Evaluating climatic and edaphic controls on dry season gas exchange at two Amazonian rain forest ...

12 Pages·2008·0.31 MB·English

agricultural and forest meteorology 148 (2008) 850–861
available at www.sciencedirect.com
journal homepage: www.elsevier.com/locate/agrformet
Evaluating climatic and soil water controls on
evapotranspiration at two Amazonian rainforest sites
Rosie A. Fishera,*, Mathew Williamsa, Maria de Lourdes Ruivob,
Antonio Lola de Costac, Patrick Meira
aSchoolofGeoSciences.UniversityofEdinburgh,Edinburgh,UK
bMuseuParaenseEmilioGoeldi,Bele´m,Para´,Brazil
cUniversidadeFederaldoPara´,Bele´m,Para´,Brazil
article i nfo abstract
Articlehistory: Interactions between the biosphere and the atmosphere have profound impacts on the
Received25July2007 functioningoftheEarthsystem.Oneofthemostimportantareasofbiosphere–atmosphere
Receivedinrevisedform interactionistheAmazonbasin,whichplaysakeyroleintheglobalcyclesofcarbon,water
28November2007 andenergy.TheAmazonisvulnerabletoclimaticchange,withincreasinglyhotanddry
Accepted3December2007 conditionsexpectedoverthenext50–100yearsinsomemodels.Theresultinglossofcarbon
fromtheAmazonbasinhasbeensuggestedasapotentiallylargepositivefeedbackinthe
climate system We investigated the differences in atmospheric demand and soil water
.
Keywords: availabilitybetweentwosites;Manaus,incentralAmazonia,whereevapotranspirationwas
Amazon limitedinthedryseason,andCaxiuana˜ ineasternAmazonia,whereitwasnot.Newsoil
Soilmoisture hydraulicdataincludingwaterreleaseandunsaturatedhydraulicconductivitycurveswere
Hydraulicconductivity collectedatCaxiauna˜usingtheinstantaneousprofilemethod(IPM),pressureplateanalysis
Gasexchange andtensioninfiltrometry.ThesedatawerecomparedtoexistingdatafromManaus.The
plantavailablesoilwaterattheCaxiuana˜sitewas2.1–3.4timeslargerthantheManaussite.
The hydraulic conductivity curves indicated the existence of a secondary macropore
structureatverylowtensions((cid:2)0.05kPato(cid:2)1kPa),potentiallycausedbybiogenicmacro-
pores, but did not vary with respect to soil water potential between sites. In addition,
differencesintheclimaticseverityofthedryseasonwereestimated.Themaximumsoil
waterdeficit,projectedusingasimplemodelofforestwateruse,wassimilarbetweenthe
sites. No difference in climatic severity between sites was found and we conclude that
below-groundsupplyofwater,ratherthanclimaticdifferences,werelikelytohavecaused
the contrasting dry season behaviour at the two sites. These findings indicate that, in
combinationwithotherfactors,heterogeneityinsoilwaterretentioncapacitymayexert
strongcontrolsonthespatialvariationinforestresponsestoclimaticchange.
#2007ElsevierB.V.Allrightsreserved.
1. Introduction feedbackbetweenthebiosphereandatmosphereislikelytobe
positive,andtoaddbetween20and200ppmtoatmospheric
Recent analyses using coupled climate and carbon cycle CO concentrationsby2100(Friedlingsteinetal.,2006;Salazar
2
models indicate that during the next 100 years, the net et al., 2007). The majority of model simulations predicted a
* Correspondingauthor.Presentaddress:DepartmentofAnimalandPlantSciences,UniversityofSheffield,Sheffield,UK.
Tel.:[phone].
E-mailaddress:rosie.fisher@sheffield.ac.uk(R.A.Fisher).
0168-1923/$–seefrontmatter#2007ElsevierB.V.Allrightsreserved.
doi:10.1016/j.agrformet.2007.12.001
agricultural and forest meteorology 148 (2008) 850–861 851
major decline in land carbon storage, located in tropical ManaussiteandasiteatCaxiuana˜,inEasternAmazonia.The
forestsandtheAmazonbasininparticular,whichleadstoan Caxiuana˜ siteisofinterestasdryseasonevapotranspiration
increasedaccumulationofcarbondioxideintheatmosphere. appears to not be limited by soil moisture deficits, as
TherecentlypublishedfourthIPCCassessmentreport(Figure demonstrated by eddy covariance data (Carswell et al.,
SPM7.IPCC,2007)suggeststhat,overlargeareasofSouthern 2002), leaf hydraulics data (Fisher et al., 2006) and sap flow
and Eastern Amazonia, more than 90% of GCM model measurements (Fisher et al., 2007). At the Manaus site, soil
predictionsagreethatdryseasonrainfallwilldeclineby10– hydraulics data (Tomasella and Hodnett, 1996) and evapo-
30%. In Northern and Central areas, there is no agreement transpiration data (Malhi et al., 1998) have previously been
betweenGCMsandinWesternAmazonia,aslightincreasein collectedandareusedhereforcomparison.
precipitation rates is predicted. In addition, increases in First, rainfall patterns and atmospheric demand at Cax-
temperatureareuniformlypredictedtoriseacrossthewhole iauna˜ andManauswereanalysedtodetectwhetherclimatic
region. Understanding the interaction between the drying differencesbetweenthesitescouldaccountforthedifferences
climate and forest carbon dioxide exchange in Amazonia is indryseasonevapotranspirationbehaviour.Simplemodelsof
therefore of fundamental importance to the prediction of potential evapotranspiration and of soil water deficit were
futurebiosphere–atmospherefeedbackintheclimatesystem constructed–toestimatehowmuchwatermustbesupplied
(Huntingfordetal.,2004;WerthandAvissar,2004). by the soil to allow unrestricted evapotranspiration to
At present, many of the forests of Southern and Eastern continue–underthedifferentclimateregimes.Theuseofa
Amazonia experience, on average, a dry season exceeding modelallowedcomparisonstobemadebetweentheclimatic
three months, where rainfall is less than 3mm day(cid:2)1 regimesindifferentsitesandfordifferentyears.
(Sombroek, 2001). High temperatures and radiation create Secondly, the soil water storage capacity was compared
an evaporative demand in the Amazonian region that is betweenthetwosites.Theamountofwaterwhichisavailable
typically3–4mmday(cid:2)1(Malhietal.,2002;WerthandAvissar, for evapotranspiration during a dry season depends funda-
2004; Cox et al., 2004; Huntingford et al., 2004). This means mentally on the amount of water stored in the soil that is
that,duringthedryseason,plantsmustutilizewaterstoredin availabletoplants.Totalsoilwateravailabilitydependsupon
soilstosupplyevapotranspirationataratewhichisunlimited thevolumeofsoilaccessedbyroots,andupontheamountof
bywateravailability(Nepstadetal.,1994;Hodnettetal.,1996). wateravailabletoplantsperunitvolumeofaccessedsoil(the
Despitethislongdryseason,itisnotyetcleartowhatextent waterholdingcapacity).Thelatterisafunctionbothofplant
the evapotranspiration of contemporary Amazonian rain- rooting density and the soil hydraulic properties, both of
forestsislimitedbywateravailability.Mostecosystemmodels which will vary with depth (Sperry et al., 1998). The soil
of Amazonian evapotranspiration predict very large reduc- hydraulic properties of relevance are the water retention
tions in water use during the dry season (Tian et al., 1998; capacity of the soil (how much water is held above the
Potter et al., 2001; Werth and Avissar, 2004). In contrast to thresholdforplantextraction)andtheunsaturatedhydraulic
thesemodelexpectations,recentEarthobservationdatahave conductivity, which determines the rate of movement from
suggested that Amazonian ecosystems can be observed to thesoilmatrixtotherootsurface.Bothwaterretentionand
‘green-up’duringthedryseason(Hueteetal.,2006)andeven soil hydraulic conductivity data are presentedin this paper.
during the recent 2005 El Nino˜ event (Saleska et al., 2007). Thehydraulicconductivitydatapresentedhereareonlythe
However,itisnotclearwhetherthesedatarepresentchanges second published set of unsaturated hydraulic conductivity
inleafareaorintheobservedchlorophyllcontentofthevisible data for Amazonian soils. To allow the use of these soil
leaves, and how they might relate to changes in photosyn- properties in dynamic models of soil water movement, the
theticrates. Furthermore,existingevidenceof forestevapo- parameters of the van Genuchten soil hydraulic model (van
transpiration from micrometeorological eddy covariance Genuchten,1980)arecalculated.Inthispaperwepresentnew
measurements at six locations show contrasting results. At soilhydraulicsdatafromanareawheredataareverysparse,
five locations, no evidence of hydraulic limitation of evapo- andattempttoexplainthedifferencesintheobservedresults
transpiration has been measured (Grace et al., 1996; Arau´jo of existing gas exchange measurements between two con-
etal.,2002;Carswelletal.,2002;Saleskaetal.,2003;daRocha trasting sites in Amazonia, understanding of which is
etal.,2004;Gouldenetal.,2004).Atonesite,Manaus,incentral currently absent. Ultimately, better understanding of how
Amazonia,theobservedevapotranspirationdeclinedbyupto climatic and edaphic factors combine to alter the rainforest
50%inresponsetolowsoilmoistureduringthe1995–96dry responses to low soil moisture will allow more detailed
season(Williamsetal.,1998;Malhietal.,1998,2002).Because predictionsofwhichareasofAmazoniaaremostvulnerableto
thesedataaretheonlydatawhereareductioningasexchange climaticchange.
inthedryseasonhasbeenobserved,theyhavebeenusedfor
calibration of global models of gas exchange (Harris et al.,
2004) and used to make predictions concerning the rate of 2. Materials and methods
Amazon die-back (Huntingford et al., 2004). However, the
cause of the different responses to rainfall seasonality 2.1. Predictionofsoilwaterdeficit
between sites is unknown, and therefore the likely spatial
and temporal extent of these different modes of behaviour It is frequently assumed that the approximate atmospheric
cannotbeestimated. demand over a month in a rainforest ecosystem is 100mm
Inthispaper,weinvestigatedtwopossibleexplanationsfor (Sombroek, 2001; Malhi and Wright, 2004) and months with
thedifferencesindryseasonevapotranspirationbetweenthe lessrainfallthandemandwillneedtoutilisestoredsoilwater.
852 agricultural and forest meteorology 148 (2008) 850–861
However,thisbasicindexdoesnotaccountforthedifference typifytheaverageresponsesofthesetwositestodryseason
in the severity of dry periods beyond this threshold, and rainfall. Instead, the analysis indicates whether soil or
assumes that evaporative demand does not vary between climateisamorelikelydriveroftheinter-sitedifferencesin
seasons. Instead of assuming an atmospheric demand of theyearsinquestion.
100mm month(cid:2)1, a model was constructed to predict daily
forestpotential evapotranspiration under well-watered con- 2.2. Measurementofsoilhydraulicproperties
ditions.ThemethodusedwassimilartothatofWilliamsetal.
(1997),whoconstructedan‘aggregatedcanopymodel’(ACM) 2.2.1. Site
ofdailycarbonuptake,byfittingasimpleempiricalmodelto TheexperimentalsiteislocatedatCaxiuana˜ NationalForest,
match the outputs of a detailed, well verified, half-hourly Para´, Brazil (184303.500S, 5182703600W), a lowland terra firme
model of forest gas exchange, the soil-plant-atmosphere rainforest.Thesoilisayellowoxisol(Brazilianclassification
model (SPA, Williams et al., 1996, 1998, 2001). In this study, latosol),witha0.3–0.4mthickstonylayerpresentbetween3
a model of potential evapotranspiration (PET) for Amazon and4mdepth.Thetextureofthesurfacesoilis75–83%sand
forests(ACM-ET)wasdeveloped.TheSPAmodelwasusedto (>0.05mmparticlediameter),12–19%clay(<0.02mm)and6–
generatearesponsesurfaceofPETtoradiation,temperature 10%silt(0.05–0.02mm)(RuivoandCuhna,2003).Thesurface
and vapour pressure deficit under a subset of observed soilconsistsofquartzinthesandfractionandpredominantly
Amazonian meteorological conditions.2503 days of meteor- kaolin in the clay fraction (Ruivo and Cuhna, 2003). Root
ologicalinputstotheSPAmodelwereassembledusingdata biomasshasbeendetectedatthissiteatdepthsof10m(the
fromfoursitesacrosstheAmazonbasin:Tapajo´s,Caxiuana˜, deepestmeasurementsmade)(Fisheretal.,2007).
ManausandRondoˆnia.Thepreciselocationandmethodolo- The area of forest investigated was delineated by the
giesofthedatacollectionaredescribedelsewhere(Malhietal., boundaries of the LBA (Large-Scale Biosphere–Atmosphere
1998; von Randow et al., 2004; Carswell et al., 2002; Saleska Experiment in Amazonia) Ecology Program (Avissar and
etal.,2003).Usingaheuristicprocess,asimplesetofequations Nobre, 2002) throughfall exclusion experiment (Fisher et al.,
was constructed to capture this response surface. A finite 2006). The experiment covered a 100m(cid:3)200m area 500m
difference gradient optimisation routine was used to adjust distant from the research station. The plot described shall
theparametersoftheequationssoastominimisetheRMSEof hereafter be referred to as the ‘experimental site’. All the
the SPA vs. ACM-ET comparison of daily potential evapo- experimentsandsamplecollectionsoccurredwithinthissite,
transpiration(UMINF,IMSLMathLibrary).Formodelstructure with the exception of the instantaneous profile experiment
andinputs,seeAppendixA. (see below), which occurred at a site 200m closer to the
ToassesshowthebalanceofPETandrainfallaffectsthe researchstation.
demand for soil water, a dynamic model of soil moisture
deficit was implemented. The maximum possiblesoil water 2.2.2. Waterretentioncurvemeasurement
deficitispredicted,assumingnoreductioninevapotranspira- Pressure plate analyses were conducted on intact samples,
tioncausedbylowsoilmoisture.Thisistoestimatehowmuch collected in April 2002 inside 55mL soil cores, at depths
water the soil must provide for the canopy to continue to correspondingtosoilhorizonsidentifiedbyRuivoandCuhna
function normally, with no hydraulic restriction. The soil (2003)infour5mdeepsoiltrencheswithintheexperimental
moisture deficit is assumed to initially be zero, as the site. Water contents were gravimetrically determined at (cid:2)6,
simulation is started in the late wet season. Each month, (cid:2)10,(cid:2)30,(cid:2)100and(cid:2)1500kPa.Thewaterretentioncurvefor
waterisremovedfromthesoilaccordingtotheoutputsofthe tensionsbetween(cid:2)8kPaand(cid:2)1kPawasdeterminedbyinsitu
ACM-ETmodel(E,inmm),andthemonthlyrainfalltotal(R,in measurements of soil water potential (C) and soil water
t t
mm) is added to the soil. The cumulative soil water deficit, content (u) by the instantaneous profile method (IPM)
CWD(mm)isalteredbythedifferencebetweenthesevalues.If described above. For the top metre of soil the two data sets
theCWDispositiveattheendofthemonththenitisresetto were combined to give continuous water retention curves
zero as the excess rain is assumed to contribute to ground- between (cid:2)1 and (cid:2)1500kPa for each of the four soil depths
water,hencesoilwaterstorageisignored: (Fig.1).
CWDtþ1¼CWDtþRt(cid:2)Et (1) 2.2.3. Unsaturatedhydraulicconductivitymeasurement
In this study, hydraulic conductivity was measured only at
IfCWD >0thenCWD =0 tensionswetterthanfieldcapacity,duetologisticaldifficulties
t t
Toestimateevaporativedemand,theACM-ETmodelwas associatedwithmeasurementofdriersoils.Twofieldbased
driven with daily meteorological data derived from Malhi methodswereused,tensioninfiltrometry(Ankenyetal.,1991)
et al. (1998) for Manaus, in 1995–96, the period when eddy andtheinstantaneousprofilemethod(Dirksen,2000)toobtain
covariancemeasurementsweremade,andfromFisheretal., estimatesofKoverdifferentrangesoftension.
2007 for Caxiuana˜ in 2002–03, the period when sap flow Firstly,measurementsofunsaturatedconductivityofsoils
measurements were made. The results were summed into inthemacroporeregion((cid:2)0.05to(cid:2)1kPatension)weremade
monthly values. The aim here is to analyse, for these two using a tension infiltrometer (TI) with a 0.2m disk (Soil
sets of observations, whether soil or climate is the likely Measurement Systems, Tucson, AZ, USA). Measurements of
driver of the differences in dry season behaviour. Because infiltration rates were made for every 0.1kPa of tension
onlylimitedtemporalrangesofmeteorologicalandfluxdata between (cid:2)0.05 and (cid:2)1kPa. It is necessary to correct the TI
are available at each site, these results do not necessarily measurements for the increase in apparent flow caused by
agricultural and forest meteorology 148 (2008) 850–861 853
Fig.1–Waterretentioncurvedataforfourdepthswithinthefirstmetreofsoilfrompressureplatesandinsitu
measurements,plusmodelapproximationstothedatamadeusingthevanGenuchten(VG)model.Errorbarsarerangesof
samplestakenfromtwodifferentsoilaccessshafts.
three-dimensionalcapillaryflow.Toachievethis,correction (mmh(cid:2)1)wascalculatedusingthemethodofDirksen(2000)
methodsuggestedbyReynoldsandElrick(1991)wasapplied. as:
Eighthydraulicconductivity(K)vs.waterpotential(C)curves q½z;t(cid:4)
K½u;z;t(cid:4)¼ (2)
were measured using TI on the surface soil. Measurements dH=dz½z;t(cid:4)
were randomly located within the experimental site. Con-
straintsto therandomselectionwerethateachsamplesite wherezisdepth(mm),tistime(h),andHisthehydraulichead
wasatleast0.5mfromanytreemorethan0.1mindiameter, (mm). The tensiometers, located at the same depth as the
andatleast0.5mfromtheedgeoftheplotorwalkwayswithin CS615 probes, also allowed the construction of in situ water
the plot, to avoid disturbed soil. In addition, three TI retention curves. The monolith was left to drain until the
measurement profiles were made by removing the soil waterpotentialofeachsensorwasbelow(cid:2)8kPa.Themethods
sequentially and measuring K vs. C at four different depths usedweresimilartothoseofTomasellaandHodnett(1996),
(0.1,0.35,0.5and0.9m).Thetensioninfiltrometermeasure- with the exception that, in this case, automatic TDR probes
ments for each profile were not located immediately below wereused,ratherthantheneutronprobeusedintheToma-
oneanothertopreventpossibleimpactsofcompactionofthe sella and Hodnett study, and the space of the probes was
soilcolumnbytheinstrument. altered as a result. The maximum difference between the
Secondly, the instantaneous profile method was used to IPM and tension infiltrometry data was 0.6mmh(cid:2)1, so con-
makemeasurementsbetween(cid:2)1kPaand(cid:2)8kPa.TheIPMis tinuousKvs.Crelationshipsbetween(cid:2)0.05and(cid:2)8kPawere
designedtomeasuretheverticalflowofwaterthroughsoilby constructedforeachdepthusingthesetwodatasets(Fig.4).
isolating a column of soil, 1m diameter by 1m depth and InthesandysoilattheCaxiuana˜ site,theapplicabilityof
wrapping it in plastic sheeting to prevent lateral water the calibration functions of the CS615 soil moisture sensors
movement. Soil water content and water potential were used in the IPM method was unknown, so a gravimetric
measured at 0.05, 0.3, 0.6 and 0.9m using CS615 TDR (time calibration was conducted to assess how the sensors
domainreflectometry)probes(CampbellScientific,Loughbor- respondedtovaryingsoilmoistureinthesandylatosolsoil.
ough, UK) and mini-tensiometers (Skye Instruments, Llan- Three0.3mlong,0.15mdiametercoresofsoilwereremoved
drindodWells,Wales),respectively,andreadingswerelogged andkeptintactwithinPVCpiping.Twocoresweretakenfrom
each minute using a Campbell Scientific CR10X datalogger thesurfacesoil(0.05–0.35mdepth)andonefrom0.35to0.65m
(Campbell Scientific, Loughborough, UK). Water was poured depth.Thelocationswerechosenatrandom(avoidinglarge
on top of the monolith until the mini-tensiometers gave roots)withintheexperimentalsite.
readingswetterthan(cid:2)0.2kPafor5min,indicatingthatnear- Each core was instrumented with a CS615 probe and a
saturationconditionshadbeenachieved.Themonolithwas mini-tensiometer. The cores were continuously saturated
covered to prevent evaporation from the soil surface, and until the automatic CS615 readings were stable for 30min,
allowedtodrain.Theflowpastthesensors(q)wascalculated andthemini-tensiometerreadingswerelessthan(cid:2)0.02kPa.
ateachdepthasthelossinstorageofthesoilabovethesensor Thecoreswereallowedtodrainandplacedinsidealightbox
inmmh(cid:2)1.Usingthisestimate,thehydraulicconductivityK at(at(cid:5)608C)andsimultaneousmeasurementsofsoilmass
854 agricultural and forest meteorology 148 (2008) 850–861
and moisture sensor output were made at gradually Natick,MA,USA).Theparameterswerefittedtominimisethe
increasing time intervals for ten days. At the end of the errorparametere,thesumoftherelativerootmeansquare
calibrationperiod,thesensorswereremoved,andthecores errorbetweentheobservedandmodelledwatercontents(u
o
weredriedfor24hat1058Candweighedtodeterminethe andu )plusthesumoftherelativerootmeansquareerrors
m
massofthedrysoil. between log of the K data (K ) and the log of the model K
o
Soil calibration curves were established by calculation of prediction(K )
m
thevolumetricwatercontentu(m3m(cid:2)3)usingthefollowing
relationship: e¼Xu0(cid:2)umþXlogK0(cid:2)logKm (4)
u0 logK0
M (cid:2)M
u¼ w d (3)
rV The log transformation of K is based on the assumption
thatthedataerrorisconstantonalogscale,andavoidsbiasof
where M and M are the masses of the wet and dry soils, thefittingroutinetowardsthelargervalues,asKvariesover
w d
respectively(kg)andristhedensityofwater(kgm(cid:2)3)andVis manyordersofmagnitude.vanGenuchtenparameterswere
thevolumeofthecoreinm3.Thismethodissimilartothat not fitted to the data from the lower layers where only
employed by Veldkamp and O’Brien (2000) to calibrate the pressureplatedatawasavailable,asthesamplesizewastoo
samesensorsforasoilinEcuador. lowtoconstrainalltheparametersofthemodel.
2.2.4. Soilhydraulicmodelparameterisation
Given the difficulties inherent in the measurement of 3. Results
unsaturated hydraulic conductivity, for most applications,
soilhydraulicmodelsareusedtoextrapolatetheK data 3.1. Predictedsoilwaterdeficit
unsat
into drier soils. To facilitate this, the parameters of the van
Genuchten (1980) model of soil hydraulic behaviour were Duringtherespectivemeasurementperiods,theCaxiuana˜site
estimatedforthetopmetreofsoil(seeAppendixB).Toallow received more rainfall in total (2350mm year(cid:2)1) than the
comparison between the measurements, the data were Manaussite(2088mmyear(cid:2)1).Bothsitesexperiencedstrong
grouped into four 0.1m depth range categories. The depth seasonality in rainfall, with wet season rainfall peaking
categoriesusedwere0.05–0.15m,0.25–0.35m,0.55–0.65mand around 400mm per month in February at Caxiuana˜ and in
0.85–0.95m. By grouping together the data in this way it is MayatManaus(Fig.2).LowrainfallbeganinJulyinbothsites
assumed that each 0.1m depth range is effectively homo- (115mm at Caxiuana˜ and 82mm at Manaus). The driest
geneous. This assumption is necessary because it is not monthinbothcaseswasAugust,however,thedryseasonat
practicallypossibletodoallthemeasurementsinthesame Caxiuana˜ lasted until December, whereas at Manaus high
location. rainfall (>100mm) only returned in February. Both the wet
The van Genuchten model required parameter optimisa- anddryseasonrainfallatManauswasclosetothelong-term
tion. This was achieved using the simplex search method averageforManausCity(Harrisetal.,2004),andtheCaxiuana˜
(Lagariasetal.,1998)embeddedin‘Matlab’(MathWorksInc., rainfallwas5mm(1%)lessinthedryseasonand80mm(5%)
Fig.2–Comparisonofrainfall(squares),modelledpotentialevapotranspiration(diamonds)andestimatedcumulativewater
deficit(circles,CWD)atCaxiuana˜ (opensymbols)andManaus(filledsymbols).
agricultural and forest meteorology 148 (2008) 850–861 855
less in the wet season than the average over 5 years of inTable2.Thesoilwatercontentat(cid:2)1500kPaisverylow(0.07
measurementsatCaxiuana˜. to 0.15m3m(cid:2)3) due to the low clay content. The water
The simple model predicted seasonality in the monthly retention curves were not measured between saturation
potentialevapotranspiration,rangingatbothsitesfrom120to and (cid:2)1kPa so the water released at these tensions is
130mminthedryseasonto70–80mminthewetseason.The unknown.
dataarefromdifferentyears,buttheseasonalcycleissimilar The water retention curves for the deeper soil layers
between sites, with a slightly greater PET at Caxiuana˜ in (between1.0and4.5mdepth,Fig.3)weresimilartothoseof
November,andlowerinSeptemberandOctober. the top metre of soil in that the amount of water released
ThelongerdryseasonatManausdidnotcreatealargersoil between (cid:2)6 and (cid:2)1500kPa was within the range found in
moisturedeficitthanatCaxiuana˜.Partlybecausetherewasa the top metre (0.164 to 0.243m3m(cid:2)3). The variation in
‘mini’wetseasoninNovember,whenthesoilmoisturewas absolute water content in the pressure plate data was
slightlyrecharged,accordingtothesimplemodel,andpartly greaterthanthevariationdetectedintheIPMasitincluded
because atmospheric demand was relatively low between bothspatialandverticalheterogeneity.Onlyoneprofilewas
November and February. The maximum predicted seasonal measuredfortheIPM,sotheeffectofspatialvariationwas
water deficit was similar in the two forests; (cid:2)210mm at not known.
Caxiuana˜ and (cid:2)214mm at Manaus. The model results thus
indicate that there were not large differences in maximum 3.2.3. Hydraulicconductivity
demand for soil moisture between the two sites. The water The hydraulic conductivity data (Fig. 4) showed good agree-
deficitwasmaintainedforonemonthlongerinManausthan mentbetweenthetwohydraulicconductivitymeasurement
inCaxiuana˜. techniques(tensioninfiltrometryandinstantaneousprofiling)
wheretheirrangesoverlappedat(cid:2)1kPa.Theobservedvalues
3.2. Soilhydraulicproperties of K varied over (cid:5)5 orders of magnitude over the measured
rangeofwatercontents,between844 and0.01mmh(cid:2)1.The
3.2.1. Soilmoisturesensorcalibration van Genuchten model, which was simultaneously fitted
The responses of the CS615 probes to water content were againstthewaterretentioncurveandtheconductivitydata,
effectivelylinear.Therewaslittlevariationbetweenthecores, isgenerallyagooddescriptoroftheshapeoftheKvs.Cdata
soallthedataweregroupedtogethertoproducethefollowing (RMS error=0.168–0.33log mmh(cid:2)1, r2=0.875 to 0.998), but
10
calibrationequation: thereissomedeviationofthemodelawayfromthedataas
saturationisapproachedat(cid:2)0.05kPa.HeretheKdataareon
u¼0:79P(cid:2)0:59 ðr2¼0:79Þ (5) average367mmh(cid:2)1higherthanthemodelpredictions.HighK
valuesclosetosaturationpotentiallyindicatetheexistenceof
whereuisthevolumetricwatercontent(m3m(cid:2)3)andPisthe secondarysystemoflargesoilpores.Thesefeatureswerealso
periodoutputoftheCS615sensorsinms(Table1). observedbyTomasellaandHodnett(1996)forrainforestsoils,
and were hypothesized to be due to the effect of biogenic
3.2.2. Waterretentioncurves macropores (Chauvel et al., 1991), which substantially
Thedatadescribeasigmoidalrelationshipwhenplottedona increase the maximum hydraulic conductivity of the soil
log–logscale,andgiveagoodfittothevanGenuchtenmodel matrix.ThevanGenuchtenmodelassumesauni-modalpore
(RMSerror=0.013m3m(cid:2)3,r2=0.88to0.98).ThevanGenuch- sizedistributionandthusisnotabletopredicttheeffectsof
tenparametersusedtointerpolatetherelationshipareshown suchsecondaryporesystems.
Table1–SummaryoftheavailablesoilhydrologydataattheCaxiuana˜ site
Datatype Crange(kPa) Depthrange(m) Kdata Cdata udata
Tensioninfiltrometer (cid:2)0.05to(cid:2)1 0–0.9 Yes Yes No
Pressureplates (cid:2)6to(cid:2)1500 0–4.5 No Yes Yes
Soilcalibrationcores (cid:2)0.1to(cid:2)6 0–0.9 No No Yes
Instantaneousprofilemethod (cid:2)0.8to(cid:2)8 0–0.9 Yes Yes Yes
Table2–vanGenuchtenparametersforfoursoildepthsattheCaxiuana˜ site
Model Parameter Depth(m)
0.50–0.15 0.25–0.35 0.45–0.55 0.90–0.10
vanGenuchten ur 0.127 0.027 0.022 0.105
us 0.516 0.421 0.384 0.413
a 5.40 1.27 1.82 2.10
N 1.27 1.23 1.10 1.20
Ksat 1011 1545 3102 2123
l (cid:2)1.25 1.88 (cid:2)1.42 (cid:2)1.03
856 agricultural and forest meteorology 148 (2008) 850–861
Fig.3–Hydraulicconductivitydata,fromtensioninfiltrometer(TI)andinstantaneousprofiling(IPM),withmodel
approximationstothedatausingthevanGenuchtenVGmodelforfourdepthswithinthefirstmetreofsoil.Errorbarsare
standarddeviationsoffourindependenttensioninfiltrometersamples.
3.2.4. Comparisonofsoilpropertiesatcaxiuana˜ andmanaus estimated. It was assumed that water held between 0 and
TheCaxiuana˜soildatawerecomparedtothedatacollectedat (cid:2)6kPaisnotavailabletoplants,asitisquicklydrainedaway
a site near Manaus by Tomasella and Hodnett (1996). The afterarainevent.Thelowerlimitofwaterextractionisnot
Caxiuana˜ andManaussoilsdifferedsignificantlyintermsof known,asitdependsuponrootdensityandosmoticpotential,
absolutewatercontent.TheCaxiuana˜soilhadlowersoilwater soil texture and the capacity of plants to resist xylem
content across all tensions (t-test p value <0.01), with the embolism (Sperry et al., 1998). Therefore, the soils were
maximum measured water content being 0.37m3m(cid:2)3 com- comparedforawiderangeofpossibleextractionlimits.The
pared to 0.52m3m(cid:2)3 at Manaus (Tomasella and Hodnett, lower limit considered was (cid:2)1500kPa, the commonly
1996).Thisistobeexpected,asclaysoilsholdmorewaterat assumed ‘wilting point’, beyond which plants are not able
verylowtensions(<1500kPa)intheirmicroporousstructure. to function (Nepstad et al., 2004). In a modelling study at
However,thesandysoilatCaxiauna˜hadamuchgreaterrange Caxiuana˜, Fisher et al. (2007) used measured root biomass
of water content over the range of tensions measured. The profiles,amongotherdata,toparameteriseamodelofforest
waterheldinthesoilwithintherangeavailabletoplantswas gas exchange. This model, which was verified against soil
watercontentdata,predictedthatinthetop5mofsoil,where
rootswereabundant,predictedextractiondidoccurdownto
(cid:2)1500kPa.However,thelowestwaterpotentialpredictedfor
soillayersbetween5and10mdepthduringtwodryseasons,
bothofwhichexhibitedveryrestrictedforestwateruse,was
(cid:5)(cid:2)100kPa.Thiswetterpredictedextractionlimitwasmainly
due to the low root density in these layers slowing the
modelledrateofwaterextractionfromthedeepsoils.Wehave
thereforealso analysed the wateravailablebetween(cid:2)6 and
(cid:2)100kPa because this is the predicted amount of water
availablefromdeeplayersundertheprevailingclimaticand
biologicalcircumstances.
At Manaus, available water content between (cid:2)6kPa and
(cid:2)100kPa averaged across all the layers tested was
0.040(cid:6)0.016m3m(cid:2)3 (Tomasella and Hodnett, 1996). Data
were not available for soils drier than (cid:2)100kPa, but extra-
polationofthevanGenuchtencurvespredictsthattherewas
Fig.4–Soilwaterretentioncurvesofallsoillayersderived anaverageof0.045(cid:6)0.011m3m(cid:2)3releasedbetween(cid:2)100kPa
frompressureplateanalyses.Errorbarscorrespondto and (cid:2)1500kPa making a total of 0.095(cid:6)0.027m3m(cid:2)3. At
rangesofsamplesfromtwodifferentsites. Caxiuana˜, the average water released between (cid:2)6kPa and
agricultural and forest meteorology 148 (2008) 850–861 857
Fig.5–(a)ComparisonofIPMdatafromManaus(closedsymbols)andCaxiuana˜ (opensymbols).Caxiuana˜ data:
0.05m=circles,0.30m=squares,0.60m=diamonds,0.90M=triangles.Manausdata:0.30m=circles,0.50m=squares,
0.75m=diamonds,1.05m=rightpointingtriangles,1.35m=leftpointingtriangles.(b)Comparisonofwaterretention
curvesfromManaus(closedsymbols)andCaxiuana˜ (opensymbols).Caxiuana˜ data:0.05m=circles,0.30m=squares,
0.60m=diamonds,0.90m=triangles.Manausdata:0.20m=circles,0.40m=squares,0.60m=diamonds,0.80m=right
pointingtriangles,1.00m=leftpointingtriangles.Notethedifferenceinscalebetweentheplots.
(cid:2)100kPaoverthetopmetreofsoil(toallowdirectcomparison The relationship between water potential C and unsatu-
with the Manaus data) was 0.138(cid:6)0.046m3m(cid:2)3 and that rated conductivity K, was similar (t-test p value=0.34)
between (cid:2)100kPa and (cid:2)1500kPa was 0.062(cid:6)0.034m3m(cid:2)3, betweentheManausandCaxiuana˜ sites(Fig.5)fortherange
givingatotalof0.200(cid:6)0.032m3m(cid:2)3.Therefore,ifthelimitof ofCoverwhichKwasmeasured((cid:2)0.1to(cid:2)8kPa)(thisisfor
extractionis(cid:2)100kPa,thentheCaxiauna˜ soilholds3.4times soilswetterthanfieldcapacity,andthusnotthesamerangeas
morewaterthanManaus.Ifitis(cid:2)1500kPa,theCaxiuana˜ soil thatoverwhichtheplantavailablewaterwascompared).We
holds2.1timesasmuchastheManaussoil. thereforeexpectthattherateofsoil-to-rootwatertransportat
858 agricultural and forest meteorology 148 (2008) 850–861
agivenCwouldbesimilarbetweenthetwosites,givensimilar mainlyimproveplantwateruptakecapabilitiesviatheuseof
rootingdensity. hydraulic lift (Lee et al., 2005; Oliveira et al., 2005). This
mechanism,wherebywateriscontinuouslytransferredfrom
wetterdeeplayerstodriershallowlayersinthedryseason,
4. Discussion mayhelptoovercomethelimitationsonwateruptakecaused
bysoil-to-rootresistance,asthisuptakemechanismactsallof
The forest at Caxiuana˜ appears to be able to withstand the thetime,notjustwhentheplantistranspiring.However,at
observeddryseasonswithoutrestrictiononevapotranspira- the Manaus site, the existence of dry season hydraulic
tioncausedbylowsoilmoisture(Carswelletal.,2002;Fisher limitationpotentiallymeansthattheimpactofhydrauliclift
et al., 2006, 2007). In contrast, the forest at Manaus islimited.
experienced substantial declines in evapotranspiration in Despitethedemonstratedexistenceofdeepwaterextrac-
the1996dryseason,asmeasuredbyeddycovariance(Malhi tion(Moreiraetal.,2000;Brunoetal.,2006)andhydrauliclift
etal.,1998).Itisimportanttounderstandthereasonsforthese (Oliveira et al., 2005) at the Tapajo´s rainfall exclusion
differentbehavioursasthecurrentwaterlimitationstatusof experiment, Nepstad et al. (2007) found a three year delay
the forest is a key uncertainty in our ability to predict the between beginning a rainfall exclusion experiment and the
magnitude of the proposed ‘Amazon dieback’ feedback onsetofincreasedtreemortalityandatleastoneyeardelay
mechanism (Cox et al., 2000; Friedlingstein et al., 2006). In before the onset of leaf water stress and LAI changes (see
thisstudy,twoabioticfactorswhichmaypotentiallyexplain Nepstadetal.,2002).Theresultsfromthisrainfallexclusion
the differences between the two forest ecosystems were experiment suggests that stored soil water at this site can
investigated; climate patternsand plant available soil water bufferlargechangesinrainfallforashortperiod,butthelong-
storagecapacity.Neitherthepotentialevapotranspirationnor term depletion of soil water reserves eventually causes
rainfallpatternsdifferedsubstantiallybetweenthetwosites, limitations in water supply. This is important, as it means
leading to only small differences of only 7mm in projected that the responses to short-term perturbations, like El Nino˜
maximum soil water deficit. However, the available soil years,donotnecessarilyprovideinsightintotheresponsesof
moistureattheCaxiuana˜ sitewasbetween2.1and3.4times the forest to long-term changes in water balance. Another
greaterpervolumeofsoilthantheManaussoil,dependingon rainfallexclusionexperiment,alsolocatedatCaxiuana˜(Fisher
the (unknown) ‘limit’ of water extraction. Therefore, we et al., 2007) responded by reducing evapotranspiration in
concludethatthereislittleevidencethatclimaticfactorsare response to the imposed drought in a single year, but this
the cause of restrictions to evapotranspiration in the dry rainfall exclusion was imposed continuously throughout
seasonatManaus,butthereissomeevidencetosuggestthat theyear,notjustinthewetseason(Nepstadetal.,2002)so
theverylowplantavailablesoilwaterstoragecapacitymay theresultsarenotdirectlycomparable(also,measurementsof
make this forest more prone than the Caxiuana˜ forest to thegasexchangeoftheTapajo´sdroughtexperimentarenot
experiencingdroughtstress. availableforcomparisonusingthemethodsemployedinthis
Oneunknowninthisstudyisthedifferenceinactiveroot study).Improvedunderstandingofrootingbehaviourandsoil
biomassprofilesbetweensites.IftheactiverootsatManaus depth, along with improvements in soil hydraulic informa-
are substantially shallower than at Caxiuana˜, and access a tion, are necessary to disentangle the effects of abiotic soil
smallervolumeof soil, then this may also contribute to the propertiesandofrootingbehaviour.
differencesinwatersupplybetweensites.AtCaxiuana˜,water Anotherunknownisthelowertensionlimitofextractionof
extraction has been observed down to 5.0m (Fisher et al., soilwater.Inthisstudy,wehaveusedtwocontrastinglimitsto
2007).Hodnettetal.(1996)observedwaterextractiondownto investigatetheconsequencesforourpredictionsofaddressing
atleast3.6mdepthatasitenearManaus.Boththesestudies this uncertainty. It is commonly assumed that plants may
detected soil water extraction at the maximum depth extractsoilwaterdownto(cid:2)1500kPa,the‘wiltingpoint’(Cox
observed.Theabsenceofmeasurementsofsoilwatercontent etal.,1998;WerthandAvissar,2004).Recently,itwasproposed
below these depths means that the active rooting depth using a modelling analysis that changes in water potential
cannotbeestimated.Forthisstudy,itisofcoursepossiblethat alonecouldnotexplainthepatternsofgasexchangeobserved
verydeeprootsatCaxiuana˜ mightbethereasonforthedry atManaus,andthatitwasalsonecessarytoinvokechangesin
seasonresilience.However,weknowrootandwateruptake soil-to-leafhydraulicconductancetoexplainthepredictions
occursdowntoatleast3.6matManaus(Hodnettetal.,1996). (Williams et al., 1998). This finding was supported by Fisher
AtCaxiuana˜,thisrootingdepthwouldeasilyprovideenough etal.(2007)usingamodelwhichpredictedsoilwateruptake
watertocoverthepredictedsoilwaterdeficitof210mm(this resistance‘apriori’fromsoilwatercontentandrootbiomass
depth of soil being predicted to provide between 496 and profiles.Animportantimplicationofthisproposedhydraulic
720mm, depending on the assumed uptake limit, but not resistance limitation mechanism is that the rate of water
allowingforchangesinsoilpropertieswithdepth),sorooting supply to leaves might be severely limited long before the
depth differences cannot easily explain the different absolute limit of extraction is reached and where roots are
responses. sparse, plant water use can be reduced at quite high water
Rootshavebeendetectedat10mdepthatCaxiuana˜,andat potentials.Therefore,todiscuss‘limits’ofwateruptakeisto
a number of other Amazonian sites (Nepstad et al., 1994; somedegreemisleading,astranspirationmaybereducedby
Moreira et al., 2000; Nepstad et al., 2004), but at very low slowwatermovementdespitenoabsolutethresholdofwater
density.Itisnotknownhowactivetheserootsareinwater contentextractionhavingbeenreached.Thiswasempirically
uptake but some studies have suggested that deep roots demonstrated by Fisher et al. (2006) who found that sub-
agricultural and forest meteorology 148 (2008) 850–861 859
stantial reduction in water use of canopy trees subjected to work was supported by a University of Edinburgh Faculty
artificiallyreducedsoilmoistureoccurredinthedryseasonat ResearchScholarship,severalNaturalEnvironmentResearch
Caxiuana˜,despitetheexistenceofonlyslightlynegativesoil Council (UK) research grants, a Natural Resources Interna-
waterpotentialsof(cid:2)0.2to(cid:2)0.6MPa.Thesewaterpotentials tional Foundation Fellowship. We would like to thank Edso
wereestimatedfrompre-dawnleafwaterpotentialmeasure- Veldkamp for advice on soil moisture sensor calibration,
ments, a proxy for the ‘average’ soil water potential of the RafaelFerreiradaCostaandLuizAragao˜ forfieldassistance
profile, assuming the system is in hydraulic equilibrium. andtheMuseuParaenseEmilioGoeldifortheuseoftheirfield
Furthermore,inthedryseason,therewasalargeincreasein stationandlaboratoryfacilities.
themeasuredbelow-groundhydraulicresistance,supporting
the hypothesis that below ground resistance to water
transportcouldbeseverelylimitingforestwateruse.Never- Appendix A. ‘ACM-ET’ Potential
theless, for the purposes of this paper, it is necessary to evapotranspiration model
demonstratethatthesoilpropertiesatCaxiauna˜ andManaus
differ substantially in terms of water retention, so it is Potentialevapotranspirationwasmodelledastheproduct
necessarytoartificiallyimposearangefortheavailableplant of the atmospheric demand and the surface conductance
water content. Here we have used two extreme extraction using a modified version of the Penman-Montieth equation
limits,anupperandlowerestimate.Soilwateravailabilitywas (Jones,1992).
substantially higher at Caxiuana˜ in both instances, so our
p ðsðI(cid:2) p Þþc g Vr
findingsare notaffectedby this uncertainty. In additionwe D¼ 1 2 p a a
sþlðg =g Þ
extended the analysis to very low soil water potentials a c
((cid:2)5000kPa, data not shown) to investigate the likelihood of where I is average daily radiation (Wm(cid:2)2), V is the average
water being released from the clay micropore matrix of the dailyvapourpressuredeficit(kPa),andg issurfaceconduc-
c
Manaussoil.Thisanalysispredictedthat,atManaustherewas tance.Listhepsychrometricconstant(0.066kPa8C(cid:2)1)c isthe
p
less than 2% additional water extraction, and at Caxiuana˜ specificheatofair(1010JKg(cid:2)1K(cid:2)1)andristhedensityofair
therewasbetween1.3and3.8%additionalwaterextraction. (1.2Kgm(cid:2)3). p is a correction between the average daily
1
Thisthereforesupportsratherthanunderminesthehypoth- radiation and the radiation observed. p is the intercept of
2
esisthattheCaxiauna˜ soilhashigherplantavailablewater. the relationship between radiation and ET. These empirical
IfKisanimportantdeterminantofforestwateruse,viaits fitting parameters are necessary because the Penman-Mon-
impactonwatertransport,thenitwillbenecessaryeventually teithisaninstantaneousequation,buttheinputparameters
topredictitataregionalscaleusinganempiricalmodelofsoil aredailyvalues.sisthechangeinslopeofthevapourpressure
properties. It is clear from the lack of data (Tomasella and curvewithtemperature,T(K):
Hodnett,1997)onthisimportantsoilproperty(whichisalso
12:27T
necessaryforinfiltrationandrunoffmodelling),thatfurther s¼6:166
273:3þT
investigation is an important priority for understanding
Amazonianforestgasexchangeprocesses. Surfaceconductanceisdefinedas:
Themainconclusionofthisstudyisthatwhiletherewere pffiffiffi
nodifferencesintotalsoilwaterdeficitbetweentheManaus gc¼ p3LCep4I R
andCaxiuana˜ sitesintheyearsconsideredforthisanalysis,
there were very large differences in the water retention where C is the difference between the minimum leaf water
capacityofthesoils.Wesuggestthatthemainfactordriving potentialandthebulksoilwaterpotential(MPa).Risthesoil-
thedifferentbehavioursislikelytobetheverylowsoilwater to-leafhydraulicresistance(sm(cid:2)2MPammol(cid:2)1)andListhe
retentioncapacityoftheclayoxisolatManauscomparedto leaf area index (m2m(cid:2)2). g is increased by high LAI and C
s
the sandy oxisol at Caxiuana˜. Some uncertainty remains values, and is limited when both radiance and soil-to-leaf
surrounding the impact of deep roots and vegetation hydraulicresistancearehigh.Thisisbecausehighirradiance
strategies regarding water stress avoidance. However, we is linked to high atmospheric demand. In times of low soil
believethatthescarcityofgoodsoilhydraulicsinformationfor wateravailability,thistendstoresultinstomatalclosure.p3
Amazoniaremainsanimpedimenttothecreationofaccurate and p4 are further fitting parameters. The parameters were
models of biosphere–atmosphere exchange, particularly for optimisedtofittheevapotranspirationdatafromtheoutputof
the purpose of identifying which forest areas may be theSPA model.These valuesincludethe effectsof soileva-
particularlysensitivetoclimaticchange. porationandleafsurfaceevaporation.Thevaluesofthefitted
parametersasdeterminedbytheoptimisationroutinewere:
Acknowledgements
p1¼2:74;p2¼6:22;p3¼4:7(cid:3)10(cid:2)5;p4¼3:0(cid:3)10(cid:2)2
The work presented in this paper would not have been
possiblewithouttheinputofthelateDrWimSombroek,and The r2 value between the SPA model and the aggregated
theauthorswouldliketodedicatethispapertohismemory. model was 0.87. The model residuals did not correlate with
Inaddition,wearegratefultoJavierTomasellaandMartin any of the input variables. The averagerelative modelerror
Hodnettformanyhelpfulcommentsonthemanuscriptand was12%.TheSPAmodeloutputhasbeenverifiedagainstsap
for the provision of soil hydraulics data from Manaus. This flow data from both well watered and drought-stressed sap