King s Research Portal Users may download and print one copy of any publication from ... of the...

Kingrsquos Research Portal

DOI105194acp-16-6961-2016

Document VersionPublishers PDF also known as Version of record

Link to publication record in Kings Research Portal

Citation for published version (APA)Pereira G Siqueira R Rosaacuterio N E Longo K L Freitas S R Cardozo F S Wooster M J (2016)Assessment of fire emission inventories during the South American Biomass Burning Analysis (SAMBBA)experiment Atmospheric Chemistry and Physics 16(11) 6961-6975 DOI 105194acp-16-6961-2016

Citing this paperPlease note that where the full-text provided on Kings Research Portal is the Author Accepted Manuscript or Post-Print version this maydiffer from the final Published version If citing it is advised that you check and use the publishers definitive version for paginationvolumeissue and date of publication details And where the final published version is provided on the Research Portal if citing you areagain advised to check the publishers website for any subsequent corrections

General rightsCopyright and moral rights for the publications made accessible in the Research Portal are retained by the authors andor other copyrightowners and it is a condition of accessing publications that users recognize and abide by the legal requirements associated with these rights

bullUsers may download and print one copy of any publication from the Research Portal for the purpose of private study or researchbullYou may not further distribute the material or use it for any profit-making activity or commercial gainbullYou may freely distribute the URL identifying the publication in the Research Portal

Take down policyIf you believe that this document breaches copyright please contact librarypurekclacuk providing details and we will remove access tothe work immediately and investigate your claim

Download date 11 May 2018

Atmos Chem Phys 16 6961ndash6975 2016wwwatmos-chem-physnet1669612016doi105194acp-16-6961-2016copy Author(s) 2016 CC Attribution 30 License

Assessment of fire emission inventories during the South AmericanBiomass Burning Analysis (SAMBBA) experimentGabriel Pereira1 Ricardo Siqueira2 Nilton E Rosaacuterio3 Karla L Longo2a Saulo R Freitas2a Francielle S Cardozo1Johannes W Kaiser4 and Martin J Wooster56

1Department of Geoscience Federal University of Sao Joao del-Rei (UFSJ) Sao Joao del-Rei Brazil2Center for Weather Forecast and Climate Studies National Institute for Space Research (INPE) Cachoeira Paulista Brazil3Environmental Sciences Department Satildeo Paulo Federal University (UNIFESP) Diadema Satildeo Paulo Brazil4Max Planck Institute for Chemistry (MPIC) Mainz Germany5Department of Geography Kingrsquos College London (KCL) London UK6NERC National Centre for Earth Observation (NCEO) Leicester UKanow at Global Modeling and Assimilation Office NASA Goddard Space Flight Center and USRAGESTARGreenbelt MD USA

Correspondence to Gabriel Pereira (pereiraufsjedubr)

Received 26 December 2015 ndash Published in Atmos Chem Phys Discuss 19 January 2016Revised 5 May 2016 ndash Accepted 23 May 2016 ndash Published 7 June 2016

Abstract Fires associated with land use and land coverchanges release large amounts of aerosols and trace gasesinto the atmosphere Although several inventories of biomassburning emissions cover Brazil there are still considerableuncertainties and differences among them While most fireemission inventories utilize the parameters of burned areavegetation fuel load emission factors and other parame-ters to estimate the biomass burned and its associated emis-sions several more recent inventories apply an alternativemethod based on fire radiative power (FRP) observationsto estimate the amount of biomass burned and the corre-sponding emissions of trace gases and aerosols The Brazil-ian Biomass Burning Emission Model (3BEM) and the FireInventory from NCAR (FINN) are examples of the firstwhile the Brazilian Biomass Burning Emission Model withFRP assimilation (3BEM_FRP) and the Global Fire Assim-ilation System (GFAS) are examples of the latter Thesefour biomass burning emission inventories were used duringthe South American Biomass Burning Analysis (SAMBBA)field campaign This paper analyzes and inter-comparedthem focusing on eight regions in Brazil and the time periodof 1 Septemberndash31 October 2012 Aerosol optical thickness(AOT550 nm) derived from measurements made by the Mod-erate Resolution Imaging Spectroradiometer (MODIS) oper-ating on board the Terra and Aqua satellites is also applied to

assess the inventoriesrsquo consistency The daily area-averagedpyrogenic carbon monoxide (CO) emission estimates ex-hibit significant linear correlations (r pgt 005 level Studentt test) between 3BEM and FINN and between 3BEM_ FRPand GFAS with values of 086 and 085 respectively Theseresults indicate that emission estimates in this region derivedvia similar methods tend to agree with one other Howeverthey differ more from the estimates derived via the alterna-tive approach The evaluation of MODIS AOT550 nm indi-cates that model simulation driven by 3BEM and FINN typ-ically underestimate the smoke particle loading in the east-ern region of Amazon forest while 3BEM_FRP estimationsto the area tend to overestimate fire emissions The daily re-gional CO emission fluxes from 3BEM and FINN have linearcorrelation coefficients of 075ndash092 with typically 20ndash30 higher emission fluxes in FINN The daily regional CO emis-sion fluxes from 3BEM_FRP and GFAS show linear corre-lation coefficients between 082 and 090 with a particularlystrong correlation near the arc of deforestation in the Ama-zon rainforest In this region GFAS has a tendency to presenthigher CO emissions than 3BEM_FRP while 3BEM_FRPyields more emissions in the area of soybean expansion eastof the Amazon forest Atmospheric aerosol optical thick-ness is simulated by using the emission inventories withtwo operational atmospheric chemistry transport models the

Published by Copernicus Publications on behalf of the European Geosciences Union

6962 G Pereira et al Assessment of fire emission inventories during the SAMBBA experiment

IFS from Monitoring Atmospheric Composition and Cli-mate (MACC) and the Coupled Aerosol and Tracer Trans-port model to the Brazilian developments on the RegionalAtmospheric Modelling System (CCATT-BRAMS) Evalu-ation against MODIS observations shows a good represen-tation of the general patterns of the AOT550 nm time seriesHowever the aerosol emissions from fires with particularlyhigh biomass consumption still lead to an underestimation ofthe atmospheric aerosol load in both models

1 Introduction

Biomass burning is a global phenomenon and an ancientpractice of human occupation as well as a natural process Itconsumes large amounts of vegetation across wide areas andmodifies Earth surface characteristics Fires still play a keyrole in ecosystem services opening areas for livestock agri-culture and pest control (Shimabukuro et al 2013) In thelast 5 decades biomass burning has been extensively and per-sistently used all over the tropics for these purposes and hasbeen involved in widespread deforestation and forest degra-dation (Crutzen and Andreae 1990 Bustamante et al 2016)Biomass burning emissions inject a considerable amount ofoxidants and aerosols into the atmosphere modifying atmo-spheric composition and reactivity and therefore disturbingthe regional climate water and biogeochemical cycles (An-dreae et al 2004 Bowman et al 2009) Moreover fire emis-sions in tropical areas are not only felt regionally but alsorapidly and strongly affect the global scale due to the effi-ciency of the atmospheric transport processes of trace gasesand aerosols emitted and the high heat release which rein-forced the intense tropical convective activity (Kaufman etal 1995 Val Martin et al 2010) Extensive fire activity alsodisturbs the environmental system producing soil depletiondamaging flora and fauna decreasing biodiversity and evenaffecting human life (Fearnside 2000)

An accurate temporal and spatial estimate of biomassburning emissions is critical to a reliable analysis of theassociated effects Nowadays efforts to quantify emissionsfrom biomass burning from space-borne instruments haveincreased considerably in scope but uncertainties in the in-put data and within the different methodologies can still leadto errors of up to an order of magnitude for trace gases andaerosol emission totals (Vermote et al 2009 Baldassarre etal 2015)

Several products of burned area (BA) and fire emissioninventories such as the Global Fire Emissions Database(GFED van der Werf et al 2010) the Brazilian BiomassBurning Emission Model (3BEM Longo et al 2010) theGlobal Fire Assimilation System (GFAS Kaiser et al 20092012) and the Fire INventory from NCAR (FINN Wiedin-myer et al 2011) have been published Most of the fire emis-sion inventories utilize active fire locations and burned area

to estimate the trace gases and aerosol emissions releasedinto the atmosphere (Mao et al 2014) However the tempo-ral and spatial distribution of biomass burning emissions isaffected by several sources of errors related eg to the lackof detection of small fires during prescribed and agriculturalburning Also global burned area products are unsuitable toestimate the burned area of small fires due to the limitationsof their algorithms (Giglio et al 2006 2010 Randerson etal 2012)

Most methodologies to estimate biomass burning emis-sions utilize the relationship proposed by Seiler andCrutzen (1980) which we will call the ldquoBA-based ap-proachrdquo

M [isin] = ABβEF[isin] (1)

where M [isin] is the emission load of species isin (g) A is theburned area (in km2) B is the fuel load (kg kmminus2) β is thecombustion completeness (unitless) and EF[isin] is the emis-sion factor released species isin (g kgminus1) In this method theburned area is one of the parameters necessary to estimatethe emission of trace gases and aerosols usually estimatedby Earth Observation (EO) satellites typically since firesmust have already occurred to produce a burned area thisis not that well suited to providing near-real-time air qualityforecasts in models that include vegetation fires Other fac-tors in Eq (1) can also be difficult to determine for exam-ple the combustion completeness typically depends on themoisture present in the biomass (and thus in part on microm-eteorology) and the fuel load (average biomass and its frac-tion above ground) is hardly homogeneous (Chuvieco et al2004 Yebra et al 2009 De Santis et al 2010)

In the last decade the development of sensors more tar-geted at active fire observations has allowed the estimationof the radiative energy flux released by fires or the fire ra-diative power (FRP in J sminus1) The FRP accuracy has beenevaluated by Kaufman et al (1998) and Wooster et al (2003)showing an average error of 16 and 65 respectively How-ever the error could be largely due to the spatial resolution ofseveral sensors basically the atmospheric transmittance andthe cloud obscuration can cause an omission error of 15 and11 in FRP estimates respectively (Schroeder et al 2008)Also according to Vermote et al (2009) the integration ofFRP over the fire cycle and its conversion to burned biomasscan cause an error of 21 and 10ndash30 respectively varyingaccording to heterogeneity of regionalzonal characteristics

Continuous acquisition of FRP over a firersquos lifetime pro-vides the fire radiative energy (FRE in MJ) released by thefire process New fire emission methods utilize the FRE tocalculate the amount of biomass burned andor the emissionof trace gases and aerosols (Wooster et al 2005 Ichoku andKaufman 2005 Ellicot et al 2009 Freeborn et al 2011Kumar et al 2011)

Atmos Chem Phys 16 6961ndash6975 2016 wwwatmos-chem-physnet1669612016

G Pereira et al Assessment of fire emission inventories during the SAMBBA experiment 6963

FREgrid(longlat) =12

nsumi=1

(FRPn+FRPn+1) (Tn+1minus Tn) (2)

M [isin] = FREgrid(longlat)γEF[isin] (3)

where FREgrid(lonlat) provides radiative energy at a geograph-ical location in terms of longitude and latitude of a spe-cific centered point of a regular grid T is the time sequenceof FRP acquisitions n represents the nth sample and γ isthe FRP-biomass conversion factor (kg MJminus1 Wooster et al2005 Kaiser et al 2012) In this method parameters suchas fuel load burning efficiency and the presence of moisturein the soil and in vegetation directly influence the observedenergy radiated by the fires and do not have to be separatelyconsidered

The South American Biomass Burning Analysis(SAMBBA) was an airborne experiment design to char-acterize the smoke physical and chemical properties inAmazonian rainforest and central area of Brazil TheSAMBBA campaign took place in September 2012 Theoperational smoke forecasting system built to supportSAMBBA flight planning utilized four fire emission in-ventories which deployed the aforementioned approaches(i) 3BEM (Longo et al 2010) (ii) Brazilian Biomass Burn-ing Emission Model with FRP assimilation (3BEM_FRPPereira et al 2009) (iii) GFAS (Kaiser et al 2012) and(iv) FINN (Wiedinmyer et al 2011) This study providesan intercomparison and evaluation of these inventories withfocus on the SAMBBA field campaign experiment

2 Data and methodology

21 Inventories description

211 Brazilian biomass burning emission models(3BEM and 3BEM_FRP)

3BEM is a model developed to estimate the daily fire emis-sions based on the location of actively burning fire ldquohotspotsrdquo(ie areas of combustion detected using active firethermalanomaly detection algorithms) derived via orbital remotesensing 3BEM utilizes the Wildfire Automated BiomassBurning Algorithm (WF_ABBA) applied in the Geostation-ary Operational Environmental Satellite (GOES) constella-tion data (Prins et al 1998) the MOD14 and MYD14 prod-ucts of the Moderate Resolution Imaging Spectroradiome-ter (MODIS) aboard the Terra and Aqua satellites (Justiceet al 2002) and the fire product developed by the En-vironmental Satellite Division (DSA) of National Institutefor Space Research (INPE) which uses the Spinning En-hanced Visible and Infrared Imager (SEVIRI) on board Me-teosat Second Generation GOES MODIS and AdvancedVery High Resolution Radiometer (AVHRR) on board the

National Oceanic and Atmospheric Administration (NOAA)constellation (Setzer et al 1994)

The 3BEM version initially developed by Longo etal (2010) estimates the emitted mass of trace gases andaerosols related to each fire detected by remote sensing asdescribed in Eq (1) Therefore the locations of the detectedfire hotspots are cross-tabulated with MODIS Land Covermap to allow the estimation of aboveground biomass densitycombustion factor and emission factor from literature values(Olson et al 2000 Andreae and Merlet 2001 Houghton etal 2001 Sestini et al 2003 Akagi et al 2011) Also themodel filters all fires located in a 1 km radius to prevent dou-ble counting between fire products

The updated 3BEM includes FRP assimilation(3BEM_FRP Pereira et al 2009) and utilizes the al-gorithm of the earlier 3BEM version but with burnedbiomass directly estimated using FRE estimates as de-scribed in Eq (3) The 3BEM_FRP model groups all FRPvalues estimated by MODIS GOES and METEOSATproducts according to their time acquisition eliminating thelow confidence fire pixels (values below 50 for MOD14MYD14 and METEOSAT products and flags 4 and 5 forWFABBAGOES product) and minimizing the impact of theMODIS bow-tie effect as described in Freeborn et al (2011)Also due to the high frequency of observations (mainlyin GOES and SEVIRI data) 3BEM needs only one firedetection to estimate the biomass burned and its associatedemissions due to the filtering process (Longo et al 2010)If the active fire has no subsequent observations in the next4 h the algorithm assumes that the fire event is over Thusmissed detections due to cloud cover have an impact on theFRE integration if the cloud persists for more than eightsatellite acquisitions

Some of these FRP estimates are compromised by sensorsaturation over larger fires particularly GOES over SouthAmerica (Xu et al 2010) Thus to not ignore importantepisodes of biomass burning by removing GOES saturatedpixels for which FRP values are not provided 3BEM_FRPutilizes Eq (4) to estimate the energy released by fires basedon the premise that emitted spectral radiance (Mλ) in spec-tral band centered at 39 microm is proportional to FRP (Woosteret al 2003)

FRPMIR =Ag

aσ

403int376

M(λT )dλminus Mb (4)

where Ag represents the area of the GOES pixel (km2) ais an empirical constant adapted for the GOES MIR spec-tral band (W mminus2 srminus1 micromminus1 Kminus4 Wooster et al 2005) σ isthe StefanndashBoltzmann constant (566times 10minus8 W mminus2 Kminus4)M is the Planck curve (W mminus2 micromminus1) λ is the wavelength(microm) T represents the temperature (K) and Mb is the radi-ance emitted by the background (110 MW) FRP values esti-mated by GOES satellites below 1000 MW are corrected by

wwwatmos-chem-physnet1669612016 Atmos Chem Phys 16 6961ndash6975 2016


+17 and FRP values higher than 1000 MW are correctedby +41 (Xu et al 2010) This procedure is also appliedto SEVIRI data but due to its spatial coverage we decidednot to include these data in the present analysis Pereira etal (2009) describe the 3BEM_FRP method in detail

A clustering process performs the combination of all de-tected fires from different sensors In this step the size of amatrix that merges FRP data can be defined according to thespatial resolution and grid configuration of the atmosphericmodel Consequently the convolution mask η(γκ) of sizeM timesN (rowstimes columns) running over the grid with FRPareal density (FRPad in W mminus2 defined by weighting theFRP values by pixel area) values estimated by different satel-lites ξ (longlat) will result in the grid (FRPgrid) containingall clustered fires for a given time step

FRPgrid(lonlatt) =

αsumγ=minusα

βsumκ=minusβ

η(γκ)ξ(long+γ lat+κ t) (5)

where the clustered grid is defined to all points wherethe mask of M timesN size overlaps the image completely(lonε[αMminusα] latε[βNminusβ]) Moreover if the interval be-tween two acquisitions is greater than 4 h (1T gt 14 400 s) atany time step of the FRP integration the observations are as-sumed to originate from two or more independent fires

212 GFAS

The GFAS is an approach used to map daily global fire emis-sion through FRP observations Therefore GFAS also as-sumes that the electromagnetic radiation emitted by fires isrelated to the consumption of burned biomass (Wooster etal 2005) In the GFAS version (11) used in this study FRPvalues of MOD14 and MYD14 fire products from Terra andAqua satellites respectively are used to estimate the averageof observed FRP areal density GFAS estimates open vege-tation fire trace gas and particulate emissions from each firedetected as described in Eq (3)

The model performs a clustering process of observed FRP(Fi) pixel area (Ai) and view zenith angle (θ) to sensorpixels with valid observations (i) to estimate the FRPad Theclustering process also takes observations of Fi = 0 ie no-fire into account Thus for each individual grid cell the esti-mated FRPad could be calculated as

FRPad =

sumk

sumikεjFik cos2(θik )sum

k

sumikεjAik cos2(θik )

where ik represents the pixel i of satellite product k (MOD14MYD14) This formulation implicitly corrects the MODISbow-tie effect and partial cloudicesnowwater cover of agrid cell Observation gaps are subsequently filled with a dataassimilation approach that currently assumes persistence ofthe fires

In GFAS the coefficient that converts the FRPad to drymatter combustion rate is based on eight land cover classes

Table 1 Land Use and Land Cover (LULC) classes based on GlobalLand Cover Characterization (GLCC)

1 Cropmixed farming 11 Semi-desert2 Short grassland 12 Ice capglacier3 Evergreen needleleaf tree 13 Bog or marsh4 Deciduous needleleaf tree 14 Inland water5 Deciduous broadleaf tree 15 Ocean6 Evergreen broadleaf tree 16 Evergreen shrub7 Tall grass 17 Deciduous shrub8 Desert 18 Mixed woodland9 Tundra 19 Forestfield mosaic10 Irrigated crop 20 Water and land mixture

(cf Heil et al 2010 Kaiser et al 2012) In addition theemission load of 40 species is calculated using emission fac-tors from Andreae and Merlet (2001) with subsequent up-dates from Christian et al (2003) Kaiser et al (2012) de-scribe the inventories method in detail

213 FINN

FINN is an approach to estimate daily fire emissions at 1 kmresolution through satellite observation of active fires TheFINN model produces global estimates of aerosols and tracegases of open vegetation fires as described in Eq (1) TheFINN utilizes the MOD14 and MYD14 fire products pro-cessed by MODIS Rapid Response (MRR) or by the MODISData Processing System (MODAPS) Collection 5 Howeversince MODIS observations do not cover the entire globedaily due to orbital gaps FINN smears MODIS detectionsof active fire over 2 days For each fire located in the equato-rial region it is assumed that the next day the fire will be halfof its original size (Wiedinmyer et al 2011)

Similarly to 3BEM the FINN model removes multiple de-tections of same fire pixel prior to estimation of the tracegases and aerosols released In addition for each active fireFINN estimates as 1 km2 the burned area for most of landuse classes with the exception of grasslandssavannas inwhich the burned area is assumed to be 075 km2 To esti-mate the emissions of trace gas and aerosol species emissionfactors derived from Andreae and Merlet (2001) and Akagiet al (2011) are used as well as the MODIS Land CoverType for 2005 and the MODIS Vegetation Continuous Fields(VCF) product Wiedinmyer et al (2011) describe the proce-dures adopted in detail

22 MODIS aerosol optical thickness (AOT)

The aerosol optical depth (AOD) product from MODIS sen-sors aboard the Aqua and Terra satellites is used to pro-vide a first order assessment of the impacts of the two dis-tinct methods to estimate biomass burning emission BA-based and FRP-based In this work the MODIS Level 20Collection 51 (051) data and Level 3 atmospheric prod-



Figure 1 (a) MCD12 Land Use and Land Cover (LULC) for South America with the dotted red line representing the Amazonian Arcof Deforestation and soybean expansion in Brazilian Cerrado (marked as X) (b) time-averaged Aqua MODIS aerosol optical thickness at550 nm during 1 Septemberndash31 October 2012 over South America

uct denominated MYD08_D3 (mean aerosol optical thick-ness at 550 nm ie AOT550 nm) are utilized to compare thefire emission inventories used during the SAMBBA cam-paign between 1 September and 31 October 2012 TheMODIS daily global product merges all MODIS acquisitionsover the globe on an equal-angle grid with 1 resolution(Kaufman and Tanreacute 1998) Figure 1a shows the MODISMCD12 Land Use and Land Cover (LULC) product forSouth America with LULC classes described in Table 1 Dur-ing the SAMBBA field campaign the highest biomass burn-ing aerosol loadings were observed over Brazilian territorymainly over the southeastern edge of the Amazonian forest(dotted red line known as Arc of Deforestation) and in thesoybean expansion area in the Brazilian savannah (markedas X) as shown in the time-averaged (1 Septemberndash31 Oc-tober 2012) AOD550 nm field derived from MODIS sensoraboard Aqua satellite (Fig 1b) The high values of AOD re-lated to fires located in eastern Mato Grosso (mainly in thesecondary forest) and in transition areas of Amazon rainfor-est and Brazilian savannah are noteworthy soybean expan-sion areas which present high concentration of fires havea lower amount of biomass than Amazon rainforest and thestrong ventilation favors the transport of smoke to west

23 Inventories configuration and analysis description

To evaluate the fire emission inventories utilized during theSAMBBA experiment we used the 3BEM preprocessor togenerate the gridded data in geographical coordinates with aspatial resolution of 01 The 3BEM preprocessor generatesdaily emission fluxes of several species but in this work weselected only carbon monoxide (CO) emission fields from1 September to 31 October 2012 to analyze The four in-

ventories are compared over eight sub-domains windowswith sizes of 10times 10 that typically cover different Brazil-ian states (Fig 2a) and different fire regimes and biomes(Fig 2b)

Two models were selected for real-time applications dur-ing SAMBBA experiment the Monitoring AtmosphericComposition and Climate (MACC) from European Cen-tre for Medium-Range Weather Forecasts (ECMWF) withGFAS fire emission inventory (described in Sect 212)and the Coupled Chemistry Aerosol and Tracer Transportmodel to the Brazilian developments on the Regional Atmo-spheric Modelling System (CCATT-BRAMS) with 3BEMfire emission inventory (described in Sect 211) These mod-els were selected due to their distinct methodologies for es-timating the biomass burning emissions (the first a FRP-based approach and the second the BA-based approach)The MACCECMWF model simulation had a global domainabout 08 (T255) horizontal resolution 60 levels verticallevels and a forecast frequency of 6 h It assimilates MODISAOD observations and the full set of meteorological obser-vations that are used for the operational weather forecastsof ECMWF The CCATT-BRAMS simulations were per-formed for the South America domain 022 of horizontalresolution 24 levels of vertical resolution forecast operationmode forecast frequency of 3 h and boundary conditionsfrom Center for Weather Forecasting and Climate Research(CPTEC) model



Figure 2 South America maps highlighting (a) the Brazilian states of Acre Amazonas Mato Grosso Para Rondonia and Tocantins whereSAMBBA field campaign took place and (b) the eight sub-domains windows for the biomass burning inventories comparison

3 Results and discussion

31 Comparing emission inventories spatially

During the SAMBBA experiment four inventories wereused in near real time within atmospheric chemistry trans-port models to support the SAMBBA flight planning (FINNGFAS 3BEM 3BEM_FRP) Each inventoryrsquos spatial distri-bution of total CO emission (104 kg mminus2) over South Amer-ica from 1 September to 31 October 2012 is depicted inFig 3 according to the respective methodologies used to es-timate the emission loading

The intercomparison shown in Fig 3 demonstrates that in-ventories that utilize the same (ie FRP-based or BA-based)methodology show similar spatial patterns in CO emissions(kg mminus2) not only in the Amazon basin but across all ofSouth America although they differ in their absolute valuesThe CO emissions estimated by the 3BEM and FINN emis-sion inventories (Fig 3a and b respectively) present highervalues than 3BEM_FRP (Fig 3c) and GFAS (Fig 3d) in theregions where the main processes of forest logging and sub-sequent agricultural expansion (grids 3 4 and 5) occur Thehighest emissions in 3BEM and FINN are located mainly inRondonia State (grid 3) and in Mato Grosso State (grid 4)where most of the SAMBBA flights took place

Cardozo et al (2014) analyzed the fires pattern in Ron-donia between 2000 and 2011 and identified that most firesresult in relatively small ldquoburn scarsrdquo on the landscape withareas of 20 to 80 ha (64 of cases) In addition only 65 of all burned areas in Rondonia are associated with recentlydeforested areas This could indicate that 3BEM and FINNare overestimating the CO emission load due to an erroneousassumption that observed fires occur in forest instead of live-stock and permanent crops which have lower aboveground

biomass Furthermore 3BEM_FRP and GFAS inventories donot display a similar pattern to 3BEM and FINN becausetheir emissions are directly related to FRP with a weakerdependency on land cover type Thus areas with low above-ground biomass will provide low values of FRE due the ob-served characteristics of the fire activity and thus low valuesof CO

The spatial distribution of the emission inventories sug-gests that fires in the region are strongly related to defor-estation activity and therefore to the general economy witha strong trend in recent years of fires in secondary forests(Cardozo et al 2014) In Fig 3c higher emission loads arelocated in the east of Brazil (grid 5) mainly in the Cerradobiome a vegetal formation composed of savanna and othertypically low-density vegetation formations which includetrees of 15 m height (as shown in Fig 1a) This region isnow economically used constituting a new agricultural fron-tier of Brazil (with more than 100 million hectares suitablefor modern mechanized crop agriculture mainly soybean) Inthis area the four fire inventories differ considerably The in-ventories that use the FRP approach show much higher emis-sion loads compared to 3BEM and FINN

The difference in South American daily area-averagedemission of CO (kg mminus2) between 1 September and 31 Oc-tober 2012 was quantified via linear correlation coefficientanalysis (all significant at plt 005) (Fig 4) The highest lin-ear correlation coefficients were found between 3BEM andFINN with 086 between 3BEM_FRP and GFAS with 085and between GFAS and FINN with 084 These high linearcorrelations indicate that inventories produced using similaremission methods tend to agree with each other The thirdcorrelation reflects that both inventories use the same activefire observations (MODIS) albeit with different data pro-cessing We highlight that all were significant at pgt 005



Figure 3 Spatial distribution of total CO emissions (104 kg mminus2) over South America from 1 September to 31 October 2012 estimated by(a) 3BEM (b) FINN (c) 3BEM_FRP and (d) GFAS

level using the Student t test To analyze the measures ofaccuracy of regression between fire inventories we used thebootstrap technique (Efron 1982) In this technique a pop-ulation of 10times 104 reconstructs the regression and providesthe parameters to create the confidence interval and erroranalysis of model estimation

The bootstrap regression among the daily area-averagedCO values to 3BEM and FINN emission inventories presenta linear correlation between 075 and 092 with a tendency ofFINN to overestimate relative to 3BEM the emission loadof CO in 20ndash30 (Fig 4) This apparent FINN overestima-tion is seen in the majority of grids with the exception ofgrids 2 5 and 7 (Fig 5a) In areas where FINN emissions arelower than 3BEM emissions vegetation is mainly composedof Brazilian Cerrado (grids 2 and 5) while the Brazilian Pan-tanal wetland biome (grid 7) is predominantly composed ofsavannah and grassland with the additional presence of de-

ciduous and semi-deciduous forest Since 3BEM and FINNimplement similar methodologies to estimate the emissionload of species released by wildfires the parameters used inEq (1) such as aboveground biomass are likely to be asso-ciated with the relatively high estimation in CO values

The intercomparison analysis of the daily area-averagedCO values to 3BEM and GFAS inventories shows a linearcorrelation coefficient between 075 and 085 with highervalues over the Arc of Deforestation in the Amazon rain-forest (Fig 1a) However GFAS has a tendency to presenthigher CO emissions (by 10ndash20 ) in grids 5 and 8 and tounderestimate by 20ndash30 in grids 3 and 4 (Fig 5b) whencompared to 3BEM In these areas the vegetation is domi-nated by the Amazon biome along with a small area of Cer-rado (located in the south of the grids) with presence of ev-ergreen broadleaf forest tropical degraded forest and crop-landlivestock areas Moreover the daily area-averaged CO



Figure 4 Linear regression between daily area-averaged emission load of CO to South America from 3BEM 3BEM_FRP FINN and GFASexamined between 1 September and 31 October 2012 In the graph x and y axis represent the CO (kg mminus2) of each fire inventory (ie thefirst line of 3BEM regressions 3BEM_FRP GFAS and FINN are x axis)

values estimated by GFAS present a considerable differencein the region of soybean expansion in grid 5

Similar to GFAS the relationship among the daily area-averaged CO values to 3BEM and 3BEM_FRP shows a lin-ear correlation coefficient between 065 and 075 However3BEM_FRP has a tendency to overestimate the CO valuesby 60ndash85 in grid 5 and to underestimate by 10ndash15 ingrids 3 and 4 (Fig 5c) The 3BEM_FRP model presented anelevated emission of CO in grid 5 possibly due the GOESviewing zenith angle (high viewing angles results in erro-neous values of infrared brightness and present a large pixelarea Roberts et al 2005 Vermote et al 2009 Petersonet al 2012) In addition in grid 5 cloud absence may in-fluence in the FRP cycle leading to an overestimation duethe large number of acquisitions in high view angles Therelationship between the daily area-averaged CO values to3BEM_FRP and GFAS shows a linear correlation between082 and 090 Accordingly the 3BEM_FRP model has a ten-dency to present higher emission fluxes of CO than GFAS

mainly in some areas of deforestation arc and in grid 5 thatcould reach 100 (Fig 5d) The high values are associatedwith assimilation of GOES FRP in 3BEM_FRP while GFASutilizes only MODIS FRP data

The fieldwork measurements acquired during theSAMBBA campaign indicate that the near-real-time airsmoke forecasts based on BA-based inventories such as3BEM used by CCATT-BRAMS typically underestimatedsmoke loading in the central to eastern region of the Ama-zon forest (near Mato Grosso and Tocantins States) asdescribed in Rosaacuterio et al (2013) In addition the emissioninventories tended to overestimate the smoke loading inthe northwestern part of Rondonia (indicated in Fig 2a)However during the SAMBBA campaign lower values ofCO emissions were identified showing an overestimationof emission flux estimates in this region In general all fireemission inventories present a good agreement with mostregressions significant at pgt 005 level according to theStudent t test (Table 2 non-significant regression marked in



Figure 5 Difference between fire inventories (a) 3BEM-FINN (b) 3BEM-GFAS (c) 3BEM-3BEM_FRP (d) 3BEM_FRP-GFAS(e) 3BEM_FRP-FINN (f) FINN-GFAS

red) The only regression that presented low values of linearcorrelation coefficients are located in grid 5 with valueslower than 030 Moreover regression between 3BEM andFINN presents the highest correlation in most of the grids

Also the absolute bias analysis indicate a high variability indaily area-averaged emission load of CO between the fourfire inventories



Figure 6 Latndashlong average time series of Aqua MODIS AOT550 nm and 3BEM 3BEM_FRP GFAS and FINN emission load of CO timeseries



Table 2 Linear correlation coefficients (r) between 3BEM3BEM_FRP FINN and GFAS daily area-averaged emission in-ventories to the eight grids and average absolute bias of CO (inkg mminus2) In the table the first value indicates the r and the secondrepresents the absolute bias (rbias) Bold value are not significant(p gt 005 level Student t test)

3BEM_FRP FINN GFAS

3BEM_G1 084933 085145 0827363BEM_FRP_G1 067minus788 076minus197FINN_G1 0825913BEM_G2 070436 075458 0792103BEM_FRP_G2 06122 078minus226FINN_G2 082minus2493BEM_G3 0781642 0941674 090minus1363BEM_FRP_G3 07732 087minus1778FINN_G3 091minus18113BEM_G4 081minus245 093996 090minus2633BEM_FRP_G4 0711241 090minus18FINN_G4 088minus12603BEM_G5 016minus6345 030minus793 0366053BEM_FRP_G5 0475552 0846950FINN_G5 08013993BEM_G6 04612 08968 088minus863BEM_FRP_G6 03255 065minus99FINN_G6 081minus1543BEM_G7 068minus14 089151 0854133BEM_FRP_G7 060165 079427FINN_G7 0852613BEM_G8 063minus138 084minus251 079minus313BEM_FRP_G8 065minus113 065106FINN_G8 091219

32 Assessment of fire inventories with AOT

Figure 6 shows the daily emission estimates from each ofthe emission inventories used in SAMBBA campaign inte-grated over the eight grids and the area-averaged values ofAOT550 nm for 1 September to 31 October 2012 During theSAMBBA campaign the grids located in the Arc of Defor-estation and in Mato Grosso state exhibited the highest val-ues of averaged AOT550 nm Generally the inventories thatuse FRP to estimate the emission of CO produce similar pat-terns Likewise the emission inventories that utilize the rela-tionship between the burned area fuel load and the com-bustion completeness lead to comparable patterns This isevident in Fig 6a in which 3BEM_FRP and GFAS havelower values during all periods with daily emission loadsof CO between 20 and 500 kg mminus2 and an average value of93 kg mminus2 In addition the 3BEM and FINN emission inven-tories presented a high variability during the analyzed periodwith a daily area-averaged CO emission flux that reaches val-ues greater than 3000 kg mminus2 The linear correlation coef-ficient of 3BEM_FRP and GFAS and of 3BEM and FINNare 085 and 076 respectively with a tendency of GFAS

Table 3 Total mean emissions (kg mminus2) of CO in the eight gridsshown in Fig 2

Total emission of CO (kg mminus2)

3BEM FINN GFAS 3BEM_FRP

Grid 1 61660 52971 17485 5636Grid 2 55234 42631 27706 29059Grid 3 155346 163552 54891 56815Grid 4 151720 167518 91905 166423Grid 5 82271 45945 129866 274896Grid 6 35199 40412 31118 29811Grid 7 53072 28266 43981 48541Grid 8 11342 13252 25674 19633

Total 605844 554547 422626 630814

and 3BEM to present higher CO emission fluxes In grid 1the total emission fluxes of CO in 3BEM and FINN from1 September to 31 October 2012 are respectively 61660and 52971 kg mminus2 (Table 3) an estimation that is 4ndash10 timesmore than GFAS and 3BEM_FRP

Figure 6b shows the area-averaged time series of COand AOT550 nm in grid 2 with the four emission invento-ries showing similar patterns with high emission values on9 September and between 20 and 30 October 2012 The lin-ear correlation coefficients of 3BEM_FRP and GFAS andof 3BEM and FINN are 078 and 075 with a tendency of3BEM (55234 kg mminus2) and 3BEM_FRP (29059 kg mminus2) topresent higher CO emission fluxes in this region Also in thegeneral pattern of temporal evolution we could observe agood agreement between CO emission load estimated by thefour emission inventories and MODIS AOT550 nm Howeverthe four emission inventories exhibit distinct CO flux distri-butions in grid 3 (Fig 6c) In September FINN and 3BEMpresented higher values than 3BEM_FRP and GFAS duringSeptember and October However 3BEM_FRP shows highervalues than GFAS in September and smaller values in Octo-ber The linear correlation coefficients of 3BEM_FRP andGFAS and of 3BEM and FINN are 087 and 094 respec-tively with a tendency of FINN and 3BEM_FRP to yieldhigher estimates of the CO emission flux The relationshipbetween GFAS and FINN presents a linear correlation coef-ficient of 091 However the total emission load of CO esti-mated by FINN (163552 kg mminus2) is 3 times more than GFAS(54891 kg mminus2)

In grid 4 all inventories showed a good agreement withlinear correlations coefficients greater than 090 Howeverduring the period from 3 to 10 September 3BEM_FRPpresents six episodes with higher emission loads which re-main undetected in the other inventories (Fig 6d) The outliervalues in 3BEM_FRP are likely to be related to GOES FRPacquisitions which suggests inconsistencies in GOES dataacquisition due to viewing angle and errors in data process-



Figure 7 Latndashlong average time series of TerraMODIS AOT550 nm (in green) AquaMODIS AOT550 nm (in black) CCATT-BRAMSsimulated AOT550 nm (in blue) and MACCECMWF (in red) AOT550 nm analysis Grids are those shown in Fig 2

ing Grid 5 shows the lowest linear correlation coefficient be-tween the 3BEM and FINN emission inventories during theSAMBBA campaign with 030 In this grid FINN presentsa low estimation of the CO emission flux when comparedwith the other emission inventories (as demonstrate in Ta-ble 3) Here FRP-based methods showed larger emissionsthan the BA-based ones The 3BEM_FRP model overesti-mated the emission values by more than 100 comparedto GFAS (Fig 6e) However the linear correlation betweenthese emission inventories is 084 In the other grids theemissions of CO provided by the four emission inventoriesshow a considerable consistency with few differences in ab-solute values indicating a good agreement across these otherregions of South America (Fig 6fndashh)

33 Evaluating the emissions against observations

During the SAMBBA campaign operational atmosphericchemistry transport models were applied in real time (i) theMACCECMWF with GFAS and (ii) CCATT-BRAMS with3BEM Figure 7 shows the simulated AOT550 nm fromMACCECMWF (red line) CCATT-BRAMS (blue line) andestimated by Terra (green line) and Aqua (black line) to theeight grid cells Although 3BEM presents the highest emis-sion flux in grid 1 and grid 2 the CCATT-BRAMS simula-tion underestimate the MODIS AOT550 nm In these grids theMACCECMWF model presents a better consistency withthe satellite measurements One potential reason for the un-derestimation of CCATT-BRAMS is related to backgroundaerosol unrelated to the fires directly being observed and thatis not included in this model causing low values of simulatedAOT550 nm for these grids Also MACCECMWF assimi-lates MODIS AOD observations and even with low emis-sion values in grids 1 and 2 the atmospheric aerosol load isstrongly constrained by the MODIS observations

The assessment of the MACCECMWF and CCATT-BRAMS models against MODIS AOT550 mm reveals a goodagreement in terms of the general pattern of the temporal evo-lution As shown in the AOT550 mm time series both meth-ods appear to be able to estimate the influence of aerosolsreleased by fires rather well However when the intensityof the biomass burning is too high the values simulated byMACCECMWF and CCATT-BRAMS for the grids 4ndash8 ap-pear underestimated possibly due to the influence of smokeon the FRP measurements lack of fire observations cloudsaboveground biomass data or fire size Moreover it is pos-sibly to identify an overestimation in grid 3 mainly due toan out-of-date land use and land cover map that inserts firesover forest areas even if the area had been burneddeforestedduring earlier years and a very noticeable underestimation ingrid 5 as demonstrated in the inventories

4 Conclusions

Full characterization of the emissions of trace gases andaerosols is often essential for assessing the atmospheric im-pacts of fire and for constructing fire inventories These in-ventories generally rely on data from environmental satel-lites and provide useful information for weather climate andair quality models In this study we analyzed data from thefour biomass burning emission inventories used during theSAMBBA airborne atmospheric sampling and remote sens-ing campaign that took place in Rondonia (Brazil) between1 September and 31 October 2012 Each inventory utilizesdistinct methodology with 3BEM and FINN deriving theemission of trace gases and aerosols through a combina-tion of BA and fuel load metrics based on vegetation mapsand field-location-specific coefficients whilst 3BEM_FRP



and GFAS estimate biomass burned more directly from FRPmeasurements made by the EO satellite instruments

The evaluation of the emission inventories focused oneight pre-defined grid cells (Fig 2) and demonstrates that in-ventories that utilize the same methodology such as 3BEMand FINN on the one hand (BA-based) and GFAS and3BEM_FRP on the other (FRP-based) show similar pat-terns in emission spatial distribution not only in the Arc ofDeforestation but also throughout South America Howeverthey can differ in their absolute values As such each in-ventory has particular characteristics with 3BEM and FINNshowing more emissions of CO in the Amazon forest log-ging area where most of the SAMBBA campaign flights oc-curred Furthermore these emission inventories typically un-derestimated the smoke loading in the southeastern regionof Amazon forest and in the northwestern region of Ron-donia where 3BEM_FRP and GFAS present larger emis-sion fluxes of CO The best overall linear correlations co-efficients were found between 3BEM and FINN with ap-proximately 086 and 3BEM_FRP and GFAS with approx-imately 085 which indicate that similar emission methodstend to agree with each other Furthermore the comparisonof the 3BEM and FINN fire inventories reveal linear corre-lation coefficients between 075 and 092 with larger totalemissions of CO in FINN than in 3BEM for grids 3 4 6and 8 3BEM_FRP and GFAS show linear correlation coeffi-cients between 065 and 095 with more emissions of CO inGFAS than in 3BEM_FRP for grids 1 6 and 8

During the SAMBBA campaign the assessment ofsimulated AOT550 nm from MACCECMWF and CCATT-BRAMS operational atmospheric chemistry transport mod-els with MODIS AOT550 nm Terra and Aqua measurementsshow a good agreement related with general pattern of thetime series Also MACCECMWF and CCATT-BRAMSmodels are capable of simulating the aerosols released byfires However when the intensity of the biomass burning istoo high the aerosol values simulated by the two models un-derestimate the MODIS measurements

Acknowledgements We would like to thank the Satildeo PauloResearch Foundation (FAPESP) for their financial support(201213575-9) and Minas Gerais State Research Foundation(FAPEMIG grant number APQ-01698-14) J W Kaiser andM J Wooster were supported by NERC in the SAMBBA project(grant number NEJ0100731) The MACCECMWF simulationswere kindly provided by the precursor of EUrsquos Copernicus Atmo-sphere Monitoring Service (httpatmospherecopernicuseu) Theauthors thank two anonymous reviewers for useful comments thathelped improve the manuscript

Edited by H Coe

References

Akagi S K Yokelson R J Wiedinmyer C Alvarado M JReid J S Karl T Crounse J D and Wennberg P O Emis-sion factors for open and domestic biomass burning for usein atmospheric models Atmos Chem Phys 11 4039ndash4072doi105194acp-11-4039-2011 2011

Andreae M O and Merlet P Emission of trace gases and aerosolsfrom biomass burning Global Biogeochem Cy 15 955ndash9662001

Andreae M Rosenfeld D Artaxo P Costa A Frank G LongoK M and Silva Dias M A F Smoking rain clouds over theAmazon Science 303 1342ndash1345 2004

Baldassarre G Pozzoli L Schmidt C C Unal A KindapT Menzel W P Whitburn S Coheur P-F Kavgaci Aand Kaiser J W Using SEVIRI fire observations to drivesmoke plumes in the CMAQ air quality model a case studyover Antalya in 2008 Atmos Chem Phys 15 8539ndash8558doi105194acp-15-8539-2015 2015

Bowman D M J S Balch J K Artaxo P Bond W J CarlsonJ M Cochrane M A DrsquoAntonio C M DeFries R S DoyleJ C Harrison S P Johnston F H Keeley J E KrawchukM A Kull C A Marston J B Moritz M A Prentice I CRoos C I Scott A C Swetnam T W van der Werf G Rand Pyne S J Fire in the Earth System Science 324 481ndash4842009

Bustamante M M C Roitman I Aide T M Alencar A An-derson L Aragatildeo L Asner G P Barlow J Berenguer EChambers J Costa M H Fanin T Ferreira L G FerreiraJ N Keller M Magnusson W E Morales L Morton DOmetto J P H B Palace M Peres C Silveacuterio D Trum-bore S and Vieira I C G Towards an integrated monitoringframework to assess the effects of tropical forest degradation andrecovery on carbon stocks and biodiversity Glob Change Biol1 92ndash109 doi101111gcb13087 2016

Cardozo F S Pereira G Shimabukuro Y E and Moraes E CAnalysis and Assessment of the Spatial and Temporal Distribu-tion of Burned Areas in the Amazon Forest Remote Sens 68002ndash8025 2014

Chuvieco E Cocero D Riano D Martinc P Martiacutenez-VegacJ Rivad J and Peacuterez F Combining NDVI and surface tem-perature for the estimation of live fuel moisture content in forestfire danger rating Remote Sens Environ 92 322ndash331 2004

Christian T J Kleiss B Yokelson R J Holzinger R CrutzenP J Hao W M Saharjo B H and Ward D E Comprehen-sive laboratory measurements of biomass-burning emissions 1Emissions from Indonesian African and other fuels J GeophysRes 108 4719 doi1010292003JD003704 2003

Crutzen P J and Andreae M O Biomass burning in the trop-ics Impact on atmospheric chemistry and biogeochemical cy-cles Science 250 1669ndash1678 1990

De Santis A Asner G P Vaughan P J and Knapp D E Map-ping burn severity and burning efficiency in California usingsimulation models and Landsat imagery Remote Sens Environ114 1535ndash1545 doi101016jrse201002008 2010

Efron B The jackknife the bootstrap and other resampling plansSociety of Industrial and Applied Mathematics CBMS-NSFMonographs No 38 Philadelphia PA 1982



Ellicott E Vermote E Giglio L and Roberts G Estimatingbiomass consumed from fire using MODIS FRE Geophys ResLett 36 L13401 doi1010292009GL038581 2009

Fearnside P M Global warming and tropical land use changeGreenhouse gas emissions from biomass burning decompositionand soils in forest conversion shifting cultivation and secondaryvegetation Climatic Change 46 115ndash158 2000

Freeborn P H Wooster M J and Roberts G Addressing thespatiotemporal sampling design of MODIS to provide estimatesof the fire radiative energy emitted from Africa Remote SensEnviron 115 475ndash498 2011

Giglio L van der Werf G R Randerson J T Collatz GJ and Kasibhatla P Global estimation of burned area usingMODIS active fire observations Atmos Chem Phys 6 957ndash974 doi105194acp-6-957-2006 2006

Giglio L Randerson J T van der Werf G R Kasibhatla PS Collatz G J Morton D C and DeFries R S Assess-ing variability and long-term trends in burned area by mergingmultiple satellite fire products Biogeosciences 7 1171ndash1186doi105194bg-7-1171-2010 2010

Heil A Kaiser J W van der Werf G R Wooster M J SchultzM G and Dernier van der Gon H Assessment of the real-timefire emissions (GFASv0) by MACC Tech Memo 628 ECMWFReading UK 2010

Houghton R A Lawrence K T Hackler J L and Brown SThe spatial distribution of forest biomass in the Brazilian Ama-zon A comparison of estimates Glob Change Biol 7 731ndash7462001

Ichoku C and Kaufman Y J A method to derive smoke emissionrates from MODIS fire radiative energy measurements IEEE TGeosci Remote Sens 43 2636ndash2649 2005

Justice C O Giglio B Korontzi S Owens J Morisette J TRoy D P Descloitres J Alleaume S Petitcolin F and Kauf-man Y The MODIS fire products Remote Sens Environ 83244ndash262 2002

Kaiser J W Suttie M Flemming J Morcrette J-J BoucherO and Schultz M G Global real-time fire emission estimatesbased on space-borne fire radiative power observations AIPConf Proc 1100 645ndash648 2009

Kaiser J W Heil A Andreae M O Benedetti A ChubarovaN Jones L Morcrette J-J Razinger M Schultz M GSuttie M and van der Werf G R Biomass burning emis-sions estimated with a global fire assimilation system basedon observed fire radiative power Biogeosciences 9 527ndash554doi105194bg-9-527-2012 2012

Kaufman J B Cummings D L Ward D E and Babbitt R Firein the Brazilian Amazon Biomass nutrient pools and losses inslashed primary forests Oecologia 104 397ndash408 1995

Kaufman Y J and Tanreacute D Algorithm for Remote Sensing ofTropospheric Aerosols from MODIS MODIS Algorithm The-oretical Basis Document Product ID MOD04 revised 26 Oc-tober 1998 available at httpmodisgsfcnasagovdataatbdatbd_mod02pdf (last access 30 October 2015) 1998

Kumar S S Roy D P Boschetti L and Kremens R Exploit-ing the power law distribution properties of satellite fire radiativepower retrievals A method to estimate fire radiative energy andbiomass burned from sparse satellite observations J GeophysRes 116 D19303 doi1010292011JD015676 2011

Longo K M Freitas S R Andreae M O Setzer A PrinsE and Artaxo P The Coupled Aerosol and Tracer Transportmodel to the Brazilian developments on the Regional Atmo-spheric Modeling System (CATT-BRAMS) ndash Part 2 Model sen-sitivity to the biomass burning inventories Atmos Chem Phys10 5785ndash5795 doi105194acp-10-5785-2010 2010

Mao Y H Li Q B Chen D Zhang L Hao W-M and LiouK-N Top-down estimates of biomass burning emissions ofblack carbon in the Western United States Atmos Chem Phys14 7195ndash7211 doi105194acp-14-7195-2014 2014

Olson J S Watts J A and Allison L J Major world ecosys-tem complexes ranked by carbon in live vegetation A databaseavailable at httpcdiacesdornlgovndpsndp017html (last ac-cess 7 August 2010) 2000

Pereira G Freitas S R Moraes E C Ferreira N JShimabukuro Y E Rao V B and Longo K M Estimatingtrace gas and aerosol emissions over South America Relation-ship between fire radiative energy released and aerosol opticaldepth observations Atmos Environ 43 6388ndash6397 2009

Peterson D Wang J Ichoku C Hyer E and AmbrosiaV A sub-pixel-based calculation of fire radiative powerfrom MODIS observations 1 Algorithm development andinitial assessment Remote Sens Environ 129 262ndash279doi101016jrse201210036 2012

Prins E M Felz J M Menzel W P and Ward D E Anoverview of GOES-8 diurnal fire and smoke results for SCAR-Band 1995 fire season in South America J Geophys Res 10331821ndash31825 1998

Randerson J T Chen Y van der Werf G R Rogers B Mand Morton D C Global burned area and biomass burningemissions from small fires J Geophys Res 117 G04012doi1010292012JG002128 2012

Roberts G Wooster M J Perry G L W Drake N RebeloL M and Dipotso F Retrieval of biomass combustion ratesand totals from fire radiative power observations Application tosouthern Africa using geostationary SEVIRI imagery J Geo-phys Res-Atmos 110 D21111 doi1010292005JD0060182005

Rosaacuterio N E Longo K M Freitas S R Yamasoe M Aand Fonseca R M Modeling the South American regionalsmoke plume aerosol optical depth variability and surface short-wave flux perturbation Atmos Chem Phys 13 2923ndash2938doi105194acp-13-2923-2013 2013

Schroeder W Csiszar I and Morisette J Quantifying the im-pact of cloud obscuration on remote sensing of active fires in theBrazilian Amazon Remote Sens Environ 112 456ndash470 2008

Seiler W and Crutzen P J Estimates of gross and net fluxes of car-bon between the biosphere and atmosphere from biomass burn-ing Climatic Change 2 207ndash247 1980

Sestini M Reimer E Valeriano D Alvalaacute R Mello E ChanC and Nobre C Mapa de cobertura da terra da Amazocircnia legalpara uso em modelos meteoroloacutegicos in Simpoacutesio Brasileiro deSensoriamento Remoto (SBSR) 11 Belo Horizonte Anais SatildeoJoseacute dos Campos INPE 2003 Artigos 2901ndash2906 CD-ROMOn-line ISBN 85-17-00018-8 2003

Setzer A W Pereira Jr A C and Pereira M C Satellite studiesof biomass burning in Amazonia some practical aspects RemoteSens Rev 10 91ndash103 1994



Shimabukuro Y E Pereira G Cardozo F S Stockler R Fre-itas S R and Coura S M C Biomass burning emission esti-mation in Amazon tropical forest in Domingo Alcaraz SeguraCarlos Marcelo Di Bella Julieta Veronica Straschnoy (Org)Earth Observation of Ecosystem Services 1st Edn Oxford UKTaylor amp Francis 1 112ndash130 2013

Val Martin M Logan J A Kahn R A Leung F-Y NelsonD L and Diner D J Smoke injection heights from fires inNorth America analysis of 5 years of satellite observations At-mos Chem Phys 10 1491ndash1510 doi105194acp-10-1491-2010 2010

van der Werf G R Randerson J T Giglio L Collatz G J MuM Kasibhatla P S Morton D C DeFries R S Jin Y andvan Leeuwen T T Global fire emissions and the contribution ofdeforestation savanna forest agricultural and peat fires (1997ndash2009) Atmos Chem Phys 10 11707ndash11735 doi105194acp-10-11707-2010 2010

Vermote E Ellicott E Dubovik O Lapyonok T Chin MGiglio L and Roberts G J An approach to estimate globalbiomass burning emissions of organic and black carbon fromMODIS fire radiative power J Geophys Res 114 205ndash2272009

Wiedinmyer C Akagi S K Yokelson R J Emmons L K Al-Saadi J A Orlando J J and Soja A J The Fire INventoryfrom NCAR (FINN) a high resolution global model to estimatethe emissions from open burning Geosci Model Dev 4 625ndash641 doi105194gmd-4-625-2011 2011

Wooster M J Zhukov B and Oertel D Fire radiative energy forquantitative study of biomass burning Derivation from the BIRDexperimental satellite and comparison to MODIS fire productsRemote Sens Environ 86 83ndash107 2003

Wooster M J Roberts G and Perry G L W Retrieval ofbiomass combustion rates and totals from fire radiative power ob-servations Calibration relationships between biomass consump-tion and fire radiative energy release J Geophys Res 110D24311 doi1010292005JD006318 2005

Xu W Wooster M Roberts G and Freeborn P New GOES im-ager algorithms for cloud and active fire detection and fire radia-tive power assessment across North South and Central AmericaRemote Sens Environ 114 1876ndash1895 2010

Yebra M Chuvieco E and Riano D Estimation of live fuelmoisture content from MODIS images for fire risk assessmentAgr Forest Meteorol 148 523ndash536 2009


Abstract

Introduction

Data and methodology

Inventories description

Brazilian biomass burning emission models (3BEM and 3BEM_FRP)

GFAS

FINN

MODIS aerosol optical thickness (AOT)

Inventories configuration and analysis description

Results and discussion

Comparing emission inventories spatially

Assessment of fire inventories with AOT

Evaluating the emissions against observations

Conclusions

Acknowledgements

References

Atmos Chem Phys 16 6961ndash6975 2016wwwatmos-chem-physnet1669612016doi105194acp-16-6961-2016copy Author(s) 2016 CC Attribution 30 License

Assessment of fire emission inventories during the South AmericanBiomass Burning Analysis (SAMBBA) experimentGabriel Pereira1 Ricardo Siqueira2 Nilton E Rosaacuterio3 Karla L Longo2a Saulo R Freitas2a Francielle S Cardozo1Johannes W Kaiser4 and Martin J Wooster56

1Department of Geoscience Federal University of Sao Joao del-Rei (UFSJ) Sao Joao del-Rei Brazil2Center for Weather Forecast and Climate Studies National Institute for Space Research (INPE) Cachoeira Paulista Brazil3Environmental Sciences Department Satildeo Paulo Federal University (UNIFESP) Diadema Satildeo Paulo Brazil4Max Planck Institute for Chemistry (MPIC) Mainz Germany5Department of Geography Kingrsquos College London (KCL) London UK6NERC National Centre for Earth Observation (NCEO) Leicester UKanow at Global Modeling and Assimilation Office NASA Goddard Space Flight Center and USRAGESTARGreenbelt MD USA

Correspondence to Gabriel Pereira (pereiraufsjedubr)

Received 26 December 2015 ndash Published in Atmos Chem Phys Discuss 19 January 2016Revised 5 May 2016 ndash Accepted 23 May 2016 ndash Published 7 June 2016

Abstract Fires associated with land use and land coverchanges release large amounts of aerosols and trace gasesinto the atmosphere Although several inventories of biomassburning emissions cover Brazil there are still considerableuncertainties and differences among them While most fireemission inventories utilize the parameters of burned areavegetation fuel load emission factors and other parame-ters to estimate the biomass burned and its associated emis-sions several more recent inventories apply an alternativemethod based on fire radiative power (FRP) observationsto estimate the amount of biomass burned and the corre-sponding emissions of trace gases and aerosols The Brazil-ian Biomass Burning Emission Model (3BEM) and the FireInventory from NCAR (FINN) are examples of the firstwhile the Brazilian Biomass Burning Emission Model withFRP assimilation (3BEM_FRP) and the Global Fire Assim-ilation System (GFAS) are examples of the latter Thesefour biomass burning emission inventories were used duringthe South American Biomass Burning Analysis (SAMBBA)field campaign This paper analyzes and inter-comparedthem focusing on eight regions in Brazil and the time periodof 1 Septemberndash31 October 2012 Aerosol optical thickness(AOT550 nm) derived from measurements made by the Mod-erate Resolution Imaging Spectroradiometer (MODIS) oper-ating on board the Terra and Aqua satellites is also applied to

assess the inventoriesrsquo consistency The daily area-averagedpyrogenic carbon monoxide (CO) emission estimates ex-hibit significant linear correlations (r pgt 005 level Studentt test) between 3BEM and FINN and between 3BEM_ FRPand GFAS with values of 086 and 085 respectively Theseresults indicate that emission estimates in this region derivedvia similar methods tend to agree with one other Howeverthey differ more from the estimates derived via the alterna-tive approach The evaluation of MODIS AOT550 nm indi-cates that model simulation driven by 3BEM and FINN typ-ically underestimate the smoke particle loading in the east-ern region of Amazon forest while 3BEM_FRP estimationsto the area tend to overestimate fire emissions The daily re-gional CO emission fluxes from 3BEM and FINN have linearcorrelation coefficients of 075ndash092 with typically 20ndash30 higher emission fluxes in FINN The daily regional CO emis-sion fluxes from 3BEM_FRP and GFAS show linear corre-lation coefficients between 082 and 090 with a particularlystrong correlation near the arc of deforestation in the Ama-zon rainforest In this region GFAS has a tendency to presenthigher CO emissions than 3BEM_FRP while 3BEM_FRPyields more emissions in the area of soybean expansion eastof the Amazon forest Atmospheric aerosol optical thick-ness is simulated by using the emission inventories withtwo operational atmospheric chemistry transport models the

Published by Copernicus Publications on behalf of the European Geosciences Union



1 Introduction













nsumi=1













FRPMIR =Ag

aσ

403int376







FRPgrid(lonlatt) =

αsumγ=minusα

βsumκ=minusβ



212 GFAS



FRPad =

sumk


k







213 FINN

















































3BEM_FRP FINN GFAS








Total 605844 554547 422626 630814











4 Conclusions








Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References



1 Introduction













nsumi=1













FRPMIR =Ag

aσ

403int376







FRPgrid(lonlatt) =

αsumγ=minusα

βsumκ=minusβ



212 GFAS



FRPad =

sumk


k







213 FINN

















































3BEM_FRP FINN GFAS








Total 605844 554547 422626 630814











4 Conclusions








Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References



nsumi=1













FRPMIR =Ag

aσ

403int376







FRPgrid(lonlatt) =

αsumγ=minusα

βsumκ=minusβ



212 GFAS



FRPad =

sumk


k







213 FINN

















































3BEM_FRP FINN GFAS








Total 605844 554547 422626 630814











4 Conclusions








Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References




FRPgrid(lonlatt) =

αsumγ=minusα

βsumκ=minusβ



212 GFAS



FRPad =

sumk


k







213 FINN

















































3BEM_FRP FINN GFAS








Total 605844 554547 422626 630814











4 Conclusions








Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References












































3BEM_FRP FINN GFAS








Total 605844 554547 422626 630814











4 Conclusions








Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References







4 Conclusions








Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References






Edited by H Coe

References






















































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References









































Abstract

Introduction




GFAS

FINN







Conclusions

Acknowledgements

References

King s Research Portal Users may download and print one copy of any publication from ... of the...

Documents

Transcript of King s Research Portal Users may download and print one copy of any publication from ... of the...