On the vertical distribution of smoke in the Amazonian atmosphere during the dry season

20 Pages·2016·7.97 MB·English

Atmos.Chem.Phys.,16,2155–2174,2016
www.atmos-chem-phys.net/16/2155/2016/
doi:10.5194/acp-16-2155-2016
©Author(s)2016.CCAttribution3.0License.
On the vertical distribution of smoke in the Amazonian atmosphere
during the dry season
FrancoMarenco1,BenJohnson2,JustinM.Langridge3,JaneMulcahy4,AngelaBenedetti5,SamuelRemy6,
LukeJones5,KateSzpek3,JimHaywood2,7,KarlaLongo8,andPauloArtaxo9
1SatelliteApplications,MetOffice,Exeter,UK
2EarthSystemandMitigationScience,MetOfficeHadleyCentre,Exeter,UK
3ObservationalBasedResearch,MetOffice,Exeter,UK
4EarthSystemCoreDevelopmentGroup,MetOfficeHadleyCentre,Exeter,UK
5EuropeanCentreforMedium-rangeWeatherForecasts,Reading,UK
6LaboratoiredeMétéorologieDynamique,UPMC/CNRS,Paris,France
7CollegeofEngineering,MathsandPhysicalSciences,UniversityofExeter,Exeter,UK
8InstitutoNacionaldePesquisasEspaciais,SãoJosédosCampos,Brazil
9InstituteofPhysics,UniversityofSãoPaulo,SãoPaulo,Brazil
Correspondenceto:FrancoMarenco(franco.marenco@metoffice.gov.uk)
Received:23October2015–PublishedinAtmos.Chem.Phys.Discuss.:12November2015
Revised:28January2016–Accepted:7February2016–Published:25February2016
Abstract. Lidar observations of smoke aerosols have been derestimatedornotcapturedinthemodelpredictions.Smoke
analysedfromsixflightsoftheFacilityforAirborneAtmo- injection heights derived from the Global Fire Assimilation
sphericMeasurementsBAe-146researchaircraftoverBrazil System (GFAS) for the region are compatible with the gen-
during the biomass burning season (September 2012). A eralheightoftheaerosollayers.
large aerosol optical depth (AOD) was observed, typically
ranging 0.4–0.9, along with a typical aerosol extinction co-
efficient of 100–400Mm−1. The data highlight the persis-
tent and widespread nature of the Amazonian haze, which 1 Introduction
hadaconsistentverticalstructure,observedoveralargedis-
tance(∼2200km)duringaperiodof14days.Aerosolswere Biomass burning is the second largest source of anthro-
foundnearthesurface;butthelargeraerosolloadwastypi- pogenic aerosols globally (Stocker et al., 2013), and South
callyfoundinelevatedlayersthatextendedfrom1–1.5to4– America features as one of the major source regions. In
6km.Themeasurementshavebeencomparedtomodelpre- Southern Amazonia, fire is often used for deforestation and
dictionswiththeMetOfficeUnifiedModel(MetUM)andthe for the preparation of agricultural fields and pasture (Brito
ECMWF-MACC model. The MetUM generally reproduced etal.,2014).ThedryseasonspansfromJulytoOctoberevery
the vertical structure of the Amazonian haze observed with year,andcontrolsthetimingoftheintensiveburningofthe
the lidar. The ECMWF-MACC model was also able to re- vegetation.Intenseprecipitationcanstilloccurinthisseason,
produce the general features of smoke plumes albeit with a due to the increase of convective available potential energy
smalloverestimationoftheAOD.Themodelsdidnotalways (CAPE) and moisture, associated with the Monsoon circu-
capturelocalisedfeaturessuchas(i)smokeplumesoriginat- lation(Gonçalvesetal.,2015).Therateofbiomassburning
ing from individual fires, and (ii) aerosols in the vicinity of intheBrazilianrainforestvariesfromyeartoyearandisaf-
clouds. In both these circumstances, peak extinction coeffi- fectedbymeteorologicalconditionsaswellassocialfactors
cients of the order of 1000–1500Mm−1 and AODs as large (Kaufmanetal.,1998).
as1–1.8wereencountered,butthesefeatureswereeitherun- The high loadings of biomass burning aerosols, with dif-
ferentdegreesofageing,canaffecttheregionalweatherand
PublishedbyCopernicusPublicationsonbehalfoftheEuropeanGeosciencesUnion.
2156 F.Marencoetal.:SmokeintheAmazonianatmosphere
climate (Sena et al., 2013; Rizzo et al., 2013). Episodes signofthedirectradiativeforcing.Ofparticularinterestare
of poor air quality and low visibility are frequent, and the the distribution of lofted layers (Mattis et al., 2003; Müller
aerosolloadingsaffecttheradiationbudgetandthecloudmi- etal.,2005;Baarsetal.,2012)andtheidentificationofcom-
crophysics (Kaufman et al., 1998). Moreover, the radiative plexscenesinvolvingbothaerosolsandclouds(Chandetal.,
balanceoftheregionisalsoaffectedbychangesinthesur- 2008).
face albedo caused by burning of the vegetation. The latter The South AMerican Biomass Burning Analysis
has an impact well beyond the burning season, as it affects (SAMBBA) campaign was an intensive field project
the regional surface energy budget all year round, and has (September–October2012),aimedatcollectinginformation
an impact on convection, cloud formation and precipitation on the atmosphere of the Amazon basin during the dry
(Senaetal.,2013). seasonandthetransitionintothewetseason(Angelo,2012).
The modified ratio of direct to diffuse radiation, and the Oneimportantfocushasbeentheimpactofbiomassburning
changesinmeteorology,inturnwillaffectthephotosynthet- aerosol on the radiation budget, and its feedback on the
ically active radiation flux and the carbon cycle (Mercado dynamicsandhydrologicalcycle,includingtheinfluenceon
et al., 2009). Given that Southern Amazonia is the Earth’s numerical weather predictions, climate, and air quality. The
largest hydrological basin, the largest carbon sink, and the partnershipinvolvedmainlyscientistsfromBrazil(National
largest tropical rainforest, the changes in the regional at- Institute for Space Research, INPE, and University of Sao
mosphereandbiosphereintroducedbybiomassburningcan Paulo) and from the United Kingdom (the Met Office and
havearelevantimpactattheglobalscale.Adetailedreview theNaturalEnvironmentResearchCouncil).
oftheliteratureonbiomassburningemissionscanbefound
inKoppmannetal.(2005)andReidetal.(2005b,a).
Thelargeamountofheatreleasedbyforestfirescangen- 2 Researchflights
erate strong updrafts and deep convection in their vicinity,
rapidly transporting aerosols to upper layers (Freitas et al., During SAMBBA, the Facility for Airborne Atmospheric
2007;Labonneetal.,2007;Sofievetal.,2012),followedby Measurements(FAAM)researchaircraftwasbasedinPorto
long-rangetransport(Kaufmanetal.,1998).Aerosolscanbe Velho, Brazil (8◦44(cid:48)S 63◦54(cid:48)W), and 20 research flights
transported for thousands of kilometres, and as they travel werecarriedoutbetween14Septemberand3October2012,
they are modified through ageing processes (Hobbs et al., totalling 65h of atmospheric research flying. Porto Velho
1997;Kaufmanetal.,1998;Fiebigetal.,2003;Vakkarietal., lies in the state of Rhondonia, where biomass burning for
2014).Thecompositionofbiomassburningaerosolsisdom- deforestation and agriculture is prevalent, and a large de-
inated by fine carbonaceous particles (organics and black forested area is evident. The flights sampled a wide range
carbon; see Brito et al., 2014), and in the first 2hours after of conditions, from very low concentrations of gas phase
emissionaerosolscatteringcanincreaseuptoafactorofsix and aerosol species over the pristine Amazonia rainforest,
duetophotochemistryandsecondaryparticleformation;this throughtomajorfireplumesemittingverylargeamountsof
isparticularlythecaseforsmoulderingfires(Vakkarietal., pollutants.Someoftheflightswerecoordinatedwithsatellite
2014).ParticlehygroscopicityandtheconcentrationofCCN overpasses,whichallowedcombiningaircraftmeasurements
arealsoenhancedduringageing(Abeletal.,2003). with spaceborne remote sensing (see, e.g., Marenco et al.,
Furtherdownwind,theseaerosolscontinuetoexertanim- 2014).
pact on cloud formation, convection, and precipitation pat- The aircraft was equipped with several probes, able to
terns (Andreae et al., 2004; Koren et al., 2008). Gonçalves sampletheatmosphereusingbothinsituandremote-sensing
et al. (2015) indicates two opposite mechanisms by which techniques. Each research flight was planned around one
biomass burning aerosols affect clouds and precipitation: of the following goals: (a) in situ characterisation of fresh
(i) in a stable atmosphere, for a given liquid water content plumes(FP),achievedbyflyingatlowlevelintheimmediate
the formation of a larger number of smaller cloud droplet vicinity of a fire and sampling the aerosols, trace gases and
induces warm rain suppression; and (ii) in an unstable at- thermodynamicstructure;(b)radiativeclosure(RC)studies,
mosphere,theaerosolsenhanceprecipitationandfavourthe achieved with a series of stacked aircraft runs and profiles
formationoflargerandlong-livedcells(convectivecloudin- abovealimitedarea,inordertotietogethertheinformation
vigoration).Moreover,Seifertetal.(2015)haveobservedan derivedbyremotesensingandtheinsituprobes;and(c)sur-
increasediceformationefficiencyforcloudsinthedrysea- vey flights (SF) at high altitude, where the properties of the
son,andacoincidencewiththeseasonalaerosolcycle. atmosphere are mainly sampled with remote-sensing tech-
KnowledgeoftheverticalstructureoftheSouthernAma- niques.BesidesPortoVelho(PV),theairportsinRioBranco
zoniasmokelayeriskeytounderstandingandassessingthe (450kmWSWofPV),Manaus(760kmNEofPV)andPal-
aerosol–cloud interactions (Baars et al., 2012). Textor et al. mas(1700kmEofPV)werealsoused.
(2006) showed that there are significant uncertainties in the The circulation in this season is typically dominated by
vertical distribution in global models, whereas this infor- moderate to strong Easterlies (trade winds), which build up
mation is critical in assessing the magnitude and even the large aerosol burdens over Western Amazonia, where the
Atmos.Chem.Phys.,16,2155–2174,2016 www.atmos-chem-phys.net/16/2155/2016/
F.Marencoetal.:SmokeintheAmazonianatmosphere 2157
low-mid tropospheric circulation is halted by the Andes. In
this season, the north-western part of the basin is charac-
terised by the development of deep convective events ac-
companied by brief but intense precipitation, whereas the
Southern and Eastern parts are typically dry. The season in
2012, however, differed somewhat from the climatology. A
Northwesterly circulation on the Southwestern part of the
basindispersedtheaerosols,andasaresultonlyamoderate
aerosol optical depth (AOD) was observed. Moreover, con-
vectiveprecipitationspreadfurtherEastthanusualduringthe
secondhalfofSeptember.
Nevertheless, burning activity continued through the ma-
jority of the campaign period, and significant aerosol load-
ing was found during most of the flights. In the majority of
cases,avarietyofmeasurementsconfirmedthattheaerosols
canbeascribedtosmokeoriginatedfromforestfires.Agen-
Figure1.GroundtracksforthesixresearchflightslistedinTable1.
eralfeaturethroughoutthecampaignwasthepersistenceof The location of the valid lidar profiles used for the computation
aerosols above the boundary layer, with plumes up to alti- of aerosol extinction is highlighted as follows: grey, profiles that
tudesof4–6km,presumablycausedbydeepconvectionand passourqualitycontroltest;blue,remainingprofiles.Thefollow-
lifting. In situ observations with wing-mounted optical par- inglocationsarealsoshown:PortoVelho(PV),RioBranco(RB),
ticle counters (PCASP and CDP; see, e.g., Liu et al., 1992; AltaFloresta(AF),Palmas(Pal),Manaus(Man),Cuiabà(Cui),and
Lanceetal.,2010;Rosenbergetal.,2012;Ryderetal.,2013) Brasilia(Bra).
showed a predominance of fine mode particles at all levels
(elevated and near-surface). Moreover, measurements with
theon-boardAL5002VUVFastFluorescenceCOAnalyser (2014)wereappliedtodeterminetheirtopheightat2sreso-
(Gerbigetal.,1996,1999;Palmeretal.,2013)showedhigh lution.
carbonmonoxideconcentrations. In order to determine the aerosol properties, further inte-
Thepresentstudyfocusesontheresultsfromtheairborne grationandverticalsmoothinghavebeenappliedduringdata
lidarduringthehigh-altitudeportionsofsixselectedflights, processing, to reduce shot noise: the aerosol data presented
between16and29September(seeTable1andFig.1).The herethereforehaveaverticalresolutionof45mandaninte-
criterionforselectingtheflightshasbeentheavailabilityof grationtimeof1min.Thisintegrationtimecorrespondstoa
sufficiently extended high-altitude and cloud-free sections, 9±2kmfootprint,attypicalaircraftspeeds.
so that the aerosol extinction vertical profile could be esti- Afirstdataselectionwasdoneasfollows:alllidarsignals
matedfromthelidar.Thesixflightsspantheregionbetween acquiredwhentheaircraftwasflyingatanaltitudelowerthan
8.5–12◦Nand46–68◦W,coveringadistanceof∼2200km 4kmhavebeenomitted,anddatahavebeendiscardedifthe
extendingalonganEast–WestaxisacrosstheBrazilianAma- lidar was pointing at more than 10◦ off the vertical (due to
zonbasin,atanapproximatemeanlatitudeof10◦S. the aircraft turning). Lidar signals within 300m of the air-
crafthavebeendiscarded,duetoincompleteoverlapbetween
theemitterandthereceiverfield-of-view,andatthefarend
3 Measurements profileshavebeentruncatedtoremovethesurfacespikeand
anydatabeyondit.Asageneralrule,averticalprofilewhere
Observations of atmospheric aerosols have been acquired acloudwasdetectedhaseitherbeenomittedcompletely,or
withtheALS-450elasticbackscatteringlidarmountedonthe hasbeenomittedintheportionbetweenthesurfaceandthe
FAAMresearchaircraft.Thisisaninstrumentmanufactured cloud top. However, in a small number of cases where the
byLeosphere;itisoperatedatawavelengthof355nm;and cloud optical depth has been considered sufficiently small,
it is mounted in a nadir-looking geometry (Marenco et al., so as to not affect the derivation of aerosol properties, data
2011).ThesystemspecificationsaresummarisedinMarenco below a cloud have been kept but the cloud layer itself has
etal.(2014)andamoredetaileddescriptionoftheinstrument beenrejected.
canbefoundinLollietal.(2011)andChazetteetal.(2012). All assumptions have been reviewed manually, on a
Lidarsignalswereacquiredwithanintegrationtimeof2s profile-by-profile basis, with the possibility to override the
andaverticalresolutionof1.5m.Theanalogueandphoton cloud exclusion criteria and to set the reference height in-
counting signals are merged at pre-processing by normali- tervalnecessaryforthederivationofaerosolextinction.Af-
sation in an overlap area chosen based on photon-counting ter the data selection discussed above, 334 vertical profiles
thresholds.Cloudsignalintheverticalprofileswasdetected have been retained. Processing of the data followed a dou-
as a “large spike”, and the thresholds given in Allen et al. bleiteration,firsttodeterminethelidarratio(extinction-to-
www.atmos-chem-phys.net/16/2155/2016/ Atmos.Chem.Phys.,16,2155–2174,2016
2158 F.Marencoetal.:SmokeintheAmazonianatmosphere
Table1.Researchflightsconsideredinthisarticle.TimeisUTC.
∗
Flight Date Takeoff Landing Latitude Longitude Type
◦ ◦
B733 16Sep RioBranco,13:51 PortoVelho,14:45 8.9–9.8 S 64.5–67.6 W briefSF
◦ ◦
B734 18Sep PortoVelho,12:05 PortoVelho,16:01 8.9–11.9 S 61.6–64.4 W RC
◦ ◦
B741 26Sep PortoVelho,12:53 Palmas,16:08 8.8–10.2 S 48.7–63.9 W SF
◦ ◦
B742 27Sep Palmas,12:52 Palmas,16:17 10.2–11.5 S 46.8–48.1 W FP
◦ ◦
B743 27Sep Palmas,18:08 PortoVelho,21:34 9.0–10.2 S 48.4–63.6 W SF
B746 29Sep PortoVelho,12:54 PortoVelho,16:38 8.7–9.4◦S 58.2–63.7◦W FP+SF
∗Flighttype:FP:freshplumesampling,mainlyinsitu;RC:radiativeclosure,combininginsituandremotesensing;SF:high-altitude
surveyflight.
backscatter ratio), and subsequently to process the data set
todeterminetheextinctioncoefficient.Themethoddetailed
inMarenco(2013)isatthebasisofthisprocessing,anditis
basedonsettingthereferencewithintheaerosollayer,rather
thanonaRayleigh-scatteringportionoftheatmosphere.The
slopemethodisusedforafirstestimateoftheextinctionco-
efficient at the reference, based on the lidar measurements
themselves.Asillustratedinthatpaper,inthisgeometryand
atthiswavelengththeforwardsolutiontothelidarequation
is unstable and cannot be used when the aerosol layers are
deep and their extinction is large; hence the need for using
thismethod.Marencoetal.(2014)illustratestheapplication
ofthismethodincomparisonwithCALIPSOretrievalsand
toconstrainedretrievalsusingAERONET.
3.1 Lidarratio
Aninitialsubsetofthelidarprofileshasbeenselected,where
the signature of Rayleigh scatteringhas beenclearly identi-
fied above the aerosol layers. This circumstance permits it-
eration using the method described in Marenco (2013) by
varying the lidar ratio (assumed constant with height), until
a good match to the overlying Rayleigh scattering layer is
reached: in this way, the lidar ratio itself can be estimated.
Out of these lidar profiles, 270 indicate at least a moderate
aerosolload(AOD>0.25),andtheyhavebeenkepttocom-
puteadistribution:resultsaredisplayedinFig.2b.Thedata
setfollowsaGaussiandistribution,andischaracterisedbya
meanandstandarddeviationof73.1and6.3sr,respectively.
Moreover, Fig. 2a shows that the distribution is not signifi-
cantly affected by how we choose the acceptance threshold Figure2.(a)LidarratioandAOD,determinedforeachlidarpro-
(AOD>0.25).Thelidarratiodeterminedinthiswayisnot file(seetext).Thedatapointsarecolour-codedwiththeflightnum-
substantially affected by the choice of the lower reference ber.Thehorizontalbluesolidlineindicatesthemean,thebluedot-
extinction,andisinsteadmainlyaffectedbythehigherlay- ted lines indicate one standard deviation from the mean, and the
ers, where the transition between a large extinction coeffi- dashed red line indicates the median. The vertical dashed line in-
cientandamolecularlayerisencountered.Thisestimateof dicates the threshold (AOD>0.25) that has been applied to the
dataset.(b)Histogramoflidarratiodeterminations,for270profiles
the lidar ratio for biomass burning aerosols is in agreement
withAOD>0.25.Mean:73.1sr,standarddeviation:6.3sr,median:
withthefindingsreportedinOmaretal.(2009),Baarsetal.
72.5sr.Agaussiancurvewiththesamemeanandstandarddevia-
(2012),Großetal.(2012),andLopesetal.(2013).
tionisoverplotted(dashedline).
Thelidarratiosodetermined,73±6sr,hasbeencompared
toMiescatteringcomputations.Figure3adisplaysthecam-
paignmeanparticlesize-distribution(PSD)determinedwith
Atmos.Chem.Phys.,16,2155–2174,2016 www.atmos-chem-phys.net/16/2155/2016/
F.Marencoetal.:SmokeintheAmazonianatmosphere 2159
indices;seeFig.3b.Theresultinglidarratioishighlydepen-
dent on the real and imaginary parts; refractive index esti-
mates from the literature are also shown in the figure. The
lidarderivationof73±6sriscompatible,forinstance,with
refractiveindicesfromRizzoetal.(2013)andDuboviketal.
(2002).Notethattheestimatescomputedwithrefractivein-
dices from Reid and Hobbs (1998) and Guyon et al. (2003)
alsodonotfalltoofaroff.
3.2 Estimateoftheaerosolextinctioncoefficient
Followingtheresultofthefirstiterationonthelidardata,a
lidarratioof73srhasbeenadoptedforthefulldataset,and
a second iteration with the method introduced in Marenco
(2013) has been applied to determine the aerosol extinc-
tion coefficient for all the 334 profiles. This method (slope-
Fernald method) is a variant of the Fernald–Klett method
(Fernald, 1984; Klett, 1985), where the reference is taken
withinanaerosollayer:thispermitsusingthestable(inward)
solution to the lidar equation in the unfavourable geometry
represented by a nadir-looking lidar. Note that this choice
is necessary if, as found during this campaign, no aerosol-
free region below the aerosol layers is available. Figure 4
shows typical resulting estimates of the aerosol extinction
coefficient, for a subset of the vertical profiles (this selec-
tion is purely illustrative in purpose). For each profile, an
estimateoftheuncertaintythatresultsfromtheretrievalas-
sumptions has been computed, by repeating the derivation
afterhavingvariedthelidarratioby±6sr(thisbeingtheun-
certaintyadoptedabove),andafterhavingvariedtheextinc-
tion value at the reference by ±50% (1σ statistical errors).
Thelattervaluereflectsthelargeuncertaintythatarisesfrom
Figure3.(a)SAMBBAcampaignmeanparticlesize-distribution,
theMarenco(2013)method,sincereferenceistakenwithin
determinedwiththewing-mountedPCASPopticalparticlecounter
anaerosollayerinsteadofinRayleighscatteringconditions.
(black dots). The fit with a bimodal lognormal is also shown; the
Note however, how quickly the uncertainty decreases when
parameters of the two lognormals are as follows: accumulation
mode: Dp=0.184µm, σ =1.47, and Nt =868.5. Coarse mode: moving upwards from the reference height; the opposite is
Dp=0.387µm,σ =2.13,andNt =2.16.(b)Contoursofthelidar unfortunately also true, i.e. where the reference height is
ratiocomputedforthecampaignmeanparticlesize-distribution,by takenatanaltitude,thenuncertaintyincreasesupto±100%
varyingtherefractiveindex.Theblacksolidanddottedlinesindi- nearthesurface.Insummary,verylargeuncertaintiesexistin
catethemeanandstandarddeviationofthelidarratiodetermined the bottom part of the vertical profiles, but they are quickly
bylidar(73±6sr).ThereddotsshowestimatesoftheAmazonian damped when moving towards the higher layers. At the top
biomassburningaerosolrefractiveindextakenfromtheliterature: oftheprofiles,uncertaintyisinsteaddrivenbythelidarratio,
(a)1.5−0.02i (ReidandHobbs,1998);(b)1.42−0.006i (Guyon
andisgenerallysmall.
etal.,2003);(c)(1.47±0.07)−(0.008±0.005)i(Rizzoetal.,2013);
(d)(1.47±0.03)−(0.0093±0.003)i(Duboviketal.,2002).
4 Observedaerosoldistribution
the PCASP, and its fit using two lognormals, each of which
isintheform Figures5and6displaythecross-sectionsofaerosolextinc-
tion coefficient and of its estimated uncertainty, as a func-
N e−12(cid:16)lnDln/σDp(cid:17)2 tion of along-track distance and height. Generally, all six
n(D)= √ t , (1) flights show a similar structure, with a moderate magnitude
2πlnσ D of aerosol extinction, of the order of 150–200Mm−1, be-
whereD isdiameter,andD ,σ,andN arethreefittedpa- tweenthesurfaceandanupperaltitudeof4–6km,withsome
p t
rameters(Johnsonetal.,2016).Wehavecomputedthelidar localised patches showing higher magnitudes. This general
ratio for this size-distribution and for a range of refractive vertical structure was broadly coherent over distances of
www.atmos-chem-phys.net/16/2155/2016/ Atmos.Chem.Phys.,16,2155–2174,2016
2160 F.Marencoetal.:SmokeintheAmazonianatmosphere
8 14:12 16 Sep B733 8 12:44 18 Sep B734 8 14:50 26 Sep B741
9(cid:176)22’S 66(cid:176)09’W 11(cid:176)19’S 63(cid:176)29’W 9(cid:176)59’S 54(cid:176)32’W
6 6 6
m)
k
e (
d 4 4 4
u
Altit
2 2 2
0 0 0
0 200 400 600 0 200 400 600 0 200 400 600
8 13:12 27 Sep B742 8 20:34 27 Sep B743 8 13:48 29 Sep B746
11(cid:176)11’S 47(cid:176)11’W 9(cid:176)40’S 59(cid:176)33’W 9(cid:176)06’S 59(cid:176)57’W
6 6 6
m)
k
e (
d 4 4 4
u
Altit
2 2 2
0 0 0
0 200 400 600 0 200 400 600 0 200 400 600
Aerosol extinction (Mm-1) Aerosol extinction (Mm-1) Aerosol extinction (Mm-1)
Figure 4. A sample of the lidar vertical profiles of aerosol extinction coefficient, discussed in this paper. The green lines indicate the
estimateofuncertainty.Theredlinesindicatethereferenceheightintervalused(differentforeachprofile).Thetime,date,flightnumberand
coordinatesareindicatedinthetitletoeachplot.Eachprofilecorrespondstoanintegrationtimeof1min.
thousands of kilometres and persisted over the 2-week pe- alsohaveoriginatedfromsomeothernearbyfire;andfinally
riodstudiedhere. itmayhavebeentransportedoveralongerdistance.
At smaller spatial scales, some noticeable features were Moreover, in flights B741 (first part) and B746 the pres-
observed, and are described as follows. Flight B742 shows enceofcloudswithtopsat2–4kmobscuresthebottompart
fourfeatureswherealargeextinctioncoefficient(approach- of the aerosol layer; above these clouds, large extinction
ing 1000Mm−1) is detected at an altitude of 1.25km, at coefficients are detected, peaking 1000–1500Mm−1. These
along-track distances of 115, 135, 310 and 360km. These largevaluesarelikelytobeeitherdirectlycausedbynearby
correspond to plumes from single fires that were seen from fires(hiddenbythecloudsthemselves),orasaresultofcon-
thecockpit.Sincetheaircraftwasflyingbackandforthover vective lifting and detraining of smoke into a layer around
thesamearea,thesesmokeplumeswerealllocatedwithina thecloud-top.
maximum distance of ∼25km from each other, and in fact From the aerosol extinction coefficient described above,
the ones observed at 310 and 360km along-track distance a few quantities have been computed. The layer extinction
wereatthesamelocation. is computed as the vertically averaged extinction, and the
Flight B743 also shows a plume from a single fire, cen- aerosol optical depth (AOD) as the vertically integrated ex-
tred at an along-track distance of 1260km; it extends from tinction. The layer height has been defined, for each verti-
the surface to 2km altitude and has a size of ∼50km in cal profile, as the weighted average√of the aerosol vertical
thealong-trackhorizontaldirection;inthisplumeapeakex- distribution, and the layer depth as 2×(AOD)/(peak ex-
tinction of 1270±40Mm−1 was encountered. Moreover, a tinction).Notethatthedefinitionoflayerdepthcanbequite
higher altitude feature is observed, well above the aerosol arbitrary;however,theabovedefinitionsareconsistentwith
layer,andco-locatedwiththisintenseplumebutapparently Marenco et al. (2011). The layer height, layer depth, layer
disconnected from it: its altitude is 3.7–4km, with a depth extinctionandaerosolopticaldepthhavebeencomputedfor
varyingbetween200and400m(FWHM).Itshorizontalex- eachverticalprofileinthedataset.Note,however,thatthese
tentisof270kmalong-track,itsaerosolopticaldepth(AOD) derivedquantitiescanbeaffectedbytheverticalextentofthe
peaks 0.09, and its extinction coefficient peaks 300Mm−1. available data, which in turn is affected by aircraft altitude,
Theoriginofthishigheraltitudefeatureisuncertain:itcould terrain height, and the presence of low clouds. As a qual-
havebeenreleasedbythesamefireatanearliertime,i.e.if itycontroltest,profilesforwhichtherelativeerroronAOD
thefireradiativepowerhadbeenatanytimestronger;itmay was larger than 50%, and profiles that were truncated (due
Atmos.Chem.Phys.,16,2155–2174,2016 www.atmos-chem-phys.net/16/2155/2016/
F.Marencoetal.:SmokeintheAmazonianatmosphere 2161
Figure5.Cross-sectionsoftheaerosolextinctioncoefficientdeterminedfromthelidarforthesixresearchflightswitha1-minintegration
time. The black lines indicate the aircraft altitude and the surface elevation from a digital elevation model, respectively. The green dots
indicatecloudtopsdetectedwiththelidarat2sresolution.Therednumberedarrowsindicatetheselectedsectionsforthecharacterisation
oftheaerosollayer(seeTable2).
www.atmos-chem-phys.net/16/2155/2016/ Atmos.Chem.Phys.,16,2155–2174,2016
2162 F.Marencoetal.:SmokeintheAmazonianatmosphere
Figure6.Cross-sectionsoftheaerosolextinctioncoefficientuncertainty.
Atmos.Chem.Phys.,16,2155–2174,2016 www.atmos-chem-phys.net/16/2155/2016/
F.Marencoetal.:SmokeintheAmazonianatmosphere 2163
Table2.Flightsectionsconsideredforthecharacterisationoftheaerosollayer,displayedwithredarrowsinFig.5.Foreachsection,the
layerheight,layerdepth,layerextinctionandaerosolopticaldeptharelisted(seetext).Foreachquantity,theaverageandstandarddeviation
areshown;forthelayerextinctionandaerosolopticaldepth,maximumvaluesareshownaswell(inparentheses).Theresultsforthewhole
datasetarelistedaswell.
Section Flight Time Numberof Layerheight Layerdepth Layerextinction Aerosolopticaldepth
number number profiles (km) (km) (Mm−1) (–)
1 B733 13:56–14:27 28 2.04±0.28 2.27±0.23 113±31(224) 0.68±0.18(1.02)
2 B734 12:35–13:04 23 2.39±0.13 3.05±0.47 93±12(117) 0.66±0.09(0.84)
3 B734 15:10–15:26 7 2.53±0.08 2.86±0.37 102±9(117) 0.61±0.06(0.70)
4 B741 13:02–13:56 5 2.63±0.07 1.47±0.35 132±22(173) 0.63±0.17(0.78)
5 B741 14:43–15:53 61 1.85±0.12 2.37±0.38 91±11(117) 0.55±0.07(0.71)
6 B742 13:02–13:17 14 2.26±0.26 2.82±0.50 212±48(311) 0.89±0.22(1.36)
7 B743 18:39–19:35 29 2.05±0.16 2.72±0.53 109±37(227) 0.68±0.25(1.43)
8 B743 20:20–21:18 50 1.69±0.20 2.09±0.41 75±35(195) 0.48±0.24(1.29)
9 B746 13:06–14:11 39 2.07±0.29 2.17±0.51 161±58(366) 0.92±0.25(1.83)
10 B746 15:52–16:23 20 2.43±0.60 1.68±0.50 125±43(228) 0.61±0.21(0.90)
Alldata 276 2.03±0.36 2.34±0.57 112±49(366) 0.65±0.24(1.83)
to cloud) at a lower boundary which was 2.5km or higher the vertical distribution of the aerosols does not vary much,
above mean sea level have not been included in the discus- despitethelargedistancetravelledbytheaircraft(morethan
sionofthederivedquantitiesdescribedabove. 2200km between the eastmost and westmost lidar profiles)
Inordertocharacterisetheaerosollayerintermsofrepre- and the relatively long time between the first and the last
sentativeproperties,thedatasethasbeendividedinthesec- flight(14days).
tions listed in Table 2, numbered 1–10, and also displayed Thequantitativeproperties,i.e.meanextinctionandAOD,
with red arrows in Fig. 5. For each of the shorter flights, a display a larger variability, as expected; however, this vari-
single section has been considered, whereas when the dis- ability is not huge. The per-section average of layer extinc-
tance travelled exceeded 1000km two flight sections have tion varies between 75 and 200Mm−1 and the per-section
beenconsidered.ForflightB742,sincetheaircrafttravelled averageofAODisbetween0.5and0.9,eachofthesequan-
backandforthoverthesamegroundtrackseveraltimes,only titiesshowingastandarddeviationof10–50%ineachflight
the first part of the lidar cross-section has been considered. section.Whencomputedoverallsixflights,theaverageand
Due to the above quality criteria and to the fact that some standarddeviationofthesequantitiesis112±49Mm−1and
flightportionshavenotbeenincluded(e.g.thesecondpartof 0.65±0.24, respectively, and the maximum values encoun-
flightB742),thenumberofretainedprofilesisreducedfrom teredoverthedatasetwereaboutthreetimeslargerthanthe
334to276.Table2summarizestheflightsectionsaverages average. The distribution of the layer properties, derived by
and standard deviations for the considered quantities; note airborne lidar for the six flights considered in this paper, is
that in this context, standard deviation is a measure of vari- showninFig.7.
abilityforeachgivenquantity.Themaximumofthelayerex- The mean vertical distributions of aerosol extinction for
tinctionandoftheaerosolopticaldepthisalsolistedforeach each of the ten sections are shown in Fig. 8. The average
section;themaximumofthelayerextinctionisingeneraldif- overthetensectionsisdisplayedinFig.9,andshowsagen-
ferentandlowerthanthemaximumvalueofextinctionthat eral structure that can be summarised as follows. Near the
isencounteredineachsection(layerextinctionbeingaver- surface, and up to an altitude of ∼1km, a surface layer of
ticallyaveragedquantity). extinction coefficient ∼200Mm−1 is observed. Above this
The geometrical properties of the aerosol vertical distri- layer, an elevated layer is found which has a slightly larger
bution, i.e. the layer height and layer depth, show a lim- extinction coefficient (peaking ∼250Mm−1) and a signif-
ited variability within each section, with standard devia- icant depth, extending from ∼1 to ∼5km altitude. When
tionaround10–15%forlayerheightand15–25%forlayer lookingattheindividualsections(Fig.8),variationsaround
depth.FlightB746representsanexceptionandshowslarger this general structure can be observed: the lower layer in
variability in its second part (Sect. 10); however, for this someoftheflightsextendsabithigher(upto∼1.5km)and
flight a large proportion of profiles are truncated due to can show a magnitude of the aerosol extinction coefficient
low cloud, and therefore the remaining data may possibly of 150–300Mm−1; and the aerosols above can extend, de-
not provide a representative sample. Averaged over all six pendingontheflightsection,uptoanaltitudebetween4and
flights,thelayerheightis2.0±0.4km,andthelayerdepthis 6km.Theelevatedaerosolsshowasasinglewell-definedel-
2.3±0.6(averageandstandarddeviation).Thisindicatesthat evated layer in Sects. 2, 5 and 7 and as a more structured,
www.atmos-chem-phys.net/16/2155/2016/ Atmos.Chem.Phys.,16,2155–2174,2016
2164 F.Marencoetal.:SmokeintheAmazonianatmosphere
60 valid for the forecast base time. The GFAS data are a daily
50 80 productbasedonalltheModerate-ResolutionImagingSpec-
troradiometer (MODIS) overpasses, over the course of any
y 40 60
nc givenday.Assimilationusingthisinventoryisknowntolead
ue 30
q 40 to an underestimation of AOD, due to the lack of detec-
e
Fr 20 tionofsmallsmoulderingfires,andfiresbelowcanopiesand
20
10 clouds. Studies show that for a better agreement it is there-
0 0 forenecessarytoscaleuptheemissions(Kaiseretal.,2012).
0.0 0.5 1.0 1.5 2.0 0 100 200 300 400 WhilsttheECMWF-MACCmodelusedascalefactorof3.4
Aerosol Optical Depth Layer extinction (Mm-1)
(Kaiser et al., 2012), this was reduced to a factor of 1.7 for
100
40 the MetUM based on an initial assessment of AODs with
80
AERONETandMODISdataearlierintheseason.
ncy 60 30 In the MetUM simulations, biomass burning aerosol was
e
qu 20 simulatedon-lineusingtheCLASSICaerosolscheme(Bel-
e 40
Fr louinetal.,2011),whileallotheraerosolspecieswererep-
20 10 resented by climatological averages. Direct aerosol effects
0 0 were included in the simulations, but indirect effects were
1.0 1.5 2.0 2.5 3.0 3.5 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 not. There are a number of uncertainties on some of the as-
Layer height (km) Layer depth (km)
sumptions used in the simulations; in particular associated
Figure7.DistributionsofaircraftlidarobservationsofAOD,layer withthepotentialtransportofaerosolsfromoutsidethedo-
extinction,layerheightandlayerdepth,forthewholedatasetcon- mainboundaries,therainoutofbiomassburningaerosolsand
sideredinthispaper(276verticalprofiles).AGaussiancurvewith theirageing,andthesourceemissions.Aerosolinjectionwas
the mean and standard deviation of the data set is superimposed prescribed between 0.1 and 3km in height; this is likely to
(dashedline). affecttherepresentationoftheverticalextentoftheaerosol
plumes, in particular from larger fires. Moreover, emissions
werenotupdatedduringthemodelsimulation,andtherefore
multi-layer atmosphere in the other sections. The signature
an assumption is made on persistence of the emission field.
of the individual fire plumes described above can be found
TheCLASSICaerosolschemeusesaprescribedaerosolsize
intheseaverageprofiles;seee.g.themaximumatanaltitude
distribution and refractive index, based on Haywood et al.
of ∼1.25km in Sect. 6, and at an altitude of ∼1.6km in
(2003).Aclimatologicalhygroscopicgrowthcurvebasedon
Sect. 8. These layers also show a larger standard deviation,
Magi and Hobbs (2003) is included in the model, and this
reflectingthevariabilitybetweenin-plumeandout-of-plume
information enables the calculation of aerosol optical prop-
conditions.NotealsothatSects.4,9and10areaffectedby
erties,includingextinctioncoefficient.
lowcloudswithlargesmokeconcentrationsabove;thisisre-
The ECMWF-MACC model issued by ECMWF is pro-
flectedinthelargevaluesofthemean+1standarddeviation
vided as part of the EU-funded projects Monitoring Atmo-
(upto600–800Mm−1).
spheric Composition and Climate, MACC, MACC-II and
MACC-III (Morcrette et al., 2009; Benedetti et al., 2009).
5 Modelsimulations The initial package of ECMWF physical parameterisations
dedicatedtoaerosolprocessesmainlyfollowsthetreatment
The lidar data have been used to evaluate aerosol simula- of the Laboratoire de Météorologie Dynamique general cir-
tions from two prediction models: (i) a limited area model culationmodel,LMD-Z(Boucheretal.,2002;Reddyetal.,
(LAM)configurationoftheMetOfficeUnifiedModel(Me- 2005). Five types of tropospheric aerosols are considered
tUM),and(ii)aerosolforecastsissuedbytheEuropeanCen- in the model, and are fully coupled with the meteorology:
treforMedium-rangeWeatherForecasts(ECMWF-MACC). sea salt, dust, organic carbon, black carbon, and sulphate.
TheMetUMlimitedareamodelwassetupfortheSAMBBA Prognostic aerosols of natural origin, mineral dust and sea-
campaignovertheAmazoniadomain(latitude25◦S–18◦N, salt, are described using three size bins, and their emis-
longitude 85–32◦W), and has a resolution of 12km, with sions depend on model parameters (surface winds among
70levelsinthevertical(Kolusuetal.,2015).Lateralbound- others). Anthropogenic emissions are specified using cur-
aryconditionsforthemeteorologicalfieldsweredrivenpro- rent emission inventories, and biomass burning emissions
videdbytheoperationalglobalconfigurationoftheMetUM are taken from the GFAS inventory. The simulations pre-
(GlobalAtmosphere3.1,Waltersetal.,2011).TheECMWF- sented here were carried out using an experimental version
MACCsimulationswereglobal,althoughanalysedhereover of the ECMWF-MACC model, which emits biomass burn-
theAmazonianregiononly.Bothmodelswereinitialisedus- ingaerosolsataninjectionheightprovidedbyaPlumeRise
ingnear-realtimeemissionsfromtheGlobalFireAssimila- Model(PRM)thathasbeenembeddedintoGFAS(Paugam
tion System (GFAS) emission data set (Kaiser et al., 2012), et al., 2015). The PRM derives injection heights from
Atmos.Chem.Phys.,16,2155–2174,2016 www.atmos-chem-phys.net/16/2155/2016/