Persistent effects of a severe drought on Amazonian forest canopy

20 Pages·2012·6.29 MB·English

Persistent effects of a severe drought on Amazonian
forest canopy
SassanSaatchia,b,1,SalviAsefi-Najafabadyb,YadvinderMalhic,LuizE.O.C.Aragãod,LianaO.Andersonc,e,
RangaB.Mynenif,andRamakrishnaNemanig
aJetPropulsionLaboratory,CaliforniaInstituteofTechnology,Pasadena,CA91109;bInstituteofEnvironment,UniversityofCalifornia,LosAngeles,CA90045;
cEnvironmentalChangeInstitute,SchoolofGeographyandtheEnvironment,UniversityofOxford,OxfordOX13QY,UnitedKingdom;dCollegeofLifeand
EnvironmentalSciences,UniversityofExeter,DevonEX44RJ,UnitedKingdom;eRemoteSensingDivision,NationalInstituteforSpaceResearch(INPE),Sao
JosedosCampos,SaoPaulo12227-010,Brazil;fDepartmentofGeographyandEnvironment,BostonUniversity,Boston,MA02215;andgBiospheric
SciencesBranch,NationalAeronauticsandSpaceAdministration/AmesResearchCenter,MoffettField,CA94035
EditedbyStevenC.Wofsy,HarvardUniversity,Cambridge,MA,andapprovedNovember12,2012(receivedforreviewMarch19,2012)
RecentAmazoniandroughtshavedrawnattentiontothevulner- the Amazonian rainforest and its functioning are not known.
abilityoftropicalforeststoclimateperturbations.Satelliteandin Measurementsofforeststructureanddensityfrominventoryplots
situobservationshaveshownanincreaseinfireoccurrenceduring overhumidtropicalforestshaveshownanincreaseintreemor-
droughtyearsandtreemortalityfollowingseveredroughts,butto talityandadeclineintheabovegroundbiomassthatmaypersist
datetherehasbeennoassessmentoflong-termimpactsofthese forseveralyears(4,14,15).Arelationshipfoundbetweenasimple
droughts across landscapes in Amazonia. Here, we use satellite measureofmoisturestressandchangesinforestbiomasswasused
microwaveobservationsofrainfallandcanopybackscattertoshow recentlytopredictthepotentialimpactsofdroughtsontheAm-
thatmorethan70millionhectaresofforestinwesternAmazonia azoncarbondynamics(6).Directevidenceoflong-termimpacts
experiencedastrongwaterdeficitduringthedryseasonof2005 of droughts on the Amazon vegetation has been demonstrated L
and a closely corresponding decline in canopy structure and onlyincontrolledsmall-scale(1-haplot)fieldexperiments(16). NTAS
moisture. Remarkably, and despite the gradual recovery in total The study showed the mostimportant forest response to severe MECE
rpaeirnsfiastlledinusnutbilsetqhueennetxytemaras,jotrhedrdoeucgrheat,seinin20c1a0n.opThyebadcekcslicnaetteinr dcaronougphytwswheanstphleanmt-oarvtaailliatyboleflsaoriglewtarteeersdweicthlincerodwbnesloinwthaecuriptipcearl VIRONSCIEN
N
backscatterisattributedtochangesinstructureandwatercontent threshold(16,17).Similardroughteffectshavebeenobservedin E
associated with the forest upper canopy. The persistence of low Amazoniaandotherregionsinresearchplots(4,18).
backscatter supports the slow recovery (>4 y) of forest canopy Sensitivityofsatellitespectralobservationstotheforest’supper-
structureaftertheseveredroughtin2005.Theresultsuggeststhat canopycharacteristics(greenness,leafarea),particularlyatoptical
theoccurrenceofdroughtsinAmazoniaat5–10yfrequencymay wavelengths,potentiallymayprovidethenecessaryinformationto
leadtopersistentalterationoftheforestcanopy. assess the long-term impacts of droughts (18). However, recent
results from optical satellites monitoring changes in vegetation
| | | |
radar canopywatercontent rainforest QSCAT canopydisturbance greennessafterthe2005droughthavebeencontradictorybecause
ofsevereimpactsofcloudsandatmosphericaerosolsonspectral
Inthepastdecade,Amazoniahasexperiencedtwomajordroughts, observations (3, 19–22) over Amazonia. No study has examined
as highlighted by the water level of the Rio Negro recorded at the potential changes of vegetation detectable at microwave
Manaus in central Amazonia, the longest (109 y) available time frequencies.
seriesrecord.Thefirstoccurredin2005(1,2),withtheminimum Here, we analyze data from two microwave satellite sensors
riverlevelat14.75m,thelowestinthepast40y,andthesecondin measuring precipitation and canopy water content to quantify
2010,withtheriverat13.63m,thelowestintherecord(3).Severe therelativeseverityofrecentdroughtsandpotentialimpactson
droughts, often associated with the ElNiño–Southern Oscillation Amazonian vegetation (Methods). First, we characterize the
(ENSO),causeadeclineinsoilmoisture,pushingtheplant-avail- droughtoverAmazoniabycalculatingthreeindicesderivedfrom
ablewaterbelowacriticalthresholdlevelforaprolongedperiod, monthlyprecipitationmeasuredbytheTropicalRainfallMeasur-
resulting in higher rates of tree mortality and increased forest ing Mission (TRMM; 1998–2010): the dry-season precipitation
flammability(4–7).Thedroughtof2005wasunliketheENSO-re- anomaly(DPA),dryseasonwaterdeficitanomaly(DWDA),and
lateddroughtsbecauseofitstemporalandspatialextent:itspeakof maximum climatological water deficit (MCWD) (SI Methods).
intensity during the dry season and its center of impact in south- Theseindicesarecomplementaryintheirinformationcontentand
western Amazonia, rather than the central and eastern regions, providespatiallyspecificindicatorsabouttheextentandseverityof
whichareassociatedmorewithElNiñodroughts(1). moisturedeficitinAmazonia.
Warming of the tropical North Atlantic sea surface tempera- Second,weexaminetheimpactofthewaterdeficitontheAm-
tureisconsideredamajorcontributingfactorinthe2005drought, azonforestbyusingobservationsfromtheSeaWindsScatterometer
whichresultedinthelowestriverlevelsrecordedtothatdatein onboard QuickSCAT (QSCAT:2000–2009).QSCAT operatesin
southern and western tributaries (1, 8, 9). Observations from microwavefrequency(13.4GHz),providingbackscattermeasure-
groundstationsshowthatprecipitationoverthesouthernregion ments strongly affected by the temporal and spatial variations of
of Amazonia declined by almost 3.2% per year in the period watercontentandstructureoftheforestcanopy(Fig.S1)(13,22).
before this decade (1970–1998) (10). The same region experi-
enced several negative precipitation anomalies during the last
decade,indicatinganincreaseindryconditionsthatculminated
Authorcontributions:S.S.,L.E.O.C.A.,R.B.M.,andR.N.designedresearch;S.S.,S.A.-N.,and
insevere2005drought(1–3,11).Climatemodelpredictionsalso L.O.A.performedresearch;S.S.,Y.M.,L.E.O.C.A.,R.B.M.,andR.N.contributednew
suggestthattheintensityofdryseasonsandextremedryevents reagents/analytictools;S.S.andS.A.-N.analyzeddata;andS.S.andY.M.wrotethepaper.
mayincreasewithclimatechange,affectingtheecosystemfunc- Theauthorsdeclarenoconflictofinterest.
tionandhealthofforestsinAmazonia(11,12). ThisarticleisaPNASDirectSubmission.
Theshort-termconsequencesofdroughteventsarewellestab- 1Towhomcorrespondenceshouldbeaddressed.E-mail:[email protected].
lishedthroughgroundandsatelliteobservations(3–6,13).How-
Thisarticlecontainssupportinginformationonlineatwww.pnas.org/lookup/suppl/doi:10.
ever,theextentandseverityoflonger-termimpactsofdroughtson 1073/pnas.1204651110/-/DCSupplemental.
www.pnas.org/cgi/doi/10.1073/pnas.1204651110 PNASEarlyEdition | 1of6
QSCAT signal propagating at 2.1-cm wavelength and incidence Patterns of Drought Impact on Forest Canopy. The impact of this
anglesofabout50°penetratesafewmeters(1–5m)intotheforest extensivewaterdeficitontheAmazonforestwascapturedbythe
canopy, depending on forest gaps, and scatters from leaves and QSCAT (2000–2009) backscatter time series (SI Materials and
branchesoftheuppercanopyoftrees.Thebackscattermeasure- Methods).Thedryseasonstandardizedanomalyin2005showed
mentscapturebiophysicalpropertiesofforests,suchasthewater widespread (2.1 M/km2) decline in forest canopy backscatter
contentinleavesandbranches,andcanopystructure(i.e.,volume (anomalylessthan−1.0σ)insouthwesternAmazonia(Fig.1A).
orbiomass)(SIMaterialsandMethods).Temporalchanges(diurnal Nearly40%ofthisarea(0.77M/km2)indicatedamajordecline
andseasonal)ofcanopywatercontent(i.e.,leavesandbranches) in backscatter (anomaly less than −2.0 σ). The backscatter
andseasonalleafphenologyhavethelargestimpactontheradar anomalyonamonthlyorseasonalbasiscalculatedfor2000–2009
backscatter(23).Structuralchangessuchaslarge-scaleforestdeg- showsastrongspatialcorrelationwiththewaterdeficitanomaly
radationanddeforestationthatmaychangethecanopyroughness (WDA)observedbyTRMMforthesameperiod,indicatingthat
(layeringoftreecrowns),creategaps,andaffectthewatercontent water stress is the likely cause of the change in forest canopy
or biomass of the upper canopy of forests can change the back- properties.TheregionaffectedbytheQSCATanomalycovered
scatter signal significantly (Fig. S1). However, because of the in- a variety of old-growth forests, from transitional semideciduous
cidenceangleandlargefootprintofQSCATradar,otherfactors, andbambooforestsinsouthwesternBrazil,northernBolivia,and
suchassoilmoistureandvariationsinleafclumpingororientation areas in southern Peru and along the Andean flank in western
ofbranches,havelessimpactontheQSCATbackscatter(23). Amazoniatoavarietyofinundatedandterrafirmeforestsinthe
In November 2009, QSCAT sensor scanning capability failed north (24). All nonforested areas were excluded from the anal-
and the sensor stopped collecting systematic data globally, lim- ysisusingagloballandcover type(SIMaterials andMethods).
iting our analysis of the changes in canopy characteristics when We performed cross-correlation between the QSCAT and
the2010droughtoccurred.However,beforeitsscanningfailure, TRMManomalyaveragedoverAmazoniaandfoundcorrelation
throughout its mission (1999–2009) QSCAT provided reliable was significant with 1–3 mo lag, but varied over the basin de-
data over ocean and land surface without any bias or sensor pending on the rainfall patterns (Fig. 1 B and C). In the south-
degradation and continued providing limited data along its or- westernregion,withalongerdryseason,thecorrelationwaslagged
bital passes (SI Materials and Methods). To examine the impact significantlybyabout3mo.However,inthenortheasternregion,
of the 2010 drought, we analyzed TRMM precipitation radar where the dry seasonismoderate and short,the correlation was
(TRMM-PR) backscatter data operating at the same frequency strong,withnotimelag.SpatialvariationsofQSCATandTRMM
as QSCAT but with nadir-looking incidence angle and surface anomaly in 2005 show that areas captured by highly negative
backscatter measurements for periods of no rain. TRMM-PR QSCATanomaly(lessthan−3.0σ)arelargerinextentthansimilar
backscatter responds to the surface moisture by penetrating areas captured by WDA. The difference is explained by closely
deeper into the canopy and scattering from soil and understory examiningareaswheremaximumwaterdeficit(MCWD)in2005
vegetationthroughforestgaps(SIMaterialsand Methods). exceeded300mmand/ortherewasastrongwaterdeficitduring
We used time series of QSCAT backscatter data from dawn theentiredriestquarter(DWDAlessthan−3.0σ).Thereduction
orbits to monitor vegetation in its least-stressed time of day by inQSCATbackscatter,causingtheanomalyinthesouthwestre-
studyingitsmonthlyandseasonalnormalizedanomalyandspatial gion, is closely associated with the WDA gradually developing
variations over Amazonia. Throughout the time-series analysis, throughthedryseasonin2005(Fig.S3).
the QSCAT backscatter measurement was used asa direct rep-
resentationoftheupper-canopyforeststructureandwatercon- Slow Recovery of Forest Canopy. We used the time series of
tent to avoid any indirect estimation and validation of water TRMMandQSCATanomalyaveragedovertheareaaffectedby
contentorstructure(13). the drought in southwestern Amazonia (4°S–12°S, 76°W–66°W)
to examine the temporal patterns of the 2005 drought and its
Results impacts. After 2005, the area affected by the drought had a re-
PatternsofWaterDeficit.ThethreeindicesderivedfromTRMM coveryoftotalrainfall,butWDAstayednegativeontheaverage
datashowastrongnegativeanomalyoversouthwesternAmazonia duringthe2006and2007dryseasons.Recoveryofwaterdeficit
in2005(Fig.S2).OfthetotalcurrentforestedareaoftheAmazon startedin2008followedbyananomalouslywetyearin2009that
basin (∼5.5 M/km2), about 30% (1.7 M/km2) experienced stan- extended over all of Amazonia except the northeastern region
dardizedDWDAlessthan−1.0σ(σ:SD)in2005,andmorethan (Fig. S4). From late 2009, the water deficit increased before it
5% of total area (0.27 M/km2) was subject to severe anomalies rapidly reached its highest value in the decade in southwestern
(DWDAlessthan−2.0σ).Boththespatialextentandtheseverity region (Fig. 2A). However, most remarkably, forest pixels af-
of drought increased in 2010, resulting in more than 48% (2.6 fected by the water deficit over southwestern Amazonia contin-
M/km2)oftheforestareasubjecttoDWDAlessthan−1.0σ,and ued to show low values in the QSCAT backscatter (about 20%
about20%(1.1M/km2)atDWDAlessthan−2.0σ(Fig.S2).In below previous mean) from 2005 through to the end of the re-
south and southwestern regions of Amazonia, this anomaly was cord in November 2009 (Fig. 2B). We used an autoregressive
superimposedonadryseasonthatisfairlystronginnormalyears, moving-average (ARMA) model with an order of about 5% of
resulting in the forests experiencing a very large water deficit thedatapoints(>6mo)tohighlightthelonger-termtrendsand
(MCWDlessthan−300mm)bytheendofthedryseason.The cycles in the data. The time seriesof the QSCAT anomaly sug-
generally wetter forests in central Amazonia with the largest gests that the 2005 drought caused a step change in the back-
negative DPA and DWDA experienced low to moderate water scatter properties of the canopy, with little recovery in the
deficit(MCWDlessthan−100mm)in2005and2010.Datashow subsequent years (SI Materials and Methods). The response is
that precipitation anomalies over these regions lasted only over very localized to regions in western Amazonia that experienced
arelativelyshorttimespanwithinthelastdecade.However,the thestrongestwaterdeficitanomalies,hencecannotbeattributed
spatial extent of precipitation anomaly (DPA) and MCWD in to hypothetical changes in sensor performance. We tested the
southwesternAmazoniareachedtothefoothillsoftheAndesin sensorperformanceinotherregionsofAmazoniaandtheworld
2005 and extended to northern regions of Peru, Ecuador, and toensurethestabilityofthebackscattersignalanditscalibration
Colombia,suggestingapatternapproximatelyconsistentwiththe (Fig.S5).SpatialpatternsofannualQSCATanomalyforthedry
low river stage measured in Rio Negro and other rivers in the season support the long-term reduction in backscatter after the
southwesternAmazonbasin(2,9). 2005droughtuntiltheendof2009,whenthepositiveanomalyof
2of6 | www.pnas.org/cgi/doi/10.1073/pnas.1204651110 Saatchietal.
A B C
Fig.1. Spatialextentandseverityofthe2005Amazoniandroughtusingseasonal(JAS)standardizedanomalyofQSCATbackscatterdataatHpolarization
forascendingorbits(acquiredatdawn),capturingtheforestcanopywaterstressandspatialpatterns.(A)MagnitudesofQSCATanomalybeyond±1.0σ.(B)
Spatialcross-correlationbetweentheTRMMmonthlyWDAandtheQSCATmonthlyanomalywith1molagovertheperiod2000–2009.(C)ACFdeveloped
betweenthetwodatasetsaveragpeffiffidffiffioverforestedpixelsinAmazoniawithtimelagsrangingfrom0to18moineitherdirection.Dashedlinesrepresentthe
95%confidenceintervalas ±2= N.
precipitationslightlychangedthetrendbeforethe2010drought (approximately −1.5 σ), relatively significant water deficit anom- L
(Fig.S5). alies in summer of 2006 and 2007 (approximately −1.2 σ), and NTAS
The intensity of the anomaly in 2005and its gradual shift to- astronganomalyin2010(peakedbetween−1.5and−2.0σ). MECE
dwiasrtrdibruetcioonvesroyflangogrimngalitzheedwanaotemradlieefiscfiotratlshoeidsryevsiedaesnotninwattheer droTuhgehtvawriaastiotnesotefdQtShCroAuTghansotamtiasltyicableftoirmeea-snedrieasftearntahlyesi2s00to5 NVIRONSCIEN
deficit and the canopy backscatter power over pixels in south- quantifysignificanceofthestepchangeandtrendsinthedata(SI E
western Amazonia (Fig. S6). The distribution of the QSCAT Materials and Methods). Using an ARMA analysis over 120 mo
anomaly peaked at a value between −2 and −2.5 σ in 2005 but (2000–2009), we found that the QSCAT time series was piece-
stayed significantly negative and different from the distribution wise stationary because of changes in the mean and, to some
of the entire time series for years before 2005 that peaked at extent, the variance over time caused bythe 2005drought. The
aboutzero(P<0.01fromatwo-sidedttest).Duringthisperiod, autocorrelation function (ACF) and partial ACF (PACF) sug-
thedistributionofwaterdeficitoverthesameregionalternated gestedthatthetimeserieshadalagof1–3mowithPACFcutoff
between negative and positive, with a strong negative in 2005 afterlag1(95%confidenceinterval),allowingtheprocesstobe
represented by autoregressive (AR) model (1) on the monthly
but not annual time scale (Fig. S7). The detection of the step
change in the QSCAT time series was performed using the
A BreaksinAdditiveSeasonandTrend(BFAST)algorithmbased
on the iterative decomposition of time-series data (25). The
Jul. 2005 010 drought rd1theemestueodlctatsuetandshd(ioeFnrwigJt.huth3ena)ea.tsT2as0hus0emi5gnpnwotiifiiiostchenalnoreotfvoe9sttl7-emd.p5i%edcahnn(ao-3nstqgσiuen)afloirunfeetQnehrceSreContrAohi(eTsReRMdrMaaStnaSEgEe)wooainsff
2
thedetection,indicatingalowcommissionerrorinthedetection
performance(SIMaterialsandMethods).TheBFASTalgorithm
also detected the seasonality of QSCAT anomaly and showed
thatitdidnotinfluencetheaccuracyofdetectingthebreakpoint
inthetime-seriesdata.Thepost-2005trendinthetimeserieswas
not significant, although it showed the slow recovery of QSCAT
signalthatlastedabout4yafterthe2005drought(Fig.3).
B To demonstrate the changes in QSCAT signal relative to
TRMM water deficit, we used the average monthly normalized
anomalies for southwestern Amazonia and calculated the rela-
tive difference between QSCAT anomaly and TRMM monthly
WDA (both are normalized and unitless). On average, the dif-
ferenceinanomaliesstayedatzero(zeroslope)beforethe2005
droughtandhadanegativeslopeafter2005,suggestingalagin
recovery of QSCAT anomaly relative to the TRMM WDA in
post-2005
canopy recovery lag southwestern Amazonia. The largest decline in QSCAT back-
scatter occurred in September 2005 from gradual development
of negative anomalies during the driestquarter in July, August,
andSeptember(JAS).Weextendedtheanalysisovertheentire
Amazoniabymapping thespatialdistributionofthepixelswith
negativeanomaliesinbothQSCATandTRMMdata(lessthan
Fig. 2. Time series of (A) TRMM and (B) QSCAT monthly anomaly over
western Amazonia (window: 4°S–12°S, 76°W–66°W). Solid lines show the −1.0 σ) and with significantly (P < 0.01) negative slopes (SI
resultoftheARMAoftheorderof6mo. Materials and Methods) between QSCAT anomaly and WDA
Saatchietal. PNASEarlyEdition | 3of6
layerofthecanopy,consistingofemergentcrownsthatoftenare
exposed to higher vapor-pressure deficits and consequently are
moresensitivetodroughts(4,16).Theoretically,thewidespread
decline in radar backscatter suggests changes in canopy water
content and structure (e.g., fresh biomass), unlike what was ob-
served in optical sensing (3, 20). The decline in backscatter
suggests a reduction of upper-canopy biomass or canopy rough-
ness attributed to potential drought-driven disturbance. The
magnitude of the decline in 2005 is significantly larger than
seasonal amplitude of the QSCAT backscatter caused by phe-
nologyandchangesofcanopywatercontentfromnaturalcycles
ofdryandwetseasons.
Second, the QSCAT anomaly remained negative after the
2005droughtoveralargeareainwesternAmazonia,suggesting
the persistent effect of the drought on the forest canopy. The
severity of the disturbance caused a slow recovery of the forest
canopy to its predrought condition in terms of biomass or
roughness (canopy layering), lagging the precipitation recovery
ofsubsequentyearsuntilthe2010drought.Notably, morethan
0.6 × 106 km2 of areas affected by the 2005 drought (QSCAT
anomaly less than −2.0 σ) coincided with the areas affected by
the 2010 water deficit (TRMM DWDA less than −2.0 σ), sug-
gestingapotentiallywidespreadexacerbationofstressonforests
ofsouthandwesternAmazonia.
Without extensive surveys and perhaps airborne observations
and validations, the interpretation of the decline of backscatter
Fig. 3. (A) BFAST seasonal trend decomposition of the QSCAT monthly and its direct relation to the forest disturbance remain chal-
normalizedanomalytimeseriesofsouthwesternAmazonia(window:4°S– lenging. A simple wilting or shedding of leaves during the peak
12°S,76°W–66°W)intoseasonal,trend,andremaindercomponents.(B)The drought, resulting in a temporary decline of net primary pro-
seasonalcomponentisestimatedbytakingthemeanofallseasonalsub- duction,wouldbeexpectedtobefollowedbyrecoveryofcanopy
componentsstartingfromJanuary2000.Therangeofseasonalamplitudeis properties within a year. Such recovery is apparent in central
lessthan20%oftherangeofQSCATanomaly.(C)Oneabruptchangeinthe
Amazonia, where the backscatter anomaly recovered rapidly,
trendcomponentofthetimeseriesisdetectedonJune2005.Theshadedbar
indicatesthe97.5%(3σ)confidenceinterval.(D)Theremaindershowsthe despiteexperiencingastrongwaterdeficitandQSCATanomaly
variationofthesignalaftertheremovalofthetrendcapturingthetemporal in 2005 (Figs. S2 and S3). The delayed recovery of QSCAT in
variationsinthetimeseries. southwestern Amazonia suggests a decline in more long-lived
aspectsofcanopystructurewithrecoverytimescalesgreaterthan
3–4y,suchaslossofleavesordiebackofbranches,orpotential
after the 2005 drought (Fig. 4A). Pixels with larger negative treefalls creating largegaps.
slopes represent areas with a longer lag in canopy recovery rel- We found no in situ observations over the region affected by
ativetotherecoveryofthewaterdeficitandshowingpersistent the2005droughtthatcouldbeusedtodirectlyverifyourresults.
drought impacts on canopy characteristics. Regions A and B, However, in most tropical drought studies, there is strong evi-
respectively, show the old-growth forests of southern Peru and dence that large-diameter or emergent trees have significantly
the states of Acre and Rondônia in the western Amazon of highermortality thansmall trees(16–18,26,27). Theeffects of
Brazil. Region C covers a mosaic of undisturbed and disturbed extremedroughtsontheunderstorylightenvironmentoftropical
forests in the state of Mato Grosso and southern Pará (both in forests may be compared with the effects of tree-fall gaps, al-
Brazil).Allthreeregionswerereportedtohaveananomalously thoughonmuchlargerscales(27).Withtherecoveryofrainfall
highernumberoffiresin2005andsubsequentyears,suggesting and potential increase in light availability due to gaps from
a potential lower canopy water content and higher fuel loads canopy disturbance, the understory vegetation andpioneer spe-
(4, 22). The 2005–2009 forest loss and degradation from de- cies may increase productivity a few months after severe
forestation and fire impacts did not change the persistent droughts (22). These changes may affect nadir-looking optical
QSCAT negative anomaly (Fig. S8). However, a significantly satelliteobservationswithsensitivitytovegetationgreennessand
large number of fires during 2005–2009 (>35%) occurred in photosynthesis capacity (3, 21, 22). In situ measurements show
QSCATpixelswithstrongnegativeanomaly(lessthan−1.0σ)in that the mortality of large trees remains elevated even a few
2005,andmorethan78%offiresin2010occurredinpixelswith years after the drought (4), suggesting a decline in canopy
large negative slopes (less than −0.01). The occurrence of fires structureorbiomassfollowedbygradualdevelopmentofcanopy
andtheareaswithslowrecoveryafterthedroughtcoincidedwith emergent trees (4, 22). The recovery of canopy trees after the
regions with large seasonality of backscatter in QSCAT data, drought event is a slower process andmay takelonger to reach
pointing predominantly to transitional and seasonal forests of the predrought state (18,22, 25,26). Based onthe resultsfrom
Amazonia(Fig.4B)(SIMaterials andMethods). this study and the evidence reported in the literature, we hy-
pothesize that western Amazonia experienced a large-scale
Discussion canopy disturbance from the 2005 drought, resulting in the de-
Overall, the results of QSCAT analysis indicate two important cline of emergent and canopy tree structure and biomass that
conclusions.First,theQSCATanomalycapturestheextentand continuedwithaslowrecoveryforthenextfewyears.Weexpect
intensityofthe 2005droughtimpact ontheAmazon forestand future field campaigns directed to examine the effect of severe
provides patterns consistent with areas that experienced the droughts and analysis of existing in situ data from permanent
largest water deficit in the driest quarter. Changes in QSCAT researchplotsinwesternAmazonia(4)totestthehypothesisand
backscatteraretheresultofchangesinthepropertiesofthetop potentially verifythe resultsofsatelliteobservations.
4of6 | www.pnas.org/cgi/doi/10.1073/pnas.1204651110 Saatchietal.
A B
Fig.4. (A)SpatialrepresentationoftherateofrecoveryofpixelsaffectedbyQSCAT-negativeanomaly(lessthan−1.0σ)calculatedbytheslopeofthe
differenceinQSCATandTRMMmonthlyanomaliesfromSeptember2005toNovember2009(SIMaterialsandMethods).Pixelswithsignificantlylarge AL
negativeslopesrepresentforestswithslowerrecoveryandcoveranareaof∼4×105km2.AreasdelineatedinhighlightedareasasA,B,andCrepresent NTS
EE
ro(2ef0gt0iho0en–s220i0n009s5o).durtohueagshtte.r(nBP)eSrpua,ttihalersetpatreesoefnAtactrieoninoBfraazreila,asnwditahresatrsoinngtehsetssteaatesoonfaMlitaytoinGcraonsosop,yrepsrpoepcetritvieelsy,dsehtoewctiendgbayretahsewQitShCpAoTtebnatcikaslclyatttheerlmaregaessutriemmpeancttss VIRONMSCIENC
N
E
A large-scale drought disturbance of forest structure from this region may be witnessing the first signs of potential large-
mortality of large trees and, to some extent, a drop in the leaf scaledegradationofAmazonianrainforestfromclimatechange
area of the forest (16, 28) may lead to a sustained efflux of (10,11,30).
carbon dioxide from the decay of wood, with the process sig-
nificantly perturbing the net ecosystem exchange and carbon MaterialsandMethods
fluxes (11, 12, 14). Results from climate analysis for the period Thisstudyisbasedontheuseofvariousmicrowavesatelliteobservationsof
1995–2005 demonstrate a steady decline in plant water avail- Amazoniatodetecttheregionalandpotentialseverityoftheimpactofthe
ability over the same region, suggesting a decade of moderate 2005droughtontheAmazonforests.Ourapproachincludesfivesteps:(i)
waterstressbeforethe2005drought(10,22),helpingtotriggera spatialanalysisofmonthlyTRMMrainfalldatatocalculatethestandardized
anomaly of rainfall during the dry season, maximum water deficit, and
large-scalecanopydisturbanceafterthe2005drought.Ahigher
water deficit in subsequent years, together with another strong anomaly of monthly water deficit; (ii) spatial analysis of monthly QSCAT
backscatter data to computer pixel-level standardized anomaly from the
localdroughtin2007,suggeststhatsoilsfromalargeportionof satellitedawnorbitstomonitorvegetationinitsleast-stressedtimeofday
southwesternAmazoniamaynothavereachedthefieldcapacity, anditsspatialcorrelationwithTRMMwaterdeficit;(iii)time-seriesanalysis
which would favor canopy recovery (29). Other factors, such as ofQSCATmonthlybackscatteranomalyoverwesternAmazoniausingthe
adeclineinrainfallandlargervariabilityinthedryseasonover ARMAmodel,testingACFandPACFwithtimelags,anditerativeapplication
southwestern Amazonia since the late 1980s and early 1990s of the additive decomposition algorithm BFAST to detect a significant
(Fig.S9),mayhavecontributedtoanincreasinglydriercondition breakpointandtrendintheQSCATdataassociatedwiththe2005drought;
(iv) quantifying the impact of deforestation and fire occurrence on the
inthisregion.Weshowthattheserecentnegativeanomaliesand
QSCAT anomaly and trend results and showing the independence of the
year-to-year variations are strongly linked to both the warming results from canopy disturbances that may have been caused byfire and
and variations in the sea surface temperature (1, 2). The most degradationduringandafterthe2005drought;and(v)testingtheregional
recent droughts are related to higher temperatures in the trop- impactsofvariationsinthehistoricalclimatedataonthepatternsofrainfall
icalAtlantic,showingastrongregionalsensitivityofWDAtothe anomaly inAmazoniatoexplaintheclimaticcauseofrecent droughts in
tropicalNorthAtlanticindex(Fig.S10). Amazonia.OuranalysisoftheQSCATdatawaslimitedtotheperiod2000–
Our analysis ends in 2009. It seems likely that the observed 2009andcouldnot provideinformationaboutthe2010drought.Wein-
canopy response was repeated in the more severe drought of cludedtheTRMM-PRbackscatteranomalytodemonstratethechangesin
surface moisture during the 2005 and 2010 droughts as evidence of the
2010(Fig.S11),forwhichQSCATdataarenotavailable;hence,
potentialimpactofthe2010droughtonforestsinsouthwesternAmazonia
anewwaveofdisturbancemayhaveaffectedforestcanopiesnot
alreadyaffected.
yetrecoveredfromthepreviousdroughtsandwaterdeficit.The
WeusedtheModerateResolutionImagingSpectroradiometer(MODIS)
TRMM-PR backscatter anomaly suggests that the surface landcovermapof2005,GlobCoverlandcovermapof2009,andMODIS-
moisture in western and southern Amazonia dropped signifi- derivedpixelfirecountsfrom2001to2010toexcludenonforestpixelsfrom
cantly in 2010and lastedlonger than the dry season (Fig. S11), ouranalysisandtoquantifythepercentageofpixelswithalargeQSCAT-
potentiallycausingmorestressontheforestcanopy.Ifdroughts negativeanomalyaffectedbyfiresafterthe2005drought(2005–2009)and
continuetooccurat5–10-yfrequency,orincreaseinfrequency, duringthe2010drought.Wehadnoindependentgroundmeasurementsto
verifyourresultsbecauseofthelargepixelsizeofthesatelliteobservations
largeareasofAmazonianforestcanopylikelywillbeexposedto
andtherecentnessofthedroughtevent.However,weprovidedbiophysical
thepersistenteffectofdroughtsandtheslowrecoveryofforest
interpretation of the satellite observations and evidence from in situ
canopystructureandfunction.In particular,areasofsouth and measurements and ecological studies to corroborate our findings of the
western Amazonia have been shown to be affected severely by persistent effect of droughts on forest canopy. We provide detailed in-
increasing rainfall variability in the past decade, suggesting that formationaboutthedatainSIMaterialsandMethods.
Saatchietal. PNASEarlyEdition | 5of6
ACKNOWLEDGMENTS. We thank Ryan Harrington from the University of aspartoftheTRMMprojectjointlysponsoredbytheJapanNationalSpace
CaliforniaatLosAngeles(UCLA)CenterforTropicalResearchforhiscontribu- DevelopmentAgency(NASDA)andNASAOfficeofEarthSciences.Theresearch
tion in time-series analysis. The enhanced-resolution QSCAT data were pro- waspartiallysupportedbyNASAgrantsattheJetPropulsionLaboratoryand
ducedbyProf.DavidLongatBrighamYoungUniversityaspartoftheNational the UCLA Institute of Environment and by Natural Environment Research
Aeronautics and Space Administration (NASA)-sponsored Scatterometer Cli- Council(London)GrantsNE/F015356/2andNE/l018123/1attheUniversityof
mateRecordPathfinder.Theprecipitationdatausedinthisstudywereacquired ExeterandGrantNE/F005806/1attheUniversityofOxford,UnitedKingdom.
1. MarengoJA,etal.(2008)ThedroughtofAmazoniain2005.JClim21:495–516. 17.daCostaACL,etal.(2010)Effectof7yrofexperimentaldroughtonvegetation
2. MarengoJA,etal.(2011)Thedroughtof2010inthecontextofhistoricaldroughtsin dynamicsandbiomassstorageofaneasternAmazonianrainforest.NewPhytol187
theAmazonregion.GeophysResLett38:L1270310.1029/2011GL047436. (3):579–591.
3. XuL,etal.(2011)WidespreaddeclineingreennessofAmazonianvegetationdueto 18.NakagawaM,etal.(2000)Impactofseveredroughtassociatedwiththe1997-1998El
the2010drought.GeophysResLett38:L0740210.1029/2011GL046824. NinoinatropicalforestinSarawak.JTropEcol16:355–367.
4. PhillipsOL,etal.(2009)DroughtsensitivityoftheAmazonrainforest.Science323 19.AsnerGP,AlencarA(2010)DroughtimpactsontheAmazonforest:Theremote
(5919):1344–1347. sensingperspective.NewPhytol187(3):569–578.
5. AragãoLEOC, et al. (2008) Interactions between rainfall, deforestation and fires 20.SaleskaSR,DidanK,HueteAR,daRochaHR(2007)Amazonforestsgreen-upduring
duringrecentyearsinBrazilianAmazonia.PhilosTransRSocLondBBiolSci363 2005drought.Science318(5850):612.
(1498):1779–1785. 21.SamantaA,etal.(2010)Amazonforestsdidnotgreen-upduringthe2005drought.
6. LewisSL,BrandoPM,PhillipsOL,vanderHeijdenGM,NepstadD(2011)The2010 GeophysResLett37:L05401–L05406.
Amazondrought.Science331(6017):554. 22.AndersonLO,etal.(2010)RemotesensingdetectionofdroughtsinAmazonianforest
7. NepstadD,etal.(2004)Amazondroughtanditsimplicationsforforestflammability canopies.NewPhytol187(3):733–750.
andtreegrowth:Abasin-wideanalysis.GlobChangeBiol10:704–717. 23.Frolking S, et al. (2006) Evaluation of the SeaWinds scatterometer for regional
8. DavidsonEA,etal.(2012)TheAmazonbasinintransition.Nature481(7381):321–328.
monitoringofvegetationphenology.JGeophysRes111,10.29/2005JD006588.
9. ZengN,etal.(2008)Causesandimpactsofthe2005Amazondrought.EnvironRes
24.PiresJM,PranceGT(1985)ThevegetationtypesoftheBrazilianAmazon.Key
Lett3(1),10.1088/1748-9326/3/1/014002.
Environments:Amazonia,edsPranceGT,LovejoyTE(Pergamon,NewYork),pp
10.LiW,FuR,JuárezRI,FernandesK(2008)Observedchangeofthestandardizedpre- 109–145.
cipitationindex,itspotentialcauseandimplicationstofutureclimatechangeinthe
Amazonregion.PhilosTransRSocLondBBiolSci363(1498):1767–1772. 25.VerbesseltJ,HyndmanR,ZeileisA,CulvenorD(2010)Phenologicalchangedetection
whileaccountingforabruptandgradualtrendsinsatelliteimagetimeseries.Remote
11.MalhiY,etal.(2009)Exploringthelikelihoodandmechanismofaclimate-change-
SensEnviron114:2970–2980.
induced dieback of the Amazon rainforest. Proc Natl Acad Sci USA 106(49):
20610–20615. 26.ConditR,HubbleSP,FosterRB(1995)Mortalityratesof205neotropicaltreeand
12.CoxPM,etal.(2008)IncreasingriskofAmazoniandroughtduetodecreasingaerosol shrubspeciesandtheimpactofaseveredrought.EcolMonogr65:419–439.
pollution.Nature453(7192):212–215. 27.SlikJW(2004)ElNiñodroughtsandtheireffectsontreespeciescompositionand
13.FrolkingS,etal.(2010)TropicalforestbackscatteranomalyevidentinSeaWinds diversityintropicalrainforests.Oecologia141(1):114–120.
scatterometermorningoverpassdataduring2005droughtinAmazonia.Remote 28.BrandoPM,etal.(2010)Seasonalandinterannualvariabilityofclimateandvege-
SensEnviron115:897–907. tation indices across the Amazon. Proc Natl Acad Sci USA 107(33):14685–14690,
14.PhillipsOL,etal.(2010)Drought-mortalityrelationshipsfortropicalforests.New 10.1073/pnas.0908741107.
Phytol187(3):631–646. 29.MeirP,etal.(2009)TheeffectsofdroughtonAmazonianrainforests.Amazoniaand
15.Williamson GB, et al. (2000) Amazonian tree mortality during the 1997 El Niño GlobalChange,GeophysicalMonographSeries,edKellerM,etal.(AmericanGeo-
drought.ConservBiol14:1538–1542. physicalUnion,Washington,DC),Vol186,pp429–449.
16.NepstadDC,TohverIM,RayD,MoutinhoP,CardinotG(2007)Mortalityoflargetrees 30.ZelazowskiP,MalhiY,HuntingfordC,SitchS,FisherJB(2011)Changesinthepo-
and lianas following experimental drought in an Amazon forest. Ecology 88(9): tentialdistributionofhumidtropicalforestsonawarmerplanet.PhilosTransactA
2259–2269. MathPhysEngSci369(1934):137–160.
6of6 | www.pnas.org/cgi/doi/10.1073/pnas.1204651110 Saatchietal.
Supporting Information
Saatchi et al. 10.1073/pnas.1204651110
SIMaterialsandMethods ascending and descending orbits with equator crossing at 0600
Tropical Rainfall Measuring Mission Data Analysis. Monthly time and 1800 hours local standard time (LST), respectively. We
series (1998–2010) of Tropical Rainfall Measuring Mission choseQSCATasanalternativetoopticalModerateResolution
(TRMM)-merged precipitation data at 0.25° spatial resolution ImagingSpectroradiometer(MODIS)datatostudytheAmazon
(3B43-v6: Goddard Distributed Active Archive Center (GES vegetation response to droughts (4) because (i) the radar back-
DISK DAAC) with cumulative values estimated in millimeters scatter at microwave frequency (13.4 GHz) and high incidence
per month were used to calculate the monthly precipitation angle (∼46° and 54° from zenith) over dense forest cover is
anomaly and dry-season [July, August, and September (JAS)] strongly sensitive to the upper-canopy (predominately leaf and
precipitation anomaly (DPA) of 2005 and 2010 (Fig. S2). The branch) structure and water content through the canopy di-
anomaly on a pixel-by-pixel basis (i, j) for each year (y) was cal- electric properties (Fig. S1) (5, 6), and (ii) being an active mi-
culatedasadeparturefromthe1998–2010mean(TRMM1998–2010), crowavesensor,QSCATimagesovertropicalforestshavealmost
excluding the measurement from year (y) and normalized by the no effects from the presence of clouds and aerosols, and no
standarddeviation(STD): sensitivitytoseasonalvariationstoincomingsolarradiation(5).
The effect of atmospheric water vapor on the signal also is
TRMM ði;jÞ¼TRMMyði;jÞ− <TPMM1998−2010ði;jÞ>: negligible and has no impact on the data quality (5, 7). We
y anomaly STDðTRMM ði;jÞÞ compared QSCAT data with other similar sensors, such as the
1998−2010
TRMMprecipitationradar(TRMM-PR)surfacebackscatterfor
the continuation of the data record to examine the changes in
As a complementary measure of drought, we also calculated
the monthly water deficit (WD) and cumulative WD (CWD) surface moisture during the 2010 drought. We found no other
radar measurements with strong sensitivity to forest canopy
valuesatthepixellevelfromtheTRMMtimeseries.Forthis,we
firstcalculatedthemonthlyWDsbasedontheapproximationthat characteristics (Fig.S1) (8).
We used QSCAT backscatter data (σ0) at H polarization in
amoisttropicalcanopytranspiresabout100mm/mo.Thisvalue
ascendingnodefrommorningpasses(0600LST)tocreateboth
was obtained from ground measurements in different locations
monthly and dry-season (JAS) anomalies for the entire time
andseasonsinAmazonia(1).Ifthemonthlyprecipitationisless
seriesusing the equation
than100mm,theforestentersWD;otherwise,theWDissetto
zero.First, wecalculate WDfor eachmonth (n)from (2):
σ0ði;jÞ− <σ0 ði;jÞ>
σ0anomalyði;jÞ¼ y (cid:1) 1999−2009 (cid:3) :
if  WDn−1ði;jÞ−Eði;jÞþTRMMnði;jÞ<0; STD σ01999−2009ði;jÞ
then  WDnði;jÞ¼WDn−1ði;jÞ−Eði;jÞþTRMMnði;jÞ;
else  WDnði;jÞ¼0 TheHandVpolarizationsprovidedsimilarresults.However,
the morning passes showed stronger anomaly than the late af-
where E(i,j) is the evapotranspiration and TRMMn(i,j) is the ternoon passes, probably because of the higher canopy water
monthlyprecipitationateachpixel.TheWDascalculatedinthe content in the morning with trees naturally inthe least-stressed
above equation is the same as the CWD of any month repre- condition because of overnight hydrological recharge (higher
senting the sum of WD values up to and including that month QSCAT backscatter) and hence greater sensitivity to the water
(3);the maximum climatological WD(MCWD) inayearis the content status of the canopy (5). Note that in all analyses, we
maximum value of CWD recorded in that year. The WD will used the QSCAT backscatter power values in linear scale (m2/
always reset to zero when monthly rainfall is greater than 100 m2) and not in decibels to allow compatible analysis between
mm.Thiswillallowarelativelywetdryseasontohavenoimpact rainfall andQSCAT.
on the maximum WD. The seasonal WD is obtained by im-
plementing the above formula for every quarter and using the QSCAT Performance. QSCAT backscatter data (σ0 in decibels)
WDfromJuly toSeptember asthe WDfor the dry season. were collected on a sun-synchronous orbit with twice-daily
We used the DPA, dry-season WD anomaly (DWDA), and swaths over a given point, providing morning passes (∼0600
MCWD as three measures to map the extent and intensity of LST) in ascending orbits and evening passes (∼1800 LST) on
droughts. Together, these measures are strong predictors of descending orbits. We acquired aggregated data covering July
droughtintensityandprovidecomplementaryspatialinformation 1999 through November 2009 from the NASA Scatterometer
ontheextentofdroughtsthatcorrelatewithtreemortality.DPA Climate Record Pathfinder project. The data were reprocessed
andDWDAfor2005and2010showthepatternsofanomalous fromthenativesensorresolutionof∼25km,withroughlydaily
wateravailabilityextendingfromsouthwesternAmazoniatothe coverage, by combining multiple orbit passes togenerate image
interiorofthebasin,whereforestsdonotoftenexperienceadry productswithimprovedspatialresolution(4.45km)andreduced
seasoninnormalyears(Fig.S2).However,MCWDprovidesthe temporal resolution (4D) (9). The QSCAT backscatter data
magnitude of WD related to the intensity of drought and its werecalibrated throughouttheinstrumentlifetime usinginsitu
spatial footprint in south andsouthwestern Amazonia. measurementsand datafromother sensors (9).
InNovember2009,theQSCATantennaceasedtospinafterits
QuickSCAT Data Analysis. The entire record of QuickSCAT continuous operation over more than a decade. However, the
(QSCAT) time-series data (1999–2009) was obtained from the sensor continued to collect valid backscatter data over narrow
National Aeronautics and Space Administration (NASA) Scat- tracksandatafixedazimuthangle;thesedataarebeingusedto
terometer Climate Record Pathfinder (www.scp.byu.edu/). The calibrate and validate Oceansat, a new sensor with similar con-
datasetcontainstheenhancedresolution(4.45-kmpixelgridand figuration from the Indian Space Agency. There has not been
8–9-km effective resolution) four-dimensional (4D) composites anyevidenceofQSCATbackscattercalibrationfailureorsignal
at both H (horizontal) and V (vertical) polarizations and for degradation during its operation for global coverage. QSCAT
Saatchietal.www.pnas.org/cgi/content/short/1204651110 1of14
has provided reliable data products of wind speed over oceans understory vegetation visible to the scatterometer through can-
(10), soil moisture over land (11), and sea ice cover over polar opygaps (8).
regions (12).To demonstrate the stability of the QSCAT signal ThepenetrationdepthoftheKu-bandsignalintothetropical
over land after the 2005 drought, we developed the backscatter forest canopy between the two sensors may be simulated using
anomaly over the northwest region of Amazonia that was not asimplewatercloudmodelorforestcanopy3Dmodelsusedfor
affected by the 2005 drought (Fig. S4). Similar patterns may be radarbackscatteratdifferentfrequencies(18–21).Theone-way
observedreadily over other regions ofthe landand ocean. attenuation can be simulated using the following model (21):
Biophysical Information in QSCAT Backscatter. Space-borne scat- τ¼exp½−κ(cid:129)GðθÞ(cid:129)V(cid:129)en(cid:129)t=cosðθÞ(cid:2);
terometers have provided continuous microwave coverage of
theearthforapproximatelytwodecades.Thesescatterometers whereG(θ)isthegapfractionseenatanyincidenceangle;Vis
originally were designed to measure oceanic surface winds and
a vegetation parameter such as the canopy water content,
land surface parameters such as soil moisture. However, their sometimes expressed as the LAI; κ is the adjusted Ku-band
data also are extremely useful in a broad range of ice and land propagation constant into the canopy (2π/λ, λ:wavelength) (18,
applications, including the use of extensive scatterometer time 19);eisthecomplexpartoftheleafdielectricconstantrelatedto
series to determine seasonal and interannual variability and itswatercontent(6);andtrepresentstheopacityofasingleleaf
possible relationships to climate change (9, 13). To date, there
adjustedbyleafthickness(20).Usingalandscape-scaleLAI=6
havebeenfivespace-bornescatterometerswithlongtime-series forrainforestcanopies(22)asV,varyingthegapfractionseenat
measurementsthat have flownondifferent international space-
differentincidenceangles(23),wecansimulatetheattenuation
crafts. These include the SeaWinds scatterometer onboard
and the half-power (when incidence energy of the microwave
QSCAT (13.4 GHz), which launched in 1999 and continued
reaches its half-power) penetration depth (17) to examine the
collectingdatauntilNovember2009;the5.3-GHzscatterometer
relative sensitivity of Ku-band backscatter to different layers of
fromtheEuropean SpaceAgency(ESA),carriedonboardboth
the forest canopy (Fig. S1). The overall variations of the pene-
theERS-1andERS-2satellitessince1991andcontinuingasthe
tration depth may be larger than what has been simulated be-
ESA advanced scatterometer (ASCAT) since 2009 (9); and
causemicrowavesignalpropagatesfartherintothecanopythan
TRMM-PR, operating since its launch in 1997 at K band (14
u its half-power level, and at large footprints (5 km), there often
GHz) and used primarily for rain measurements over the land
surface andanon–sun-synchronous orbit (14,15). are largegaps dueto variations offorest cover.
Insummary,factorsaffectingtheQSCATdataareasfollows:
Thescatterometersystemtransmitsradarpulsesandreceives
(i)Thereareseasonalchangesincanopywatercontentthatalso
the backscatter energy from the surface with its intensity repre-
sentedasthenormalizedradarbackscattercoefficient(σ0)(Fig. may be related to phenology in temperate forests and seasonal
S1). The magnitude of σ0 depends on the roughness and di- tropicalforests(18).(ii)Becauseofthewavelengthoftheradar
signal (about 2.1 cm), the penetration depth, and the course
electric properties of the particular target under observation,
indicating a strong sensitivity of σ0 to the surface soil moisture resolutionimagecharacteristics,thewatercontentandstructure
of branches and leaves affect the scattering as they form the
over bare surfaces or open and low-density vegetation (e.g.,
upper-canopy roughness and dielectric properties. Leaf clump-
grass,shrublands)(13)andcanopystructureandwatercontent
ing, the orientation of scatterers, and other minor structural
overdensely covered vegetation (e.g.,tropical rainforests) (8).
characteristics have less impact on the QSCAT radar backscat-
In this study, we chose to use QSCAT data over all other
ter.(iii)Structuralchanges offorestcanopy,suchaslarge-scale
existing datasets to monitor changes of canopy properties over
Amazonia for several reasons: (i) QSCAT provides one of the degradation and deforestation, that may change the canopy
roughness,creategaps,andaffectthewatercontentorbiomass
longest records of continuous measurement, with consistent
oftheforestcanopycanchangethebackscattersignal.(iv)Inour
calibrationglobally.Othersensors,suchastheEuropeanremote-
sensingsatellite(ERS)seriesandASCAT,providemeasurement monthly time-series analysis, the impact of direct rain and
from three different spacecrafts, with potential differences in morningdeworothershort-termclimateeventsisnegligible.(v)
calibrations that may affect long-term time-series analyses. (ii) Soil moisture effect is relatively small on the QSCAT data, ex-
QSCATmeasurementsareperformedathigherfrequency(13.4 ceptinareaswheretheforestcoverispatchyandtherearelarge
GHz) than ERS and ASCAT sensors (5.3 GHz), with less pen- areasof exposed soilwithin the QSCAT pixel.
etration into the forest canopy and relatively no impact of soil ResultsfromtheTRMM-PRanalysisovertheAmazonbasin
moistureonthebackscatterindenselyvegetatedsurfacessuchas showasimilarandstrongdependenceofincidenceangleonradar
Amazonia. (iii) QSCAT observations are performed at high in- backscatter,causing16–18-dBdifferencesinbackscatterfroma0°
cidence angles (46° and 54° from zenith), suggesting a strong to5°incidenceangle(16).Ingeneral,theTRMM-PRdatapro-
sensitivitytoonlythetopfewmetersofforestcanopy(Fig.S1). cessed with elimination of the signal during the rain event may
To date, there is clear evidence that QSCAT backscatter data carryinformationaboutthesurfacemoisturesmoothedovertime
can monitor forest canopy structure and water content with without showing the rain events (14). We used the TRMM-PR
significant correlations to seasonal and spatial variations of the data and developed the backscatter anomaly over all of Ama-
canopyleafareaindex(LAI)indenselyforestedareas(8).Ithas zoniatodemonstratethestronganomalyforthe2005and2010
been demonstrated that QSCAT time-series data can capture droughts(Fig.S11).Althoughbothdatasetsshowsimilarregions
phenological variations over different biomes globally(16). for the 2005 drought in Amazonia, the QSCAT signal is much
TRMM-PR measurements also have been collected continu- stronger and widespread as it shows only the changes in the
ouslyoveralongperiod(1997topresent)atarelativelyhigher canopypropertiesandhasrelativelynoinformationfromthesoil
temporal frequency than QSCAT twice a day. TRMM-PR moisture. The 2010 backscatter anomaly shows the widespread
standard backscattered product PR-2A21 is the σ0 value with changes in surface moisture over Amazonia extending from
a rain flag over the range of incidence angles that may be pro- the early part of the year and continuing through the end of
cessed to eliminate data when it rains and to develop surface 2010. The overall behavior of the TRMM-PR anomaly closely
backscatter similar to QSCAT data. The TRMM-PR ob- follows the rainfall anomaly shown in Fig. 2 without the strong
servationsareperformedatoff-nadirangles(−17°to+17°)(17), signalofrainevents,suggestingthatthebackscatterisresponding
making the backscatter measurements over Amazonia strongly tothesurfacemoisture(representingsoilandcanopymoisture)
influenced by the surface soil moisture and scattering from ratherthantheupper-canopymoistureandstructurechange.
Saatchietal.www.pnas.org/cgi/content/short/1204651110 2of14
P(cid:1) (cid:3)(cid:1) (cid:3)
Time-SeriesAnalysis.WeusedtheQSCATbackscattervaluesfrom x −x y −y
i i
January2000toDecember2009(repeatingtheNovembervalue r¼rffiffiffiffiiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi;
for the missing December measurements) to create a complete P(cid:1)x −x(cid:3)2P(cid:1)y −y(cid:3)2
10-y (120-mo) time series for the analysis. We applied an i i
i i
autoregressive moving-average (ARMA) model to analyze the
QSCAT time-series anomaly. ARMA is based on the autore-
whererisknownastheSpearmanrhobetweenrankedvariables
gressive(AR)modelthatrepresentsthecurrentvalueofaprocess
byafunction of previous values, plussomenoise: xi and yi at a pixel location, with x and y being the average of
variables over the time interval. In this case, xi and yi are the
XP QSCATandTRMManomaliesovertheperiod2000–2009.The
X ¼kþ ϕX þe; correlationcoefficients thatare significantlydifferent fromzero
t i t−i t
i¼1 are then shown for each pixel. We performed the correlation
withtimelagsextendingupto18mo(15%ofthetotalpointsin
where k is a constant, et is a white noise with zero mean and the timeseries) inbothdirections.
constant variance, and ϕ1,......ϕp are the parameters of the OversouthwesternAmazonia,thepost-2005backscattervalues
model.TheorderofAR(p)processisdefinedbythetimelagof and monthly anomalies stayed lower relative to the pre-2005
the time-series process p using an autocorrelation function values, as shown in Fig. 2. We calculated the average QSCAT
(ACF).Oncetheorderofthemodelisdefined,theparameters monthlyanomaliesforallpixelswithinaTRMMpixel(0.25°)and
are estimated using a least-squares regression approach. The computed the difference between QSCAT anomaly (less than
time series would follow a moving-average AR process if we −1.0 σ) and TRMM WDA. The negative difference in normal-
found that the present observation depended less on the pre- ized anomaly suggests a lag in recovery of QSCAT anomaly
vious observation and more on how the previous value differed relativetoTRMMWDA.Wemappedthespatialdistributionof
from its average value (24). The order of the AR model is de- changes in pixels that were affected by the 2005 drought using
fined using ACF or partial ACF (PACF) functions, with auto- the slope of the QSCAT – TRMM WDA monthly anomaly at
correlation withlag-p defined as: eachpixelfromSeptember2005toNovember2009.Pixelswith
significantly (P < 0.01) larger negative slopes represent areas
(cid:4) (cid:5)(cid:4) (cid:5)
NP−P withalongerlaginQSCATbackscatterrecovery.Themapwas
X−X X −X
tþp coloredbased onthenegativeslopeofdifferenceoverthetime
ρp¼ t¼1 PN(cid:4)X −X(cid:5)2 ; speerrifeosrmsiendceonthtehe20Q0S5CdArToudgahtta. tTohqeusapnatitfiyalthaenaalvyesirsagaelsaomwpalis-
t
t¼1 tude of backscatter seasonality over the entire QSCAT time
series(2000–2009)toidentifyareaswiththegreatestseasonality
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P incanopywater content (Fig.4B).
withSE: SEρ¼ ð1þ2 Pi¼−11ρ2iÞ=N.
The SE is used to test the significance of the lag time and to Impact of Deforestation and Fire. Time-series analysis of the
approximatetheconfidenceboundsontheestimate.ThePACFis QSCATdataandthepersistentpost-2005negativeanomalymaybe
the same as the ACF when any linear dependence in the time- affectedbytheannualdeforestationsoccurringparticularlyinthe
series observations is removed. We use a combination of pro- southern and southwestern regions of Amazonia. We included
grams developed in Interactive Data Language (IDL) and R threedatasetstoremovetheeffectofnonforestlandcovertypes
(programming language) toperform the analysis. anddeforestationfromtheanalysisoftheQSCATdata.First,we
Weused the BFAST (Breaks in Additive Season and Trend) used the Moderate Resolution Imaging Spectroradiometer
algorithm developed in R (http://bfast.R-Forge.R-project.org/). (MODIS) level 3 global 0.05° Land Cover Type Yearly Climate
Thealgorithmisbasedonanadditivedecompositionmodelthat ModelingGridproduct (MCD12C1, V051)fortheyear 2005to
iterativelyfitsapiecewiselineartrendandseasonalmodeltothe identify theforested areas and mask out nonforested vegetation
data(25): including deforested landscapes classified as pasture and crops
(https://lpdaac.usgs.gov/products/). Forest areas were identified
Xt ¼TtþStþRt; t¼1;...;N; using five main forest classes in the MODIS land cover map:
evergreenneedleleafforest,evergreenbroadleafforest,deciduous
whereXtistheobservationdataattimetandTt,St,andRtare, needleleafforest,deciduousbroadleafforest,andmixedforests.
respectively, the trend, seasonal, and remainder variation com- Second,weincludedtheMODISVegetationContinuousFields
ponentsofthedata.ThealgorithmassumesthatTtispiecewise (VCF)product(http://glcf.umiacs.umd.edu/data/vcf/)producedat
linear with potential breakpoints that can be determined by fit- 500-mresolution annually (2001–2005). We used the 2005 VCF
tingthelinearmodelsiterativelytodifferentsectionsofthedata product, aggregated it to the QSCAT resolution of 0.05°, and
in a moving window. The abrupt changes are detected by mini- developedaforestcovermaskbyeliminatingallpixels<50%of
mizingtheresidualsumofsquares,andtheiroptimalpositionsin treecover.WeappliedthemasksgeneratedfromtheMODISland
time series can be determined based on the Bayesian in- cover data and VCF to exclude all nonforested pixels from the
formationcriterion (25). QSCATbackscatteranomalyanalysis.
WealsousedtheEuropeanSpaceAgency’s(ESA)GlobCover
SpatialAnalysis.Weused the QSCAT monthly anomalies repre- 2009landcoverdataat300-mresolutionproducedbyESA2010
senting the changes in canopy properties and TRMM monthly and Université Catholique de Louvain. The map was derived
waterdeficitanomaly(WDA)representingthedroughtintensity from Envisat’s Medium Resolution Imaging Spectrometer
toexaminehowthetwoarerelatedandchangeacrosslandscape (MERIS) data and was used to mask all potential pixels repre-
andtime. Toexamine therelationship betweenthe two, wede- sentingthenonforesttypesbeforetheendof2009(www.esa.int/
veloped a spatiotemporal cross-correlation analysis between esaCP/SEM5N3TRJHG_index_1.html). The GlobCover forest
QSCATandTRMManomaliesbyresamplingtheQSCATdata mask was produced at 300-m resolution using all forest cover
to match the TRMM pixel and performing pixel-by-pixel corre- types and later was resampled to 0.05°, and all pure pixels of
lation.WecalculatedthePearsoncorrelationcoefficientbetween forest masks were used as a new mask in the QSCAT anomaly
theranksof variables onapixel-to-pixel basis using analysis. In general, the GlobCover 2009 data might have been
Saatchietal.www.pnas.org/cgi/content/short/1204651110 3of14
the only mask to eliminate the deforested pixel from our anal- computed the accumulated number of fires in 2010 to examine
ysis.However,usingthethreemasksinouranalysisensuresthat what percentage of the pixels affected by the 2005 drought and
all deforestations occurring from 2000 to the end of 2009 are having persistent effects on canopy structure and water content
completely eliminated from the anomaly analysis and any ob- wassusceptibletofuturefires.
servedvariationsinQSCATdataaredirectlyrelatedtochanges
incanopyproperties in themore intact forests. Climate Analysis. We extended the analysis of rainfall over the
The effect of fire has been quantified using the active fire southwesternandentireAmazonbasinbyincludingthelong-term
product derived from monthly accumulated data, from January monthlydatafromtheClimaticResearchUnit(CRU;www.cru.
2001toDecember2010,at1-kmspatialresolution,fromMODIS– uea.ac.uk/; 1930–2005). We computed the DWDA from CRU
TerrasensorproductMCD14MLcollection5.Thisproductde- andcombineditwithTRMM(1998–2010)toshowthelong-term
tects flaming and smoldering fires approximately 1 km2 in size variationsinprecipitationanomaliesovertheAmazonbasin(Fig.
under cloud-free conditions; for particularly “good” observa- S9). We developed spatial correlation between TRMM (1998–
tionalconditions(near-nadirandreducedsmoke),flamingfiresof 2010) and CRU (1970–2009) monthly anomalies and the two
100 m2 can be detected (26). Underestimation of fire detection climate indices: the Tropical North Atlantic Index (TNA) and
may occur in situations in which the fire has started and ended the Southern Oscillation Index (SOI; downloaded from www.
between the satellite overpasses, in cloudy conditions, or under esrl.noaa.gov/psd/data/timeseries/). These indices represent,
theforestcanopy,orwhenitistoosmallortoocoolforthe1-km2 respectively, the fluctuation in sea surface temperature in the
footprint(26).Weusedonlyhigh-confidencefirepixelsforpro- tropical North Atlantic and the difference between sea level
cessingthe data, allowing a >80% confidence level indetecting pressuresinthePacificOcean(atTahiti,Darwin,andAustralia)
fires. The monthly fire pixel count was then aggregated at 0.1° (27,28).Thecorrelationsareperformedonapixel-to-pixelbasis
spatialresolution.Pre-2005droughtfireeffectsweremaskedby for the entire region using the Spearman rank correlation
summingthemonthlydatafromJanuary2001toJune2005,when method (above). The TRMM WDA over southwestern Ama-
thebreakpointinQSCATdataduetothe2005droughtoccurred. zoniaandTNAindexwereplottedtogethertodemonstratethe
The post-2005 drought effects were masked by summing the clearout-of-phasebehaviorofrainfallpatternswithTNAinthe
number of fires from June 2005 to December 2009. We also region(Fig.S10).
1. ShuttleworthWJ(1989)Micrometerologyoftemperateandtropicalforest.Philos 15.FrisonPL,MouginE(1996)MonitoringglobalvegetationdynamicswithERS-1wind
TransRSocLondonSerB324:299–334. scatterometerdata.IntJRemoteSens17(16):3201–3218.
2. AragãoLEOC,etal.(2007)SpatialpatternsandfireresponseofrecentAmazonian 16.SatakeM,HanadoH(2004)DiurnalchangeofAmazonrainforestσ0observedbyKu-
droughts.GeophysResLett34:L07701–GL028946. bandspaceborneradar.IEEETransGeosciRemSens42(6):1127–1134.
3. Dracup JA, Lee KS, Paulson EG, Jr. (1980) On the definition of droughts. Water 17.KawanishiT,etal.(2000)TRMMprecipitationradar.AdvSpaceRes25:969–972.
ResourcesResearch16(20):297–302. 18.MoranMS,VidalA,TroufleauD,InoueY,MitchellTA(1998)Ku-andC-bandSARfor
4. AsnerGP,AlencarA(2010)DroughtimpactsontheAmazonforest:Theremote discriminating agricultural crop and soil conditions. IEEE Trans Geosci Rem Sens
sensingperspective.NewPhytol187(3):569–578. 36:265–272.
5. FrolkingS,etal.(2010)TropicalforestbackscatteranomalyevidentinSeaWinds 19.AttemaEPW,UlabyFT(1978)Vegetationmodeledasawatercloud.RadioSci13(2):
scatterometermorningoverpassdataduring2005droughtinAmazonia.Remote 357–364.
SensEnviron115:897–907. 20.SaatchiS,MoghaddamM(2000)Estimationofcrownandstemwatercontentand
6. UlabyFT,MooreRK,FungAK(1986)MicrowaveRemoteSensing—ActiveandPassive biomassofborealforestusingpolarimetricSARimagery.IEEETransGeosciRemSens
(ArtechHouse,Norwood,MA),VolIII. 38:697–709.
7. SlikJW(2004)ElNiñodroughtsandtheireffectsontreespeciescompositionand 21.SchneebeliM,etal.(2011)RelatingX-bandopacityofatropicaltreecanopyto
diversityintropicalrainforests.Oecologia141(1):114–120. sapflow, rain interception, and dew formation. Remote Sens Environ 115(8):
8. Frolking S, et al. (2006) Evaluation of the SeaWinds scatterometer for regional 2116–2125.
monitoringofvegetationphenology.JGeophysRes111:D17302. 22.ClarkDB,OlivasPC,OberbauerSF,ClarkDA,RyanMG(2008)Firstdirectlandscape-
9. LongDG,DrinkwaterM,HoltB,SaatchiS,BertoiaC(2001)Globaliceandlandclimate scalemeasurementoftropicalrainforestLeafAreaIndex,akeydriverofglobal
studiesusingscatterometerimagedata.EosTransAGU82(43):503. primaryproductivity.EcolLett11(2):163–172.
10.RicciardulliL,WentzF(2011)ReprocessedQuikSCAT(V04)WindVectorswithKu-2011 23.CanhamCD,etal.(1990)Lightregimesbeneathclosedcanopiesintemperateand
Geophysical Model Function (Remote Sensing Systems, Santa Rosa, CA), Remote tropicalforests.CanJForRes20:620–631.
SensingSystemsTechRep043011. 24.HamiltonJ(1994)TimeSeriesAnalysis(PrincetonUnivPress,Princeton,NJ).
11.Nghiem SV (2012) Microwave remote sensing of soil moisture—science and 25.Pires JM, Prance GT (1985) The vegetation types of the Brazilian Amazon. Key
applications.RemoteSensingofDrought:InnovativeMonitoringApproaches,eds Environments: Amazonia, eds Prance GT, Lovejoy TE (Pergamon, New York),
WardlowBD,AndersonMA,VerdinJP(CRC,BocaRaton,FL). pp109–145.
12.BartschA(2012)TenyearsofseawindsonQuikSCATforsnowapplications.Remote 26.GiglioL(2010)MODISCollection5ActiveFireProductUser’sGuide,Version2.4(Science
Sens2:1142–1156. SystemsandApplications,Inc.,UniversityofMaryland,DepartmentofGeography).
13.KimballJS,McDonaldKC,KeyserAR,FrolkingS,RunningSW(2001)Applicationof 27.EnfieldDB,etal.(2001)TheAtlanticMultidecadalOscillationanditsrelationto
theNASAscatterometer(NSCAT)fordeterminingthedailyfrozenandnonfrozen rainfallandriverflowsinthecontinentalU.S.GeophysResLett28:2077–2080.
landscapeofAlaska.RemoteSensEnviron75:113–126. 28.MalhiY,WrightJ(2004)Spatialpatternsandrecenttrendsintheclimateoftropical
14.WagnerW,ScipalK(2000)VariabilityinERSscatterometermeasurementsoverland. rainforestregions.PhilosTransRSocLondBBiolSci359(1443):311–329.
IEEETransGeosciRemSens38(4):1777–1782.
Saatchietal.www.pnas.org/cgi/content/short/1204651110 4of14