Content uploaded by Paulo M Brando
Author content
All content in this area was uploaded by Paulo M Brando on Jun 26, 2019
Content may be subject to copyright.
Glob Change Biol. 2019;00:1–14. wileyonlinelibrary.com/journal/gcb
|
1
© 2019 John Wiley & Sons Ltd
Received:31October2018
|
Revised:13M arch2019
|
Accepted:31March2019
DOI : 10.1111/gcb .14659
PRIMARY RESE ARCH ARTICLE
Prolonged tropical forest degradation due to compounding
disturbances: Implications for CO2 and H2O fluxes
Paulo M. Brando1,2 | Divino Silvério2,3 | Leonardo Maracahipes‐Santos2 |
Claudinei Oliveira‐Santos2,4 | Shaun R. Levick5,6,7 | Michael T. Coe1 | Mirco Migliavacca7 |
Jennifer K. Balch8 | Marcia N. Macedo1,2 | Daniel C. Nepstad9 | Leandro Maracahipes2 |
Eric Davidson10 | Gregory Asner11 | Olaf Kolle7 | Susan Trumbore7
1WoodsHoleResearchC enter,Falmouth,
Massachusetts
2InstitutodePesquisaAmbient alda
Amazônia(IPAM),Brasília,Brazil
3EcologyDepar tment,UniversityofBrasí lia,
Brasília,Brazil
4FederalUniversityofG oiás,Goiânia,Brazil
5Charle sDarwinUniversity,Dar win,NT,
Australia
6CSIROTropicalEcosystemsRe search
Centre,Darwin,NT,Austr alia
7MaxPlanckInstituteforB iogeochemistry,
Jena,Germany
8GeographyDepartm ent,Universityof
Colorado‐Boulder,Boulder,Colorado
9EarthInnovat ionInstitute,SanFrancisco,
California
10AppalachianLabor atory,Universityof
MarylandCenterforEnvironmentalScience,
Frostburg,Mar ylan d
11CenterforGlobalDiscover yand
Conser vationScience,ArizonaState
University,Tempe,Ar izona
Correspondence
PauloM.Brando,WoodsHoleResearch
Center,149WoodsHoleRd,Falmouth,MA
02540,USA.
Email:pbrando@whrc.org
Funding information
Max‐Planck‐Gesellschaft;Conselho
NacionaldeDesenvolvimentoCie ntífico
eTecnológico,Grant /AwardNumber:
441703/2016‐0an d460494/2014‐7;
Gordona ndBettyMooreFoundation;
DivisionofEnvironmentalBiology,Grant /
AwardNumber:1146206;Nationa lScience
Foundation,Grant/AwardNumber:
1146206;Borders,Gr ant/AwardNumber :
40580 0;USA ID;DepartmentofState;
EMBRAPA;USForestSer vice
Abstract
Drought, fire,and windstormscaninteracttodegradetropicalforestsandtheeco‐
system ser vices they provide, but how these forests recover after catastrophic
disturbanceeventsremainsrelativelyunknown.Here,weanalyzemulti‐yearmeas‐
urementsof vegetation dynamics and function (fluxes of CO2andH2O)in forests
recoveringfrom7yearsofcontrolledburns,followedbywinddisturbance.Located
insoutheastAmazonia,theexperimentalforestconsistsofthree50‐haplotsburned
annually,triennially,ornotatallfrom2004to 2010.During thesubsequent6‐year
recoveryperiod,postfiretreesurvivorshipandbiomasssharplydeclined,withabove‐
groundCstocksdecreasingby70%–94%alongforestedges(0–200mintotheforest)
and 36%–40% inthe forest interior.Vegetation regrowth in theforestunderstory
triggere d partial can opy closure (70%–80%) fro m 2010to 2015. T he composition
andspatialdistributionofgrassesinvadingdegr adedforestevolvedrapidly,likelybe‐
causeofthedelayedmortality.Fouryearsaftertheexperimentalfiresended(2014),
thebur nedpl otsas si milated36%lessc arb ontha nt heControl,bu tnetCO2exchange
andevapotranspiration(ET)hadfullyrecovered7yearsaftertheexperimentalfires
ended (2017). Carbon upta ke recovery occurred largely in respo nse to increased
light‐useefficiency andreduced postfire respiration, whereas increased wateruse
associated withpostfire growthofnew recruits andremaining trees explained the
recovery in E T.Alt hough the effe cts of interac ting disturba nces (e.g., fires, fo rest
fragmentation, andblowdown events)onmortalityandbiomass persist over many
years,therapidrecoveryofcarbonandwaterfluxescanhelpstabilizelocalclimate.
KEY WORDS
disturbance,recovery,resilience,tropical,wildfires
2
|
BRANDO et Al .
1 | INTRODUCTION
Episodic droughts andwindthrowevents strongly shapethe struc‐
ture,dynamics, and diversity oftropicalforestsby killingtrees and
altering c ompetition for l imiting resour ces (Davidson et al. , 2012;
Zarin, Davidson, & Brondizio, 2005). As people clear, thin, and ex‐
ploit tropical forests, natural disturbances increasingly interact
with und erstory f ires, logging a ctivities , and forest fr agmentatio n
(Brandoetal., 2014; Cochrane et al., 1999;Davidsonetal., 2012).
Combinedwithclimatechange,disturbancesmayoccurtoofastfor
plantcommunitiesto adapt acrosslargeforestedareas.Despitere‐
silience tooccasionalevents, tropicalforestsmaynot recovertheir
structure,composition,andfunctionfollowingmultipleandfrequent
disturbances(Trumbore,Brando,&Hartmann,2015).
In southern Amazonia, extreme weather events and agricul‐
turalpracticeshavealready intensifiedforestfireregimes(Alencar,
Brando, Asner, & Putz, 2015; Morton, Page, DeFries, Collatz, &
Hurtt, 2013). Recent severe droughts have triggered forest fires
that are lar ger and more fr equent (Al encar et al., 2 015), spanning
awiderrangeofdry‐seasonmonths (Jollyet al., 2015).Evenas de‐
forestationratesdeclined,forestfireshavecontinuedtoburn large
areas, including bothhistorically fragmented landscapes(Aragão &
Shimabu kuro, 2010) and lar ge tract s of protected fo rests (Br ando
etal., 2014).In the 200 0s, forexample, more than 85,000 km2 of
forestsinsouthernAmazoniaburnedduetosynergisticinteractions
betweenhuman activities and droughts(Morton et al.,2013). The
capacity of these fire‐disturbed forests torecover remains poorly
unders tood, par tially due to t he complexi ty of process es involved
(Floresetal.,2017),whicharedeterminedbyinteractionsbetween
local site factors, landscape history and structure, regional spe‐
ciespools, andspecieslifehistories (Chazdonet al., 2007;Derroire
etal.,2016).
BurnedforestsoftheAmazoncanfollowawiderangeofrecov‐
erytrajectories,butthetwoextremesarefullrecoveryortransition
toanewstate(Trumboreetal.,2015).Withnofundamentalchanges
in climate, most burned forests recover biomass within decades
(Saldarriaga,West,Tharp,&Uhl,1988)andclimate‐relatedfunctions
(e.g.,CO2andH2Ofluxes)withinafewyears(Chazdonetal.,20 07).
Despite theoverall high resilience of tropical forests,catastrophic,
compounding disturbanceevents can slow or prevent recovery by
killing ad ult trees (Bar low, Peres, Lag an, & Haugaase n, 2003) and
depletingseedandresproutbanks(Kennard&Putz,2005).Although
severalstudiesreportedincreasednutrientavailabilityshortlyafter
disturbances,thereisalsoevidenceofprolongedandseveredistur‐
bances re ducing soil nut rient pools thr ough soil gas emis sions, as
wellashydrologicleachingor fire‐related volatilizationofnutrients
(Davidsonetal.,2004;Davidsonetal.,2007;McGrath,Smith,Gholz,
&Oliveira,20 01;Zarin et al., 2005).Inaddition,increasedpostfire
mortality ofcanopy trees creates niche opportunities forlight‐de‐
manding pasture grasses and lianas that can outcompete woody
nativespecies(D'Antonio,Hughes,&Vitousek,2001).Finally,wind‐
storm‐related tree snapping, uprooting, and canopy damage can
elevate postfire mortality rates of largetreesforseveralyearsand
delayrecovery(Barlowetal.,2003;Silvérioetal.,2019).
In the years following catastrophic disturbances, successional
dynamics likely influence ecosystem carbon and water fluxes.
Disturbancesassociatedwithblowdownsandwildfiresreducecom‐
petitionforlight, nutrients,and water (van derSande et al.,2017),
favoring “p ioneer” sp ecies capa ble of assimilat ing and using t hose
extra resourcesmoreefficiently (Saldarriagaet al., 1988).Although
theseshort‐livedspeciestranspiremorewatertosupporttherapid
produc tion of wood and l eaves, they may l ack the deep ro ot sys‐
tems that normally support high evapotranspiration (ET) in sea‐
sonally dry forests(e.g., dry season > months; Nepstad, Carvalho,
Davidson,&Nature,1994).Large,deep‐rootedtreessurvivingmajor
distur bances may benef it from reduced co mpetition for d eep soil
water(Nepstadetal.,2001).However,disturbance‐relateddamages
to their crow ns, stems, a nd roots co uld limit reso urce assimila tion
and use‐efficiency. Over time, competitionamong pioneer andre‐
mainingtreesislikely to increase; carbon‐ and light‐useefficiency
(LUE)arelikelytodecrease;andwateruseislikelytobecomemore
conservative (Chazdon et al., 2007). How long postdisturbance re‐
covery wi ll take remains un clear. Studies that us e remote sensing
imagery,chronosequences,vegetationmodels,andeddycovariance
(EC) techniques predict recovery of productivity and ET ranging
fromafewyearstoseveraldecades(Arroyo‐Rodríguezetal.,2017;
Chazdon et al.,20 07;Miller etal., 2011).Rates of changecanvary
considerably accordingtotheintensit y,duration, andfrequencyof
thedisturbance(Chazdonetal.,2007).
Tounderstand how compounding disturbancesinfluence forest
struc ture and funct ions (e.g., net ecosys tem exchange [NEE] and
ET),westudiedpost firedynamicsinasetofneotropicalforestplots
(50ha)thatwereexperimentallyburned(annually[B1yr],triennially
[B3yr] or un burned cont rol) from 20 04 to 2010, and imp acted by
ablowdownevent in 2012(Silvério et al., 2019; Figure S1).These
highly de graded fore st areas coul d follow diffe rent trajec tories of
recovery,leadingtoalternatestates—eitheradegraded,derived‐sa‐
vanna envi ronment or re covery of a close d‐canopy fo rest ecosy s‐
tem(e.g.,lowresilience).Wehypothesizedthatdelayedmortalityof
large trees would perpetuate postfire degradationand grass pres‐
enceduringthe7‐yearpostfireperiod(Barlowetal.,2003),causing
reductions in ecosystem‐level ET and net CO2 uptake (e.g.,higher
NEE). The alternative hypothesis is thatbiomass and canopy cover
would recover rapidly, driving (a) the replacement of grasses by
sha de‐tolerantspecie sand(b)r ecoveryofE Ta ndCO2uptaketolev‐
elssimilartotheunburnedControltreatment.
Overall , we expecte d differe nces in fores t struct ure and func‐
tioningtodecreaseovertimebetweenburnedandunburnedareas,
albeit moreslowly where fireand winddamage weremore severe
(e.g.,B3yrtreatmentandforestedges;Brandoetal.,2014).Wefirst
reviewpreviousfindingsfromtheinitialphasesoftheexperimental
firesandpresentnewresultsquantifyingpostdisturbanceforestre‐
coveryanditsconsequencesforclimaticecosystemservicesassoci‐
atedwithNEEandET.
|
3
BRAN DO et Al .
2 | MATERIAL AND METHODS
2.1 | Description of area and fire experiment
The stu dy area is locate d on the Fazenda Tanguro (83, 000 ha) in
Mato Grosso state, 30 km north ofthe southern boundary ofthe
AmazonrainforestinBrazil(FigureS1).Thesiteislocatedwithinthe
driest portionof the basin (13°04′S, 52°23′Ν), a region character‐
izedbyaprolongeddryseason(4–5months),withannualprecipita‐
tionof1,770mm(Balchetal.,2008).
The experimental area consisted of three adjacent 50 ‐ha plots
burnedannually(B1yr),triennially(B3yr),ornotatall(Control)from
2004to2010.Inburnedplot sareas,weignitedfireswithkerosene
drip torchesalong transects spaced50mapart (details in Balchet
al., 2008). Prior to the first experimental fire (2004), aboveground
biomass (ABG) in the Control was 11.1% and 14.1% higher than
B1yrandB3yr,respectively,whereascanopygreennesswassimilar
amongthethreetreatmentplots(Figure1).Incontrast,the burned
plots ho used more tr ee species a nd had higher l itterf all along th e
forestedges.
2.2 | Ecological measurements
Weconducted pre‐ and postfire inventoriesacross the threetreat‐
mentplots.InJuly2004,wetagged, mapped,andmeasuredheight
(m)and diameterat breastheight(dbh)oftreesand lianasindiffer‐
entstrataforeachtreatment(Balchetal.,2008).Withineach50‐ha
treatmentplot, we sampled all trees ≥40 cm dbh (~930 individuals
perplot).Wesampledtreesandlianaswith20–39.9cmdbhalongsix
transects,whichranparallelwiththeedgeoftheagriculturalfieldat
10,30,100,250,500,and750mfromtheedge(500×20m;5.5ha
sampledpertreatmentplot;~880individualsperplot).Nestedsub‐
samplin g was conducted w ithin these tra nsects to mea sure trees
andliana swith10–19.9cmdbh(50 0×4m;1. 2hasamp le dpert re at‐
mentplot;~490individualsperplot).Werepeatedtheseinventories
annuall y within each 5 0‐ha plot (det ails in Balc h et al., 200 8). We
mappedwoodyspeciesthatenteredthe10cmdbhsize‐classinour
inventoryin2008,2009,2010,2012, 2014, and 2016. During this
period,we also remeasuredtree dbh of individualsusing a diame‐
ter tape to c alculate tr ee growth . Toguar antee the sa me point of
measurementineachvisit,weuseda1.3‐mruler.Theseinventories
servedasABGcensusesofwoodyspecies(treesandlianas)≥10cm
indbh.WeusedtheallometricequationfromChaveetal.(2014)for
dry forests toestimatetree biomass fromtree dbh,wood density,
andheight. Data on ABG biomasspublished in Brandoetal. (2014)
wereextendedfrom2009to2016.
We estimated postfire leaf area index (LAI) each year of the
study(200sitesperplot;dryandwetseasons)usingtwoLiCor200 0
PlantCanopyAnalyzers(LI‐CORBiosciencesInc,Lincoln)from2005
to 2018. One in strument was p laced in an adjace nt open field to
measure incoming radiation with no canopy influence; the other
instrument wassimultaneously used to take understory measure‐
ments . The two instr uments were int ercalibrate d before each set
of measurementsin the open field. The light fieldof each sensor
wasreducedto90%usingopaquesensorcaps.Measurementswere
taken 1 m fro m the ground du ring diffus e light conditi ons—either
before08:0 0hoursorafter17:00hours.LAIcalculationsweremade
usingtheinnerfourquantumsensorrings.After weendedtheex‐
perimentalfires,wecontinuedtoconductLAImeasurementsevery
3 months. B rando et al. (2014) publis hed LAI data fr om 2005 to
2009.Here,weex tendthistimeseriestoDecember2017.
Litter fall was coll ected biwee kly from Augus t 2004 to Au gust
2018using0.5m2screentraps(N:90pertreatmentplotfrom2004
to 2012; N: 70 per treatmentplot from2013to 2018) placedsys‐
tematicallythroughouttheplots.Wesuspendedthelittertraps1m
abovetheforestfloor.Woodydebris(>1cm)wasexcludedfromour
samples(Clark et al., 2001).Litter wasoven‐dried(65ºCfor48 hr)
andweighedforcalculationsofdrymass.
Toobtainratesofgrassinvasionovertimeinourexperiment,we
mappedin2012and2015theadvanceofgrassesandareainvaded
byeachspeciesfromtheedgeintotheforestinterior.Thismapping
consistedofthepresenceofgrasseswithinsubplot s(5×5m)where
thegrasscoverwasgreaterthan50%(Silverioetal.,2013).
AirborneLiDARdatawerecollectedbytheGEOIDLtda.Company
(Belo Horizonte, MG) as par t of the SustainableLandscapes Brazil
project (https://www.paisagenslidar.cnptia.embrapa.br/webgis/).
Twosu rveys were con ducted, t he first in A pril 2012 and th e sec‐
ondinOctober2014.The studyarea was flownat an averagealti‐
tudeof850ma.s.l.andcoveredanareaofapproximately1,0 05ha.
AnOptechALTM‐3100laserscannerinstrumentwas usedin 2012
and an Opte ch ORION‐0 9SEN243 in 2014; average return d ensi‐
ties were13.7 per m2in2012and41.05perm2 in 2014.Full and
reduced‐densitydat asetswere processed to generatet hecanopy
height mod el (CHM) abovegrou nd raster layers at 1‐m reso lution.
CHM produ cts were gen erated using t he G‐LiHT algor ithm (Cook
etal.,2013)byselectingthehighestLiDARreturninevery1‐mgrid
FIGURE 1 Annualpatternsinenhancedvegetationindex(EVI;
MODIS13Q1collectionsix)from2003to2017fortheControland
burnedplots.Thesolidlinerepresentsasecond‐orderpolynomial
model;thedashedlineathresholdlinearmodel;pointstheaverage
EVIforagivenplot;andredlinesthedeviationfromtheobserved
andpredictedEVI(basedonthethresholdmodel).Negativevalues
representhighervaluesintheburnedplotsthanintheControl
15
10
5
0
5
03 04 05 06 07 08 09 10 11 12 13 14 15 16 17
Year
Burned Control (% difference in EVI)
Drought Drought
Pre-Fire Burning period Recovery period
Blowdown
xx x x xx x xx
4
|
BRANDO et Al .
cell,building aTINbasedonthesepoints,andinterpolatingcanopy
heights ona1‐mrastergrid(moredetailsinLongoetal.,2016).We
alsoestimatedleafareadensity(LAD)acrosstheforestprofileusing
theLADfunctioninthe R packagelidR, usingmethodsinBouvier,
Durrieu,Fournier,andRenaud(2013).
2.3 | Eddy covariance flux
Tomeasure how forestrecoveryinfluencedcarbon,water,anden‐
ergy fluxes, we established two36‐m EC towersinlate 2014, one
inthe Control (Figure S1 and Figure 2) andanotheron the border
betweenthetwo burnedplots (FigureS1).Bothtowersuse identi‐
calinstrumentsetups consistingofa3D‐sonicanemometer(USA1;
METEK, Elmshorn, GER) and an infrared gas analyzer (LI7200;
LI‐COR Biosciences), measuring the three‐dimensional wind com‐
ponents (u, v,w), sonic temperature (Ts),a nd mixing rat ios of CO2
andH2O.Thesamplingfrequencywas20Hzandthemeasurement
height36mabovegroundand17maboveaveragecanopyheight,re‐
spectively.Half‐hourlyfluxes werecomputedusingEddyPro(6.2.0,
LI‐COR Biosciences).Raw data were despiked ( Vickers & Oceanic,
1997),block averaged, and lag correctedwithinadefined window
oftime,subjecttoadoublerotationprocedure.Qualitycontrolfol‐
lowedpreviouslypublishedmethods(Foken& Wichura,1996).We
discardedalldatawithlowfric tionvelocity(u*<0.14m/s).Weeval‐
uatedtheenergybalanceclosure(Foken,Aubinet,&Leuning,2011)
byregressingtheavailableenergy(Rnet)fromoutgoingandincoming
short‐andlong‐waveradiationagainsttheRnetfromlatentandsen‐
sibleheatfluxes(FigureS2).
Themarginaldistribution sampling methodwas used to gap‐fill
half‐hourly fluxes(Reichsteinetal.,2005)withseveralmeteorolog‐
icalvariablescollectedbyourECsystems, includingphotosyntheti‐
cally activeradiation(PAR), vaporpressure deficit,global radiation,
air temperature, and relative humidity. The nighttime partition‐
ing method implemented in the REddyProc R Package (Wutzler,
Lucas‐Moffat, Migliavacc a, & Knauer, 2018) was applie d to com‐
pute ecos ystem respirat ion (Reco) and gross ecos ystem produc tiv‐
ity(GEP).Wealso calculated GEP and Recousingdaytimedataonly
(Lasslopetal.,2009),whichyieldedsimilarresults,sowereportonly
thenighttimemethod here.LUE wasestimatedastheratioofGEP
and the fr action of ab sorbed PAR, w hich was esti mated from L AI
andPARfollowingWuetal.(2016).
Two‐dimensional(2D) flux footprints were calculated for each
halfhourusing the footprintmodel ofHsieh, Katul,andChi(2000)
withthelateral dispersionterm accordingto Detto,Katul,Mancini,
Montaldo,and Albertson (2008) and then rotatedinto the respec‐
tive mean w ind direction . 2D footprints wer e calculated only f or
half‐hourly data whenfriction velocitywashigherthan0.14m/sto
ensurethatameaningfulturbulenttransportwaspresent.Foreach
half‐hourfrom08:00hoursto18:00hours,we computedthefoot‐
printprobabilitydensityfunction(PDF).Foragivenx and ylocation,
wecomputedthefootprintclimatologyasthemeanoftheindividual
footprintPDFsforthemeasurement period.The resulting 2DPDF
isthefootprintclimatology,providinginformationontheportionof
theecosystemsampledinthemeasurementperiod.Duringtheday
(Rg > 50 W/m),th e burned and C ontrol plot s contribut ed most of
thefluxessampledbytheeddyfluxtowers.Giventheprevalenceof
northwinds, the eddy flux towerlocated intheburnedplots sam‐
pled mostly the vegetation growing close to the edge, amore de‐
graded environment;thetowerlocatedinthe Controlplotsampled
amorepristinearea.LAIaroundthetowerlocatedintheburnedplot
was subst antially lower t han in the Contr ol (Figure S3). Howeve r,
duringcalmnights,weobservedsomeoverlapbetweentheburned
andControlplots(Figure3).
2.4 | Soil moisture
Soilmoisturewasmeasuredmonthlyindeep(0.5–8m)soilpitsfrom
2010to2018,usingtimedomainreflectometry(TDR;Jipp,Nepstad,
Cassel,&ReisDeCarvalho,1998).Thenumberofsoilpitspertreat‐
mentvariedovertime.Initially,weinstalledTDRsystemsintwosoil
pits, onein B1yrand one in theControl. Weadded t wo more pits
per treatment inthe following year (2011), but one of them failed
in 2015 and was r eplaced in 2016. We us ed soil moist ure data to
estimatetranspirationduringrainlessperiods(10ormoreconsecu‐
tivedays with no precipitation), including most dry‐seasonmonths
(June–October). Weassumedthatthechangeinsoilmoisturefrom
one day to the next(summed from 0.1 to 8 mdepth) represented
theETratesduring dryperiod. Because soil evaporation and base‐
flowaccountforasmallfractionofthetotalchangeinsoilmoisture
during th e dry season (Dav idson et al., 2011; Markewi tz, Devine,
Davidson,Brando,& Nepstad, 2010), thisestimateof ET probably
representsmostlyplanttranspiration.
2.5 | Statistical analysis
We used linear mixed models (Bates, Mächler, Bolker, & Walker,
2014) to test for dif ferences a mong treat ments. Th ese model s in‐
clu de dt heresponsev ar iableofinter es t(e.g.,treegrowth,L AI ,l it te r‐
fall)andpredictorssuchas“treatment”andyear.Weusedindividual
trees as o ur sample unit s for tree wood incre ment and mort ality
FIGURE 2 Viewoftheeddycovarianceopen‐closedsystem
locatedinsouthernAmazonia
|
5
BRAN DO et Al .
models.Treegrowthmodelsincludedrandomeffectofspeciesand
subplot s to minimize unwante d sources of varia bility. Toev aluate
theprobabilityoftreemortality,weusedageneralizedlinearmixed
model (family: binomial) that included tree mortality/survivorship
astheresponsevariable; treesize,wooddensity,and treatment as
predictors;andrandomeffectsof speciesandsubplot. Toestimate
confidenceintervalsonthepredictedvariables,wegeneratedanew
datasetinwhichallvariableswerefixedtotheiraveragevalues.We
thenbootstrappedthemodelpredictions using the func tion “boot‐
Mer”(Batesetal.,2014).Previousstudiesconductedatthissite(e.g.,
Brandoetal.,2014)showedthattheforestinteriorandedgesofthe
burnedplotsdifferedintreemortalityrates,canopycover,andgrass
invasion rates. We therefore included a categoricalvariable repre‐
sentingforestinteriorandedgeinmostofourmodels.Allerrorbars
representbootstrappedconfidenceintervals.
3 | RESULTS
3.1 | Experimental fires
Our previ ous studies cond ucted in the regio n during nondrou ght
yearsshowedthatthefirstprescribedfiresreleasedsmallamounts
ofenergy and caused modestchanges in foreststructure (Balchet
al.,20 08;Brandoetal.,2014).Incontrast,theprescribedfirescon‐
ducted during the 20 07droughtweremore intense (Brandoet al.,
2014),killed a higher proportion of woodyindividuals, andcaused
sharpdeclinesinLAI(B1yr:50%;B3yr:49.4%),litterfall(B1yr:19.8%;
B3yr: 3 6.3%), and ABG ( B1yr:12 .8%; B3yr : 17.7%) d uring the fol‐
lowing year ( 2008), par ticularly alo ng the forest edg es (Figure 4).
Silvérioetal.(2013)foundthat,ascanopycoverdeclinedandmore
lightreached the forestunderstoryafter thesefires,grassesbegan
invadingtheburnedforestedges,creatingaflammableenvironment
even during nondrought years (e.g., 2009). In 2010, the southeast
Amazon regionexperienced anothersevere drought(Brandoetal.,
2014).Weshowthatfire‐inducedtreemortalityabruptlyincreased
inthe yearfollowingthis drought,especiallyalongtheforestedges
andinB3yr(Figure4;TableS1).ThesefiresfurtherreducedLAI,lit‐
terfall,and AGB biomass(Figure 4). Combined withthe availability
ofpropagules,fire‐relatedchangesinforeststructurefacilitatedthe
invasion of gr asses along m uch of the fores t edges of the bu rned
plots(Silvérioetal.,2013).
3.2 | Forest recovery period
With the e nd of the prescribe d fires in 2010, the experiment al
forest s entered a recove ry phase c haracter ized by two contr ast‐
ingpat terns.First,postfiremortalit yoflargetreescontinuedtobe
substantiallyhigherthanintheControl(Figure4;TableS1).Awind‐
storm th at impacted th ee xperiment al plots in 2012 pun ctuated
thispattern.Fire‐damagedtreeshadtrunksthatweremorevulner‐
abletowindbre akageandcrow nst hatweremoreexposedtowind
andassociateduprooting(moredetailsin Silvério et al., 2019).As
largetreesdied,AGBbiomasscontinuedtodropacrosstheburned
plots, reaching their lowestlevels by the endoftherecover ype‐
riod.T hesepatter nsweremorepron ouncedalongthefores tedges
oftheburnedplots,particularlyofB3yr(Figure4;TableS1).
The second pattern we observed duringthe recovery period
was vigorou s vegetation g rowth. Bet ween 2011 and 2016, 63%
ofthe species common to all three treatment grewfasterin the
burned plotsthanin the Control(Figures S4 and S5),andgrowth
ratesincreasedwithtreesizeintheburnedplots(FigureS6;Table
S2).Thisresultedinhigheroverallgrowthandpostfirecarbonac‐
cumulationperindividualtreeintheburnedplots(Figure5;Table
S3),althoughgrowthandcarbonaccumulationweresimilarduring
someperiods (e.g.,between2012 and2014).Newrecruit senter‐
ing our inventory during the recovery period also grew faster in
theburnedplots,mainlyduetofourrapid‐growthspecies(Mabea
fistulifera, Tachigali vulgaris, Cordia bicolor, Casearia grandiflora;
Figure 6). In co ntrast, new r ecruits sa mpled bet ween 2004 a nd
20 0 8(i .e .,b urn ing per iod )we recom pri sed ofamo rediversegr oup
ofslow‐growthspeciesinallthreetreatmentplots,especiallyin
theControl(Figure6).
Theoverall increasedpostfire growthcaused ashift in the tra‐
jectoryofcanopydynamics,asrepresentedbyincreasesinLAI(late
2014), litter fall (mid‐2013), and EVI (e arly 2012; Figu res 1 and 4).
Althoughlitt er fallandEV Ir ec overedtov al ue ssimilartot heCo nt ro l
FIGURE 3 Estimatedfootprint
probabilitiesforourtwoeddycovariance
towers,representingdaytime(bet ween
07:00hoursand17:00hours;leftpanel)
andnighttime(between18:00hours
and06:0 0hours;rightpanel)fluxes
measuredbetweenearly2014andlate
2017.Two‐dimensionalfootprintswere
calculatedonlyforhalf‐hourlydatawhen
frictionvelocitywashigherthan0.14m/s
toensurethatameaningfulturbulent
transportwaspresent
0.00 0.05 0.10 0.15 0.20
Forest
Daytime Nighttime
ControlB3yrB1yr ControlB3yrB1yr
Probability
6
|
BRANDO et Al .
by2016 (Figures1and4),L AI remainedlower inthe burnedplots
than in th e Control alon g both forest ed ges (Control: 4.1 m2 m‐2 ;
B3yr :2.3m2 m‐2 ;B1yr:1.1m2 m‐2 )an di nteriors(Control:4. 6m2 m‐2,
B3yr:3.4m2 m‐2;B1yr:2.0m2 m‐2 ),withsharpdeclinesinlate2016.
Comparison of repeated LiDAR measurements in 2012 and
2014capturedcontrastingpatternsinvegetationstructureduring
awindowoftherecoveryperiod.First,lossofunderstoryvegeta‐
tion(height<3m)wasobservedas adecreasein height between
2012 and 2014, as rep resented by change s in frequency d istri‐
butions of L iDAR returns f rom the vegetat ion canopy (F igure 7;
Figure S7 ). This occurre d across 4.9% (B1yr) and 7.9%( B3yr) of
the experimentally burned plot s, but only 1.5% of the Control
plot.Second,vegetationregrowth—detectedasanincreaseinthe
forest understory height—wasobserved across 19.5%(B1yr) and
17.5%(B3yr)ofthe burnedplotarea,comparedwith2.5%of the
Control.Onaverage,lossesinforestheightoccurredat18m,and
postfiregrowthbetween2 and3 m. In general, our estimates of
LADderivedfromLiDARsuggestthatby2014,canopycoverwas
70%loweri ntheburnedpl ot sthanint heContr ol,mo st lybecause
LADwaslowerb et ween3a nd20mintheburnedplot s(F igure8).
Therecoveryofcanopycoverintheburnedplotswasexpected
toreducegrass cover afterthefires ended,butourrepeatedmap‐
ping of grasses in 2015 indicated that grasses covered roughly
thesame area as in 2012: 43% (B1yr)and 49%(B3yr) of the forest
edgearea(Figure 9;FigureS8), comparedwith <1%in the Control.
FIGURE 4 Temporalpatternsof
environmentalvariablesmeasuredalong
theforestedge(lef tpanels)andinthe
forestinterior(rightpanels)ofthethree
experimentalplots(Control, B3yr,and
B1yr).(a)Abovegroundlivebiomass
(N=6).(b)Leafareaindex(N = 70;
LAI).(c)Litterfall(N=70).(d)predicted
mortalityfortreeswithdiameter
atbreastheight(dbh)of20cm(see
MaterialandMethodsfordetails).Red
xswithintheplottingsymbolsindicate
theyearswhentheexperimentalfires
wereconducted(Augustofeachyear),
endingin2010.Theblowdownevent
occurredinOctober2012(represented
by*inthex‐axis).Thecoloredbands
representbootstrappedconfidence
intervals
Edge Forest
2004 2008 2012 2016 2004 20082012 2016
0
50
100
150
1
2
3
4
5
1
2
3
4
5
0.0
0.2
0.4
0.6
Ye ar
AGL biomass
(Mg/ha)
LAI
(m2 m–2)
Litterfall
(Mg ha–1 year–1)
(a)
(c)
(b)
(d)
Predicted
mortality
Treat
B1yr
B3yr
Control
**
FIGURE 5 Stembiomassaccumulationofindividualsinthe
threetreatmentplots(Control,B3yr,andB1yr),accordingtoour
linearmixedmodel.Theannualstembiomassincrementusedinthe
modelwascalculatedasthedifferenceintreebiomassmeasured
betweentwosamplingintervals(Julyofeachyear).Formore
details,seetheMaterialandMethodssection
08-04 10-08 11-10 12-11 14-12 16-14
Biomass increment
(kg ind–1 year –1)
ControlB3yrB1yr
0
5
10
15
20
Period between sampling intervals
|
7
BRAN DO et Al .
Withthepostfiremortality oflargetreesandvegetationregrowth,
grassesshiftedspatially,withonly46%(B3yr)and36%(B1yr)ofthe
areacoveredbygrassesin2012remainingsoin2015,implyingthat
grassescolonizednewareasoftheburnedplotsduringthistimepe‐
riod.ThedominantgrassspeciesalsoshiftedfromA. longifolia(aC3
nativeCerradospecies)toA. gayanus (a C4 African species)andamix
ofgrasses.Reductionsincanopytreeheightwereanimportantpre‐
dictorofgrassinvasionsuchthatfor a10‐mreduction inoverstor y
height(e.g.,from15to5m)therewasanincreaseof12%–20%inthe
probabilityofgrassinvasionintheburned plots,butonly4%in the
Control (Figure S9).Incontrast, grasses in the Control coveredless
than1%oftheforestedgein2012and2015.
3.3 | CO2 and water fluxes
Overall,ETderivedfromourECsystemsinburnedandControlplots
werestrikingly similar.Averaged across the entire period,postfire
ETwasonly1.1%higherthanin theunburnedControl, which aver‐
aged 0.45 mm/hr during daytime (i.e., global radiation ≥50 W/m2).
Thenotableexceptionwasin2015, when ETwas 7% lower in the
burned plot sthanintheControl (Figure10).Theoverallhighpost‐
fire water fluxesapparently resultedfrom increased ET perunitof
leafarea.In2014,forexample,ETnormalizedbyLAIwas2.4times
higherintheburnedplotsthanintheControlduringrainlessperiods
(sixormoredayswithprecipitation<1mm;FigureS10),whentran‐
spirationprobablydominatedthewaterfluxes(Knaueretal.,2018).
While postfire ET normalized per LAIdeclinedduring the recover y
period,bytheendof2017,plantsintheburnedplotsstilltranspired
75%moretha ntheControlperunitofL AI.Ourestimatesofecosys‐
tem ET from soilmoisture measurements also point tovegetation
intheburnedplotsusinglargeamountsofsoilwatertogrow,given
thatET was higherinthe burnedplotscompared withthe Control
(FigureS11).However,theECsystemrepresentedamuchlargerarea
thanthesoilpitsusedtocalculateETduringdr yperiods.
NEEfluxes indicate that theburned plots assimilated 36%less C
th a nthe C o ntro l did i n20 14, 3ye a r saf te rth e e xper i m e ntal f i rese n d ed
(Figure10).Thislowercapacitytotakeupc arbonwasassociatedwith
a26%increaseinpostfireecosystemrespiration(Reco)anda10%drop
FIGURE 6 Averagetreegrowthfor
eachspecies,asafunctionofaverage
numberofnewrecruitsreachingthe
minimumsizetoentertheinventory
(10cmdbh).Theleftandrightpanels
representtheburning(2004–2011)
andrecover yperiods(2012–2016),
respectively
T
. vulgaris
g
ar
T
T
I. hete
I. hete
t
r
r
ophylla
l
ophylla
r
r
r
r
P
. pilosissimum
p
m
P
P
C. bicolor
o
C. bicolor
i
r
T
. vulgaris
u
T
T
A. guianensis
n
.
A. guianensis
g
g
s
A. guianensis
n
.
V
V
. vismiifolia
m
i iif li
V
V
V
T
. guianensis
n
T
T
S. sinemariensis
S
a
S
ema
C. schomburkianus
m
C
u
mburk
Unknown
o
Unknown
A. guianensis
a
X. amazonica
i
a
0.0 2.5 5.0 7.5 0.0 2.5 5.0 7.5
0
1
2
3
Treat
B1yr
B3yr
Control
Burning period Recovery period
Number of new recruits (ha/year)
Tree growth (mm/year)
FIGURE 7 LiDAR‐basedmapsshowingforestheightchangesbetween2012and2014fortheunderstoryvegetation(<3m)and
overstor yvegetation(>3m)acrossthethreeexperimentaltreatmentplot s(Control,B3yr,andB1yr)
0250125
Meters
2012– 2014 dynamics
Removed
Unchanged
Growth (<3 m)
Growth (>3 m)
8
|
BRANDO et Al .
inpostfireGEP(NEE–Re co)compared with theControl. After2014,
differences in NEE between burned andControl plots averaged less
than1%,poi nt ingtoa rapidrecoveryof CO2fluxesdespitedifferences
inLAIandABG.TosustainacomparableGEPtotheControl,theveg‐
etationintheburnedplotapparentlyusedmoreoftheabsorbedPAR
tofixcarbon.In2014,forexample,LUEwas65%higherintheburned
plots th an in the Cont rol. As L AI increas ed, however, diffe rences in
LUEbetweentreatmentsplotsdecreasedto13%in2017(Figure10).
4 | DISCUSSION
The intensification of disturbances in Amazonia could push for‐
ests tonewstates, with potentialreductionsinecosystem ser vices
regulatingtheregion'sclimate(e.g.,ET,CO2uptake,andcarbonstor‐
age;Davidsonetal.,2012;Nobreetal.,2016).Theforestsmostsus‐
ceptibletoapotentialshiftintoapermanentlydegradedstategrow
alongthedriestsouthernandeasternmarginsoftheAmazon,where
drought, fires, and fragmentation already interact synergistically
(Alencar et al.,2015;Mortonetal., 2013). However,theresilience
oftheseforeststomultiple,severedisturbancesremainunclear,es‐
peciallythetimescalesforrecoveryofdifferentecosystemservices
(Chazdonetal.,2007).
In our experimental forest, we expected biomass loss to begin
recoveryafterthecessationofexperimentalfires.Instead,increased
mortalityoflargetreesandrelatedlossesinABGandcanopycover
charac terized the 7‐year post fire period. T hus, in contrast to o ur
original hypothesis,forestdegradationcontinuedformultipleyears
after the experimental manipulation ended. Par ticularly along the
forest edges, where introduced grasses persisted throughout the
recovery period, a potential consequence of legacies bet ween
high‐intens ity fire and a wind throw event. Despi te these striki ng
differences betweenburnedandunburned areas,their CO2 uptake
andETwere unexpectedlysimilar 5years aftertheend ofhigh‐in‐
tensityfires,likelyreflectingoptimizedresourceusebypostfirein‐
coming(includinggrasses)andfire‐survivingvegetation(Berenguer
etal.,2018;Brando,Oliveria‐Santos,Rocha,Cur y,&Coe,2016).Yet,
becaus e gross CO2 exchange w as equivale nt between t he burned
andControlplots, thenet recover y ofcarbonstocksfollowing the
observed disturbanceswill probably beslow, as observed inother
burnedAmazonforests(Silvaetal.,2018).
4.1 | Ecosystem structure and diversity
Surfacetropicalfiresareexpectedtokillmostlysmall‐andmedium‐
sizedtrees,becausethelarger,thickerbarkedindividualscanavoid
heat‐related da mage to cambium cells d uring fires (Br ando et al.,
2012).Althoughdelayed postfiremortalityoflarge treeshasbeen
obser ved across the tro pics, the mech anisms driving t his process
remain unclear (Ba rlow et al., 2003). In our stud y, the persistent
postfiremortalityoflargetreesmayhavebeenassociatedwiththeir
FIGURE 8 Leafareadensity(L AD)
profileestimatedfromairborneLiDAR
measurementsfortheunburnedControl
andtheplotsburnedtriennially(B3yr)
andannually(B1yr)from2004to2010
(withtheexceptionof20 08).TheLiDAR
overflightswereconductedinOctober
2014.(a)LADperforestheightclass,
showingthatLADwasgreatlyreduced
intheburnedplots.(b)CumulativeLAD
throughouttheforestprofile,showing
thatLADintheControlwas~70%higher
thanintheburnedplots
0.0 0.2 0.4 0.6
012345
0
10
20
30
Leaf area density (m
2
m
2
)
Forest height (m)
(a) (b)
Control
B3yr
B1yr
FIGURE 9 Mapsofgrassinvasionsfor2012(upperpanel)and
2015(lowerpanel)intreatmentplotssubjectedtoannual(B1yr)or
triennialburns(B3yr)andintheunburnedControlbetween2004
and2010.Differentcolorsrepresentdif ferentgrassspeciesor
absenceofgrasses(i.e.,forests,showninblack)
B1yr B3yr Control
|
9
BRAN DO et Al .
increasedvulnerability to windstorms.Specifically,thefires weak‐
enedtrunksandincreasedcrownexposure oflarge treesto strong
winds thatimpacted the experimental forest (Silvério et al., 2019).
Thispostfiredisturbancecausedgreaterlossesincanopycoverand
biomass inareasimpactedbyour experimentalfires. Theobser ved
rapidgrowthofresproutsandsaplingswasexpectedtopartiallyre‐
store forest structure (Chazdon et al., 2007), but ABG reached its
lowest le ve ls7y ear saftert he exper ime nt alf ir eshaden ded ,p art icu‐
larlyalongtheforestedges.Ourresultssuggestthatfragmentation
of tropical forests—and the associated increase in forest edges—
could dra matically i ncrease the n egative impa cts of fir es and lead
topersistentlossesofbiomass(Silvaetal.,2018).Thisisparticularly
true whe re fires co‐ occur with ex treme dro ughts that c an furt her
increasefire intensity andfire‐related tree mortality by triggering
accumulationanddesiccationoffuels(Aragãoetal.,20 08).
Apermanent transition toa new,derived‐savanna state in this
ecotone would requirepersistenceofgrasscoveranda moreopen
canopy. While grasses couldpersist along the forest edges due to
delayed po stfire tre e mortalit y (Higgins, Bo nd, & Trollope, 200 0),
we instead observed widespread postfire resprouting of woody
vegetation—includinginareasthathadbeenpreviouslycoveredby
grasse s. However, the total area o ccupied by grasse s did not de‐
crease,becausegrassesinvadednewareasbetween2012and2015.
Furthermore,wefoundnomajorreplacementofC4 by C3 species,
aprocessexpectedtooccuraslightintheunderstoryoftheburned
plots be came scarce. I nstead, light‐deman ding African C4 g rasses
became more abundant,particularly wherefires in previous years
hadbeenmoreintenseandsevere,andwheretheblowdownevent
had topkilled more trees (e.g., B3yr). The delayed tree mortalit y
drove much of this processby increasing lightand probably water
availability as canopy cover decreased. Another likely driver was
the greatercapacity of African grasses to produce, germinate, and
disperseseeds compared withthe Cerrado native grasses (Higgins
etal.,2000).Theseresultsraisethequestionofwhethertheedgeof
FIGURE 10 Eddycovariance‐based
measurementsofseveralproperties
andprecipitationmeasuredinsoutheast
Amazonia.Thepanelsshowtemporal
variationofweeklyevapotranspiration
(ET;a),netecosystemexchange(NEE;
b),ecosystemrespiration(Reco;c),gross
primar yproductivity(GEP;d),andlight‐
useefficiency(LUE;e).Thebottompanel
representsmonthlyprecipitation(PPT;
f).Thesolidlinesrepresentgeneralized
additivemodels(GAM)foreachoneof
thevariablesfortheburned(orange)
andcontrol(green)plots.Thedashed
linerepresentsalinearmodel.Thered
semitransparentredbarsrepresentdry
seasonmonths(June–September)
0.2
0.3
0.4
0.5
0.6
2
0
2
4
0
3
6
9
5
10
15
0.01
0.02
0.03
0.04
0.05
Treat
Burned
Control
2014 2015 2016 2017 2018
ET
(mm/hr)
NEE
(gC m–2 day–1)
LUE
(CO2/photons)
GEP
(gC m–2 day–1)
PPT
(mm/month)
Reco
(gC m–2 day–1)
0
100
200
300
400
500
(a)
(b)
(c)
(d)
(e)
(f)
Date
10
|
BRANDO et Al .
theexperimentalareaisundergoinga permanent shift toalow‐di‐
versity,grass‐dominatedvegetationenvironment.Wespeculatethat
overthenextfewyearsordecades,woodyvegetationwillprobably
replacegrasses,butonlyiftherearenorecurrentfires,fundamental
changes in c limate, and/or deple tion of nutrie nts, which h as been
shown to slow d own recovery in othe r tropical fores ts (Chazdon
etal.,2007).Ourexperimentalsitereceives over1,700mmof pre‐
cipitation,whichissufficientto supportdense forestsacross much
ofthetropics(Staver,Archibald,&Levin,2011).
4.2 | Ecosystem functions
Large tracts of Amazon forests burn during droughts (Mor ton
et al., 2013) or ar e affected by epi sodic blowdown eve nts (Rifai
et al., 2016), but the magnitude of the associated changes in
water cycl ing is poorly q uantified . Degrade d Amazon fores ts are
expec ted to transpir e substantia lly less beca use of the potenti al
red uc tionsinL AIan droot in gdepth(Silvérioe tal.,2015).However,
ETinourstudy was similarbetweentreatment sthroughoutmost
ofthestudyperiod,evenwhenpostdisturbanceLAIwashalfofthe
Control . There are at le ast four likel y ecologica l explanati ons for
thisfinding.First,early‐successional,fast‐growingspeciescoloniz‐
ing the bur ned areas (e.g ., M. fistulifera,T. vulgaris, C. grandiflora,
and C. bicolor)probablyusedmorewater per unitof LAIthan the
late‐successional ones they replace(Chiariello, Field, & Mooney,
1987),apattern commonlyobservedinslashand burnagriculture
(Sommer e t al., 2002). O ur LiDAR measure ments showe d annual
height growth reaching 2 m, a process probably requiring large
amounts of water. Second, transpiration of large fire‐surviving
treesmayhaveincreasedaspostfirecompetitiondecreased,given
thatgrow th wa sdis pr oportionallyhigh er fo rl ar ge ri ndividu al s(e.g.,
Figure S6).Also, recentstudies haveshown that afewlargetrees
can accountforalarge proportionof ecosystem‐level ET (Kunert
et al., 2017), be cause of thei r large canop ies, large ste m storage
capacity,and ability to take up deepsoil water throughout much
of the year (Ne pstad et al ., 1994).Th ird, fire‐ind uced reduc tions
inLAIhighinthecanopylikelyallowedmoremixingwithairaloft,
decreased overall relative humidity,an dprob ably caused associ‐
atedincreasesinevaporativedemand(Kunert,Aparecido,Higuchi,
Santos, &Trumbore, 2015).Fourth,grassesmay havecontributed
tothehighETduringthewetseason,whengrasseshavemoreac‐
cess to soil moisture. However,grasses coveredless than20% of
theove ra lltowerfootprintandlessth an3%ofth emostlikel yfoot‐
print are a (i.e., sampling p robability > 0.1).Fur thermore , had this
been the case,we wouldhaveexpectedhigher‐than‐observed ET
during th e wet season in t he burned p lots. Over all, we conclu de
thatacombinationof theseprocesseslikely allowedplants in the
burned plots to maximize water use andto maintain high E T and
cycled60%–75%oftotalrainfallforthisregion.
Postdisturbance CO2uptakeoftropicalforestscaneitherin‐
creaseas vegetationrecovers,decreaseas necromass decomposes,
or remain unchanged as emissions and uptake cancel each other
out(Malhi,2011).Althoughnetcarbonuptakeintheexperimentally
burned forestsinitially was lowerthan in the Control, it apparently
recovered toControl levels withina few years afterthe prescribed
fire s.T hi sr ecoverywaspa r ti all ya ss ociat ed witht heve ge t at ioninth e
burned plot sinvestingmoreof the absorbed PARtofixcarbon.For
example,postfireLUE(0.021 mol CO2/mol photons)between2014
and 2017 was much h igher than the u nburned Co ntrol (0.015 mol
CO2/mol photons), but comparable to an east‐central Amazonian
moistforest(0.019molCO2/molphotons;Wuetal.,2016).Although
carbonfluxes between treatmentplots weresimilar after 2014, we
observed important seasonal differences between our treatment
plots. Fo r instance , the burned p lots exceede d GEP in the Cont rol
duringthewet‐seasonmonthsof2015,aresultpotentiallyrelatedto
higherthroughfallinthemoreopenburnedplots,increasingsoilwater
availabil ity and allowing p lants to photosynt hesize longer into th e
dry‐season(Honda & Durigan, 2016)andtogrowfaster (Berenguer
etal.,2018;Brandoetal., 2016).Inadditiontorecovering GEP,the
vegetatio n growing in the burn ed plots also incr eased net carbo n
uptake due to a sharp reduction in Recostartingin2015,mostlyin
thedry season.Althoughthehighrespirationintheburned plot sin
2014 could be a resu lt of necromass d ecompositi on from the fi res
andtheblowdown,fast‐growingspeciescolonizingtheareaprobably
contributedtothisprocessaswell.Wespeculatethatfrom2014and
2015,Recodecreasedascompetitionforresourcesincreasedbutalso
inresponse to lower necromass decomposition. Rochaet al. (2014)
andMetcalfeetal.(2018),forinstance,observedfastpostfirerecov‐
ery of eco system‐level c arbon flu xes across the ex perimenta l area
due to reduced Reco, which wasderivedfrom bottom‐up field mea‐
surement s(Ma lh ie tal. ,2009) .O ve ra ll,o urre sultspo in ttorapidpost‐
fire restoration of carbon cycling, albeit not large enough to offset
thebiomasslossesassociatedwithincreasedmort alityoflargetrees,
aprocessthatmaytakeseveraldecadestooccur(Silvaetal.,2018).
Our results suggest that the vegetation in the burned plots
rapidly recovered fluxes of CO2andH2O.However,legacy fire‐ef‐
fectsmayhave occurred withinthe first 3.5 years of the recovery
period,whenno eddyfluxtowerswereinplace.Someofourmea‐
surement ssuggestthattheburnedplots appeartohave recovered
CO2fluxestoControllevelbylate2014orearly2015.Forexample,
our10–12yearsofmeasurementsofLAI‐litterfall also support this
view.PostfireLAIaroundtheeddyfluxtowerrecoveredslowly,but
litterfall becamemoresimilar bet ween treatmentsstartingin2013
and 2014. This sug gests that by inc reasing leaf turn over (primar‐
ily) and lea f area (second arily), plant s growing in the b urned plot s
rapidly recovered postfire canopy productivity and water fluxesto
theControllevels,aninterpretationsupportedbyMODIS‐EVImea‐
surement s.Rochaet al.(2014)alsofoundarapidrecoveryinforest
productivity along atransect of the experimental area, largely be‐
causeofincreasedcarbon‐useefficiencyintheburnedplots.Similar
patterns in soil moisture dynamics between the two experimental
plots re inforce the not ion that ET ra pidly recover ed following t he
experimental fires.However,wecannotdiscardthepossibility that
differencesinNEEandETexistedbeforetheexperimentalburns.
Overall,our results support the hypothesisthatevenhighly dis‐
turbedforestscanrapidlyrecoverfluxesofwaterandcarbondueto
|
11
BRAN DO et Al .
canopyclosureasearlysuccessional,fastgrowingspeciesbecomethe
dominanttrees. Asvegetationrecovers in the nearfuture,competi‐
tionforlight, water,and nutrients maychangehowplantsoptimize
resources(Everham&Brokaw,1996).Inparticular,theburnedforest
is expected to become light‐saturatedas LAI increases, causing as‐
sociated reductions in ecosystem‐levelLUE (Li, Bian, Lei, & Huang,
2012). Similarly, as competitionfor water increases and throughfall
decreases,plantsgrowingintheburnedplotsprobablywilltranspire
less waterper unit of LAI. Our results suggest that these two pro‐
cesses maybe already occurring. We also expect slow‐growthtree
speciestobemorecommonamongnewrecruits,asobservedinthe
Control.
4.3 | Broader implications
One of the mostimportant questionsinAmazon conservation is
whetherthe impactsofsevere,repeateddisturbances occurring
in the region could surpass the ecosystem'scapacity to recover.
Althou gh several stu dies have shown th at much of the Amazo n
forest is resilient to moderate disturbance, we found that this
resilien ce, in terms of car bon stocks a nd forest cover, has be en
temporarilyimpededbyrecurrentfiresinteractingwithdroughts,
forest fragmentation, and blowdown events. The observed de‐
layed postfire mortality of large trees reduced forest carbon
stocksandcreatedopportunitiesfortheestablishmentofinvasive
grasses,pushingthissystemtoanewenvironment.However,we
cannotdetermineyetwhetheratippingpointwasreached,lead‐
ingthissystemtoanewstable,moresavanna‐likestate.Previous
studies have shown that even severely disturbed forests may
mostlyrecoverpredisturbancebiomassandfluxeswithindecades
(Chazdon et al.,2016).Yet,we speculate thatin the nearfuture,
the continued or even increased vulnerability of large surviving
tr e e scan incr e a set h eris k sof n ewd i stur b a nce s impa c t i ngt h eare a
before recover yoccurs.For example, climate changeis expected
to increas e the freque ncy and intens ity of distur bances such as
windsto rms and droug hts (Duf fy, Brando, A sner, & Field, 2015).
Evenif higherlevels of atmospheric CO2 leadtoincreasedrates
ofbiomassrecovery,morefrequentdisturbanceswouldresultin
chronicimpoverishmentofbiomassandbiodiversity,especiallyin
landscapesbecomingmorefragmentedbydeforestation.
However,amajorfinding here isthatclimateservicessuchas
evaporation are recovered even in highly degraded, low biomass
forests.Intropicalregions,landcoverchangehasdramaticallyal‐
teredsurfaceenergyandwaterbalanceatregionalscales(Silvério
et al., 2015). These changes, especially reductions in recycling
ofevaporated water as a source of precipitation, havebeen hy‐
pothesizedtoexacerbatedroughtintheAmazonandneighboring
regions (Nobreet al., 2016).Thesimilarityweobservedbetween
ET in a highly degraded forest and an undisturbed one agrees
with other recent studies of fluxes in agroforests and natural
forests (Sabajoet al., 2017).Relativelyrapid recovery of climate
serv ices indicate s that, even th ough highly dis turbed for ests, in
our case s with stru cture simila r to a derived sav anna, they may
playanimportant climateservicesrole bymaintaining ET fluxes.
Yet,widespread disturbance eventsoccurringoverthousandsof
hectares(e.g.,Witheyetal.,2018)mayconstrainETbyincreasing
regionallandsurfacetemperatures,andperhapsinfluencingpre‐
cipitationpatterns.
As regionalclimate changes, forest resilience is expected to
d e cr e a s e (S c h w a l me t a l . ,2 0 1 7 ), w h e r e as t h e f r eq u e n c y a n di n t e n ‐
sityofdisturbancesareexpectedtoincrease (De Faria, Brando,
& Environme ntal, 2017). Transitional fo rests growing be tween
tropicalfor es tsandsavannasarelikelytob ethemostvulnera ble
to a potentia l intensificat ion in disturb ance regimes c aused by
wildfires,blowdowns,andfragmentation.Somestudiessuggest
ashiftin stablestatestowardanecosystemwithlower biomass
and more gr ass dominance, es pecially when pre cipitation fall s
below1,500mm(Staveretal.,2011).Forthetransitionforestof
thepresents tudy,we fo un devide nc eforresilienceofthecarb on
fluxan dE Tp rocesse s,butst il ld el ayed re co ve ryofcar bonstocks
and forestcover and delayed decline in grasscover.Hence, we
cannotdetermine yetwhetheratippingpointwasreached that
willpermitmorefireandwindthrowdisturbance,fur therexacer‐
bated bymajor droughts, whichwould lead to derived savanna,
orwhetherforestcoverwilleventuallyrecoverandreducethose
risks.
ACKNOWLEDGEMENTS
We thank the cr ew from IPAM for thei r help with dat a collectio n
andanalysis,A.Maggiforprovidingaccesstothefieldsite,andTara
Massad and Martin Hertel helping toinstall theeddy flux towers.
The manuscript was improved by comments of R. Houghton, G.
Durigan,F.Putz,L.Paolucci,L.Rattis,andT.El‐Madany.Weappre‐
ciate the fi nancial suppor t from the Nation al Science Foundat ion
(#1146206),Max PlanckInstituteforBiogeochemistry,theGordon
and Betty Moore Foundation, and the Conselho Nacional de
Desenvolvimento Científico e Tecnológico (CNPq‐Brazil) through
aProductivity Grant forP.Brando, PELD‐Tang (#441703/2016‐0),
through a P VE‐Science without Bo rders funding for S . Trumbore
andD.Silvério(#405800/2013‐4), and through a postdoctoral fel‐
lowshipforL.Maracahipes.LiDARdatawereacquiredwithsuppor t
from USAID, the USDepar tmentof State,EMBRAPA, and the US
ForestServiceOfficeofInternationalPrograms.
CONFLICT OF INTEREST
Theauthorshavenoconflictofinteresttodeclare.
ORCID
Paulo M. Brando https://orcid.org/0000‐0001‐8952‐7025
Divino Silvério https://orcid.org/0000‐0003‐1642‐9496
Leonardo Maracahipes‐Santos https://orcid.
org/0000‐0002‐8402‐1399
12
|
BRANDO et Al .
REFERENCES
Alencar,A.A.,Brando,P.M.,Asner,G.P.,&Putz,F.E.(2015).Landscape
fragmentation, severe drought , and the new Amazon forest
fire regime. Ecological Applications, 25, 1493–1505. https://doi.
org /10.1890/14 ‐1528.1
Aragão, L. E. O. C ., Malhi, Y., Barbier, N., Lima, A., Shimabukuro, Y.,
Anderson,L.,&Saatchi,S.(2008).Interactionsbetweenrainfall,de‐
forest ation and fires duringrecent yearsintheBrazilianAmazonia.
Philosophical Transac tions of the Royal Society of Londo n. Serie s
B, Biological Sciences, 363, 1779–1785. htt ps://doi.o rg /10.10 98/
rstb.2007.0026
Aragão,L.E.O.C., &Shimabukuro,Y.E.(2010).Theincidenceoff irein
AmazonianforestswithimplicationsforREDD.Science,328, 1275–
1278.https://doi.org/10.1126/science.1186925
Arroyo‐R odríguez, V., Mel o, F. P.L ., Martí nez‐Ramos, M ., Bongers, F.,
Chazdon , R. L., Meave , J. A., … Tabarelli, M . (2017). Multipl e suc‐
cessional pathways in human‐modified tropical landscapes: New
insightsfrom forestsuccession,forestfragmentationandlandscape
ecology research. Biological Reviews, 92 (1), 326–340. ht tps://doi.
org /10.1111/b rv.122 31
Balch, J. K., Nepstad, D. C., Brando, P. M., Curr an, L. M., Portela,
O., de Car valho, J. O., & Lefebvre, P. (2008). Negative fire
feedback in a transitional forest of southeastern Amazonia.
Global Change Biology, 14(10), 2276–2287. https://doi.
org /10.1111/j .136 5‐2486.200 8. 01655.x
Barlow, J., Peres, C. A., Lagan, B . O., & Haugaasen, T. (2003).
Large tree mor talit y and the decline of forest biomass follow‐
ing Amazonian wildfires. Ecology Letters, 6(1), 6–8. https://doi.
org/10.1046/j.1461‐0248.2003.00394.x
Bates, D., Mächler, M., Bolker, B., & Walker, S. (2014). Fitting linear
mixed‐ef fectsmodelsusinglme4.arXiv. org.
Berenguer,E.,Malhi,Y.,Brando,P.,C ardosoNunesCordeiro,A.,Ferreira,
J.,França,F.,…Barlow, J.(2018).Treegrowth and stemc arbonac‐
cumulationinhuman‐modifiedAmazonianforestsfollowingdrought
and fire. Philosophical Transactions of the Royal Society of London.
Series B, Biological Sciences,373, 20170308–20170318. https://doi.
org /10.1098/r st b. 2017.030 8
Bouvier, M., Durrieu, S., Fournier, R. A ., & Renaud, J.‐P. (2013).
Generalizing predictive models of forest inventory attributes using
anarea‐basedapproachwithairborneLiDARdata.Remote Sensing of
Environment,156,1–13.
Brando,P.M.,Balch,J.K.,Nepst ad,D.C.,Morton,D.C.,Putz,F.E.,Coe,
M. T., … Soares‐F ilho, B. S. ( 2014). Abrupt in creases in A mazonian
tree mor tality due to drought‐fire interac tions. Proceedi ngs of the
National Academy of Sciences of the United States of America, 111,
6347–6352 .h tt ps ://doi.or g/10 .1073/pnas.1305499111
Brando,P.M.,Nepstad, D.C ., Balch,J. K.,Bolker,B.,Christman, M. C.,
Coe, M.,&Putz,F.E. (2012). Fire‐inducedtree mortality inaneo‐
tropical forest: The roles of bark trait s, tree size, wood density
and fire b ehavior. Global Change Biology,18 , 630–6 41. https://doi.
org /10.1111/j .136 5‐2486.2011. 02533 .x
Brando,P.M.,Oliveria‐Santos,C.,Rocha,W.,Cury,R.,&Coe,M.T.(2016).
Effectsofexperimentalfueladditionsonfireintensityandseverity:
Unexpectedcarbonresilienceofaneotropicalforest.Global Change
Biology,22(7),2516–2525.https://doi.org/10.1111/gcb.13172
Chave,J.,Réjou‐Méchain,M.,Búrquez,A .,Chidumayo,E.,Colgan,M.S.,
Delit ti,W.B.C .,…V ieilled en t, G.(2014).Im prove dallo me tr icmodels
toestimatetheabovegroundbiomassoftropicaltrees.Global Change
Biology,20,3177–3190.https://doi.org/10.1111/gcb.12629
Chazdon, R. L., Broadbent, E. N., Rozendaal, D. M. A ., Bongers, F.,
Zambr ano, A. M. A ., Aide, T. M., … Poo rter, L. (2016). Ca rbon se‐
quest ration poten tial of second‐g rowth fore st regenerat ion in the
LatinAmerican tropics. Science Advances, 2, e1501639.https://doi.
org /10.1126/scia dv.1501639
Chazdon, R. L., Letcher, S. G., van Br eugel, M., Martínez‐Ra mos, M.,
Bongers, F.,& Finegan, B.(2007). Rates ofchange intree commu‐
nities of secondary Neotropical forests following major distur‐
bances. Philosophical Transactions of the Royal Society of London.
Series B, Biological Sciences,362, 2 73–2 89. ht tp s: //doi .o rg /10 .10 98 /
rstb.2006.1990
Chiariello, N. R ., Field, C. B., & Mooney, H. A . (1987). Midday wilt‐
ing in a trop ical pioneer t ree. Functional Ecology, 1, 3. http s://doi.
org/10.2307/2389351
Clark,D.A., Brown,S .,Kicklighter,D. W.,Chambers,J.Q., Thomlinson,
J.R., & Ni, J. (2001). Measuring net primary production in forest s:
Concepts and field methods.Ecological Applications, 11, 356–370.
https://doi.org/10.1890/1051‐0761(2001)011[0356:MNPPI
F]2.0.CO;2
Cochrane,M.,Alencar,A.,Schulze,M.,Souza,C.,Nepstad,D.,Lefebvre,
P.,& Davidson, E. (1999). Positivefeedbacks in thefire dynamic of
closedcanopytropicalforests.Science,284,1832–1835.https://doi.
org/10.1126/science.284.5421.1832
Cook,B.,Corp,L.,Nelson,R.,Middleton,E.,Morton,D.,McCorkel,J.,…
Montes ano, P.(2013). N ASA Godd ard’s LiDAR, H yperspe ctral and
Therma l (G‐LiHT) a irborne ima ger.Remote Sensing, 5, 4045– 4066.
https://doi.org/10.3390/rs5084 045
D'Ant o n io, C . M .,H u g h es,R . F. ,&V i t o use k , P.M.( 2 0 01) . Fa c t o r si n f l uenc ‐
ingdynamicsof twoinvasive C4grassesin seasonally dryHawaiian
woodlands.Ecology,82,82.ht tps://doi.org/10.2307/2680088
Davidson,E.A.,deAraújo, A. C.,Artaxo,P.,Balch,J.K .,Brown, I. F.,C.
Bustam ante,M.M.,…Wofs y,S.C .(2 012).TheAmazonbasi nint ran‐
sition.Nature,4 81,321–328.https://doi.org /10.1038/nature10717
Davidson,E.A.,deCarvalho,C.J.R.,Figueira,A.M.,Ishida,F.Y.,Ometto,
J. P.H. B ., Nardoto, G . B., … Mar tinelli , L. A. (2 007). Rec uperati on
ofnitrogencyclinginAmazonianfores tsfollowingagriculturalaban‐
donment. Nature, 447, 995–998. https://doi.org/10.1038/natur
e059 00
Davidson,E.,Lefebvre,P.,Brando,P.M.,Ray,D.,Trumbore,S.,Solorzano,
L., … Ne pstad, D. (2 011).C arbon in puts an d water upta ke in deep
soilsofaneasternAmazonforest.Forest Science,57(1),51–58.
Davidson,E.A.,Neill,C.,Krusch,A.V.,Ballester,V.V.R.,Markewitz,
D., & Figuei redo, R. d e O. (2004). L oss of nutri ents fro m terres‐
trial eco systems to stre ams and the atmosp here following la nd
use changein Amazonia. In R. DeFries, G. Asner, & R. Houghton
(Eds.), Ecosystems and land use change. Geophysical Monograph
Series153( pp .147–158) .Was hi ngt on ,D.C . :A me ric anG eo physica l
Union.
De Faria, B.L., Brando, P. M., & Environmental,M. M. (2017). Current
andfuturepatternsof fire‐inducedforestdegradationinAmazonia.
Environmental Research Letters,12(9),095005.
Derroire,G.,Balvanera,P.,Castellanos‐Castro,C .,Decocq,G.,Kennard,
D. K., Leb rija‐Trejos, E., … Heal ey,J. R . (2016). Resilien ce of tropi‐
caldr yforests–Ameta‐analysisofchangesinspeciesdiversityand
composition duringsecondar ysuccession. Oikos, 125, 1386–1397.
https ://doi.or g/10.1111/oik.03229
Detto,M.,Katul,G.,Mancini,M.,Mont aldo,N.,&Albertson,J.D.(2008).
Surface heterogeneity and its signature in higher‐order scalar sim‐
ilarity relationships. Agricultural and Forest Meteorology,14 8, 902–
916.https://doi.org/10.1016/j.agrformet.2007.12.008
Duff y,P. B., Br ando, P., Asner, G. P., & Field, C . B. (2015). Pr ojectio ns
of future m eteorologi cal drought a nd wet period s in the Amazon .
Proceedi ngs of the National Academy of Sciences of the United States
of America, 112, 13172–13177. https://doi.org/10.1073/pnas.14210
10112
Everham, E. M., &Brokaw, N. V.L. (1996).Forestdamage and recover y
fromcat astrophicwind.The Botanical Review,62(2),113–185.
Flores, B. M., Holmgren, M., Xu, C., van Nes, E. H., Jakovac, C . C.,
Mesquita,R .C.G.,&Scheffer,M.(2017).FloodplainsasanAchilles'
heel of Amazonian forest resilience. Proceedings of the National
|
13
BRAN DO et Al .
Academy of Sciences of the United States of America,114,4442–4446.
h t t p s : / / d o i . o r g / 1 0 . 1 0 7 3 / p n a s . 1 6 1 7 9 8 8 1 1 4
Foken,T.,Aubinet,M.,&Leuning,R.(2011).TheEddycovariancemethod.
InM.Aubinet, T.Vesala,& D.Papale (Eds.),Eddy covariance: A prac‐
tical guide to measurement and d ata analysis(pp.1–19).Netherlands,
Dordrecht:Springer.
Foken,T.,&Wichura,B.(1996).Toolsforqualityassessmentofsurface‐
based flux measurements. Agricultural and Forest Meteorology, 78,
83–105.https://doi.org /10.1016/0168‐1923 (95) 02248‐1
Higgins, S. I., Bond, W. J., & Trollope, W. S. W. (2000). Fire, re‐
sprouting and variability: A recipe for grass–tree coexis‐
tence in savanna. Jo urnal of Eco logy, 88, 2 13 –2 29. h t tps : //doi .
org/10.1046/j.1365‐2745.2000.00435.x
Honda, E.A., & Durigan,G. (2016). Woody encroachment andits con‐
sequencesonhydrologicalprocessesinthesavannah.Philosophical
Transactions of t he Royal Society of Lon don. Series B, B iological Scie nces,
371,20150313–20150319.https://doi.org/10.1098/rstb.2015.0313
Hsieh, C .‐I., Katul, G ., & Chi, T.‐W. ( 2000). An a pproximate ana lytical
modelforfootprintestimationofscalarfluxesinthermallys tratified
atmosphericflows.Advances in Water Resources,23,765–772.https
://doi.org/10.1016/S030 9‐1708(99)00 042‐1
Jipp, P. H., Nepst ad, D. C., C assel, D. K. , Reis De Car valho, C. (1998).
Deepsoilmoisturestorageandtranspirationinforestsandpastures
ofseasonally‐dr yAmazonia.Climatic Change,39(2–3),395–412.
Jolly,W.M.,Cochrane, M. A.,Freeborn,P.H., Holden, Z. A.,Brown, T.
J.,Williamson,G.J.,& Bowman,D.M. J. S.(2015).Climate‐induced
variations in global wildf ire danger from 1979 to 2013. Nature
Communications,6,7537.https://doi.org/10.1038/ncomms8537
Kennard, D.K ., & Putz, F. E. (2005). Differential responses of Bolivian
timber species to prescribed fire and other gap treatments. New
Forests,30,1–20.https://doi.org/10.1007/s11056‐004‐0762‐y
Knauer, J., Zaehle, S., Medlyn, B. E., Reichstein, M., Williams, C. A .,
Migliavacca, M., … Linderson,M.‐L. (2018). Towards physiologically
meaningful water‐use efficiency estimates from eddy covariance
data. Global Change Biology,24 , 694 –710 . htt ps://doi. org/10 .1111/
gcb.13893
Kunert,N.,Aparecido,L.M.T.,Higuchi,N.,dosSantos,J.,&Trumbore,S.
(2015).Highertreetranspirationduetoroad‐associatededgeeffects
inatropicalmoistlowlandforest.Agricultural and Forest Meteorology,
213,183–192.https://doi.org/10.1016/j.agrformet.2015.06.009
Kunert,N .,Aparecido, L. M. T.,Wolff,S.,Higuchi, N., dos Santos,J.,de
Araujo, A. C ., & Trumbore, S. (2017). A revised hydrological model
fortheCentralA mazon: The importance ofemergentcanopy trees
intheforest waterbudget.Agricultural and Forest Meteorology, 239,
47–57.https://doi.org/10.1016/j.agrformet.2017.03.002
Lasslop,G.,Reichstein,M.,Papale,D.,Richardson,A.D.,Arneth,A.,Barr,
A., … Wohlfahrt, G. (2009). Separation of net ecosystem exchange
into assimilation and respiration using a light r esponse curve a p‐
proach: Criticalissues andglobal evaluation. Global Change Biology,
16,187–208.https://doi.org/10.1111/j.1365‐2486.2009.02041.x
Li,A., Bian,J.,Lei,G., &Huang,C.(2012).Estimating the maximal light
use efficienc y for different vegetation through the CASA model
combine d with time‐ser ies remote sensi ng data and groun d mea‐
surements.Remote Sensing,4,3857–3876.https://doi.org/10.3390/
r s 4 1 2 3 8 5 7
Longo, M., Keller, M., dos‐Santos, M. N., Leitold, V., Pinagé, E . R.,
Baccini,A.,…Morton,D.C .(2016).Abovegroundbiomassvariability
across intact and degraded forestsin the Brazilian Amazon. Global
Biogeochemical Cycles, 30, 1–63. https://doi.org/10.1002/2016G
B005 465
Malhi,Y.(2011).Theproductivity,metabolismandcarboncycleoftrop‐
ical forest veget ation. Journal of Ecology, 100, 65–75. https://doi.
org /10.1111/j .136 5‐2745.2011. 01916 .x
Malhi,Y.,Aragao,L.E.,Metcalfe,D.B.,Paiva,R.,Quesada,C.A.,Almeida,
S., … Hutyra, L . R. (2009). Comprehensive assessment of carbon
produc tivit y, allocation and storage in three Amazonian forest s.
Global Change Biology,15(5),1255–1274.
Markewit z,D.,Devine,S.,Davidson, E.A.,Brando,P.M.,&Nepstad,D.
C. (2010). So il moisture de pletion unde r simulated dro ught in the
Amazon:Impact sondeeprootuptake.New Phytologist,187(3),592–
607.https://doi.org/10.1111/j.1469‐8137.2010.03391.x
McGra th, D. A., Sm ith, C. K ., Gholz, H. L., & Ol iveira, F. D. A. (20 01).
Effec ts of land‐u se change on s oil nutrie nt dynami cs in Amazôn ia.
Ecosys tems,4,625–645.
Metcalfe,D. B.,Rocha, W.,Balch, J.K .,Brando,P.M.,Dought y,C. E., &
Malhi,Y.(2018).Impact soffireonsourcesofsoilCO2ef fluxinadry
Amazonrainforest.Global Change Biology,24(8),3629–3641.
Miller,S.D.,Goulden,M.L.,Hutyra,L.R.,Keller,M.,Saleska,S.R.,Wofsy,
S.C.,…deCamargo,P.B.(2011).Reducedimpactloggingminimally
alterstropicalrainforestcarbonandenergyexchange.Proceedings of
the National Academy of Sciences of the United States of America,10 8,
19431–19435.ht tp s://doi.o rg/10.1073/pnas.1105068108
Morto n, D. C., Le Page, Y., DeFrie s, R., Collat z, G. J., & Hurt t, G. C.
(2013). Understorey fire frequency and the f ate of burned for‐
ests in southern Amazonia. Philosophical Transactions of the Royal
Society B: Biological Sciences,368,20120163–20120163.https://doi.
org /10.1098/rstb.2012.0163
Nepstad,D.C.,deCarvalho,C .R.,Davidson,E.A.,&Nature,P.J.(1994).
The role of deep root s in the hydrologic al and carbon c ycles of
Amazonianforestsandpastures.Nature,372(6507),666.
Nepst ad, D., Georgi a, C., Ana, C . B., Ane, A. , João, P.C ., Josh, B., …
Elaine, P. (2001). Ro ad paving, fi re regime feed backs, and t he fu‐
tureofAmazonforests.Forest Ecology and Management,154(3),395
–4 07.
Nobre, C . A., Sampai o, G., Borma, L . S., Castill a‐Rubio, J. C., Sil va, J.
S., & Cardoso, M. (2016). Land‐use and climate change risks in the
Amazonandtheneedofanovelsustainabledevelopmentparadigm.
Proceedi ngs of the National Academy of Sciences of the United States
of America,113 , 10759–10768. https://doi.org /10.1073/pnas.16055
16113
Reichstein,M.,Falge,E.,Baldocchi,D.,Papale,D.,Aubinet,M.,Berbigier,
P.,… Vale ntini, R. ( 2005). On th e separati on of net ecosy stem ex‐
changeintoassimilationandecosystemrespiration:Reviewandim‐
provedalgorithm.Global Change Biology,11,1424–1439.htt ps://doi.
org /10.1111/j .136 5‐2486.2005.00100 2. x
Rifai, S.W.,Urquiza Muñoz, J.D.,Negrón‐Juárez,R.I., RamírezArévalo,
F.R.,Tello‐Espinoza, R., Vanderwel,M. C.,… Bohlman,S.A.(2016).
Landscape‐scale consequences of differential tree mortality from
catastrophicwinddisturbanceintheAmazon.Ecological Applications,
1–39.https://doi.org/10.1002/eap.1368
Rocha, W., Metcalfe, D. B., Doughty, C. E., Brando, P., Silvério, D.,
Halladay,K., …Malhi,Y.(2014).Ecosystemproductivity and carbon
cyclinginintactandannuallyburntforestatthedrysouthernlimitof
theAmazonrainforest(MatoGrosso,Brazil).Plant Ecol ogy & Diversity,
7(1–2),25–40.
Sabajo, C . R., le Maire, G.,June, T., Meijide, A.,Roupsard, O., & Knohl,
A. (2017). E xpansion of oil palm and othe r cash crops caus es an
increas e of the land sur face temp erature in t he Jambi provi nce in
Indonesia. Biogeosciences, 14, 4619–4 635 . ht tps: //doi.o rg/1 0.5194/
bg ‐14 ‐ 4 619‐ 2017
Saldarriaga,J. G., West, D.C., Tharp,M. L.,&Uhl, C. (1988).Long‐term
chronosequence of forest succession in the upper Rio Negro of
ColombiaandVenezuela.The Journal of Ecology,76 ,938.
Schwalm,C.R .,Anderegg,W.R.L.,Michalak,A.M.,Fisher,J.B.,Biondi,
F.,Koch,G.,…Tian,H.(2017).Global patternsofdroughtrecover y.
Nature,548,202–205.https://doi.org/10.1038/nature23021
Silva, C . V.J., A ragão, L . E. O. C., B arlow, J., Espi rito‐Santo, F., Young,
P. J., Anderson , L. O., … Xaud, H. A . M. (2018). Drough t‐in duced
Amazoni an wildfires ins tigate a decad al‐scale disr uption of forest
carbon dynamics. Philosophical Transactions of the Royal Society of
14
|
BRANDO et Al .
London. Series B, Biological Sciences, 373, 20180043–20180112.
https://doi.org/10.1098/rstb.2018.0043
Silveri o, D. V.,Br ando, P. M., Balch , J. K., Put z, F.E ., Nepst ad, D. C.,
Oliveira‐Santos, C., & Bustamante, M. M. C . (2013). Testing
the Amazon savannization hypothesis: Fire effects on inva‐
sion of a neot ropical fo rest by nati ve cerrado a nd exotic pas ture
grasses. Philosophical Transactions of the Royal Society B: Biological
Sciences, 368, 20120427–20120427. https://doi.org/10.1098/
rstb.2012.0427
Silvério,D.V.,Brando, P.M., Bus tamante, M. M.C., Put z, F.E., Marra,
D. M., Levic k, S. R., & Trumb ore, S. E. (2019). Fir e, fragmen tation,
andwindstorms: Arecipe for tropical forestdegradation.Journal of
Ecology,107(2),656–667.
Silvério,D.V.,Brando,P.M.,Macedo,M. N.,Beck, P.S.A.,Bustamante,
M., & Coe, M . T. (2015). Agricultural expansion dominates cli‐
mate changes in southeastern Amazonia: The overlooked non‐
GHG forcing. Environmental Research Letters, 10, 1–8. https://doi.
org /10.1088/1748‐9326/10/10/104 015
Sommer, R., de Abreu, T.D., Vieljauer,K.,deA raújo,A .C.,… Vlek, P.L.
G.(2002).Transpirationandcanopyconductanceofsecondaryveg‐
etatio n in the east ern Amazon . Agricultural and Forest Meteorology,
112(2),103–121.
Staver,A . C., Archibald,S.,&Levin, S. A.(2011).Theglobal extent and
determinants of savanna and forest as alternative biome states.
Science,334,230–232.https://doi.org/10.1126/science.1210465
Trumbore, S., Brando, P., & Hartmann, H. (2015). Forest health and
globalchange.Science,349,814–818.https://doi.org/10.1126/scien
ce.aac6759
vanderSande,M.T.,Poorter,L.,Kooistra,L.,Balvanera,P.,Thonicke,
K., Th ompson, J., … Peñ a‐Claros, M . (2017). Biodiver sity in spe‐
cies, traits, andstructure determinescarbon stocksanduptakein
tropicalforest s. Biotropica,49, 593–603. ht tp s://doi .org/10.1111/
btp.12453
Vickers, D., & Oceanic, L. M. J. O. A. A . (1997). Quality con‐
trol and flux sampling problems for tower and aircraf t data.
Atmospheric and Oceanic Technology, 14, 512–526. https://doi.
org /10.1175/1520 ‐0 426 (1997 )014<0512:Q CAFSP>2.0.CO ;2
Withey,K.,Berenguer,E.,Palmeira,A.F.,Espírito‐Santo,F.D.B.,Lennox,
G.D.,Silva, C. V.J.,…Barlow, J.(2018).Quantifyingimmediate car‐
bon emissions from El Nino‐mediated wildfires in humid tropical
forest s. Philosophica l Transactions of the Royal Society B: Biological
Sciences,373(1760),20170312.
Wu,J.,Guan,K.,Hayek,M.,Restrepo‐Coupe,N.,Wiedemann,K.T.,Xu,
X., … Sal eska, S. R . (2016). Part itioning c ontrols on A mazon fores t
photosynthesisbetweenenvironmentalandbioticfactorsathourly
to interannual timescales. Global Change Biology, 23, 1240–1257.
https ://doi.or g/10.1111/gcb.1350 9
Wutzler, T., Lucas‐Moffat, A., Migliavacca, M., & Knauer, J. (2018).
Basic and e xtensibl e post‐processi ng of eddy covar iance flux dat a
with REddyProc. Biogeosciences, 15(16), 5015–5030. https://doi.
org/10.5194/bg‐15‐5015‐2018
Zarin,D.J.,Davidson, E.A., &Brondizio,E. (2005).Legacy offireslows
carbon accumulation in Amazonian forest regrowth. Frontiers in
Ecology and the Enviro nment,3(7),365–369.
SUPPORTING INFORMATION
Additional supporting information may be found online in the
Suppor tingInformationsectionattheendofthearticle.
How to cite this article:BrandoPM,SilvérioD,Maracahipes‐
SantosL,etal.Prolongedtropicalforestdegradationdueto
compoundingdisturbances:ImplicationsforCO2andH2O
fluxes.Glob Change Biol. 2019;00:1–14. h t t p s : / / d o i .
org /10.1111/gc b.14 659
