two (a power of one yields similaraverages CV lt 11 ) using package gstat version 11ndash6 inthe R version 341 software (Pebesma 2004 R Core Team2017) Ebullition at the reservoir scale was calculated as theaverage of measured reservoir ebullition rates applied to thelittoral area (lt 3 m deep)

The estimated GHG emissions of Batang Ai based on mea-sured data was compared to values derived from the G-resmodel (UNESCOIHA 2017) and the model presented inBarros et al (2011) Both models predict surface CO2 andCH4 diffusion as a function of age and account for the effectof temperature using different proxies the G-res uses effec-tive temperature while the Barros et al model uses latitude(an indirect proxy that integrates other spatial differences)In terms of CO2 surface diffusion the G-res uses reservoirarea soil C content and TP to quantify the effect of C in-puts fuelling CO2 production while the Barros et al modeluses DOC inputs directly (based on in situ DOC concentra-tion) For CH4 surface diffusion both models account formorphometry using the fraction of littoral area (G-res) or themean depth (Barros et al model) Overall both models pre-dict surface diffusion based on the same conceptual frame-work but use different proxies CH4 ebullition and degassingare modelled only by the G-res being the sole model avail-able to this date Details on model equations and input vari-ables are presented in the Supplement (Tables S2 and S3)

3 Results and discussion

31 Water chemistry

The reservoir is stratified throughout the year with a ther-mocline at a depth around 13 m and mostly anoxic condi-tions in the hypolimnion of the main basin (Fig S1 in theSupplement) The system is oligotrophic with very low con-centrations of DOC TP TN and Chl a averaging 09 (SDplusmn02) mgLminus1 59 (SDplusmn24) microgLminus1 011 (SDplusmn004) mgLminus1and 13 (SDplusmn 07) microgLminus1 respectively (Table S1) and highwater transparency (Secchi depth gt 5 m) In the reservoirinflows concentrations of measured chemical species areslightly higher but still in the oligotrophic range (Table S1)

however the transparency is much lower due to turbidity(Secchi lt 05 m) The oligotrophic status of the reservoirlikely results from nutrient-poor soils (Wasli et al 2011)and a largely undisturbed forested catchment in the protectedBatang Ai National Park The reservoirrsquos low Chl a concen-trations are comparable to the neighbouring Bakun reservoir(Ling et al 2017) and its DOC concentrations are on the lowend of the wide range of measured values in nearby rivers(Martin et al 2018)

32 Surface diffusion

Measured CO2 diffusion in the reservoir averaged 77 (SDplusmn182) mmolmminus2 dminus1 (Table S1) which is on the low endcompared to other reservoirs (Deemer et al 2016) and evento natural lakes (Sobek et al 2005) but similar to CO2 fluxesmeasured in two reservoirs in Laos (Chanudet et al 2011)CO2 diffusion across all sites ranged from substantial up-take to high emissions (from minus308 to 5939 mmolmminus2 dminus1Table S1) reflecting a large spatial and temporal variabil-ity Spatially CO2 fluxes measured in the main basin andbranches had similar averages of 79 and 73 mmolmminus2 dminus1

respectively (overall SDplusmn182) contrasting with higher andmore variable values in the inflows with a mean of 1373(SDplusmn 1924) mmolmminus2 dminus1 (Fig 2) Within the reservoirCO2 fluxes varied (SDplusmn182 mmolmminus2 dminus1) but did not fol-low a consistent pattern and might reflect pre-flooding land-scape heterogeneity (Teodoru et al 2011) Temporally high-est average reservoir CO2 fluxes were measured in AprilndashMay 2017 when no CO2 uptake was observed contraryto other campaigns especially FebruaryndashMarch and Au-gust 2018 when CO2 uptake was common (Fig S2) andaverage Chl a concentrations were the highest This reflectsthe important role of metabolism (namely CO2 consumptionby primary production) in modulating surface CO2 fluxes inBatang Ai

All CH4 surface diffusion measurements were positive andranged from 003 to 1134 mmolmminus2 dminus1 (Table S1) Spa-tially CH4 fluxes were progressively higher moving furtherupstream (Figs 2 and S3) with decreasing water depth andincreasing connection to the littoral This gradient in mor-phometry induces an increasingly greater contact of the waterwith bottom and littoral sediments where CH4 is producedexplaining the spatial pattern of CH4 fluxes CH4 surface dif-fusion also varied temporally but to a lesser extent than CO2being on average highest in August 2018 in the reservoir andin NovemberndashDecember 2016 in the inflows

Reservoir N2O surface diffusion (measured with a limitedspatial resolution) averaged minus02 (SDplusmn 21) nmolmminus2 dminus1

(Table S1) The negative value indicates that the systemacts as a slight net sink of N2O absorbing an estimated21 gCO2 eqmminus2 yrminus1 (Table 2) Atmospheric N2O uptakehas previously been reported in aquatic systems and linkedto low oxygen and nitrogen content conducive to completedenitrification which consumes N2O (Soued et al 2016

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 517

Figure 1 Map of Batang Ai showing the location of sampled sites and reservoir sections Represents the reservoir inflow sites

extraction by hot 90 ethanol solution (Sartory and Grobbe-laar 1984)

23 Surface diffusion

Surface diffusion is the flux of gas between the water sur-face and the air driven by a gradient in gas partial pressureSurface diffusion of CO2 and CH4 to the atmosphere wasmeasured at 36 sites in the reservoir and 3 sites in the inflowrivers (Fig 1) and sampling of the same sites was repeatedin each campaign (with a few exceptions) Fluxes were mea-sured using a static airtight floating chamber connected ina closed loop to an ultraportable gas analyser (UGGA fromLGR) Surface diffusion rates (Fgas) were derived from thelinear change in CO2 and CH4 partial pressures (continu-ously monitored at 1 Hz for a minimum of 5 min) throughtime inside the chamber using the following Eq (1)

Fgas =sV

mA (1)

where s is the gas accumulation rate in the chamber V =25 L the chamber volume A= 0184 m2 the chamber sur-face area andm the gas molar volume at current atmosphericpressure

N2O surface diffusion was estimated at seven of the sam-pled sites (Fig 1) using the following Eq (2) (Lide 2005)

Fgas = kgas(CgasminusCeq

) (2)

where kgas is the gas exchange coefficient Cgas is the gasconcentration in the water and Ceq is the theoretical gas

concentration at equilibrium given measured water temper-ature atmospheric pressure and ambient gas concentrationCN2O was measured using the headspace technique with a112 L sealed glass serum bottle containing surface water anda 012 L headspace of ambient air After shaking the bottlefor 2 min to achieve airndashwater equilibrium the headspace gaswas extracted from the bottle with an airtight syringe andinjected in the previously evacuated 9 mL glass vial cappedwith an airtight butyl stopper and aluminium seal Three ana-lytical replicates and a local sample of ambient air were takenat each site and analysed by gas chromatography using a Shi-madzu GC-2040 with a Poropaq Q column to separate gasesand an electron capture detector (ECD) calibrated with 031 and 3 ppm of N2O certified standard gas After analysis theoriginal N2O concentration of the water was back-calculatedbased on the water temperature before and after shaking (forgas solubility) the ambient atmospheric pressure the ratioof water to air in the sampling bottle and the headspace N2Oconcentration before shaking kN2O was derived from mea-sured kCH4 values obtained by rearranging Eq (2) for CH4with known values of Fgas Cgas and Ceq The kCH4 to kN2Otransformation was done using the following Eq (3) (Coleand Caraco 1998 Ledwell 1984)

kN2O =

(ScN2O

ScCH4

)minus067

kCH4 (3)

where Sc is the gas Schmidt number (Wanninkhof 1992)CH4 and CO2 concentrations in the water were mea-

sured using the headspace technique Surface water was col-lected in a 60 mL gas-tight plastic syringe in which a 30 mL

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

518 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

Figure 2 Boxplots of measured CH4 (on a log axis) and CO2fluxes grouped according to spatial position Boxes are bounded bythe 25th and 75th percentile and show medians (solid lines) andwhiskers show 10th and 90th percentiles Grey circles show singledata points

headspace was created (using either ambient air or carbon-free air) The syringe was shaken for 2 min to achieve gasequilibrium between air and water The gas phase was theninjected in a 12 mL airtight pre-evacuated vial and subse-quently analysed through manual injection on a ShimadzuGC-8A gas chromatograph with a flame ionization detectorfollowing a calibration curve with certified gas standards (0ndash10 000 ppm for CO2 and 0ndash50 000 ppm for CH4) The sam-ples were also analysed for isotopic δ13CO2 and δ13CH4 sig-natures by manually injecting 18 mL of gas in a cavity ring-down spectrometer (CRDS) equipped with a small sampleisotopic module (SSIM A0314 Picarro G2201-i analyser)set in a non-continuous mode with a three-point calibrationcurve based on certified gas standards (minus40 permil minus39 permil and253 permil for δ13CO2 and minus665 permil minus383 permil and minus239 permilfor δ13CH4)

Figure 3 Concentrations (black symbols and solid line) and δ13C(grey symbols and dotted lines) of CO2 and CH4 from right up-stream of the dam (grey band) to 19 km downstream in the out-flow river Circles squares diamonds and triangles represent val-ues from NovemberndashDecember 2016 AprilndashMay 2017 FebruaryndashMar 2018 and August 2018 respectively

24 Ebullition

Ebullition is the process through which gas bubbles formedin the sediments rise through the water column and are re-leased to the atmosphere Sediment gas ebullition was mea-sured at four sites in the reservoir and two sites in the inflows(Fig 1) by deploying 0785 m2 underwater inverted funneltraps at 2 to 3 m deep for approximately 20 d in the reservoirand 1 h in the inflows The top part of a closed plastic sy-ringe was fixed to the narrow end of the funnel trap wherethe emerging bubbles accumulated Upon recovery bubblegas volume was measured collected from the syringe and in-jected in 12 mL pre-evacuated airtight vials for CO2 and CH4concentration analyses (using the aforementioned method)Ebullition rate was calculated assuming the original bubblecomposition was similar to bubbles collected almost right af-ter ascent in the inflows sites which was 100 CH4 Hencewe considered CO2 and N2O ebullition to be null

In order to estimate the potential for sediment accumula-tion fuelling ebullition in the littoral zone we calculated themud energy boundary depth (EBD in metres below whichfine-grained sediment accumulation occurs) using the reser-voir surface area (E in km2) as the exposure parameter in thefollowing Eq (4) (Rowan et al 1992)

EBD= 2685E0305 (4)

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 519

Table 1 CO2 and CH4 dynamics downstream of the dam gas export rate from upstream to downstream of the dam degassing result ofCH4 oxidation (CO2 production and CH4 consumption) downstream emissions and total emissions to the atmosphere below the damUncertainties based on variation coefficients are reported in parentheses Units are in mmolmminus2 dminus1 of reservoir surface area

GHG downstream of the dam (mmolmminus2 dminus1)

Exported Degassed Gainloss by oxidation Downstream emissions Total emissions

CO2

NovndashDec 2016 4062 (plusmn227) 1526 (plusmn085) 090 (plusmn013) 1267 (plusmn071) 2793 (plusmn156)AprndashMay 2017 3780 (plusmn211) 1491 (plusmn083) 059 (plusmn008) 983 (plusmn055) 2470 (plusmn138)FebndashMar 2018 3798 (plusmn212) 958 (plusmn054) 180 (plusmn026) 970 (plusmn054) 1930 (plusmn108)Aug 2018 3807 (plusmn213) 2167 (plusmn121) 038 (plusmn005) 831 (plusmn046) 3000 (plusmn168)

CH4

NovndashDec 2016 1484 (plusmn210) 1156 (plusmn164) 090 (plusmn013) 219 (plusmn031) 1376 (plusmn195)AprndashMay 2017 732 (plusmn104) 400 (plusmn057) 059 (plusmn008) 190 (plusmn027) 590 (plusmn084)FebndashMar 2018 1247 (plusmn177) 492 (plusmn070) 180 (plusmn026) 399 (plusmn057) 891 (plusmn126)Aug 2018 1071 (plusmn152) 954 (plusmn135) 038 (plusmn005) 051 (plusmn007) 1005 (plusmn142)

25 Degassing downstream emissions and CH4oxidation

Degassing of CO2 and CH4 right after water discharge(Fdeg) and downstream emissions of the remaining reservoir-derived GHGs in the outflow river (Fdwn) were calculatedusing the following Eqs (5) and (6)

Fdeg =Q(CupminusC0

) (5)

Fdwn =Q(C0minusC19+Cox) (6)

where Q is the water discharge and Cup C0 and C19 themeasured gas concentrations upstream of the dam at the wa-ter withdrawal depth at the powerhouse right after water dis-charge and in the outflow 19 km downstream of the dam re-spectively Cox is the net change in gas concentration due tooxidation (loss for CH4 and gain for CO2) For downstreamemissions we considered that after a river stretch of 19 kmall excess gas originating from the reservoir was evaded andgas concentration was representative of the outflow riverbaseline This assumption potentially underestimates actualdownstream emissions (in the case of remaining excess gasafter 19 km) However given the observed exponential de-crease in gas concentration along the outflow (Fig 3) emis-sions after 19 km are expected to be small compared to thosein the 0 to 19 km river stretch consistent with observations inother reservoirs (Gueacuterin et al 2006 Kemenes et al 2007)

Gas concentrations upstream and downstream of the damwere obtained by measuring in each campaign CO2 andCH4 concentrations in a vertical profile right upstream ofthe dam at a 1 to 3 m interval from 0 to 32 m and at fourlocations in the outflow at 0 (power house) 06 27 and19 km downstream of the dam (Fig 1) Sampling was doneusing a multi-parameter probe equipped with depth oxygenand temperature sensors (Yellow Spring Instruments YSImodel 600XLM-M) attached to a 12 V submersible Tornado

pump (Proactive Environmental Products) for water collec-tion Gas concentration and δ13C were measured as describedin Sect 23 Water withdrawal depth ranged from 20 to 23 mand was estimated based on known values of elevations ofwater intake and water level compared to sea level Gas con-centration in the water exiting the reservoir was defined asthe average measured gas concentrations in the plusmn1 m rangeof the withdrawal depth

Estimates of downstream CH4 oxidation were obtainedfor each sampling campaign by calculating the fraction ofCH4 oxidized (Fox) using the following Eq (7)

Fox =

minus(ln

(δ13CH4resid+1000)minusln(δ13CH4source+1000

))middot

(1minus [CH4]resid

[CH4]source

)(1minus 1

α

)ln

([CH4]resid

[CH4]source

) (7)

Equation (7) is based on a non-steady-state isotopic modeldeveloped considering evasion (emission to the atmosphere)and oxidation as the two loss processes for CH4 in the out-flow river assuming negligible isotopic fractionation for eva-sion (Knox et al 1992) and a fractionation of α = 102 foroxidation (Coleman et al 1981) (see derivation in Supple-ment) [CH4]source [CH4]resid δ13CH4source and δ13CH4residare the concentrations of CH4 and their corresponding iso-topic signatures at the beginning of the outflow (km 0)and 19 km downstream representing the source and residualpools of CH4 respectively The amount of CH4 oxidized toCO2 along the 19 km of river stretch for each sampling cam-paign was calculated as the product of Fox and [CH4]sourceThe resulting loss of CH4 and gain of CO2 in the outflowwere accounted for in downstream emissions (Cox in Eq 6)Note that downstream N2O emissions were considered nullsince N2O concentrations measured in the deep reservoirlayer were lower than concentrations in the outflow

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

520 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

26 Ecosystem-scale C footprint

Batang Ai annual C footprint was calculated as the sumof surface diffusion ebullition degassing and downstreamemissions of CO2 CH4 and N2O considering a greenhousewarming potential of 1 34 and 298 respectively over a 100-year lifetime period (Myhre et al 2013) For each flux path-way annual flux was estimated as the average of the sam-pling campaigns Ecosystem-scale estimate of surface diffu-sion was calculated for each campaign as the average of mea-sured flux rates applied to the reservoir area for N2O and forCO2 and CH4 it was obtained by spatial interpolation of mea-sured fluxes over the reservoir area based on inverse distanceweighting with a power of two (a power of one yields similaraverages CV lt 11 ) using package gstat version 11ndash6 inthe R version 341 software (Pebesma 2004 R Core Team2017) Ebullition at the reservoir scale was calculated as theaverage of measured reservoir ebullition rates applied to thelittoral area (lt 3 m deep)

The estimated GHG emissions of Batang Ai based on mea-sured data was compared to values derived from the G-resmodel (UNESCOIHA 2017) and the model presented inBarros et al (2011) Both models predict surface CO2 andCH4 diffusion as a function of age and account for the effectof temperature using different proxies the G-res uses effec-tive temperature while the Barros et al model uses latitude(an indirect proxy that integrates other spatial differences)In terms of CO2 surface diffusion the G-res uses reservoirarea soil C content and TP to quantify the effect of C in-puts fuelling CO2 production while the Barros et al modeluses DOC inputs directly (based on in situ DOC concentra-tion) For CH4 surface diffusion both models account formorphometry using the fraction of littoral area (G-res) or themean depth (Barros et al model) Overall both models pre-dict surface diffusion based on the same conceptual frame-work but use different proxies CH4 ebullition and degassingare modelled only by the G-res being the sole model avail-able to this date Details on model equations and input vari-ables are presented in the Supplement (Tables S2 and S3)

3 Results and discussion

31 Water chemistry

The reservoir is stratified throughout the year with a ther-mocline at a depth around 13 m and mostly anoxic condi-tions in the hypolimnion of the main basin (Fig S1 in theSupplement) The system is oligotrophic with very low con-centrations of DOC TP TN and Chl a averaging 09 (SDplusmn02) mgLminus1 59 (SDplusmn24) microgLminus1 011 (SDplusmn004) mgLminus1and 13 (SDplusmn 07) microgLminus1 respectively (Table S1) and highwater transparency (Secchi depth gt 5 m) In the reservoirinflows concentrations of measured chemical species areslightly higher but still in the oligotrophic range (Table S1)

however the transparency is much lower due to turbidity(Secchi lt 05 m) The oligotrophic status of the reservoirlikely results from nutrient-poor soils (Wasli et al 2011)and a largely undisturbed forested catchment in the protectedBatang Ai National Park The reservoirrsquos low Chl a concen-trations are comparable to the neighbouring Bakun reservoir(Ling et al 2017) and its DOC concentrations are on the lowend of the wide range of measured values in nearby rivers(Martin et al 2018)

32 Surface diffusion

Measured CO2 diffusion in the reservoir averaged 77 (SDplusmn182) mmolmminus2 dminus1 (Table S1) which is on the low endcompared to other reservoirs (Deemer et al 2016) and evento natural lakes (Sobek et al 2005) but similar to CO2 fluxesmeasured in two reservoirs in Laos (Chanudet et al 2011)CO2 diffusion across all sites ranged from substantial up-take to high emissions (from minus308 to 5939 mmolmminus2 dminus1Table S1) reflecting a large spatial and temporal variabil-ity Spatially CO2 fluxes measured in the main basin andbranches had similar averages of 79 and 73 mmolmminus2 dminus1

respectively (overall SDplusmn182) contrasting with higher andmore variable values in the inflows with a mean of 1373(SDplusmn 1924) mmolmminus2 dminus1 (Fig 2) Within the reservoirCO2 fluxes varied (SDplusmn182 mmolmminus2 dminus1) but did not fol-low a consistent pattern and might reflect pre-flooding land-scape heterogeneity (Teodoru et al 2011) Temporally high-est average reservoir CO2 fluxes were measured in AprilndashMay 2017 when no CO2 uptake was observed contraryto other campaigns especially FebruaryndashMarch and Au-gust 2018 when CO2 uptake was common (Fig S2) andaverage Chl a concentrations were the highest This reflectsthe important role of metabolism (namely CO2 consumptionby primary production) in modulating surface CO2 fluxes inBatang Ai

All CH4 surface diffusion measurements were positive andranged from 003 to 1134 mmolmminus2 dminus1 (Table S1) Spa-tially CH4 fluxes were progressively higher moving furtherupstream (Figs 2 and S3) with decreasing water depth andincreasing connection to the littoral This gradient in mor-phometry induces an increasingly greater contact of the waterwith bottom and littoral sediments where CH4 is producedexplaining the spatial pattern of CH4 fluxes CH4 surface dif-fusion also varied temporally but to a lesser extent than CO2being on average highest in August 2018 in the reservoir andin NovemberndashDecember 2016 in the inflows

Reservoir N2O surface diffusion (measured with a limitedspatial resolution) averaged minus02 (SDplusmn 21) nmolmminus2 dminus1

(Table S1) The negative value indicates that the systemacts as a slight net sink of N2O absorbing an estimated21 gCO2 eqmminus2 yrminus1 (Table 2) Atmospheric N2O uptakehas previously been reported in aquatic systems and linkedto low oxygen and nitrogen content conducive to completedenitrification which consumes N2O (Soued et al 2016

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

518 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

Figure 2 Boxplots of measured CH4 (on a log axis) and CO2fluxes grouped according to spatial position Boxes are bounded bythe 25th and 75th percentile and show medians (solid lines) andwhiskers show 10th and 90th percentiles Grey circles show singledata points

headspace was created (using either ambient air or carbon-free air) The syringe was shaken for 2 min to achieve gasequilibrium between air and water The gas phase was theninjected in a 12 mL airtight pre-evacuated vial and subse-quently analysed through manual injection on a ShimadzuGC-8A gas chromatograph with a flame ionization detectorfollowing a calibration curve with certified gas standards (0ndash10 000 ppm for CO2 and 0ndash50 000 ppm for CH4) The sam-ples were also analysed for isotopic δ13CO2 and δ13CH4 sig-natures by manually injecting 18 mL of gas in a cavity ring-down spectrometer (CRDS) equipped with a small sampleisotopic module (SSIM A0314 Picarro G2201-i analyser)set in a non-continuous mode with a three-point calibrationcurve based on certified gas standards (minus40 permil minus39 permil and253 permil for δ13CO2 and minus665 permil minus383 permil and minus239 permilfor δ13CH4)

Figure 3 Concentrations (black symbols and solid line) and δ13C(grey symbols and dotted lines) of CO2 and CH4 from right up-stream of the dam (grey band) to 19 km downstream in the out-flow river Circles squares diamonds and triangles represent val-ues from NovemberndashDecember 2016 AprilndashMay 2017 FebruaryndashMar 2018 and August 2018 respectively

24 Ebullition

Ebullition is the process through which gas bubbles formedin the sediments rise through the water column and are re-leased to the atmosphere Sediment gas ebullition was mea-sured at four sites in the reservoir and two sites in the inflows(Fig 1) by deploying 0785 m2 underwater inverted funneltraps at 2 to 3 m deep for approximately 20 d in the reservoirand 1 h in the inflows The top part of a closed plastic sy-ringe was fixed to the narrow end of the funnel trap wherethe emerging bubbles accumulated Upon recovery bubblegas volume was measured collected from the syringe and in-jected in 12 mL pre-evacuated airtight vials for CO2 and CH4concentration analyses (using the aforementioned method)Ebullition rate was calculated assuming the original bubblecomposition was similar to bubbles collected almost right af-ter ascent in the inflows sites which was 100 CH4 Hencewe considered CO2 and N2O ebullition to be null

In order to estimate the potential for sediment accumula-tion fuelling ebullition in the littoral zone we calculated themud energy boundary depth (EBD in metres below whichfine-grained sediment accumulation occurs) using the reser-voir surface area (E in km2) as the exposure parameter in thefollowing Eq (4) (Rowan et al 1992)

EBD= 2685E0305 (4)

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 519

Table 1 CO2 and CH4 dynamics downstream of the dam gas export rate from upstream to downstream of the dam degassing result ofCH4 oxidation (CO2 production and CH4 consumption) downstream emissions and total emissions to the atmosphere below the damUncertainties based on variation coefficients are reported in parentheses Units are in mmolmminus2 dminus1 of reservoir surface area

GHG downstream of the dam (mmolmminus2 dminus1)

Exported Degassed Gainloss by oxidation Downstream emissions Total emissions

CO2

NovndashDec 2016 4062 (plusmn227) 1526 (plusmn085) 090 (plusmn013) 1267 (plusmn071) 2793 (plusmn156)AprndashMay 2017 3780 (plusmn211) 1491 (plusmn083) 059 (plusmn008) 983 (plusmn055) 2470 (plusmn138)FebndashMar 2018 3798 (plusmn212) 958 (plusmn054) 180 (plusmn026) 970 (plusmn054) 1930 (plusmn108)Aug 2018 3807 (plusmn213) 2167 (plusmn121) 038 (plusmn005) 831 (plusmn046) 3000 (plusmn168)

CH4

NovndashDec 2016 1484 (plusmn210) 1156 (plusmn164) 090 (plusmn013) 219 (plusmn031) 1376 (plusmn195)AprndashMay 2017 732 (plusmn104) 400 (plusmn057) 059 (plusmn008) 190 (plusmn027) 590 (plusmn084)FebndashMar 2018 1247 (plusmn177) 492 (plusmn070) 180 (plusmn026) 399 (plusmn057) 891 (plusmn126)Aug 2018 1071 (plusmn152) 954 (plusmn135) 038 (plusmn005) 051 (plusmn007) 1005 (plusmn142)

25 Degassing downstream emissions and CH4oxidation

Degassing of CO2 and CH4 right after water discharge(Fdeg) and downstream emissions of the remaining reservoir-derived GHGs in the outflow river (Fdwn) were calculatedusing the following Eqs (5) and (6)

Fdeg =Q(CupminusC0

) (5)

Fdwn =Q(C0minusC19+Cox) (6)

where Q is the water discharge and Cup C0 and C19 themeasured gas concentrations upstream of the dam at the wa-ter withdrawal depth at the powerhouse right after water dis-charge and in the outflow 19 km downstream of the dam re-spectively Cox is the net change in gas concentration due tooxidation (loss for CH4 and gain for CO2) For downstreamemissions we considered that after a river stretch of 19 kmall excess gas originating from the reservoir was evaded andgas concentration was representative of the outflow riverbaseline This assumption potentially underestimates actualdownstream emissions (in the case of remaining excess gasafter 19 km) However given the observed exponential de-crease in gas concentration along the outflow (Fig 3) emis-sions after 19 km are expected to be small compared to thosein the 0 to 19 km river stretch consistent with observations inother reservoirs (Gueacuterin et al 2006 Kemenes et al 2007)

Gas concentrations upstream and downstream of the damwere obtained by measuring in each campaign CO2 andCH4 concentrations in a vertical profile right upstream ofthe dam at a 1 to 3 m interval from 0 to 32 m and at fourlocations in the outflow at 0 (power house) 06 27 and19 km downstream of the dam (Fig 1) Sampling was doneusing a multi-parameter probe equipped with depth oxygenand temperature sensors (Yellow Spring Instruments YSImodel 600XLM-M) attached to a 12 V submersible Tornado

pump (Proactive Environmental Products) for water collec-tion Gas concentration and δ13C were measured as describedin Sect 23 Water withdrawal depth ranged from 20 to 23 mand was estimated based on known values of elevations ofwater intake and water level compared to sea level Gas con-centration in the water exiting the reservoir was defined asthe average measured gas concentrations in the plusmn1 m rangeof the withdrawal depth

Estimates of downstream CH4 oxidation were obtainedfor each sampling campaign by calculating the fraction ofCH4 oxidized (Fox) using the following Eq (7)

Fox =

minus(ln

(δ13CH4resid+1000)minusln(δ13CH4source+1000

))middot

(1minus [CH4]resid

[CH4]source

)(1minus 1

α

)ln

([CH4]resid

[CH4]source

) (7)

Equation (7) is based on a non-steady-state isotopic modeldeveloped considering evasion (emission to the atmosphere)and oxidation as the two loss processes for CH4 in the out-flow river assuming negligible isotopic fractionation for eva-sion (Knox et al 1992) and a fractionation of α = 102 foroxidation (Coleman et al 1981) (see derivation in Supple-ment) [CH4]source [CH4]resid δ13CH4source and δ13CH4residare the concentrations of CH4 and their corresponding iso-topic signatures at the beginning of the outflow (km 0)and 19 km downstream representing the source and residualpools of CH4 respectively The amount of CH4 oxidized toCO2 along the 19 km of river stretch for each sampling cam-paign was calculated as the product of Fox and [CH4]sourceThe resulting loss of CH4 and gain of CO2 in the outflowwere accounted for in downstream emissions (Cox in Eq 6)Note that downstream N2O emissions were considered nullsince N2O concentrations measured in the deep reservoirlayer were lower than concentrations in the outflow

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

520 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

26 Ecosystem-scale C footprint

Batang Ai annual C footprint was calculated as the sumof surface diffusion ebullition degassing and downstreamemissions of CO2 CH4 and N2O considering a greenhousewarming potential of 1 34 and 298 respectively over a 100-year lifetime period (Myhre et al 2013) For each flux path-way annual flux was estimated as the average of the sam-pling campaigns Ecosystem-scale estimate of surface diffu-sion was calculated for each campaign as the average of mea-sured flux rates applied to the reservoir area for N2O and forCO2 and CH4 it was obtained by spatial interpolation of mea-sured fluxes over the reservoir area based on inverse distanceweighting with a power of two (a power of one yields similaraverages CV lt 11 ) using package gstat version 11ndash6 inthe R version 341 software (Pebesma 2004 R Core Team2017) Ebullition at the reservoir scale was calculated as theaverage of measured reservoir ebullition rates applied to thelittoral area (lt 3 m deep)

The estimated GHG emissions of Batang Ai based on mea-sured data was compared to values derived from the G-resmodel (UNESCOIHA 2017) and the model presented inBarros et al (2011) Both models predict surface CO2 andCH4 diffusion as a function of age and account for the effectof temperature using different proxies the G-res uses effec-tive temperature while the Barros et al model uses latitude(an indirect proxy that integrates other spatial differences)In terms of CO2 surface diffusion the G-res uses reservoirarea soil C content and TP to quantify the effect of C in-puts fuelling CO2 production while the Barros et al modeluses DOC inputs directly (based on in situ DOC concentra-tion) For CH4 surface diffusion both models account formorphometry using the fraction of littoral area (G-res) or themean depth (Barros et al model) Overall both models pre-dict surface diffusion based on the same conceptual frame-work but use different proxies CH4 ebullition and degassingare modelled only by the G-res being the sole model avail-able to this date Details on model equations and input vari-ables are presented in the Supplement (Tables S2 and S3)

3 Results and discussion

31 Water chemistry

The reservoir is stratified throughout the year with a ther-mocline at a depth around 13 m and mostly anoxic condi-tions in the hypolimnion of the main basin (Fig S1 in theSupplement) The system is oligotrophic with very low con-centrations of DOC TP TN and Chl a averaging 09 (SDplusmn02) mgLminus1 59 (SDplusmn24) microgLminus1 011 (SDplusmn004) mgLminus1and 13 (SDplusmn 07) microgLminus1 respectively (Table S1) and highwater transparency (Secchi depth gt 5 m) In the reservoirinflows concentrations of measured chemical species areslightly higher but still in the oligotrophic range (Table S1)

however the transparency is much lower due to turbidity(Secchi lt 05 m) The oligotrophic status of the reservoirlikely results from nutrient-poor soils (Wasli et al 2011)and a largely undisturbed forested catchment in the protectedBatang Ai National Park The reservoirrsquos low Chl a concen-trations are comparable to the neighbouring Bakun reservoir(Ling et al 2017) and its DOC concentrations are on the lowend of the wide range of measured values in nearby rivers(Martin et al 2018)

32 Surface diffusion

Measured CO2 diffusion in the reservoir averaged 77 (SDplusmn182) mmolmminus2 dminus1 (Table S1) which is on the low endcompared to other reservoirs (Deemer et al 2016) and evento natural lakes (Sobek et al 2005) but similar to CO2 fluxesmeasured in two reservoirs in Laos (Chanudet et al 2011)CO2 diffusion across all sites ranged from substantial up-take to high emissions (from minus308 to 5939 mmolmminus2 dminus1Table S1) reflecting a large spatial and temporal variabil-ity Spatially CO2 fluxes measured in the main basin andbranches had similar averages of 79 and 73 mmolmminus2 dminus1

respectively (overall SDplusmn182) contrasting with higher andmore variable values in the inflows with a mean of 1373(SDplusmn 1924) mmolmminus2 dminus1 (Fig 2) Within the reservoirCO2 fluxes varied (SDplusmn182 mmolmminus2 dminus1) but did not fol-low a consistent pattern and might reflect pre-flooding land-scape heterogeneity (Teodoru et al 2011) Temporally high-est average reservoir CO2 fluxes were measured in AprilndashMay 2017 when no CO2 uptake was observed contraryto other campaigns especially FebruaryndashMarch and Au-gust 2018 when CO2 uptake was common (Fig S2) andaverage Chl a concentrations were the highest This reflectsthe important role of metabolism (namely CO2 consumptionby primary production) in modulating surface CO2 fluxes inBatang Ai

All CH4 surface diffusion measurements were positive andranged from 003 to 1134 mmolmminus2 dminus1 (Table S1) Spa-tially CH4 fluxes were progressively higher moving furtherupstream (Figs 2 and S3) with decreasing water depth andincreasing connection to the littoral This gradient in mor-phometry induces an increasingly greater contact of the waterwith bottom and littoral sediments where CH4 is producedexplaining the spatial pattern of CH4 fluxes CH4 surface dif-fusion also varied temporally but to a lesser extent than CO2being on average highest in August 2018 in the reservoir andin NovemberndashDecember 2016 in the inflows

Reservoir N2O surface diffusion (measured with a limitedspatial resolution) averaged minus02 (SDplusmn 21) nmolmminus2 dminus1

(Table S1) The negative value indicates that the systemacts as a slight net sink of N2O absorbing an estimated21 gCO2 eqmminus2 yrminus1 (Table 2) Atmospheric N2O uptakehas previously been reported in aquatic systems and linkedto low oxygen and nitrogen content conducive to completedenitrification which consumes N2O (Soued et al 2016

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 519

Table 1 CO2 and CH4 dynamics downstream of the dam gas export rate from upstream to downstream of the dam degassing result ofCH4 oxidation (CO2 production and CH4 consumption) downstream emissions and total emissions to the atmosphere below the damUncertainties based on variation coefficients are reported in parentheses Units are in mmolmminus2 dminus1 of reservoir surface area

GHG downstream of the dam (mmolmminus2 dminus1)

Exported Degassed Gainloss by oxidation Downstream emissions Total emissions

CO2

NovndashDec 2016 4062 (plusmn227) 1526 (plusmn085) 090 (plusmn013) 1267 (plusmn071) 2793 (plusmn156)AprndashMay 2017 3780 (plusmn211) 1491 (plusmn083) 059 (plusmn008) 983 (plusmn055) 2470 (plusmn138)FebndashMar 2018 3798 (plusmn212) 958 (plusmn054) 180 (plusmn026) 970 (plusmn054) 1930 (plusmn108)Aug 2018 3807 (plusmn213) 2167 (plusmn121) 038 (plusmn005) 831 (plusmn046) 3000 (plusmn168)

CH4

NovndashDec 2016 1484 (plusmn210) 1156 (plusmn164) 090 (plusmn013) 219 (plusmn031) 1376 (plusmn195)AprndashMay 2017 732 (plusmn104) 400 (plusmn057) 059 (plusmn008) 190 (plusmn027) 590 (plusmn084)FebndashMar 2018 1247 (plusmn177) 492 (plusmn070) 180 (plusmn026) 399 (plusmn057) 891 (plusmn126)Aug 2018 1071 (plusmn152) 954 (plusmn135) 038 (plusmn005) 051 (plusmn007) 1005 (plusmn142)

25 Degassing downstream emissions and CH4oxidation

Degassing of CO2 and CH4 right after water discharge(Fdeg) and downstream emissions of the remaining reservoir-derived GHGs in the outflow river (Fdwn) were calculatedusing the following Eqs (5) and (6)

Fdeg =Q(CupminusC0

) (5)

Fdwn =Q(C0minusC19+Cox) (6)

where Q is the water discharge and Cup C0 and C19 themeasured gas concentrations upstream of the dam at the wa-ter withdrawal depth at the powerhouse right after water dis-charge and in the outflow 19 km downstream of the dam re-spectively Cox is the net change in gas concentration due tooxidation (loss for CH4 and gain for CO2) For downstreamemissions we considered that after a river stretch of 19 kmall excess gas originating from the reservoir was evaded andgas concentration was representative of the outflow riverbaseline This assumption potentially underestimates actualdownstream emissions (in the case of remaining excess gasafter 19 km) However given the observed exponential de-crease in gas concentration along the outflow (Fig 3) emis-sions after 19 km are expected to be small compared to thosein the 0 to 19 km river stretch consistent with observations inother reservoirs (Gueacuterin et al 2006 Kemenes et al 2007)

Gas concentrations upstream and downstream of the damwere obtained by measuring in each campaign CO2 andCH4 concentrations in a vertical profile right upstream ofthe dam at a 1 to 3 m interval from 0 to 32 m and at fourlocations in the outflow at 0 (power house) 06 27 and19 km downstream of the dam (Fig 1) Sampling was doneusing a multi-parameter probe equipped with depth oxygenand temperature sensors (Yellow Spring Instruments YSImodel 600XLM-M) attached to a 12 V submersible Tornado

pump (Proactive Environmental Products) for water collec-tion Gas concentration and δ13C were measured as describedin Sect 23 Water withdrawal depth ranged from 20 to 23 mand was estimated based on known values of elevations ofwater intake and water level compared to sea level Gas con-centration in the water exiting the reservoir was defined asthe average measured gas concentrations in the plusmn1 m rangeof the withdrawal depth

Estimates of downstream CH4 oxidation were obtainedfor each sampling campaign by calculating the fraction ofCH4 oxidized (Fox) using the following Eq (7)

Fox =

minus(ln

(δ13CH4resid+1000)minusln(δ13CH4source+1000

))middot

(1minus [CH4]resid

[CH4]source

)(1minus 1

α

)ln

([CH4]resid

[CH4]source

) (7)

Equation (7) is based on a non-steady-state isotopic modeldeveloped considering evasion (emission to the atmosphere)and oxidation as the two loss processes for CH4 in the out-flow river assuming negligible isotopic fractionation for eva-sion (Knox et al 1992) and a fractionation of α = 102 foroxidation (Coleman et al 1981) (see derivation in Supple-ment) [CH4]source [CH4]resid δ13CH4source and δ13CH4residare the concentrations of CH4 and their corresponding iso-topic signatures at the beginning of the outflow (km 0)and 19 km downstream representing the source and residualpools of CH4 respectively The amount of CH4 oxidized toCO2 along the 19 km of river stretch for each sampling cam-paign was calculated as the product of Fox and [CH4]sourceThe resulting loss of CH4 and gain of CO2 in the outflowwere accounted for in downstream emissions (Cox in Eq 6)Note that downstream N2O emissions were considered nullsince N2O concentrations measured in the deep reservoirlayer were lower than concentrations in the outflow

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

520 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

26 Ecosystem-scale C footprint

Batang Ai annual C footprint was calculated as the sumof surface diffusion ebullition degassing and downstreamemissions of CO2 CH4 and N2O considering a greenhousewarming potential of 1 34 and 298 respectively over a 100-year lifetime period (Myhre et al 2013) For each flux path-way annual flux was estimated as the average of the sam-pling campaigns Ecosystem-scale estimate of surface diffu-sion was calculated for each campaign as the average of mea-sured flux rates applied to the reservoir area for N2O and forCO2 and CH4 it was obtained by spatial interpolation of mea-sured fluxes over the reservoir area based on inverse distanceweighting with a power of two (a power of one yields similaraverages CV lt 11 ) using package gstat version 11ndash6 inthe R version 341 software (Pebesma 2004 R Core Team2017) Ebullition at the reservoir scale was calculated as theaverage of measured reservoir ebullition rates applied to thelittoral area (lt 3 m deep)

The estimated GHG emissions of Batang Ai based on mea-sured data was compared to values derived from the G-resmodel (UNESCOIHA 2017) and the model presented inBarros et al (2011) Both models predict surface CO2 andCH4 diffusion as a function of age and account for the effectof temperature using different proxies the G-res uses effec-tive temperature while the Barros et al model uses latitude(an indirect proxy that integrates other spatial differences)In terms of CO2 surface diffusion the G-res uses reservoirarea soil C content and TP to quantify the effect of C in-puts fuelling CO2 production while the Barros et al modeluses DOC inputs directly (based on in situ DOC concentra-tion) For CH4 surface diffusion both models account formorphometry using the fraction of littoral area (G-res) or themean depth (Barros et al model) Overall both models pre-dict surface diffusion based on the same conceptual frame-work but use different proxies CH4 ebullition and degassingare modelled only by the G-res being the sole model avail-able to this date Details on model equations and input vari-ables are presented in the Supplement (Tables S2 and S3)

3 Results and discussion

31 Water chemistry

The reservoir is stratified throughout the year with a ther-mocline at a depth around 13 m and mostly anoxic condi-tions in the hypolimnion of the main basin (Fig S1 in theSupplement) The system is oligotrophic with very low con-centrations of DOC TP TN and Chl a averaging 09 (SDplusmn02) mgLminus1 59 (SDplusmn24) microgLminus1 011 (SDplusmn004) mgLminus1and 13 (SDplusmn 07) microgLminus1 respectively (Table S1) and highwater transparency (Secchi depth gt 5 m) In the reservoirinflows concentrations of measured chemical species areslightly higher but still in the oligotrophic range (Table S1)

however the transparency is much lower due to turbidity(Secchi lt 05 m) The oligotrophic status of the reservoirlikely results from nutrient-poor soils (Wasli et al 2011)and a largely undisturbed forested catchment in the protectedBatang Ai National Park The reservoirrsquos low Chl a concen-trations are comparable to the neighbouring Bakun reservoir(Ling et al 2017) and its DOC concentrations are on the lowend of the wide range of measured values in nearby rivers(Martin et al 2018)

32 Surface diffusion

Measured CO2 diffusion in the reservoir averaged 77 (SDplusmn182) mmolmminus2 dminus1 (Table S1) which is on the low endcompared to other reservoirs (Deemer et al 2016) and evento natural lakes (Sobek et al 2005) but similar to CO2 fluxesmeasured in two reservoirs in Laos (Chanudet et al 2011)CO2 diffusion across all sites ranged from substantial up-take to high emissions (from minus308 to 5939 mmolmminus2 dminus1Table S1) reflecting a large spatial and temporal variabil-ity Spatially CO2 fluxes measured in the main basin andbranches had similar averages of 79 and 73 mmolmminus2 dminus1

respectively (overall SDplusmn182) contrasting with higher andmore variable values in the inflows with a mean of 1373(SDplusmn 1924) mmolmminus2 dminus1 (Fig 2) Within the reservoirCO2 fluxes varied (SDplusmn182 mmolmminus2 dminus1) but did not fol-low a consistent pattern and might reflect pre-flooding land-scape heterogeneity (Teodoru et al 2011) Temporally high-est average reservoir CO2 fluxes were measured in AprilndashMay 2017 when no CO2 uptake was observed contraryto other campaigns especially FebruaryndashMarch and Au-gust 2018 when CO2 uptake was common (Fig S2) andaverage Chl a concentrations were the highest This reflectsthe important role of metabolism (namely CO2 consumptionby primary production) in modulating surface CO2 fluxes inBatang Ai

All CH4 surface diffusion measurements were positive andranged from 003 to 1134 mmolmminus2 dminus1 (Table S1) Spa-tially CH4 fluxes were progressively higher moving furtherupstream (Figs 2 and S3) with decreasing water depth andincreasing connection to the littoral This gradient in mor-phometry induces an increasingly greater contact of the waterwith bottom and littoral sediments where CH4 is producedexplaining the spatial pattern of CH4 fluxes CH4 surface dif-fusion also varied temporally but to a lesser extent than CO2being on average highest in August 2018 in the reservoir andin NovemberndashDecember 2016 in the inflows

Reservoir N2O surface diffusion (measured with a limitedspatial resolution) averaged minus02 (SDplusmn 21) nmolmminus2 dminus1

(Table S1) The negative value indicates that the systemacts as a slight net sink of N2O absorbing an estimated21 gCO2 eqmminus2 yrminus1 (Table 2) Atmospheric N2O uptakehas previously been reported in aquatic systems and linkedto low oxygen and nitrogen content conducive to completedenitrification which consumes N2O (Soued et al 2016

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

520 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

26 Ecosystem-scale C footprint

Batang Ai annual C footprint was calculated as the sumof surface diffusion ebullition degassing and downstreamemissions of CO2 CH4 and N2O considering a greenhousewarming potential of 1 34 and 298 respectively over a 100-year lifetime period (Myhre et al 2013) For each flux path-way annual flux was estimated as the average of the sam-pling campaigns Ecosystem-scale estimate of surface diffu-sion was calculated for each campaign as the average of mea-sured flux rates applied to the reservoir area for N2O and forCO2 and CH4 it was obtained by spatial interpolation of mea-sured fluxes over the reservoir area based on inverse distanceweighting with a power of two (a power of one yields similaraverages CV lt 11 ) using package gstat version 11ndash6 inthe R version 341 software (Pebesma 2004 R Core Team2017) Ebullition at the reservoir scale was calculated as theaverage of measured reservoir ebullition rates applied to thelittoral area (lt 3 m deep)

The estimated GHG emissions of Batang Ai based on mea-sured data was compared to values derived from the G-resmodel (UNESCOIHA 2017) and the model presented inBarros et al (2011) Both models predict surface CO2 andCH4 diffusion as a function of age and account for the effectof temperature using different proxies the G-res uses effec-tive temperature while the Barros et al model uses latitude(an indirect proxy that integrates other spatial differences)In terms of CO2 surface diffusion the G-res uses reservoirarea soil C content and TP to quantify the effect of C in-puts fuelling CO2 production while the Barros et al modeluses DOC inputs directly (based on in situ DOC concentra-tion) For CH4 surface diffusion both models account formorphometry using the fraction of littoral area (G-res) or themean depth (Barros et al model) Overall both models pre-dict surface diffusion based on the same conceptual frame-work but use different proxies CH4 ebullition and degassingare modelled only by the G-res being the sole model avail-able to this date Details on model equations and input vari-ables are presented in the Supplement (Tables S2 and S3)

3 Results and discussion

31 Water chemistry

The reservoir is stratified throughout the year with a ther-mocline at a depth around 13 m and mostly anoxic condi-tions in the hypolimnion of the main basin (Fig S1 in theSupplement) The system is oligotrophic with very low con-centrations of DOC TP TN and Chl a averaging 09 (SDplusmn02) mgLminus1 59 (SDplusmn24) microgLminus1 011 (SDplusmn004) mgLminus1and 13 (SDplusmn 07) microgLminus1 respectively (Table S1) and highwater transparency (Secchi depth gt 5 m) In the reservoirinflows concentrations of measured chemical species areslightly higher but still in the oligotrophic range (Table S1)

however the transparency is much lower due to turbidity(Secchi lt 05 m) The oligotrophic status of the reservoirlikely results from nutrient-poor soils (Wasli et al 2011)and a largely undisturbed forested catchment in the protectedBatang Ai National Park The reservoirrsquos low Chl a concen-trations are comparable to the neighbouring Bakun reservoir(Ling et al 2017) and its DOC concentrations are on the lowend of the wide range of measured values in nearby rivers(Martin et al 2018)

32 Surface diffusion

Measured CO2 diffusion in the reservoir averaged 77 (SDplusmn182) mmolmminus2 dminus1 (Table S1) which is on the low endcompared to other reservoirs (Deemer et al 2016) and evento natural lakes (Sobek et al 2005) but similar to CO2 fluxesmeasured in two reservoirs in Laos (Chanudet et al 2011)CO2 diffusion across all sites ranged from substantial up-take to high emissions (from minus308 to 5939 mmolmminus2 dminus1Table S1) reflecting a large spatial and temporal variabil-ity Spatially CO2 fluxes measured in the main basin andbranches had similar averages of 79 and 73 mmolmminus2 dminus1

respectively (overall SDplusmn182) contrasting with higher andmore variable values in the inflows with a mean of 1373(SDplusmn 1924) mmolmminus2 dminus1 (Fig 2) Within the reservoirCO2 fluxes varied (SDplusmn182 mmolmminus2 dminus1) but did not fol-low a consistent pattern and might reflect pre-flooding land-scape heterogeneity (Teodoru et al 2011) Temporally high-est average reservoir CO2 fluxes were measured in AprilndashMay 2017 when no CO2 uptake was observed contraryto other campaigns especially FebruaryndashMarch and Au-gust 2018 when CO2 uptake was common (Fig S2) andaverage Chl a concentrations were the highest This reflectsthe important role of metabolism (namely CO2 consumptionby primary production) in modulating surface CO2 fluxes inBatang Ai

All CH4 surface diffusion measurements were positive andranged from 003 to 1134 mmolmminus2 dminus1 (Table S1) Spa-tially CH4 fluxes were progressively higher moving furtherupstream (Figs 2 and S3) with decreasing water depth andincreasing connection to the littoral This gradient in mor-phometry induces an increasingly greater contact of the waterwith bottom and littoral sediments where CH4 is producedexplaining the spatial pattern of CH4 fluxes CH4 surface dif-fusion also varied temporally but to a lesser extent than CO2being on average highest in August 2018 in the reservoir andin NovemberndashDecember 2016 in the inflows

Reservoir N2O surface diffusion (measured with a limitedspatial resolution) averaged minus02 (SDplusmn 21) nmolmminus2 dminus1

(Table S1) The negative value indicates that the systemacts as a slight net sink of N2O absorbing an estimated21 gCO2 eqmminus2 yrminus1 (Table 2) Atmospheric N2O uptakehas previously been reported in aquatic systems and linkedto low oxygen and nitrogen content conducive to completedenitrification which consumes N2O (Soued et al 2016

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 521

Webb et al 2019) These environmental conditions matchobservations in Batang Ai with a low average TN concentra-tion of 011 (004) mgLminus1 (Table S1) and anoxic deep waters(Fig S1)

33 Ebullition

We calculated that CH4 ebullition rates in Batang Airsquos lit-toral area ranged from 002 to 084 mmolmminus2 dminus1 whichcontrasts with rates measured in its inflows which are sev-eral orders of magnitude higher (52 to 103 mmolmminus2 dminus1)Similar patterns were observed in other reservoirs where in-flow arms where bubbling hot spots due to a higher organicC supply driven by terrestrial matter deposition (DelSontro etal 2011 Grinham et al 2018) Since ebullition rates are no-toriously heterogeneous and were measured at only four sitesin the reservoir they may not reflect ecosystem-scale ratesHowever our attempt to manually provoke ebullition at sev-eral other sites (by physically disturbing the sediments) didnot result in any bubble release confirming the low potentialfor ebullition in the reservoir littoral zone Moreover we cal-culated that fine-grained sediment accumulation is unlikelyat depths shallower than 97 m (estimated EBD) in Batang AiThis combined with the reservoir steep slope prevents thesustained accumulation of organic material in littoral zones(Blais and Kalff 1995) hence decreasing the potential forCH4 production and bubbling there Also apparent littoralsediment composition in the reservoir and dense clay withlow porosity may further hinder bubble formation and emis-sion (de Mello et al 2018)

34 Degassing and downstream emissions

Emissions downstream of the dam expressed on a reservoir-wide areal basis ranged from 193 to 300 mmolmminus2 dminus1

for CO2 and from 59 to 138 mmolmminus2 dminus1 for CH4 (Ta-ble 1) The amount of CO2 exiting the reservoir varied littlebetween sampling campaigns (CV= 3 ) contrary to CH4(CV= 28 Table 1 and Fig 3) Higher temporal variabilityof CH4 concentration in discharged water is likely modulatedby microbial CH4 oxidation in the reservoir water columnupstream of the dam Evidence of high CH4 oxidation is ap-parent in reservoir water column profiles showing a sharpdecline of CH4 concentration and increase in δ13CH4 rightaround the water withdrawal depth (Fig S1) This verticalpattern results from higher oxygen availability when movingup in the hypolimnion (Fig S1) promoting CH4 oxidation atshallower depths

Once GHGs have exited the reservoir a large fraction(40 and 65 for CO2 and CH4 respectively) is immedi-ately lost to the atmosphere as degassing emissions (Table 1)which is in line with previous literature reports (Kemenes etal 2016) Along the outflow river CO2 and CH4 concen-trations gradually decreased δ13CO2 remained stable andδ13CH4 steadily increased (Fig 3) Given the very small

isotopic fractionation (09992) of CH4 during gas evasion(Knox et al 1992) the only process that can explain theobserved δ13CH4 increase is CH4 oxidation (Bastviken etal 2002 Thottathil et al 2018) We estimated that river-ine CH4 oxidation ranged from 038 to 180 mmolmminus2 dminus1

(expressed per squared metre of reservoir area for compari-son) transforming 18 to 32 (depending on the samplingcampaign) of the CH4 to CO2 within the first 19 km of theoutflow Riverine oxidation rates did not co-vary temporallywith water temperature oxygen availability or CH4 concen-trations (known as typical drivers Thottathil et al 2019)hence they might be regulated by other factors like lightand microbial assemblages (Murase and Sugimoto 2005Oswald et al 2015) Overall riverine oxidation of CH4 toCO2 (which has a 34 times lower warming potential) re-duced radiative forcing of downstream emissions (excludingdegassing) by on average 21 and the total annual reser-voir C footprint by 7 Despite having a measurable impacton reservoir GHG emissions CH4 oxidation downstream ofdams was only considered in three other reservoirs to ourknowledge (DelSontro et al 2016 Gueacuterin and Abril 2007Kemenes et al 2007) Accounting for this process is par-ticularly important in systems where downstream emissionsare large which is a common situation in tropical reservoirs(Demarty and Bastien 2011) While additional data on thesubject are needed our results provide one of the first basisfor understanding CH4 oxidation downstream of dams andeventually integrating this component to global models (fromwhich it is currently absent)

35 Importance of sampling resolution

High spatial and temporal sampling resolution have been re-cently highlighted as an important but often lacking aspectof reservoir C footprint assessments (Deemer et al 2016Paranaiacuteba et al 2018) Reservoir-scale fluxes are usuallyderived from applying an average of limited flux measure-ments to the entire reservoir area For Batang Ai this methodoverestimates CO2 and CH4 surface diffusion by 14 (130 gCO2 eqmminus2 yrminus1) and 64 (251 gCO2 eqmminus2 yrminus1)respectively compared to spatial interpolation This is dueto the effect of extreme values that are very constrained inspace but have a disproportionate effect on the overall fluxaverage Also reducing temporal sampling resolution to onecampaign instead of four changes the reservoir C footprintestimate by up to 33 An additional source of uncertaintyin reservoir flux estimates is the definition of a baselinevalue representing natural river emissions in order to cal-culate downstream emissions of excess gas in the outflowattributable to damming In Batang Ai downstream emis-sion was estimated assuming the GHG concentration 19 kmdownstream of the dam is a representative baseline for theoutflow however measured values in the pre-impoundedriver would have substantially reduced the estimate uncer-tainty Results from Batang Ai reinforce the importance of

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

522 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

pre- and post-impoundment sampling resolution and upscal-ing methods in annual reservoir-scale GHG flux estimates

36 Reservoir C footprint and potential mitigation

Most of the Batang Ai emissions occur downstream of thedam through degassing (642 ) and downstream emissions(250 ) while surface diffusion contributed only 106 and ebullition 014 (Table 2) In all pathways radiativepotential of CH4 fluxes was higher than CO2 and N2O (es-pecially for degassing) accounting for 790 of Batang AiCO2 eq emissions This distribution of the flux can be at-tributed mostly to the accumulation of large quantities ofCH4 in the hypolimnion combined with the fact that thewithdrawal depth is located within this layer allowing theaccumulated gas to escape to the atmosphere Previous stud-ies on reservoirs with similar characteristics to Batang Ai(tropical climate with a permanent thermal stratification anddeep water withdrawal) have also found degassing and down-stream emissions to be the major emission pathways espe-cially for CH4 (Galy-Lacaux et al 1997 Kemenes et al2007)

Overall we estimated that the reservoir emits on aver-age 2475 (plusmn327) g CO2 eqmminus2 yrminus1 which corresponds to0169 TgCO2 eqyrminus1 over the whole system In comparisonthe annual areal emission rate (diffusion and ebullition) ofthe inflows based on a more limited sampling resolutionis estimated to range from 108 to 525 kgCO2 eqmminus2 yrminus1mainly due to extremely high ebullition When applied to theapproximated surface area of the river before impoundment(152 km2) this rate translates to 0016ndash0080 TgCO2 eq (Ta-ble 2) assuming similar flux rates in the current inflows andpre-impoundment river While the emission rate of the riverper unit of area is an order of magnitude higher than for thereservoir its estimated total flux remains 21 to 106 timeslower due to a much smaller surface Higher riverine emis-sions rates are probably due to a shallower depth and higherinputs of terrestrial organic matter both conducive to CO2and CH4 production and ebullition Changing the landscapehydrology to a reservoir drastically reduced areal flux ratesespecially ebullition however it widely expanded the vol-ume of anoxic environments (sediments and hypolimnion)creating a vast new space for CH4 production The new hy-drological regime also created an opportunity for the largequantities of gas produced in deep layers to easily escape tothe atmosphere through the outflow and downstream river

One way to reduce reservoir GHG emissions is to ensurelow CO2 and CH4 concentrations at the water withdrawaldepth In Batang Ai maximum CO2 and CH4 concentra-tions are found in the reservoir deep layers and rapidly de-crease from 20 to 10 m for CO2 and from 25 to 15 m for CH4(Fig S1) This pattern is commonly found in lakes and reser-voirs and results from thermal stratification and biologicalprocesses (aerobic respiration and CH4 oxidation) Knowingthis concentration profile degassing and downstream emis-

Table2E

stimated

reservoirandinflow

arealandtotalG

HG

fluxesto

theatm

osphere(plusmn

standarderrorform

easuredvaluesor95

confidence

intervalbasedon

modelstandard

errorforG

-resvalues)from

differentpathways

basedon

measured

andm

odelledapproaches

Diffusion

Ebullition

Degassing

Dow

nstreamriver

Total

CO

2C

H4

N2 O

CH

4C

O2

CH

4C

O2

CH

4

Fluxrate

(gC

O2

eqmminus

2yrminus

1)

Reservoir

Measured

113(plusmn

22)153

(plusmn22)

minus21

(plusmn4)

34(plusmn

19)

247(plusmn

14)1342

(plusmn190)

163(plusmn

9)456

(plusmn65)

2475(plusmn

327)G

-resm

odel577

(509ndash655)161

(132ndash197)N

A52

(32ndash83)N

A468

(266ndash832)N

AN

A1258

(1041ndash1636)B

arrosetalm

odel4671

176N

AN

AN

AN

AN

AN

A4847

Inflows

Measured

156ndash9538248ndash22

510N

A10

377ndash20498

00

00

10781ndash52

546

Totalflux(T

gC

O2

eqyrminus

1)

Reservoir(m

eas)0008

0010minus

00001

000020017

00920011

00310169

River a

0ndash00140ndash0034

NA

0016ndash00310000

00000000

00000016ndash008

NA

notavailable aR

epresentsthe

estimated

pre-impounded

riverfluxesassum

ingthey

were

similarto

currentfluxesfrom

thereservoirinflow

s

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 523

sions could have been reduced in Batang Ai by elevating thewater withdrawal depth to avoid hypolimnetic gas releaseWe calculated that elevating the water withdrawal depth by1 3 and 5 m would result in a reduction of degassing anddownstream emissions by 1 11 and 22 for CO2 andby 28 92 and 100 for CH4 respectively (Fig S4)Consequently a minor change in the dam design could havedrastically reduced Batang Airsquos C footprint This should betaken into consideration in future reservoir construction es-pecially in tropical regions

37 Measured versus modelled fluxes

Based on measurements Batang Ai emits on average 113(plusmn22) gCO2 eqmminus2 yrminus1 via surface CO2 diffusion Thisvalue is 41 times lower than predicted by the Barros etal model (4671 gCO2 eqmminus2 yrminus1 Table 2) based on reser-voir age DOC inputs (derived from DOC water concentra-tion) and latitude (Barros et al 2011) The high predictedvalue for Batang Ai being a relatively old reservoir with verylow DOC concentration is mainly driven by its low latitudeWhile reservoirs in low latitudes globally have higher aver-age CO2 fluxes due to higher temperature and often denseflooded biomass (Barros et al 2011 St Louis et al 2000)our results provide a clear example that not all tropical reser-voirs have high CO2 emissions by simple virtue of their ge-ographical location Despite high temperature Batang Airsquosvery low water organic matter content (Table S1) offers lit-tle substrate for net heterotrophy and its strong permanentstratification creates a physical barrier potentially retainingCO2 derived from flooded biomass in the hypolimnion Theonly three other sampled reservoirs in Southeast Asia (NamLeuk and Nam Ngum in Laos and Palasari in Indonesia)also exhibited low organic C concentration (for reservoirs inLaos) and low to negative average surface CO2 diffusion de-spite their low latitude (Chanudet et al 2011 Macklin et al2018) This suggests that while additional data are neededlow CO2 diffusion may be common in Southeast Asian reser-voirs and is likely linked to the low organic C content

In comparison the G-res model predicts a CO2 sur-face diffusion of 577 (509ndash655) gCO2 eqmminus2 yrminus1 whichincludes the flux naturally sustained by catchment C in-puts (397 gCO2 eq mminus2 yrminus1 predicted flux 100 years afterflooding) and the flux derived from organic matter flooding(180 gCO2 eqmminus2 yrminus1) While the predicted G-res value ismuch closer than that predicted from the Barros et al modelit still overestimates measured flux mostly the natural base-line (catchment derived) part of it The G-res predicts base-line CO2 effluxes as a function of soil C content a proxyfor C input to the reservoir While Batang Ai soil is richin organic C (sim 50 gkgminus1) it also has a high clay content(gt 40 ) (ISRIC ndash World Soil Information 2019 Wasli etal 2011) which is known to bind with organic matter and re-duce its leaching to the aquatic environment (Oades 1988)This may explain the unusually low DOC concentration in

the reservoir and its inflows (03 to 18 mgLminus1 Table S1)that are among the lowest reported in freshwaters globally(Sobek et al 2007) Clay-rich soils are ubiquitous in tropi-cal landscapes (especially in Southeast Asia Central Amer-ica and central and eastern Africa) (ISRIC ndash World Soil In-formation 2019) however their impact on global-scale pat-terns of aquatic DOC remains unknown This may be due toa lack of aquatic DOC data with the most recently publishedglobal study on the subject featuring only one tropical systemand a heavy bias towards North America and Europe (Sobeket al 2007) Exploring the global-scale picture of aquaticDOC and its link to watershed soils characteristics would bea significant step forward in the modelling of reservoir CO2diffusion Indeed had the G-res model been able to capturethe baseline emissions more correctly in Batang Ai (closeto zero given the very low DOC inputs) predictions wouldhave nearly matched observations Finally note that the G-res model is not suitable to predict CO2 uptake which wasobserved in 32 of flux measurements in Batang Ai due toan occasionally net autotrophic surface metabolism favouredunder low C inputs (Bogard and del Giorgio 2016) Improv-ing this aspect of the model depends on the capacity to pre-dict internal metabolism of aquatic systems at a global scalewhich is currently lacking Overall reservoir CO2 diffusionmodels may be less performant in certain regions like South-east Asia due to an uneven spatial sampling distribution anda general lack of knowledge and data on C cycling in someparts of the world

Our field-based estimate of Batang Ai CH4 surface diffu-sion is 153 (plusmn22) gCO2 eqmminus2 yrminus1 which differs by only5 and 15 from the G-res and Barros et al modelledpredictions of 161 (132ndash197) and 176 gCO2 eqmminus2 yrminus1 re-spectively (Table 2) Both models use as predictors age aproxy for water temperature (air temperature or latitude)and an indicator of reservoir morphometry ( littoral areaor mean depth) and Barros et al (2011) also uses DOC in-put (Table S3) Similar predictors were identified in a recentglobal literature analysis which also pointed out the role oftrophic state in CH4 diffusion with Batang Ai falling wellin the range of flux reported in other oligotrophic reservoirs(Deemer et al 2016) Overall our results show that globalmodelling frameworks for CH4 surface diffusion capture thereality of Batang Ai reasonably well

Measured estimate of reservoir-scale CH4 ebullition aver-aged 34 (plusmn19) gCO2 eqmminus2 yrminus1 (Table 2) which is oneof the lowest reported globally in reservoirs (Deemer etal 2016) and is an order of magnitude lower than the 52(32ndash83) gCO2 eqmminus2 yrminus1 predicted by the G-res model (theonly available model for reservoir ebullition) This contrastswith the perception that tropical reservoirs consistently havehigh ebullitive emissions and supports the idea that the sup-ply of sediment organic matter rather than temperature is theprimary driver of ebullition (Grinham et al 2018) Batang Aisediment properties and focusing patterns mentioned earliercould explain the model overestimation of CH4 ebullition

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

524 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

The G-res model considers the fraction of littoral area andhorizontal radiance (a proxy for heat input) as predictors ofebullition rate but does not integrate other catchment prop-erties Building a stronger mechanistic understanding of theeffect of sediment composition and accumulation patterns onCH4 bubbling may improve our ability to more accuratelypredict reservoir ebullition flux

Our empirical estimate shows that 409 (plusmn23) and 1798(plusmn255) gCO2 eqmminus2 yrminus1 are emitted as CO2 and CH4 re-spectively downstream of the dam (including degassing)accounting for 89 of Batang Ai GHG emissions (Ta-ble 2) Currently there is no available model predicting down-stream GHG emissions from reservoirs except the G-resmodel which is able to predict only the CH4 degassingpart of this flux Modelled CH4 degassing in Batang Ai is468 (266ndash832) gCO2 eqmminus2 yrminus1 compared to an estimated1342 (plusmn190) gCO2 eqmminus2 yrminus1 based on our measurementsPredictive variables used to model CH4 degassing are mod-elled CH4 surface diffusion (based on littoral area and tem-perature) and water retention time (Table S3) In Batang Aimain basin the strong and permanent stratification favoursoxygen depletion in the hypolimnion which promotes deepCH4 accumulation combined with a decoupling between sur-face and deep water layers The model relies strongly on sur-face CH4 patterns to predict excess CH4 in the deep layerwhich could explain why it underestimates CH4 degassingin Batang Ai Similar strong stratification patterns are ubiq-uitous in the tropics with a recent study suggesting a largemajority of tropical reservoirs are monomictic or oligomic-tic (Lehmusluoto et al 1997 Winton et al 2019) and hencemore often stratified than temperate and boreal ones Thissuggests that CH4 degassing is potentially more frequentlyunderestimated in low-latitude reservoirs The G-res effortto predict CH4 degassing is much needed given the impor-tance of this pathway and the next step would be to refinethis model and develop predictions for other currently miss-ing fluxes like CO2 degassing and downstream emissions inthe outflow Our results suggest that improving latter aspectsrequires a better capacity to predict GHG accumulation indeep reservoirs layers across a wide range of stratificationregimes

4 Conclusions

The comprehensive GHG portrait of Batang Ai highlightsthe importance of spatial and temporal sampling resolutionand the inclusion of all flux components in reservoir GHGassessments Gas dynamics downstream of the dam (de-gassing outflow emissions and CH4 oxidation) commonlynot assessed in reservoir GHG studies are major elements inBatang Ai We suggest that these emissions could have beengreatly diminished with a minor change to the dam design(shallower water withdrawal) Mitigating GHG emissionsfrom future reservoirs depends on the capacity to predict

GHG fluxes from all pathways In this regard the comparisonbetween Batang Ai measured and modelled GHG flux esti-mates allowed us to identify knowledge gaps based on whichwe propose the four following research avenues (1) Refinethe modelling of reservoir CO2 diffusion by studying its linkwith metabolism and organic matter leaching from differentsoil types (2) Examine the potential for CH4 ebullition inlittoral zones in relation to patterns of organic matter sedi-mentation linked to morphometry (3) Improve the modellingof CH4 degassing by better defining drivers of hypolimneticCH4 accumulation namely thermal stratification (4) Gatheradditional field data on GHG dynamics downstream of dams(degassing river emissions and river CH4 oxidation) in or-der to incorporate all components of the flux to the modellingof reservoir C footprint

Data availability The dataset related to this article is availableonline through Zenodo at httpsdoiorg105281zenodo3629898(Soued and Prairie 2020)

Supplement The supplement related to this article is available on-line at httpsdoiorg105194bg-17-515-2020-supplement

Author contributions CS contributed to the conceptualizationmethodology validation formal analysis investigation data cura-tion writing of the original draft review and editing and projectadministration YTP contributed to the methodology validation in-vestigation resources review and editing supervision and fundingacquisition

Competing interests The authors declare that they have no conflictof interest

Acknowledgements This is a contribution to the UNESCO chairin Global Environmental Change and the GRIL We are gratefulto Karen Lee Suan Ping and Jenny Choo Cheng Yi for their lo-gistic support and participation in sampling campaigns We alsothank Jessica Fong Fung Yee Amar Marsquoaruf Bin Ismawi Ger-ald Tawie Anak Thomas Hilton Bin John Paula Reis Sara Mercier-Blais and Karelle Desrosiers for their help on the field as well asKatherine Velghe and Marilyne Robidoux for their assistance dur-ing laboratory analyses

Financial support This research has been supported by the NaturalSciences and Engineering Research Council of Canada (DiscoveryGrant to Yves T Prairie and BES-D scholarship to Cynthia Soued)and Sarawak Energy Berhad (research and development)

Review statement This paper was edited by Ji-Hyung Park and re-viewed by one anonymous referee

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 525

References

Barros N Cole J J Tranvik L J Prairie Y T BastvikenD Huszar V L M del Giorgio P and Roland FCarbon emission from hydroelectric reservoirs linkedto reservoir age and latitude Nat Geosci 4 593ndash596httpsdoiorg101038ngeo1211 2011

Bastien J and Demarty M Spatio-temporal variation of grossCO2 and CH4 diffusive emissions from Australian reservoirsand natural aquatic ecosystems and estimation of net reser-voir emissions Lakes Reserv Res Manag 18 115ndash127httpsdoiorg101111lre12028 2013

Bastviken D Ejlertsson J and Tranvik L Measure-ment of methane oxidation in lakes A comparisonof methods Environ Sci Technol 36 3354ndash3361httpsdoiorg101021es010311p 2002

Bastviken D Tranvik L J Downing J A Crill P Mand Enrich-Prast A Freshwater Methane Emissions Off-set the Continental Carbon Sink Science 331 50ndash50httpsdoiorg101126science1196808 2011

Beaulieu J J McManus M G and Nietch C T Esti-mates of reservoir methane emissions based on a spatially bal-anced probabilistic-survey Limnol Oceanogr 61 S27ndashS40httpsdoiorg101002lno10284 2016

Blais J M and Kalff J The influence of lake morphome-try on sediment focusing Limnol Oceanogr 40 582ndash588httpsdoiorg104319lo19954030582 1995

Bogard M J and del Giorgio P A The role of metabolism inmodulating CO2 fluxes in boreal lakes Global BiogeochemCycles 30 1509ndash1525 httpsdoiorg1010022016GB0054632016

Chanudet V Descloux S Harby A Sundt H Hansen B HBrakstad O Serccedila D and Guerin F Gross CO2 and CH4emissions from the Nam Ngum and Nam Leuk sub-tropicalreservoirs in Lao PDR Sci Total Environ 409 5382ndash5391httpsdoiorg101016jscitotenv201109018 2011

Cole J J and Caraco N F Atmospheric exchange ofcarbon dioxide in a low-wind oligotrophic lake measuredby the addition of SF6 Limnol Oceanogr 43 647ndash656httpsdoiorg104319lo19984340647 1998

Coleman D D Risatti J B and Schoell M Frac-tionation of carbon and hydrogen isotopes by methane-oxidizing bacteria Geochim Cosmochim Ac 45 1033ndash1037httpsdoiorg1010160016-7037(81)90129-0 1981

Deemer B R Harrison J A Li S Beaulieu J J DelSontro TBarros N Bezerra-Neto J F Powers S M dos Santos MA and Vonk J A Greenhouse Gas Emissions from ReservoirWater Surfaces A New Global Synthesis Bioscience 66 949ndash964 httpsdoiorg101093bioscibiw117 2016

DelSontro T McGinnis D F Sobek S Ostrovsky I and WehrliB Extreme Methane Emissions from a Swiss HydropowerReservoir Contribution from Bubbling Sediments Environ SciTechnol 44 2419ndash2425 httpsdoiorg101021es90313692010

DelSontro T Kunz M J Kempter T Wuumlest A Wehrli B andSenn D B Spatial Heterogeneity of Methane Ebullition in aLarge Tropical Reservoir Environ Sci Technol 45 9866ndash9873httpsdoiorg101021es2005545 2011

DelSontro T Perez K K Sollberger S and WehrliB Methane dynamics downstream of a temperate run-

of-the-river reservoir Limnol Oceanogr 61 S188ndashS203httpsdoiorg101002lno10387 2016

Demarty M and Bastien J GHG emissions from hydroelectricreservoirs in tropical and equatorial regions Review of 20 yearsof CH4 emission measurements Energy Policy 39 4197ndash4206httpsdoiorg101016jenpol201104033 2011

de Mello N A S T Brighenti L S Barbosa F A R StaehrP A and Bezerra Neto J F Spatial variability of methane(CH4) ebullition in a tropical hypereutrophic reservoir silted ar-eas as a bubble hot spot Lake Reserv Manag 34 105ndash114httpsdoiorg1010801040238120171390018 2018

Galy-Lacaux C Delmas R Jambert C Dumestre J-FLabroue L Richard S and Gosse P Gaseous emissionsand oxygen consumption in hydroelectric dams A case studyin French Guyana Global Biogeochem Cycles 11 471ndash483httpsdoiorg10102997GB01625 1997

Grinham A Dunbabin M and Albert S Importance of sed-iment organic matter to methane ebullition in a sub-tropicalfreshwater reservoir Sci Total Environ 621 1199ndash1207httpsdoiorg101016jscitotenv201710108 2018

Gueacuterin F and Abril G Significance of pelagic aerobicmethane oxidation in the methane and carbon budget ofa tropical reservoir J Geophys Res-Biogeo 112 G3httpsdoiorg1010292006JG000393 2007

Gueacuterin F Abril G Richard S Burban B Reynouard C SeylerP and Delmas R Methane and carbon dioxide emissions fromtropical reservoirs Significance of downstream rivers GeophysRes Lett 33 L21407 httpsdoiorg1010292006GL0279292006

International Hydropower Association (IHA) A brief his-tory of hydropower available at httpswwwhydropowerorga-brief-history-of-hydropower (last access 11 July 2019) 2015

ISRIC ndash World Soil Information SoilGrids v053 availableat httpssoilgridsorglayer=ORCDRC_M_sl4_250mampvector=1 last access 1 May 2019

Keller M and Stallard R F Methane emission by bub-bling from Gatun Lake Panama J Geophys Res 99 8307httpsdoiorg10102992JD02170 1994

Kemenes A Forsberg B R and Melack J M Methane releasebelow a tropical hydroelectric dam Geophys Res Lett 34 1ndash5httpsdoiorg1010292007GL029479 2007

Kemenes A Forsberg B R and Melack J M Downstreamemissions of CH4 and CO2 from hydroelectric reservoirs (Tucu-ruiacute Samuel and Curuaacute-Una) in the Amazon basin Inl Waters6 295ndash302 httpsdoiorg101080IW-63980 2016

Knox M Quay P D and Wilbur D Kinetic isotopic fraction-ation during air-water gas transfer of O2 N2 CH4 and H2 JGeophys Res 97 20335 httpsdoiorg10102992JC009491992

Ledwell J J The Variation of the Gas Transfer Coefficient withMolecular Diffusity in Gas Transfer at Water Surfaces pp 293ndash302 Springer Netherlands Dordrecht 1984

Lehmusluoto P Machbub B Terangna N Rusmiputro SAchmad F Boer L Brahmana S S Priadi B Se-tiadji B Sayuman O and Margana A National in-ventory of the major lakes and reservoirs in Indone-sia available at httpwwwkolumbusfipasilehmusluoto210_expedition_indodanau_report1997PDF (last access 8 Novem-ber 2019) 1997

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020

526 C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir

Lehner B Liermann C R Revenga C Voumlroumlsmarty C FeketeB Crouzet P Doumlll P Endejan M Frenken K Magome JNilsson C Robertson J C Roumldel R Sindorf N and WisserD High-resolution mapping of the worldrsquos reservoirs and damsfor sustainable river-flow management Front Ecol Environ 9494ndash502 httpsdoiorg101890100125 2011

Lide D CRC Handbook of Chemistry and Physics edited byCRC Press Boca Raton FL available at httpwebdelprofesorulavecienciasisoldalibroshandbookpdf (last access 30 Au-gust 2019) 2005

Ling T Y Gerunsin N Soo C L Nyanti L Sim S F andGrinang J Seasonal changes and spatial variation in water qual-ity of a large young tropical reservoir and its downstream riverJ Chem 2017 8153246 httpsdoiorg101155201781532462017

Macklin P A Gusti Ngurah Agung Suryaputra I Maher D Tand Santos I R Carbon dioxide dynamics in a lake and a reser-voir on a tropical island (Bali Indonesia) PLoS One 13 6httpsdoiorg101371journalpone0198678 2018

Martin P Cherukuru N Tan A S Y Sanwlani N MujahidA and Muumlller M Distribution and cycling of terrigenous dis-solved organic carbon in peatland-draining rivers and coastalwaters of Sarawak Borneo Biogeosciences 15 6847ndash6865httpsdoiorg105194bg-15-6847-2018 2018

Murase J and Sugimoto A Inhibitory effect of light onmethane oxidation in the pelagic water column of a mesotrophiclake (Lake Biwa Japan) Limnol Oceanogr 50 1339ndash1343httpsdoiorg104319lo20055041339 2005

Myhre G Shindell D Breacuteon F-M Collins W FuglestvedtJ Huang J Koch D Lamarque J-F Lee D Men-doza B Nakajima T Robock A Stephens G Take-mura T and Zhang H Anthropogenic and Natural Radia-tive Forcing in Climate Change 2013 The Physical Sci-ence Basis Contribution of Working Group I to the FifthAssessment Report of the Intergovernmental Panel on Cli-mate Change pp 659ndash740 Cambridge United Kindomavailable at httpwwwclimatechange2013orgimagesreportWG1AR5_Chapter08_FINALpdf (last access 30 August 2019)2013

Oades J M The retention of organic matter in soils Biogeochem-istry 5 35ndash70 httpsdoiorg101007BF02180317 1988

Oswald K Milucka J Brand A Littmann S Wehrli BKuypers M M M and Schubert C J Light-DependentAerobic Methane Oxidation Reduces Methane Emissionsfrom Seasonally Stratified Lakes PLoS One 10 e0132574httpsdoiorg101371journalpone0132574 2015

Paranaiacuteba J R Barros N Mendonccedila R Linkhorst A IsidorovaA Roland F Almeida R M and Sobek S Spatially Re-solved Measurements of CO2 and CH4 Concentration andGas-Exchange Velocity Highly Influence Carbon-Emission Es-timates of Reservoirs Environ Sci Technol 52 607ndash615httpsdoiorg101021acsest7b05138 2018

Patton C and Kryskalla J Methods of analysis by the US Geo-logical Survey National Water Quality Laboratory evaluation ofalkaline persulfate digestion as an alternative to Kjeldahl diges-tion for determination of total and dissolved nitrogen and phos-phorus in water US Government Printing Office Denver Col-orado 2003

Pebesma E J Multivariable geostatistics in S thegstat package Comput Geosci 30 683ndash691httpsdoiorg101016jcageo200403012 2004

Prairie Y T Alm J Beaulieu J Barros N Battin T ColeJ del Giorgio P DelSontro T Gueacuterin F Harby A Har-rison J Mercier-Blais S Serccedila D Sobek S and VachonD Greenhouse Gas Emissions from Freshwater ReservoirsWhat Does the Atmosphere See Ecosystems 21 1058ndash1071httpsdoiorg101007s10021-017-0198-9 2018

R Core Team R A language and environment for statistical com-puting available at httpswwwr-projectorg (last access 9 De-cember 2019) 2017

Rowan D J Kalff J and Rasmussen J B Estimating the MudDeposition Boundary Depth in Lakes from Wave Theory CanJ Fish Aquat Sci 49 2490ndash2497 httpsdoiorg101139f92-275 1992

Rudd J Harris R and Kelly C A Are Hydroelectric ReservoirsSignficant Sources of Greenhouse Gases Ambio 22 246ndash2481993

Sarawak Government The Geography of Sarawak available athttpswwwsarawakgovmywebhomearticle_view159176last access 3 May 2019

Sartory D P and Grobbelaar J U Extraction ofchlorophyll a from freshwater phytoplankton for spec-trophotometric analysis Hydrobiologia 114 177ndash187httpsdoiorg101007BF00031869 1984

Sherman B and Ford P Methane Emissions from Two reservoirsin a steep sub- Tropical rainforest catchment in Science Forumand Stakeholder Engagement Building Linkages Collaborationand Science Quality edited by Begbie D K and Wakem SL pp 1ndash9 Urban Water Security Research Alliance BrisbaneQueensland 2011

Sobek S Tranvik L J and Cole J J Temperature independenceof carbon dioxide supersaturation in global lakes Global Bio-geochem Cycles 19 2 httpsdoiorg1010292004GB0022642005

Sobek S Tranvik L J Prairie Y T Kortelainen P and Cole JJ Patterns and regulation of dissolved organic carbon An anal-ysis of 7500 widely distributed lakes Limnol Oceanogr 521208ndash1219 httpsdoiorg104319lo20075231208 2007

Soued C and Prairie Y Carbon dioxide methane and chem-ical data from Batang Ai reservoir (Version 100) Zenodohttpsdoiorg105281zenodo3629898 2020

Soued C del Giorgio P A and Maranger R Nitrous oxidesinks and emissions in boreal aquatic networks in Queacutebec NatGeosci 9 116ndash120 httpsdoiorg101038ngeo2611 2016

St Louis V L Kelly C A Duchemin Eacute Rudd JW M and Rosenberg D M Reservoir Surfaces asSources of Greenhouse Gases to the Atmosphere A GlobalEstimate Bioscience 50 766 httpsdoiorg1016410006-3568(2000)050[0766RSASOG]20CO2 2000

Teodoru C R Prairie Y T and del Giorgio P A Spatial Hetero-geneity of Surface CO2 Fluxes in a Newly Created Eastmain-1Reservoir in Northern Quebec Canada Ecosystems 14 28ndash46httpsdoiorg101007s10021-010-9393-7 2011

Teodoru C R Bastien J Bonneville M-C del GiorgioP A Demarty M Garneau M Heacutelie J-F Pelletier LPrairie Y T Roulet N T Strachan I B and Trem-blay A The net carbon footprint of a newly created bo-

Biogeosciences 17 515ndash527 2020 wwwbiogeosciencesnet175152020

C Soued and Y T Prairie Carbon footprint of a Malaysian tropical reservoir 527

real hydroelectric reservoir Global Biogeochem Cycles 26 2httpsdoiorg1010292011GB004187 2012

Thottathil S D Reis P C J del Giorgio P A and Prairie YT The Extent and Regulation of Summer Methane Oxidationin Northern Lakes J Geophys Res-Biogeo 123 3216ndash3230httpsdoiorg1010292018JG004464 2018

Thottathil S D Reis P C J and Prairie Y T Methane oxida-tion kinetics in northern freshwater lakes Biogeochemistry 143105ndash116 httpsdoiorg101007s10533-019-00552-x 2019

UNESCOIHA The GHG Reservoir Tool (G- res) available athttpsg-reshydropowerorg (last access 1 May 2019) 2017

Venkiteswaran J J Schiff S L St Louis V L Matthews C JD Boudreau N M Joyce E M Beaty K G and Bodaly RA Processes affecting greenhouse gas production in experimen-tal boreal reservoirs Global Biogeochem Cycles 27 567ndash577httpsdoiorg101002gbc20046 2013

Wanninkhof R Relationship between wind speed and gasexchange over the ocean J Geophys Res 97 7373httpsdoiorg10102992JC00188 1992

Wasli M E Tanaka S Kendawang J J Abdu A Lat JMorooka Y Long S M and Sakurai K Soils and Vegeta-tion Condition of Natural Forests and Secondary Fallow Forestswithin Batang Ai National Park Boundary Sarawak MalaysiaKuroshio Sci 5 67ndash76 2011

Webb J R Hayes N M Simpson G L Leavitt P RBaulch H M and Finlay K Widespread nitrous oxide un-dersaturation in farm waterbodies creates an unexpected green-house gas sink P Natl Acad Sci USA 116 201820389httpsdoiorg101073pnas1820389116 2019

Wetzel R G and Likens G E Limnological Analyses 3rd edSpringer New York New York NY 2000

Winton R S Calamita E and Wehrli B Reviews and syn-theses Dams water quality and tropical reservoir stratificationBiogeosciences 16 1657ndash1671 httpsdoiorg105194bg-16-1657-2019 2019

Zarfl C Lumsdon A E Berlekamp J Tydecks L and TocknerK A global boom in hydropower dam construction Aquat Sci77 161ndash170 httpsdoiorg101007s00027-014-0377-0 2015

wwwbiogeosciencesnet175152020 Biogeosciences 17 515ndash527 2020