Factors controlling water-column respiration in rivers of the central and

southwestern Amazon Basin

Erin E. Ellis,a,* Jeffrey E. Richey,a Anthony K. Aufdenkampe,b Alex V. Krusche,c Paul D. Quay,a

Cleber Salimon,d and Hilandia Brandao da Cunhae

a School of Oceanography, University of Washington, Seattle, WashingtonbStroud Water Research Center, Avondale, Pennsylvaniac Centro de Energia Nuclear na Agricultura, Universidade de Sao Paulo, Piracicaba, Sao Paulo, BrazildCentro de Ciencias Biologicas e da Natureza, Universidade Federal do Acre, Rio Branco, Acre, Brazile Coordenacao de Pesquisas em Clima e Recursos Hıdricos, Instituto Nacional de Pesquisas da Amazonia, Manaus, Amazonas, Brazil

Abstract

We examined the factors controlling the variability in water-column respiration rates in Amazonian rivers. Ourobjectives were to determine the relationship between respiration rates and the in situ concentrations of the sizeclasses of organic carbon (OC), and the biological source (C3 and C4 plants and phytoplankton) of organic matter(OM) supporting respiration. Respiration was measured along with OC size fractions and dissolved oxygen isotopes(d18O-O2) in rivers of the central and southwestern Amazon Basin. Rates ranged from 0.034 mmol O2 L21 h21 to1.78 mmol O2 L21 h21, and were four-fold higher in rivers with evidence of photosynthetic production(demonstrated by d18O-O2 , 24.2%) as compared to rivers lacking such evidence (d18O-O2 . 24.2%; 1.35 6 0.22 vs.0.30 6 0.29 mmol L21 h21). Rates were likely elevated in the former rivers, which were all sampled during low water,due to the stimulation of heterotrophic respiration via the supply of a labile, algal-derived substrate and/or theoccurrence of autotrophic respiration. The organic composition of fine particulate OM (FPOM) of these rivers isconsistent with a phytoplankton origin. Multiple linear regression analysis indicates that [FPOC], C : NFPOC ratios,and [O2] account for a high amount of the variability in respiration rates (r2 5 0.80). Accordingly, FPOC derivedfrom algal sources is associated with elevated respiration rates. The d13C of respiration-derived CO2 indicates thatthe role of phytoplankton, C3 plants, and C4 grasses in supporting respiration is temporally and spatially variable.Future scaling work is needed to evaluate the significance of phytoplankton production to basin-wide carboncycling.

Nearly all rivers and lakes across both temperate andtropical ecosystems are net sources of carbon dioxide (CO2)to the atmosphere (Cole et al. 1994; Richey et al. 2002;Duarte and Prairie 2005). In situ respiration of organicmatter (OM) derived from upland ecosystems appears to bethe primary source of CO2 supersaturation in most of thesewaters (Richey et al. 2002; Mayorga et al. 2005a; Battinet al. 2008). Although aquatic ecosystems represent a smallfraction of the earth’s surface, over long time periods theircarbon fluxes are often larger than those of the surroundingterrestrial ecosystem, giving them the ability to affect re-gional carbon budgets (Cole et al. 2007). In some cases, gasevasion from rivers can exceed dissolved and particulatecarbon export to the ocean by an order of magnitude(Richey et al. 2002). Accordingly, it is essential to quantifyrespiration rates and to understand their variability giventhe importance of respiration in supporting gas evasionfluxes and resolving regional carbon balances.

Tropical rivers are of particular interest to biogeochemistsbecause their areal outgassing rates typically exceed theirtemperate counterparts, with rivers and streams havinghigher rates than lakes and wetlands in both biomes(Aufdenkampe et al. 2011). The most well-studied of thetropical river basins is the Amazon, where lowland riversand streams are highly supersaturated with CO2, as reflected

by partial pressures (PCO2) ranging from 57 Pa to 3399 Pa(565–33,550 matm [Mayorga 2004; Mayorga et al. 2005a]).Approximately 0.5 Pg C yr21 is lost to the atmospherethrough outgassing throughout the basin, an amountroughly equal to the total terrestrial sequestration of carbonin the Amazon Basin (Richey et al. 2002). Respiration andgas exchange control the seasonal distribution of dissolvedoxygen (O2) and CO2 in Amazonian rivers (Quay et al. 1992;Devol et al. 1995). Specifically, CO2 supersaturation hasbeen hypothesized to be sustained by water-column respira-tion of material derived from both autochthonous andallochthonous sources, in addition to CO2 dissolution fromroot respiration in forest soils, and floating and emergentmacrophyte respiration (Richey et al. 1990, 2002). Despitethe potential importance of in situ respiration in fueling gasevasion, little is known about the mechanisms behind thehigh variability of water-column respiration rates observedacross the Amazon Basin (from 0.08 mmol O2 L21 h21 to5.5 mmol O2 L21 h21 [Richey et al. 1990; Benner et al. 1995;Devol et al. 1995]).

These observations pose the question, what combinationof chemical, physical, and biological factors produces theobserved variability in water-column respiration rates, andfurther, the distribution of CO2 outgassing? The couplingbetween physical size classes of organic carbon and water-column respiration has been an area of significant researchwithin the basin over the last 20 yr; yet, our understanding ofthe relationships between carbon size classes and respiration* Corresponding author: ellise@u.washington.edu

Limnol. Oceanogr., 57(2), 2012, 527–540

E 2012, by the Association for the Sciences of Limnology and Oceanography, Inc.doi:10.4319/lo.2012.57.2.0527

527

remains unresolved. Measurements of the organic compo-sition of coarse and fine particulate organic carbon (OC;CPOC and FPOC, respectively), and dissolved OC (DOC)demonstrate that carbon becomes more degraded as its phys-ical size decreases (Hedges et al. 1994; Devol and Hedges2001). A mass balance of carbon inputs and outputs to theAmazon suggests these carbon pools are not sufficient tobalance the requirements needed to support water-columnrespiration. Instead, it is hypothesized that a rapidly-cyclingunmeasured pool of labile OC likely exists within these bulksize fractions that fuels respiration (Richey et al. 1990). Thislabile pool could exist within the dissolved component ofDOC: elevated bacterial growth and respiration wereobserved during incubations with DOC size fractions .1 kDa as compared to size fractions , 1 kDa in the Solimoes(i.e., the Amazon mainstem above the confluence of theNegro River) and the Negro rivers (Amon and Benner 1996).More recently, the average age of CO2 in equilibrium with dis-solved inorganic carbon (DIC) has been demonstrated to be, 5 yr old, although the age of bulk size fractions (CPOC,FPOC, and DOC) ranges from tens to thousands of years(Mayorga et al. 2005a). Thus, there is a growing consensusthat the bulk material is of limited bioavailability, and instead,a small, labile, young pool of carbon fuels respira-tion (Amon and Benner 1996; Mayorga and Aufdenkampe2002; Mayorga et al. 2005a).

Other research has speculated on the role of differentbiological sources of OM (i.e., C3 and C4 plants, and phy-toplankton) in supporting respiration. Isotopic work suggestsC4 vegetation (mainly grasses within river-corridors andfloodplains) may be an important substrate for respirationwithin the central Amazon Basin (Quay et al. 1992; Engleet al. 2008). For example, the d13C of respiration-derived CO2

(CO2resp) indicates that C4 grasses can support up to 40% of

water-column respiration on the Solimoes River during theearly rising-water stage (Quay et al. 1992) within the centralAmazon Basin. This finding is somewhat surprising becausethe majority of the OM transported by the river has aterrestrial C3 plant signature, and, consequently, this maysuggest that C4 grasses are inherently more biodegradable(Hedges et al. 1986; Quay et al. 1992; Mayorga et al. 2005a).

Outside the Amazon, phytoplankton and periphytonproduction are important sources of labile carbon thataffect respiration rates (Kritzberg et al. 2004; del Giorgioand Pace 2008). Dissolved oxygen isotopes (18O : 16O) indi-cate that photosynthetically produced oxygen is found inAmazonian lakes and rivers (Quay et al. 1995), yet the roleof phytoplankton in supporting respiration remains unin-vestigated. This may be due to the very low photosynthesisrates observed in the sediment-rich waters of the Amazonmainstem (Wissmar et al. 1981), which has led to the per-ception that OC from phytoplankton production is ofminimal importance in rivers throughout the basin.

Accordingly, to address these outstanding research ques-tions key to understanding carbon cycling in the AmazonBasin, this study examines the factors affecting the variabilityin water-column respiration rates across rivers and streamswith contrasting OC concentrations, highlighting a continu-um of river sizes. Specifically, we ask, is there a relationshipbetween the in situ concentrations of the physical size classes

of carbon and water-column respiration rates? Differentbiological sources (e.g., C3 vegetation, C4 vegetation, or phy-toplankton) may affect water-column respiration rates;therefore, we also examine the influence of these sources onrespiration rates. Finally, to improve our understanding ofthe coupling between respiration and basin-wide gas evasion,we assess the ability of water-column respiration to supportCO2 gas evasion in rivers and streams of different water types.

Much of the previous work examining respiration in theAmazon Basin represents the mainstem and lower maintributaries, often in a single snapshot or in controlledexperiments. Although manipulations directly determinethe effects of a particular variable on respiration, suchstudies may be biased due to bottle artifacts and theperturbation of other variables from in situ levels. Incontrast, we examine relationships between respirationrates and the in situ measurement of a given variable. Acaveat of this approach is that some correlations can bespurious, which must be differentiated from causal rela-tionships. Consequently, we present our results in contextwith the chemical, physical, and seasonal differencesbetween the sites that may concurrently affect respirationrates. We employ stable carbon and oxygen (d13C andd18O) isotopes to partition OM among sources and toidentify the relative importance of photosynthesis acrosssites. d13C is used to differentiate between OC originatingfrom C3 plants (, 228%, typically terrestrial), C4 plants(, 212%, typically floodplain macrophytes), and phyto-plankton (, 233% to 237%). Further, the d18O ofdissolved O2 provides information about the relativeimportance of heterotrophic respiration, autochthonousproduction, and gas exchange at each site (Benson andKrause 1984; Quay et al. 1995).

Methods

Study area—Samples were collected during 2005 and2006 in the Brazilian states of Acre and Amazonas (Fig. 1),which are located in southwestern and central Amazonia.Locations were chosen to represent a range of water typesand conditions found in these regions. Rivers in theAmazon Basin differ by both their dissolved and particu-late organic and inorganic chemical contents due to theextent that they drain the Andes, Andean forelands, orhighly weathered lowland shields (Sioli 1984). Due to thepredominance of whitewater rivers in both Amazonas andAcre and the logistical challenges of reaching a variety ofriver types in this large region, we sampled 10 whitewaterrivers of varying sizes. We also sampled two clear waterstreams (one each in Amazonas and Acre), and two black-water rivers in Amazonas. Watershed areas ranged from, 10 km2 to 2,910,510 km2 (Table 1), which werecalculated as in Mayorga et al. (2005b).

Not all sites were in the same stage of the hydrograph atthe time of sampling. All rivers in Acre were in the low-water stage, whereas rivers draining Amazonas were in thefalling-water stage (Table 1). We collected samples from 14sites from July to September 2005, which coincided with asevere drought in the western and southern regions of theAmazon Basin (Zeng et al. 2008). Eight of these sites were

528 Ellis et al.

resampled during August–September 2006 of the followingyear (no drought).

Sample collection—Samples were collected from thethalweg of rivers using a submersible pump placed at 6/10of the total river depth. pH, conductivity, and dissolvedoxygen were measured by immersing field probes (ThermoOrion 290 A+ pH meter, a Chek Mite CD-30 conductivitymeter, and a 55 YSI dissolved oxygen probe) in a contin-uously overflowing graduated cylinder. For small streams,samples were taken directly below the water surface to

minimize disturbance. PCO2was analyzed via headspace

equilibration following the methods of Cole et al. (1994)and modified as in Alin et al. (2010). PCO2

samples wereeither measured immediately using infra-red gas analysisvia a Li-Cor LI-820 (Alin et al. 2010) or stored in glassbottles until analysis with a Shimadzu gas chromatograph(GC-17A equipped with flame ionization and electroncapture detectors and a methanizer).

Size fractionation—Bulk size fractions (CPOC [. 63 mm],FPOC [0.7–63 mm], and DOC [, 0.7mm]) were filtered inthe field, whereas size fractionation of DOC was processedin the laboratory. Coarse suspended sediment (CSS)concentrations were measured by first passing a knownvolume of river water through a 63-mm sieve, and thenlater drying and weighing the sieved material. The mate-rial collected from a plankton net was preserved with HgCl2for later analyses of weight percentages (wt%) of C, N, andd13C of CPOC. CSS concentrations were multiplied by wt%C to determine CPOC concentrations. Sieved river water washomogenized with a churn splitter (Wilde and Radtke 2003)and then filtered, providing the fine suspended sedimentconcentration (FSS) by mass difference (Aufdenkampe et al.2001). Sieved water was also passed through precombustedglass fiber filters (GF/F), which were then analyzed for d13C,wt% of C and N, and FPOC concentrations.

The filtrate of the GF/F filter (defined as DOC) wasstored in precombusted glass vials and immediately pre-served with HgCl2 pending no further analysis. Centrifugeultrafiltration was used to size-fractionate DOC into thefollowing categories: high molecular weight (HMW; . 100kDa), medium molecular weight (MMW; 5–100 kDa), andlow molecular weight (LMW; , 5 kDa) DOC, using amethod modified from Burdige and Gardner (1998). Waterwas filtered through two GF/F filters in the field, placed on

Fig. 1. Map of study sites within the Amazon River Basin.Major tributaries are in grey. The circled sites are located in thestate of Acre, whereas the others are in the state of Amazonas.

Table 1. Site characteristics.

Name Abbreviation DateLat.(uS)

Long.(uW)

Watershedarea (km2)

Stage ofhydrograph

Watertype

Depth(m)

Campina Cm 17 Aug 06 2.589 60.033 ,10 falling black 0.3Barro Branco BB 16 Jul 05 2.930 59.974 ,10 falling clear 0.4

29 Aug 06Negro Ng 02 Aug 05 3.062 60.285 716,770 falling black 34

11 Aug 06Catuaba Ct 30 Aug 05 10.073 67.614 ,10 low clear 0.5

25 Sept 06Humaita Hm 26 Aug 05 9.751 67.672 ,10 low white 0.3

15 Sep 06Amazon Am 19 Jul 05 3.358 58.746 2,910,510 falling white 51Solimoes So 02 Aug 05 3.285 60.041 2,241,320 falling white 24

21 Aug 06Madeira Ma 19 Jul 05 3.405 58.771 1,381,590 falling white 25Acre Ac 02 Sep 05 10.011 67.843 20,750 low white 1.3

14 Sep 06Moa Mo 18 Aug 05 7.652 72.700 8710 low white 0.6Jurua Jr 18 Aug 05 7.682 72.660 34,450 low white 2.0Tarauaca Tr 20 Aug 05 8.175 70.773 7607 low white 1.0Envira En 20 Aug 05 8.170 70.391 12,570 low white 1.3Purus Pr 02 Sep 05 8.999 68.596 46,220 low white 1.5

22 Sep 06

Controls on respiration in the Amazon 529

ice in the dark, transported back to the laboratory, andrefrigerated until analyzed (within 48 h). Centrifuge tubes(Amicon Ultra-15 Centrifugal Filter Units) were precleanedby sonication with two 10% HCl rinses, followed by anAlconex rinse, and five C-free Milli-Q water rinses for anhour each. The 5-kDa filters were cleaned by centrifuging15 mL of 0.1 mol L21 NaOH through the filter unit, at 4,0003 g for 20 min twice, followed by a third NaOH rinse for10 min, and two Milli-Q water rinses for 10 min. Fifteenmilliliters of sample was then centrifuged for 40 min. The100-kDa filters were cleaned as follows: three 15-mL NaOHrinses for 15 min, followed by one Milli-Q water rinse, andtwo Milli-Q water rinses for 30 min. The sample was thencentrifuged for 30 min. Controls were run using albumin toensure that the filter was not affected by the cleaning process.The concentration factors for the retentate of the 5-kDa and100-kDa centrifugal filter units were 104.39 and 53.39,respectively (Millipore Corporation). However, we opted toanalyze the filtrate, rather than the retentate, because it was ofadequate volume for DOC analyses (. 180 mL). Thus, thefollowing equation was used to calculate the percentage ofDOC that is less than size X (%DOC,X):

%DOCvX~DOCFVF

DOC1V1

� �|100 ð1Þ

where X is either 5 kDa or 100 kDa, DOCF and DOCI arethe DOC concentrations (mg L21) in the filtrate (F) andinitial (I; i.e., unfiltered) sample, respectively, and VF andVI are the volume of the filtrate and the initial sample,respectively. The average blank value associated with thefiltrate was subtracted from DOCF prior to calculatingpercent recoveries. To determine the concentration of DOC, 5 kDa or , 100 kDa, the unfiltered DOC concentration(i.e., DOCI) was multiplied by %DOC,X. The concentrationof the size fractions of DOC was obtained by subtracting theappropriate value(s) from the original DOC concentration.

DOC concentrations were measured after acidificationand sparging with high-temperature combustion using aShimadzu TOC500A carbon analyzer (2005 samples) and aShimadzu TOC-V CPH carbon and nitrogen analyzer (2006samples). Those samples with high DIC concentrations wereacidified and sparged for an additional 20 min to ensureDIC removal.

Organic matter source: Carbon isotopic analyses andC : N ratios—Stable isotopes (13 : 12C) of carbon weremeasured in both the inorganic and organic size fractionsto aid in partitioning OM among end-member sources.Results are given in delta (d) notation with units of permil (%), and were normalized relative to Vienna PeedeeBelemnite. After drying coarse and fine materials in a 60uCoven, the samples were analyzed for d13C and C : N ratiosusing a Finnigan Delta Plus mass spectrometer coupled to aFissions EA 1110 CHN analyzer with a precision of 0.3%for the mass spectrometer. For 2005 samples, the C : N ratiowas obtained directly from a model 440 CHN analyzer madeby Exeter Analytical.

The d13C of DOC was analyzed using an automatedmethod in which DIC was sparged from the sample after

adding phosphoric acid, followed by sodium persulfateoxidation of DOC to CO2. The CO2 gas was carried to aninfra-red gas analyzer and then to a PDZ Europa-Hydra20-20 isotope ratio mass spectrometer. Only the 2006 sam-ples were analyzed.

DIC field collections and isotopic measurements (d13C ofDIC) were conducted as in Quay et al. (1992), with aprecision of 0.03% for the 2006 samples. DIC concentra-tions were calculated from pH and PCO2

for the 2005 sam-ples. We used temperature-dependent equilibrium constantvalues (K1, K2, and KH) as reported in Clark and Fritz(1997). To estimate the d13C of phytoplankton, we used anisotopic fractionation factor of 12–17% between the d13Cof H2CO3 (calculated from the d13C of DIC [as in Mayorga2004, using the equilibrium fractionation factors fromZhang et al. 1995]) and phytoplankton. This fractionationfactor is derived from the relationship between H2CO3 andPOM (predominantly phytoplankton) in the surface ocean(Goericke and Fry 1994).

In situ respiration rates—Respiration rates were calcu-lated at all sites by measuring the consumption of oxygenover a 24-h period. Five initial and final replicate sampleswere incubated in 60-mL acid-washed Biological OxygenDemand bottles in the dark in river water held at ambienttemperatures. Bottles were agitated twice daily by gentlyinverting them several times to reduce aggregate formation.Oxygen concentrations were measured by Winkler Titra-tions (Wetzel and Likens 1991) using a Hach titrator.Dissolved oxygen consumption was determined as the rateof change between the initial and final replicates over theincubation period in mmol of O2 L21 h21. To convert ourmeasurements into CO2 production values, we used arespiratory quotient of one, to be consistent with previouswork in the Amazon Basin (Devol et al. 1987; Richey et al.1988).

We obtained depth-integrated respiration rates by multi-plying the respiration rates by the depth of the river orstream. The depth was obtained from the nearest hydrolog-ical monitoring station maintained by the Brazilian nationalwater agency (Agencia Nacional de Aguas, ANA: http://hidroweb.ana.gov.br). We used the average depth for themonth of sampling as recorded in the long-term time series(. 33 yr). Because there was no hydrological monitoringstation nearby on the Amazon, Solimoes, Madeira, andNegro rivers, we calculated depth from field measurementstaken from long-term Carbon in the Amazon RiverExperiment (CAMREX) data sets (J. E. Richey unpubl.).We used Monte Carlo error analysis to obtain the error inthe depth-integrated respiration rate.

Bacterial abundance measurements—Bacterial abundancemeasurements were made by epifluorescence microscopyusing 49 69-diamidino-2-phenylindole (DAPI) optical filtersin 2006. Forty-milliliter samples were collected, preservedwith formaldehyde to a final concentration of 2%, and ana-lyzed within 2–4 months of collection. A surfactant (0.5%solution of Triton X-100 in distilled water) was added drop-wise to particle-rich samples, which were then sonicated for10 min. Next, the sample solution was stained with Acridine

530 Ellis et al.

Orange for 3 min. Samples were then filtered using 0.22-mmblack polycarbonate-membrane filters, and then stainedwith DAPI for 10 min. Because these samples had highsediment concentrations, this dual-stain technique wasnecessary to illuminate the bacterial cells against theparticle-rich background for counting purposes (Schmidtet al. 1998). Between 250 mL and 5 mL of sample were used,such that $ 200 cells were counted in 20 fields.

Measurements of d18O-O2 and d18O-H2O—Stable iso-topes of oxygen dissolved in water (d18O-O2) were measuredto assess the origin of O2 (Holtgrieve et al. 2010). Sampleswere analyzed within 3 months using a Finnigan Delta XLcontinuous-flow mass spectrometer (Thermo ElectronCorp), with an analytical error (6 1SD) of 6 0.2%. Masses32, 34, and 40 (16O : 16O, 18O : 16O, and 40Ar) weresimultaneously measured (Barth et al. 2004; Holtgrieveet al. 2010). Water isotopes (d18O-H2O), which came froma separate sample of river water, were analyzed on aMicromass Isoprime mass spectrometer. Results are givenrelative to Standard Mean Ocean Water in delta notationwith units of per mil. [O2] was calculated using the ratio of O2

to argon (Ar) and [Ar] as a function of water temperature(Weiss 1970).

Gas evasion—For most sites studied in 2005, CO2 out-gassing was simultaneously determined using a floatingchamber equipped with a fan. These rates are reported inAlin et al. (2010). We used their published rates to deter-mine the ability of respiration to support the outgassingflux. However, corresponding chamber measurements werenot available for several sites, so the evasion flux for thesemedium to large rivers (streams were excluded) were cal-culated as follows:

F~kCO2c Patm

CO2{Ps

CO2

� �ð2Þ

where F is the outgassing of CO2 in mmol m22 s21, kCO2is

the temperature-dependent gas exchange coefficient forCO2 (m d21), c is CO2 solubility (mmol m23 mPa21; Weiss1974), and PCO2

atm and PCO2

s are the partial pressures of CO2 inthe atmosphere and solution, respectively. kCO2

wasdetermined as a function of wind speed (Alin et al. 2010).A value of k600, the gas transfer velocity for fresh-water at 20uC, was selected for each wind speed based onthe relationship between k600 and u10 values presented inAlin et al. (2010). Monte Carlo error-propagation tech-niques were used to determine the error of our calculatedgas evasion flux.

Isotopic composition of CO2—To analyze the d13C ofCO2resp

, bottle incubations were conducted in which [DIC]and the d13C of DIC were measured initially and at the endof the incubation period (Quay et al. 1992). Two hundredfifty–milliliter glass bottles were incubated in the dark for3–6 d at ambient temperatures. The initial bottle wasimmediately poisoned in the field with 100 mL of HgCl2,whereas the final bottle was poisoned at the end of theincubation period. The d13C of CO2resp

was calculated us-ing an isotope ratio mass balance.

Statistical analysis—Shapiro–Wilk’s test for normalitywas used to assess whether variables came from a normallydistributed population prior to statistical analyses. Nor-mality was rejected when p , 0.01. The following variablesfailed this criterion and were then log-transformed: [CSS],[DIC], [CPOC], [DOC], [SO4], conductivity, and the d13C ofCSS. Upon transformation, variables were normallydistributed except for the d13C of CSS, which was thenexcluded from statistical analysis. Pearson correlation wasperformed between all variables. The site Campina was alsoexcluded because of problems with the Winkler method,possibly due to the high [DOC] (31 mg L21) or nitrite oriron species interference (Wetzel and Likens 1991).

Standard linear regression and step-wise backwardmultiple linear regression (MLR) were used to explore therelationship between environmental parameters and respi-ration rates. To be considered an input for the MLR,independent variables had to meet the following criteria:they must be significantly correlated with respiration rates;to avoid multicollinearity, they must not be highlycorrelated with each other (i.e., r . 0.7; Tabachnick andFidell 2001); and there must be no more than one missingdatum for each variable. A maximum of three independentvariables at a time were considered. All statistical analyseswere performed in SPSS.

Results

Variation in PCO2, outgassing rates, and respiration rates—

The PCO2of rivers and streams ranged from 2 to 30

times that of atmospheric equilibrium, consistent withprevious work in the basin (Mayorga 2004; Mayorga et al.2005a). Minimum and maximum values were found in theAcre River (87.5 Pa) and the stream, Campina (1308 Pa;Table 2). Gas evasion fluxes also varied widely, with aminimum of 1.0 mmol C m22 s21 and a maximum of 12.7mmol C m22 s21 at the Purus and Negro rivers, re-spectively (Table 3).

The variability of water-column respiration rates mea-sured in this study spanned several orders of magnitude(Fig. 2). The lowest rate (0.03 mmol O2 L21 h21) wasobserved in Barro Branco, whereas the highest rate(1.78 mmol O2 L21 h21) was measured in the Acre River.Despite the drought of 2005, there was no significantdifference in respiration rates between 2005 and 2006 forrivers studied during both years (t-test, t 5 20.129, df 5 18,p 5 0.898). Depth-integrated respiration rates ranged from0.014 6 0.007 mmol CO2 m22 s21 in Barro Branco to 4.1 60.8 mmol CO2 m22 s21 in the Amazon River (Table 3). TheCO2 outgassing fluxes were consistently higher than thedepth-integrated respiration rates. Accordingly, water-col-umn respiration accounted for between 0.11% 6 0.05% to100% 6 20% of the gas evasion flux.

Variation in d18O of O2 and dissolved oxygen concen-trations—The d18O of O2 is affected by gas exchange,photosynthesis, and respiration. Therefore, measurementsof d18O of O2 can provide information about the relativeimportance of these processes. If gas exchange dominates,d18O-O2 values are near 24.2% due to an atmospheric value

Controls on respiration in the Amazon 531

of 23.5% and a 0.7% equilibrium fractionation factorduring gas dissolution (Benson and Krause 1984). Respi-ration preferentially uptakes 16 : 16O at a greater rate than18 : 16O, causing an enrichment of the remaining dissolvedoxygen pool by 18% (Quay et al. 1995). Because there is noisotopic fractionation during photosynthesis, this process

produces O2 with the d18O of the water, which ranged from23.3% to 26.2% in this study. Thus, d18O-O2 values .24.2% with undersaturated O2 levels indicate that respira-tion exceeds photosynthesis, whereas values , 24.2% withsupersaturation suggest that photosynthesis dominates(Quay et al. 1995). d18O-O2 values equal 24.2% and

Table 2. Measurements of aquatic chemical parameters. DO 5 dissolved oxygen; DIC 5 dissolved inorganic carbon.

Site YearTemperature

(uC) pHPCO2

(Pa)DO

(mmol L21)DIC

(mmol L21)d13C DIC

(%)

Campina 2006 25.1 4.0 1308 102.6 344.7 226.6Barro Branco 2005 25.6 4.6 970 186.9 327.4 222.4Barro Branco 2006 25.8 4.7 731 196.1 279.9 224.3Negro 2005 29.2 5.1 659 101.3 208.5 225.2Negro 2006 29.1 4.8 491 93.8 222.8 225.6Catuaba 2005 23.5 6.2 201 207.7 121.0 216.9Catuaba 2006 24.0 5.1 163 217.9 108.5 220.2Humaita 2005 27.9 5.6 281 156.5 104.9 217.1Humaita 2006 25.4 6.3 251 168.9 310.1 218.6Amazon 2005 28.4 6.6 365 104.8 333.1 217.0Solimoes 2005 28.5 6.7 627 105.9 676.5 215.4Solimoes 2006 29.8 6.8 426 126.3 592.5 216.5Madeira 2005 29.7 6.9 293 200.8 450.4 213.6Acre 2005 30.5 7.3 119 201.5 415.1 212.0Acre 2006 28.8 7.3 88 213.3 910.0 211.6Moa 2005 28.4 7.5 148 221.7 747.4 211.4Jurua 2005 28.5 7.9 400 200.2 4895.7 214.7Tarauaca 2005 28.5 8.2 174 228.6 4852.3 212.9Envira 2005 28.4 8.4 195 247.9 8796.6 212.8Purus 2005 29.9 8.6 139 232.0 8263.8 212.4Purus 2006 32.0 8.3 103 225.9 4503.5 211.6Average 28.1 6.6 341 181.9 1856.0 216.6SD 2.2 1.3 244 49.6 2787.8 4.7

Table 3. Comparison between the depth-integrated respiration flux and the gas evasion flux.

Site YearDepth-integrated respirationflux (mmol CO2 m22 s21)*

Gas evasion flux(mmol CO2 m22 s21){

Percentage of gas evasion accountedfor by respiration (%)

Barro Branco 2005 0.01460.007 12.41 0.1160.05Negro 2005 1.060.4 12.760.8{ 863Negro 2006 0.860.4 4.360.5{ 20610Catuaba 2005 0.02460.008 2.87 0.860.3Humaita 2005 0.0660.03 1.13 662Amazon 2005 4.160.8 4.260.5{ 100620Solimoes 2005 2.260.4 7.760.8{ 2966Solimoes 2006 1.960.2 2.260.7{ 90630Madeira 2005 0.860.3 2.09 36616Acre 2005 0.560.2 4.07 1164Acre 2006 0.760.3 1.260.3{ 56626Moa 2005 0.2160.08 6.41 361Jurua 2005 0.860.3 9.1 963Tarauaca 2005 0.360.1 6.8 562Envira 2005 0.560.2 6.75 762Purus 2005 0.560.1 1.0 48612Purus 2006 0.660.1 0.860.1{ 70620Average 0.9 5 29SD 1 4 32

* The respiratory flux was calculated by converting measurements of oxygen consumption to CO2 production by using a respiratory quotient of 1, as inprevious work within the basin (Devol et al. 1987; Richey et al. 1988).

{ The gas evasion flux is that which is reported by Alin et al. (2010) unless otherwise indicated.{ The gas evasion rate was calculated as described in this study.

532 Ellis et al.

dissolved oxygen levels are close to saturation when gasexchange dominates.

Minimum and maximum values of d18O-O2 were in theEnvira (20.8%) and Solimoes (27.6%) rivers. O2 concen-trations ranged from a minimum of 93.8 mmol L21 in theNegro River to a maximum of 247.9 mmol L21 in theEnvira River (Table 2). The fraction of dissolved O2

saturation was negatively correlated with d18O-O2 (r 520.916, n 5 19, p , 0.0005; Fig. 3); sites with d18O-O2 .24.2% had significantly lower O2 saturation levels (0.7 60.1) than sites with d18O-O2 , 24.2% (where the fraction ofdissolved O2 saturation was 0.97 6 0.04; t-test, t 5 26.052,df 5 12.082, p , 0.0005). The latter sites consisted ofmedium-sized rivers in Acre that were in the low-waterstage of the hydrograph (with an average depth of 1.3 m).In contrast, sites with d18O-O2 . 24.2% consisted of small,unshaded streams and major tributaries. The averagerespiration rate of sites with d18O-O2 values , 24.2%was over four times higher than sites with d18O-O2 values. 24.2% (1.35 6 0.22 vs. 0.30 6 0.29 mmol L21 h21).

Variations in the concentration and organic composition ofthe physical size classes of organic carbon—Concentrationsof CPOC averaged 0.06 6 0.08 mg L21, but ranged from

0 mg L21 to 1.30 mg L21 (Table 4). FPOC containedslightly more carbon (average 5 1.15 6 0.70 mg L21), andconcentrations ranged from 0.07 to 2.40 mg L21. DOCranged from 1.5 (Catuaba) to 31.3 mg L21 (Campina), andaveraged 3.8 6 1.8 mg L21. Between 21% and 90% ofthe DOC was , 5 kDa (LMW DOC). Generally, thedistribution of carbon within the size fractions variedwidely (between 10% and 67% for the 5–100-kDa sizefraction and between 0.2% and 55% for the . 100-kDafraction.

The d13C and the C : N ratio of OM was measured topartition POM samples among end-member sources (i.e., C3

plants, C4 plants, and phytoplankton; Fig. 4). The d13C ofCPOC ranged from 235.8% to 227.9%. There was asignificant difference between the d13C of FPOC at sites withd18O-O2 values , 24.2%, as compared to sites with d18O-O2

values . 24.2% (d13C 5 233% 6 2% and 228.8% 6 0.9%,respectively; t-test, t 5 5.722, df 5 18, p 5 0.0005). The d13Cof DOC for the 2006 sites ranged from 227.2% to 229.9%.The C : N of CPOC ranged from 5.8 (Purus) to 29.6(Campina), whereas the C : N of FPOC was generally lower,and ranged from 6.0 to 23.6 (Fig. 4). The C : N of FPOC wassignificantly lower at sites with d18O-O2 values , 24.2% ascompared to sites with d18O-O2 . 24.2% (7.7 6 1.0 vs. 10.76 2.2, respectively; t-test: t 5 3.662, df 5 17, p 5 0.002).

The d13C of CO2respwas determined in the medium to

large rivers studied during 2006 to further identify the end-member sources fueling respiration. Values ranged from230% to 233% (Fig. 5), with the most enriched valuesfound on the Negro River, and the most depleted valuesfound on the Acre River. Values ranged between 228.3%and 228.7% in the small streams (data not shown).

Relationships between respiration rates and environ-mental variables—A backward-selection MLR approachwas used to develop a model for the dependence ofrespiration rates on in situ levels of environmentalparameters. Out of all variables studied, the combinationof [FPOC], the C : N ratio of FPOC, and O2 concentrationsexplained the highest amount of the variation in respirationrates (adjusted r2 5 0.80; Table 5). When standard linearregression was used for variables measured during bothyears, [FPOC] alone explained the most variation(Fig. 6A), followed by the C : N of FPOC (Fig. 6B), andthen pH (r2 5 0.70, 0.61, and 0.58, respectively). [O2] andbacterial abundance were both also positively correlatedwith respiration (r2 5 0.35 for [O2] and r2 5 0.78 forbacterial abundance; Fig. 6C and 6D, respectively). Respi-ration rates were not correlated with the concentration ofCPOC, DOC, or any of the size fractions within DOC.

pH was not included in the MLR because it was highlycorrelated with FPOC (r 5 0.818, n 5 19, p , 0.0005); therelationship between pH and respiration was not significantonce the influence of FPOC on these variables wasaccounted for (r 5 0.257, n 5 16, p 5 0.304). However,pH may be driving the positive correlation betweenrespiration and the percentage of DOC , 5 kDa (r 50.536, n 5 16, p 5 0.032): the latter was more stronglycorrelated with pH (r 5 0.757, n 5 16 p 5 0.001) thanrespiration. Accordingly, the correlation between the

Fig. 2. Respiration rates measured at various sites during2005 and 2006. See Table 1 for the abbreviation of each site.

Fig. 3. The relationship between d18O-O2 and the fraction ofdissolved O2 saturation. The horizontal line at 24.2% representsthe d18O value of dissolved oxygen in the water at atmosphericequilibrium. SMOW refers to the standard (standard meanocean water).

Controls on respiration in the Amazon 533

percentage of DOC , 5 kDa and respiration was notsignificant once the variability in pH was controlled for(r 5 20.101, n 513, p 5 0.720).

Differences between the sites, such as seasonality ordrainage basin size, affected respiration rates. Rivers in the

falling-water stage of the hydrograph had significantlylower respiration rates (0.17 6 0.11 mmol L21 h21) thanrivers in the low-water stage (1.09 6 0.46 mmol L21 h21;t-test 5 6.559, df 5 12.818, p , 0.0005). Because thewatershed area of the sites varied considerably, the effectsof drainage basin size on respiration rates was assessed. Themean respiration rate of medium-sized rivers (1.35 60.22 mmol L21 h21) was significantly different than that ofsmall (0.41 6 0.38) or large rivers (0.20 6 0.11; F2,17 540.7, p , 0.005).

Table 4. The concentration of organic carbon size fractions and total suspended solids. All units are in mg L21. TSS is totalsuspended solids; TOC represents total organic carbon; CPOC and FPOC are coarse and fine particulate organic carbon, respectively;DOC is dissolved organic carbon; — indicates the size fraction could not be calculated due to methodological problems.

Site Year TSS TOC CPOC FPOC DOC DOC.100 kDa DOC 5–100 kDa DOC,5 kDa

Campina 2006 4.0 32.8 1.30 0.18 31.3 3.6 21.1 6.6Barro Branco 2005 0.5 3.4 0.17 0.08 3.2 — — 2.2Barro Branco 2006 0.3 2.4 0.03 0.07 2.2 0.1 1.3 0.9Negro 2005 0.4 9.6 0 0.56 9.0 2.9 3.0 3.1Negro 2006 6.6 7.6 0.06 0.35 7.2 1.3 4.1 1.9Catuaba 2005 4.2 2.1 0.01 0.50 1.6 0.8 0.2 0.5Catuaba 2006 9.3 2.6 0.03 1.05 1.5 0.8 0 0.7Humaita 2005 4.6 4.2 0.05 0.48 3.7 1.1 — —Humaita 2006 13.7 4.0 0.24 1.03 2.8 1.0 0.3 1.4Amazon 2005 133.7 6.7 0.22 1.63 4.8 0 1.7 3.3Solimoes 2005 74.8 5.6 0.06 1.07 4.4 1.3 0 3.3Solimoes 2006 102.1 — 0.17 — 3.6 0.3 1.3 2.1Madeira 2005 45.4 4.3 0 0.80 3.5 — — 3.0Acre 2005 29.1 5.3 0 1.25 4.0 1.2 — —Acre 2006 87.2 5.5 0.01 2.40 3.1 0.4 0.4 2.3Moa 2005 29.7 4.6 0.01 1.78 2.8 — — 1.6Jurua 2005 25.8 6.9 0.02 2.26 4.6 — — 3.8Tarauaca 2005 77.8 4.9 0.11 1.79 3.0 0 0.4 2.6Envira 2005 22.8 5.2 0.01 1.61 3.6 0 0.3 3.2Purus 2005 30.6 5.6 0.01 1.67 3.9 0.7 — —Purus 2006 47.9 4.2 0.01 1.49 2.7 — — —Average 37.2 5.0 0.06 1.15 3.8 0.8 1.1 2.2SD 38.7 1.9 0.08 0.70 1.8 0.8 1.3 1.0

Fig. 4. The relationship between the d13C and C : N ratios ofPOC in the Amazon Basin. C : N ratios of phytoplankton, soil,and leaves range between 5–8, 10–27, and 21–106, respectively(Hedges et al. 1986; Trumbore 1993; Devol and Hedges 2001). Thed13C of C3 plants range from 226% to 231%, whereas C4 grassesfound on floodplains and in pastures are more enriched (212% to216% [Hedges et al. 1986; Bernardes et al. 2004; Mayorga et al.2005a]). The d13C of phytoplankton range from 230% to 245%in rivers and lakes (Araujo-Lima et al. 1986; Hedges et al. 1986;Mayorga et al. 2005a). The d13C of soil generally spans the rangeof the vegetation types (Bernardes et al. 2004).

Fig. 5. The d13C (%) of bulk size fractions and respiration-derived CO2 from sites studied in 2006.

534 Ellis et al.

Discussion

This study highlights the variability of water-columnrespiration rates found across rivers of the central andsouthwestern Amazon Basin: these rates vary by threeorders of magnitude, from 0.03–1.78 mmol O2 L21 h21,consistent with previous studies (Richey et al. 1990; Benneret al. 1995; Devol et al. 1995). The large variability callsinto question the environmental controls that affectrespiration rates.

As discussed below, our results indicate that together[FPOC], C : N ratios of FPOC, and [O2] explain 80% of thevariability of water-column respiration rates. Oxygenisotopes indicate that the highest respiration rates werefound in rivers with evidence for photosynthetic production.

Considering that these rivers also have the highest FPOCconcentrations, we argue below that the strong correlationbetween respiration and the aforementioned parameters isbecause each of these parameters is reflective of photosyn-thetic activity. Thus, rather than identifying a particular sizefraction that is predictive of respiration rates, we concludethat respiration rates are controlled by the interplay ofbiogeochemical processes occurring within the water col-umn. This study demonstrates that FPOC derived fromautochthonous origins is associated with high respirationrates, indicating that phytoplankton and/or periphytonproduction can significantly influence river metabolism evenin rivers carrying high sediment loads. This is in contrastwith previous work, which has assumed that high sedimentloads inhibit in situ production (Wissmar et al. 1981). This

Fig. 6. Correlations between respiration rates and some of the environmental variables analyzed in this study. Respiration rates(mmol O2 L21 h21) were significantly correlated with (A) FPOC concentration, (B) C : N ratios of FPOC, (C) dissolved O2 concentration,and (D) bacterial abundance.

Table 5. Statistical relationships between respiration rates and environmental variables.

Independent variable Model* r2 F df p

[FPOC], [O2], C : NFPOC Rsp=0.43FPOC20.084C : NFPOC+0.003O2+0.485 0.80{ 25.1 (3,15) ,0.0005Bacterial abundance{ Rsp52.9431027 BA20.22 0.78 17.7 (1,5) 0.008FPOC Rsp50.70FPOC20.084 0.7 39.1 (1,17) ,0.0005C : NFPOC Rsp520.20C : NFPOC22.6 0.61 26.6 (1,17) ,0.0005pH Rsp50.34pH21.52 0.58 25.3 (1,18) ,0.0005d13CFPOC Rsp520.15d13CFPOC23.9 0.37 10.6 (1,18) 0.004O2 Rsp50.007O220.55 0.35 9.7 (1,18) 0.006%DOC,5 kDa Rsp50.015%DOC,5kDa20.32 0.29 5.7 (1,14) 0.032

* The model in bold is the best parameterized for this study. Stepwise backward multiple linear regression (MLR) was used to examine the influence ofmultiple variables on respiration rates (Rsp), whereas standard linear regression was used to assess the dependence of respiration on only one variable.BA is bacterial abundance, and %DOC,5 kDa is the percentage of the total DOC concentration , 5 kDa. All other variables are defined in the text.

{ Refers to the adjusted r2 as it was generated from a MLR. All others are standard coefficients of determination.{ Only data from 2006 were used.

Controls on respiration in the Amazon 535

research further demonstrates that multiple sources (C3

plants, C4 grasses, and phytoplankton) serve as the substratefor respiration basin-wide, suggesting that the type of carbonpreferentially consumed during respiration is affected byhydrograph stage and/or floodplain connectivity.

Relationships between organic carbon size fractions andrespiration rates—A long-standing hypothesis for theAmazon mainstem is that a small, rapidly cycling pool oflabile OC fuels respiration (Richey et al. 1990). This labilepool likely coexists within the larger pool of refractorymaterials that comprise the bulk size fractions of CPOC,FPOC, and DOC (Richey et al. 1990; Amon and Benner1996; Mayorga et al. 2005a). Recently, measurements ofthe D14C and d13C of the bulk size fractions and CO2 gasin equilibrium with DIC suggests that the CO2 producedthrough respiration is younger and disproportionatelycomposed of C4 grasses relative to the bulk size fractions(Mayorga et al. 2005a), providing evidence that an iso-topically distinct pool of OC (relative to the bulk sizefractions) may support river metabolism. Accordingly, amajor objective of this study was to assess the relationshipbetween bulk OC size fractions, and size fractions withinthe DOC pool, with respiration rates.

FPOC was the only bulk size fraction that wassignificantly correlated with respiration. Further, out of allparameters studied during both years, it explained the high-est amount of the variability in respiration rates (Fig. 6A;Table 5), suggesting that it may contain a labile pool of OM.Previous work has shown that although the majority of theFSS carried by Amazonian lowland rivers originates in theAndes, the d13C of FPOC of these rivers reflects OM derivedfrom the lowlands. This indicates near-complete remineral-ization of Andean-derived FPOC (Mayorga et al. 2005a),suggesting it is a labile size fraction. Furthermore, the d13Cof FPOC carried by the Solimoes and Amazon rivers mirrorsthe high-altitude enrichment of terrestrial plants expected of1% for every 1,000-m change in elevation (Hedges et al.2000; Mayorga et al. 2005a).

Although FPOC may be an inherently labile size fraction,the positive correlation between respiration rates and FPOCobserved in this study is primarily due to the influence ofphytoplankton production at sites with high FPOC concen-trations. The average respiration rate at sites with d18O-O2

values , 24.2% was over four-fold greater than thatobserved at sites with d18O-O2 values . 24.2%. At theformer sites, FPOC concentrations were over twice that atsites with d18O-O2 . 24.2%, demonstrating that photosyn-thetic production of dissolved oxygen was occurring at siteswith high suspended loads. This is not unexpected becausethese rivers with high sediment concentrations were sampledduring the low-water stage, and consequently they wereshallow, unshaded, and visually tinted green. They inher-ently had high suspended loads due to the high-relief natureof their drainage basins, but the particulate carbonconcentrations were likely enhanced by autochthonousproduction. In contrast, sites with d18O-O2 . 24.2%consisted of small, shaded streams throughout the studyarea, and major tributaries (the average depth was 34 m)with deep and highly turbid water-columns.

Primary production can directly stimulate water-columnrespiration rates through multiple pathways. First, previouswork within the basin suggests that the lack of a bioavailablesubstrate limits bacterial growth and respiration in largewhitewater rivers of the Amazon (Benner et al. 1995).Autochthonous production produces labile carbon (Sunet al. 1997; del Giorgio and Pace 2008), and in Amazonian-floodplain lakes, respiration is positively correlated withchlorophyll a concentrations and phytoplankton cell carbon(Wissmar et al. 1981). Therefore, one direct pathway bywhich increased photosynthetic activity can lead to elevatedrespiration rates is via providing a labile substrate forheterotrophs, thereby increasing heterotrophic respirationrates. Our bacterial abundance measurements support thisnotion, because the highest abundances were found at siteswith d18O-O2 , 24.2%, and abundances were positivelycorrelated with respiration rates (Fig. 6D). However, pri-mary production may elevate water-column respiration dueto autotrophic respiration. This is unlikely to be significantat sites with d18O-O2 . 24.2%, which have the lowest res-piration rates observed in our study, but it may be asignificant factor at sites with d18O-O2 , 24.2%. Regardlessof the direct mechanism by which respiration is stimulated,these results indicate that the low-water stage of thehydrograph is an important time of carbon processing.

The composition of POM provides further support thatautochthonous production is occurring at sites with d18O-O2

, 24.2%. Plotting d13C values against C : N ratios demon-strates that the FPOC of these sites lie within the rangeexpected for carbon influenced by phytoplankton end-members (Fig. 4), whereas the FPOC from sites with d18O-O2 . 24.2% (i.e., shaded streams and large tributaries) islikely derived from forest soils. In contrast, CPOC appearsto be primarily sourced from forest soils and C3 vegetation inboth areas, indicating that autochthonous production ispredominantly found in the fine fraction. This is consistentwith previous work, as phytoplankton have been shown tosignificantly contribute to POM composition in the Andeanheadwaters of the Amazon during the dry season (Town-send-Small et al. 2007).

Accordingly, the results from the MLR demonstrate thatrespiration is well-predicted by a model that includes FPOC,C : NFPOC, and O2 (r2 5 0.80; Table 5; Fig. 6) because eachof these parameters reflect photosynthetic production. Thesignificant inverse relationship between C : NFPOC and res-piration demonstrates that as FPOC becomes more similarto phytoplankton end-members, respiration rates increase(Fig. 6B). Similarly, the inclusion of O2 in the MLR is likelybecause it reflects the importance of oxygen produced fromphotosynthesis. Thus, we conclude that FPOC is associatedwith high respiration rates when it can be demonstrated tohave characteristics of autochthonous production.

Conversely, any relationship between in situ concentra-tions of DOC size classes and respiration rates is less clear.It has been speculated that a labile pool of carbon may existwithin the dissolved pool that partially fuels respiration(Mayorga and Aufdenkampe 2002). For example, previouswork on the Solimoes and Negro rivers has demonstratedhigher rates of bacterial growth and respiration duringincubations amended with high molecular weight DOC

536 Ellis et al.

(i.e., DOC . 1 kDa) as compared to incubations with DOC, 1 kDa (Amon and Benner 1996). Thus, DOC . 1 kDa ishypothesized to be of more recent origin (and more bio-available) than DOC , 1 kDa (Amon and Benner 1996).As a test of this conceptual model, we examined the rela-tionship between respiration rates and the in situ concen-trations of size fractions of DOC.

We found no relationship between respiration and[DOC] or the concentration of any of the size fractions ofDOC. There are several mechanisms that could account forthe lack of a significant correlation. First, the influence ofpH, which ranged from 4.0 to 8.6 at our sites, may obscureany relationship between DOC size classes and respirationrates. pH affects the molecular weight distribution, size,and shape of fulvic acids leached from soils due toinfluences caused by changes in hydrogen bonding andvan der Waal’s forces (Schnitzer 1978; De Haan et al.1983). Therefore, the positive correlation between thepercentage of the total DOC pool that was , 5 kDa insize and pH could potentially be explained by those factors.Although respiration rates were positively correlated withthe percentage of DOC , 5 kDa, the positive correlationbetween the percentage of DOC , 5 kDa and pH wasstronger. Accordingly, after controlling for pH via partialcorrelation analyses, the relationship between respirationand the percentage of DOC , 5 kDa was no longersignificant (see Results).

The lack of correlation between respiration and DOCconcentrations (both bulk and size fractions) suggest thatother variables, such as in situ primary production, play alarger role in influencing respiration rates than the in situconcentrations of the DOC size classes examined in thisstudy. The occurrence of autochthonous respiration at siteswith d18O-O2 , 24.2% could obfuscate any correlations.Such respiration would obscure relationships betweendissolved size fractions and respiration because autochtho-nous respiration is not dependent upon these size fractions.Accordingly, the discrepancy between our results and thatof Amon and Benner (1996) likely reflects differences inmethodology (i.e., comparisons of relationships betweenrespiration rates and in situ concentrations vs. manipulatedbottle incubations) and/or the effects of other variables incontrolling the molecular weight distribution of DOC.

Potential biological sources of CO2 produced from respi-ration—In contrast to previous work that has suggestedthat C4 grasses are preferentially consumed during respi-ration (Mayorga et al. 2005a), our results demonstrate thata variety of sources are consumed across the basin. Wemeasured the d13C of CO2 produced during dark-bottleincubations at select sites to further our understanding ofthe origin of CO2 produced from respiration. The d13C ofCO2resp

was 233.2% 6 0.6% and 231% 6 2% (Fig. 5) atsites where d18O-O2 , 24.2%. If this CO2 was producedentirely from the consumption of phytoplankton, the(calculated) d13C of CO2resp

would be expected to rangebetween 232% and 237% at these sites. Conversely, if theCO2 was produced entirely from C3 or C4 substrates, thed13C of CO2resp

would be expected to range between 226%and 231.3%, and between 212% and 216%, respectively

(Hedges et al. 1986; Bernardes et al. 2004; Mayorga et al.2005a). Therefore, the similarity of the d13C of CO2resp

to thatof these potential end-members suggests that the d13C ofCO2resp

produced during bottle incubations is derived fromthe consumption of both phytoplankton and/or periphytonand C3 plants at sites with d18O-O2 , 24.2%.

In contrast, at most sites where d18O-O2 . 24.2%, the d13Cof CO2resp

appears to be coming from C3 sources (Fig. 5). Forexample, in the streams (Catuaba and Humaita), the d13C ofCO2resp

ranged from 228.3% to 228.7% (data not shown),whereas it was 230% on a major tributary (the Negro).These values are substantially more enriched than ourestimates of phytoplankton in these rivers (237% to245%).

Perhaps the most surprising result is that that the d13C ofCO2resp

from the Solimoes River was 233% 6 3%. Thed13C of phytoplankton should range from 235% to 240%,suggesting that the d13C of CO2resp

is derived from C3-plants and phytoplankton in August. The Solimoes showslittle evidence of autochthonous production based on d18O-O2 and the composition of OM (Figs. 3 and 4). However,algal material could be produced in nearby floodplain lakes(Wissmar et al. 1981). During this time period (the earlystages of falling water), these lakes drain into the riverchannel, possibly transporting labile carbon that couldstimulate riverine metabolism. Previous work has suggestedthat a rapidly cycling pool of labile carbon is needed tosupport water-column respiration along the mainstem(Richey et al. 1990; Mayorga et al. 2005a); therefore, wesuggest that phytoplankton-derived OM may provide thissubstrate during certain times of the year.

The d13C of CO2respreported here for the Solimoes River

differs significantly from the one measurement that hasbeen reported previously (222%). It was taken during theearly rising-water stage, and implied that 40% of the water-column respiration on the Solimoes River was supportedby the consumption of floodplain grasses (Quay et al.1992). Given the discrepancy between these results, wemeasured the d13C of CO2resp

on the Solimoes at the samelocation during the rising-water stage (Apr 07). It was223% 6 3% (data not shown), consistent with the resultsof Quay et al. (1992). Accordingly, the CO2 producedthrough respiration appears to be highly dynamic giventhat its isotopic signature may vary by as much as 10%depending upon the time of the year that sampling occurs.This variability may be explained by differences in theproduction of C4 grasses on the floodplain. For example,the biomass of grasses in an Amazonian floodplain lake inApril was demonstrated to be over twice that in August,with a higher monthly biomass loss to the mainstemoccurring in April relative to August (Engle et al. 2008).

In conclusion, our bottle-incubation results suggest thatphytoplankton and C3 plants are important substrates forrespiration at sites where d18O-O2 , 24.2%. In contrast, atsites with d18O-O2 . 24.2%, C3 plants typically play alarger role fueling respiration rates than phytoplankton,and respiration rates are reduced. However, all of theseend-members (C3 and C4 plants, and phytoplankton and/orperiphyton) are important substrates for respiration through-out the basin, and their role in supporting respiration likely

Controls on respiration in the Amazon 537

reflects temporal variability caused by changes in hydrographstage.

Comparisons between the d13C of respiration-derived CO2

and organic carbon size fractions—Several studies havesuggested that the d13C of CO2 produced throughrespiration is isotopically enriched relative to the bulk sizefractions carried by rivers in the Amazon Basin (Quay et al.1992; Mayorga et al. 2005a), suggesting that the C4 grasseslining river corridors are inherently more biodegradablethan the bulk OM transported by the Amazon River, whichresembles C3 plants. To assess whether this finding holdsunder the continuum of rivers examined in this study, wecompared the d13C of CO2resp

to that of the bulk size frac-tions. We found that the error associated with the d13C ofCO2resp

often overlaps with the isotopic signature of bulkOM (Fig. 5), suggesting that this material (which isisotopically consistent with C3-plant origins) supportsrespiration in rivers ranging from major blackwater andwhitewater tributaries (i.e., the Negro River and the PurusRiver, respectively) to small forested streams (Catuaba andHumaita: data not shown). However, isotopically depleted Cis respired relative to the bulk size fractions found in theAcre and Solimoes rivers, suggesting that phytoplanktonmay be preferentially consumed in certain rivers. Thus, thecoupling between the bulk size fractions and respired carbonis also substantially more variable than suggested previously.

Role of water-column respiration in fueling CO2 out-gassing—Finally, to what extent does water-column respi-ration support CO2 gas evasion fluxes in Amazonianwaters? The role of respiration is highly variable, withdepth-integrated respiration rates accounting for between0.1% and 100% of the outgassing flux during the low- andhigh-falling water seasons (Table 3). Water-column metab-olism is typically , 3.5% of the total ecosystem metabolismin streams draining the Hudson River watershed (Bott et al.2006), so it may be expected that water-column respirationwould comprise a small percentage of the outgassing flux insmall streams. Indeed, respiration fuels , 6% of theoutgassing flux for all streams in this study. In suchstreams, inputs of CO2 from groundwater via soil and rootrespiration are likely significant contributors to the gasevasion flux, and work within the Amazon Basin hasdemonstrated that 90% of CO2 transported by groundwa-ter is evaded in the headwater reaches of first- and second-order Amazonian streams (Johnson et al. 2008). As streamsscale in size, the influence of respiration increases in thisstudy: it accounted for an average of 26% 6 27% of theoutgassing flux in the medium-sized rivers.

The contribution of water-column respiration to gasevasion is also highly variable in large tributaries, and theextent to which respiration fuels gas evasion appears to beinfluenced by water-chemistry type. For example, the aerialrespiration flux accounts for between 8% and 18% of theoutgassing flux during the falling-water stage on the Negro,suggesting that additional CO2 production mechanisms maybe important in blackwater rivers, such as photo-oxidation(Remington et al. 2011). In contrast, 62% 6 34% of out-gassing CO2 is accounted for by water-column respiration in

large whitewater rivers (Amazon, Madeira, and Solimoes;Table 3). We hypothesize this variability is due to changes inwind speed affecting gas evasion rates or due to fluctuationof respiration rates. Our results suggest that in situ water-column respiration rates are capable of supporting the highoutgassing rates in large whitewater rivers.

Implications toward our understanding of carbon cyclingin Amazonian rivers—A significant implication of this workis that the coupling between OM sources and respiration ishighly variable. At any point in time, rivers within theAmazon Basin vary dramatically in terms of their drainagebasin size, chemical composition, and stage of the hydro-graph. Therefore, it may be expected that respiration rateswill vary significantly throughout the basin as lightavailability, floodplain connectivity, and land-water cou-pling dynamics differ across river types.

One of the primary objectives of this work was to assessthe relationship between respiration rates and size fractionsof OC. Our results indicate that respiration is controlled bya variety of biogeochemical processes, rather than a singlesize fraction. We found that when FPOC is derived fromalgal sources, it is associated with high water-column res-piration rates as both autotrophic respiration and hetero-trophic respiration are stimulated (the latter being due tothe production of a labile substrate).

Relative to regional carbon budgets, how consistent are theresults here with the partitioning of basin outgassing into thepotential sources terms estimated by Richey et al. (2002)?They viewed outgassing as direct carbon losses from uplandforests and floating and emergent macrophytes. Based oncalculations of floodplain production of macrophytes and thearea of the floodplain in the central Amazon, they estimatedthat about 25% of the flux could be due to C4 decompositionand respiration. The results here are consistent, in that asignificant C4 signal is detected. However, given that we havedemonstrated that this signal is highly variable, this estimatelikely needs revision. Furthermore, the presence of autoch-thonous production at certain times of the year does notchange our perspective of Amazonian rivers as processors ofterrestrial carbon (Quay et al. 1992; Richey et al. 2002;Mayorga et al. 2005a). As phytoplankton fix inorganiccarbon generated via weathering and respiration, CO2

produced from the consumption of phytoplankton (or fromautotrophic respiration) should still be considered to be acarbon loss from upland ecosystems.

Future studies should strive to incorporate temporal andspatial variability in their sampling designs to further ourunderstanding of the diverse array of processes that controlcarbon cycling in tropical rivers. Specifically, studies areneeded to quantify the temporal and spatial variability ofautochthonous production, including scaling work to evalu-ate its significance to the regional carbon budget. Previously,autochthonous production has been discounted as a signif-icant source of OC in sediment-rich rivers of the AmazonBasin (Wissmar et al. 1981). Here we provide evidence thatOC derived from autochthonous production is found even inrivers with high sediment loads, suggesting that it couldpotentially be a source of rapidly-cycling labile carbon thatstimulates respiration in other rivers throughout the basin.

538 Ellis et al.

AcknowledgmentsWe thank J. Souza, V. Neu, E. Souza de Silva, and J. Cerdeira

for their exceptional efforts in the field. R. Victoria and the staffwithin the Large-Scale Biosphere-Atmosphere Experiment inAmazonia (LBA) program provided invaluable help with logis-tical support for field work. We are further grateful to G. GobetBaldi, X. Montebello, J. Stutsman, M. Haught, and S. Carpenterfor help with sample analyses. We thank S. Alin, G. Holtgrieve, A.Ingalls, and M. Logsdon for their encouragement and seminaldiscussions. We acknowledge P. Martin and E. Mayorga for theircomments on the manuscript, and in particular E. Mayorga forproviding the calculations on drainage basin size. We would alsolike to thank the two anonymous reviewers for their criticalcomments that have significantly improved this manuscript.

Financial support for this research was provided by the NationalAeronautic Space Administration’s LBA program (NCC5-345 andNCC5-689), the Division of Environmental Biology at the NationalScience Foundation (NSF; 0213585), and an NSF GraduateResearch Fellowship to EEE. This is Carbon in the Amazon RiverExperiment (CAMREX) publication number 157.

References

ALIN, S. R., M. F. F. L. RASERA, C. I. SALIMON, J. E. RICHEY,G. W. HOLTGRIEVE, A. V. KRUSCHE, AND A. SNIDVONGS. 2010.Physical controls on carbon dioxide transfer velocity and fluxin low-gradient river systems and implications for regionalcarbon budgets. J. Geophys. Res.-Biogeosci. 116: GO1009,doi:10.1029/2010JG001398

AMON, R. M. W., AND R. BENNER. 1996. Bacterial utilization ofdifferent size classes of dissolved organic matter. Limnol.Oceanogr. 41: 41–51, doi:10.4319/lo.1996.41.1.0041

ARAUJO-LIMA, C., B. R. FORSBERG, R. VICTORIA, AND L. MARTINELLI.1986. Energy sources for detrivorous fishes in the Amazon.Science 234: 1256–1258, doi:10.1126/science.234.4781.1256

AUFDENKAMPE, A. K., J. I. HEDGES, J. E. RICHEY, A. V. KRUSCHE,AND C. A. LLERENA. 2001. Sorptive fractionation of dissolvedorganic nitrogen and amino acids onto fine sediments withinthe Amazon Basin. Limnol. Oceanogr. 46: 1921–1935,doi:10.4319/lo.2001.46.8.1921

———, AND OTHERS. 2011. Riverine coupling of biogeochemicalcycles between, land, oceans, and atmosphere. Front. Ecol.Environ. 9: 53–60, doi:10.1890/100014

BARTH, J. A., C. A. TAIT, AND M. BOLSHAW. 2004. Automatedanalyses of 18O/16O ratios in dissolved oxygen from 12-mL watersamples. Limnol. Oceanogr.: Methods 2: 35–41, doi:10.4319/lom.2004.2.35

BATTIN, T. J., AND OTHERS. 2008. Biophysical controls on organiccarbon fluxes in fluvial networks. Nat. Geosci. 1: 95–100,doi:10.1038/ngeo101

BENNER, R., S. OPSAHL, G. CHINLEO, J. E. RICHEY, AND B. R.FORSBERG. 1995. Bacterial carbon metabolism in the AmazonRiver system. Limnol. Oceanogr. 40: 1262–1270, doi:10.4319/lo.1995.40.7.1262

BENSON, B. B., AND D. KRAUSE. 1984. The concentration andisotopic fractionation of oxygen dissolved in freshwater andseawater in equilibrium with the atmosphere. Limnol.Oceanogr. 29: 620–632, doi:10.4319/lo.1984.29.3.0620

BERNARDES, M. C., AND OTHERS. 2004. Riverine organic mattercomposition as a function of land use changes, SouthwestAmazon. Ecol. Appl. 14: S263–S279, doi:10.1890/01-6028

BOTT, T. L., AND OTHERS. 2006. Ecosystem metabolism instreams of the Catskill Mountains (Delaware and HudsonRiver watersheds) and Lower Hudson Valley. J. N. Am.Benthol. Soc. 25: 1018–1044, doi:10.1899/0887-3593(2006)025[1018:EMISOT]2.0.CO;2

BURDIGE, D. J., AND K. G. GARDNER. 1998. Molecular weightdistribution of dissolved organic carbon in marine sediment porewaters. Mar. Chem. 62: 45–64, doi:10.1016/S0304-4203(98)00035-8

CLARK, I., AND P. FRITZ. 1997. Environmental isotopes inhydrogeology. Lewis.

COLE, J. J., N. F. CARACO, G. W. KLING, AND T. K. KRATZ. 1994.Carbon dioxide supersaturation in the surface waters of lakes.Science 265: 1568–1570, doi:10.1126/science.265.5178.1568

———, AND OTHERS. 2007. Plumbing the global carbon cycle:Integrating inland waters into the terrestrial carbon budget.Ecosystems 10: 171–184, doi:10.1007/s10021-006-9013-8

DE HAAN, H., G. WERLEMARK, AND T. DEBOER. 1983. Effect of pHon molecular weight and size of fulvic acids in drainage waterfrom peaty grassland in NW Netherlands. Plant. Soil. 75:63–73, doi:10.1007/BF02178614

DEL GIORGIO, P. A., AND M. L. PACE. 2008. Relative independenceof dissolved organic carbon transport and processing in alarge temperate river: The Hudson River as both pipe and reactor.Limnol. Oceanogr. 53: 185–197, doi:10.4319/lo.2008.53.1.0185

DEVOL, A. H., B. R. FORSBERG, J. E. RICHEY, AND T. P. PIMENTEL.1995. Seasonal variation in chemical distributions in theAmazon (Solimoes) River—a multiyear time series. Glob.Biogeochem. Cycles 9: 307–328, doi:10.1029/95GB01145

———, AND J. I. HEDGES. 2001. Organic matter and nutrients inthe mainstem Amazon River, p. 275–306. In M. E. McClain,R. L. Victoria, and J. E. Richey [eds.], The biogeochemistry ofthe Amazon Basin. Oxford Univ. Press.

———, P. D. QUAY, J. E. RICHEY, AND L. A. MARTINELLI. 1987.The role of gas exchange in the inorganic carbon, oxygen, and222Rn budgets of the Amazon River. Limnol. Oceanogr. 32:235–248. , doi:10.4319/lo.1987.32.1.0235

DUARTE, C. M., AND Y. T. PRAIRIE. 2005. Prevalence of hetero-trophy and atmospheric CO2 emissions from aquatic ecosys-tems. Ecosystems 8: 862–870, doi:10.1007/s10021-005-0177-4

ENGLE, D. L., J. M. MELACK, R. D. DOYLE, AND T. R. FISHER.2008. High rates of net primary production and turnover offloating grasses on the Amazon floodplain: Implications foraquatic respiration and regional CO2 flux. Glob. Change Biol.14: 369–381, doi:10.1111/j.1365-2486.2007.01481.x

GOERICKE, R., AND B. FRY. 1994. Variations of marine plankton d13Cwith latitude, temperature, and dissolved CO2 in the world ocean.Glob. Biogeochem. Cycles 8: 85–90, doi:10.1029/93GB03272

HEDGES, J. I., W. A. CLARK, P. D. QUAY, J. E. RICHEY, A. H.DEVOL, AND U. D. SANTOS. 1986. Compositions and fluxes ofparticulate organic material in the Amazon River. Limnol.Oceanogr. 31: 717–738, doi:10.4319/lo.1986.31.4.0717

———, G. L. COWIE, J. E. RICHEY, P. D. QUAY, R. BENNER, M.STRONG, AND B. R. FORSBERG. 1994. Origins and processing oforganic matter in the Amazon River indicated by carbohy-drates and amino acids. Limnol. Oceanogr. 39: 743–761,doi:10.4319/lo.1994.39.4.0743

———, AND OTHERS. 2000. Organic matter in Bolivian tributariesof the Amazon River: A comparison to the lower mainstem.Limnol. Oceanogr. 45: 1449–1466, doi:10.4319/lo.2000.45.7.1449

HOLTGRIEVE, G. W., D. E. SCHINDLER, T. A. BRANCH, AND Z.TERESA A’MAR. 2010. Simultaneous quantification of aquaticecosystem metabolism and reaeration using a Bayesianstatistical model of oxygen dynamics. Limnol. Oceanogr. 53:1047–1063, doi:10.4319/lo.2010.55.3.1047

JOHNSON, M. S., J. LEHMANN, S. J. RIHA, A. V. KRUSCHE, J. E.RICHEY, J. P. H. B. OMETTO, AND E. G. COUTO. 2008. CO2

efflux from Amazonian headwater streams represents asignificant fate for deep soil respiration. Geophys. Res. Lett.35: L17401, doi:10.1029/2008GL034619

Controls on respiration in the Amazon 539

KRITZBERG, E. S., J. J. COLE, M. L. PACE, W. GRANELI, AND D. L.BADE. 2004. Autochthonous versus allochthonous carbonsources of bacteria: Results from whole-lake C-13 additionexperiments. Limnol. Oceanogr. 49: 588–596, doi:10.4319/lo.2004.49.2.0588

MAYORGA, E. 2004. Isotopic constraints on sources and cycling ofriverine dissolved inorganic carbon in the Amazon Basin.Ph.D. thesis. Univ. of Washington,

———, AND A. K. AUFDENKAMPE. 2002. Processing of bioactiveelements in the Amazon River System, p. 124. In M. E.McClain [ed.], The ecohydrology of South American riversand wetlands. IAHS Spec. Pub no. 6, IAHS Press.

———, AND OTHERS. 2005a. Young organic matter as a source ofcarbon dioxide outgassing from Amazonian rivers. Nature436: 538–541, doi:10.1038/nature03880

———, M. G. LOGSDON, M. V. BALLESTER, AND J. E. RICHEY.2005b. Estimating cell-to-cell land surface drainage paths fromdigital channel networks, with an application to the Amazonbasin. J. Hydrol. 315: 167–182, doi:10.1016/j.jhydrol.2005.03.023

QUAY, P. D., D. O. WILBUR, J. E. RICHEY, A. H. DEVOL, R. BENNER,AND B. R. FORSBERG. 1995. The 18O/16O of dissolved oxygen inrivers and lakes in the Amazon Basin: Determining the ratio ofrespiration to photosynthesis rates in freshwaters. Limnol.Oceanogr. 40: 718–729, doi:10.4319/lo.1995.40.4.0718

———, ———, ———, J. I. HEDGES, A. H. DEVOL, AND R. VICTORIA.1992. Carbon cycling in the Amazon River: Implications fromthe 13C compositions of particles and solutes. Limnol. Oceanogr.37: 857–871, doi:10.4319/lo.1992.37.4.0857

REMINGTON, S., A. KRUSCHE, AND J. RICHEY. 2011. Effects ofDOM photochemistry on bacterial metabolism and CO2

evasion during falling water in a humic and a whitewater riverin the Brazilian Amazon. Biogeochemistry 105: 185–200,doi:10.1007/s10533-010-9565-8

RICHEY, J. E., A. H. DEVOL, S. C. WOFSY, R. VICTORIA, AND M. N. G.RIBERIO. 1988. Biogenic gases and the oxidation and reductionof carbon in Amazon river and floodplain waters. Limnol.Oceanogr. 33: 551–561, doi:10.4319/lo.1988.33.4.0551

———, J. I. HEDGES, A. H. DEVOL, P. D. QUAY, R. VICTORIA,R. MARTINELLI, AND B. R. FORSBERG. 1990. Biogeochemistryof carbon in the Amazon River. Limnol. Oceanogr. 35:352–371, doi:10.4319/lo.1990.35.2.0352

———, J. M. MELACK, A. K. AUFDENKAMPE, V. M. BALLESTER,AND L. L. HESS. 2002. Outgassing from Amazonian rivers andwetlands as a large tropical source of atmospheric CO2.Nature 416: 617–620, doi:10.1038/416617a

SCHMIDT, J. L., J. W. DEMING, P. A. JUMARS, AND R. G. KEIL.1998. Constancy of bacterial abundance in surficial marinesediments. Limnol. Oceanogr. 43: 976–982, doi:10.4319/lo.1998.43.5.0976

SCHNITZER, M. 1978. Some observations on chemistry of humicsubstances. Agrochimica 22: 216–225.

SIOLI, H. 1984. The Amazon: Limnology and landscape ecologyof a mighty tropical river and its basin. Kluwer Academic.

SUN, L., E. M. PERDUE, J. L. MEYER, AND J. WEIS. 1997. Use ofelemental composition to predict bioavailability of dissolvedorganic matter in a Georgia river. Limnol. Oceangr. 42:714–721, doi:10.4319/lo.1997.42.4.0714

TABACHNICK, B. G., AND L. S. FIDELL. 2001. Using multivariatestatistics, 4th ed. Harper and Rowe.

TOWNSEND-SMALL, A., J. L. NOGUERA, M. E. MCCLAIN, AND J. A.BRANDES. 2007. Radiocarbon and stable isotope geochemistryof organic matter in the Amazon headwaters, PeruvianAndes. Glob. Biogeochem. Cycles 21: GB2029, doi:10.1029/2006GB002835

TRUMBORE, S. E. 1993. Comparison of carbon dynamics in tropicaland temperate soils using radiocarbon measurements. Glob.Biogeochem. Cycles 7: 275–290, doi:10.1029/93GB00468

WEISS, R. F. 1970. Solubility of nitrogen, oxygen and argon inwater and seawater. Deep-Sea Res. 17: 721–735.

———. 1974. Carbon dioxide in water and seawater: The solubilityof a non-ideal gas. Mar. Chem. 2: 203–215, doi:10.1016/0304-4203(74)90015-2

WETZEL, R. G., AND G. E. LIKENS. 1991. Limnological analyses,2nd ed. Springer-Verlag.

WILDE, F. D., AND D. B. RADTKE. 2003. National field manualfor the collection of water-quality data, U.S. GeologicalSurvey.

WISSMAR, R. C., J. E. RICHEY, R. F. STALLARD, AND J. M. EDMOND.1981. Plankton metabolism and carbon processes in theAmazon River, its tributaries, and floodplain waters, Peru-Brazil, May–June 1977. Ecology 62: 1622–1633, doi:10.2307/1941517

ZENG, N., J.-H. YOON, J. A. MARENGO, A. SUBRAMANIAM, C. A.NOBRE, A. MARIOTTI, AND J. D. NEELIN. 2008. Causes andimpacts of the 2005 Amazon drought. Environ. Res. Lett. 3:014002. , doi:10.1088/1748-9326/3/1/014002

ZHANG, J., P. D. QUAY, AND D. O. WILBUR. 1995. Carbon isotopefractionation during gas-water exchange and dissolution ofCO2. Geochim. Cosmochim. Acta 59: 107–114, doi:10.1016/0016-7037(95)91550-D

Associate editor: George W. Kling

Received: 29 November 2010Accepted: 08 November 2011Amended: 19 December 2011

540 Ellis et al.