NASA Technical Reports Server (NTRS) 20120017007: Desert Dust Satellite Retrieval Intercomparison

8.8 MB·English

by  NASA Technical Reports Server (NTRS)

Atmos.Meas.Tech.Discuss.,5,691–746,2012 Atmospheric Dis
www.atmos-meas-tech-discuss.net/5/691/2012/ Measurement c
u
doi:10.5194/amtd-5-691-2012 Techniques ss
©Author(s)2012.CCAttribution3.0License. Discussions io
n
P
a
p
Thisdiscussionpaperis/hasbeenunderreviewforthejournalAtmosphericMeasurement e
r
Techniques(AMT).PleaserefertothecorrespondingfinalpaperinAMTifavailable.
|
D
is
Desert dust satellite retrieval c
u
s
s
intercomparison io
n
P
a
p
e
E.Carboni1,G.E.Thomas1,A.M.Sayer1,2,11,R.Siddans2,C.A.Poulsen2, r
R.G.Grainger1,C.Ahn3,D.Antoine4,S.Bevan5,R.Braak6,H.Brindley7, |
S.DeSouza-Machado8,J.L.Deuze´9,D.Diner10,F.Ducos9,W.Grey5,C.Hsu11, D
O.V.Kalashnikova10,R.Kahn11,P.R.J.North5,C.Salustro11,A.Smith1, isc
D.Tanre´9,O.Torres11,andB.Veihelmann6,* us
s
io
1Atmospheric,OceanicandPlanetaryPhysics,ClarendonLaboratory,UniversityofOxford, n
P
Oxford,UK a
p
2SpaceScienceandTechnologyDepartment,RutherfordAppletonLaboratory,Harwell er
ScienceandInnovationCampus,Didcot,UK
3ScienceSystemsandApplications,Maryland,USA |
4Laboratoired’Oce´anographiedeVillefranche(LOV),CentreNationaldelaRecherche D
is
Scientifique(CNRS)andUniversite´ PierreetMarieCurie,Paris06,Villefranche-sur-Mer, c
u
France ss
5GeographyDepartment,CollegeofScience,SwanseaUniversity,UK ion
6RoyalNetherlandsMeteorologicalInstitute(KNMI),TheNetherlands P
a
7ImperialCollege,London,UK pe
r
691 |
D
is
c
u
s
8UniversityofMarylandBaltimoreCounty,USA sio
9LOA,Universite´ Lille-1,France nP
10JetPropulsionLaboratory,CaliforniaInstituteofTechnology,Pasadena,California,USA ap
e
11NASAGoddardSpaceFlightCenter,Greenbelt,Maryland,USA r
∗ now at: Science, Applications and Future Technologies Department, ESA/ESTEC, The |
Netherlands D
is
c
Received:30November2011–Accepted:28December2011–Published:17January2012 u
s
s
Correspondenceto:E.Carboni([email protected]) io
n
P
a
PublishedbyCopernicusPublicationsonbehalfoftheEuropeanGeosciencesUnion. pe
r
|
D
is
c
u
s
s
io
n
P
a
p
e
r
|
D
is
c
u
s
s
io
n
P
a
p
e
r
692 |
Abstract D
is
c
u
ThisworkprovidesacomparisonofsatelliteretrievalsofSaharandesertdustaerosol ss
optical depth (AOD) during a strong dust event through March 2006. In this event, a ion
large dust plume was transported over desert, vegetated, and ocean surfaces. The Pa
p
5 aim is to identify and understand the differences between current algorithms, and er
hence improve future retrieval algorithms. The satellite instruments considered are
|
AATSR,AIRS,MERIS,MISR,MODIS,OMI,POLDER,andSEVIRI.Aninterestingas-
pectisthatthedifferentalgorithmsmakeuseofdifferentinstrumentcharacteristicsto D
is
obtain retrievals over bright surfaces. These include multi-angle approaches (MISR, cu
s
AATSR), polarisation measurements (POLDER), single-view approaches using solar s
10 io
wavelengths(OMI,MODIS),andthethermalinfraredspectralregion(SEVIRI,AIRS). n
P
Differencesbetweeninstruments,togetherwiththecomparisonofdifferentretrievalal- a
p
gorithmsappliedtomeasurementsfromthesameinstrument,provideauniqueinsight e
r
intotheperformanceandcharacteristicsofthevarioustechniquesemployed. Aswell
|
astheintercomparisonbetweendifferentsatelliteproducts,theAODshavealsobeen
15
D
comparedtoco-locatedAERONETdata. Despitethefactthattheagreementbetween is
c
satellite and AERONET AODs is reasonably good for all of the datasets, there are u
s
significant differences between them when compared to each other, especially over sio
n
land. Thesedifferencesarepartiallyduetodifferencesinthealgorithms, suchasas-
P
20 sumptionsaboutaerosolmodelandsurfaceproperties. However,inthiscomparisonof ap
e
spatiallyandtemporallyaverageddata,atleastassignificantasthesedifferencesare r
samplingissuesrelatedtotheactualfootprintofeachinstrumentontheheterogeneous |
aerosolfield,cloudidentificationandthequalitycontrolflagsofeachdataset.
D
is
c
u
s
1 Introduction s
io
n
P
25 Desert dust is one of the most abundant and important aerosols in the atmosphere. a
p
Dust grain size and composition make it radiatively active over a wide spectral range e
r
693 |
(from the ultraviolet to the thermal infrared) and so airborne dust has a significant D
is
direct radiative forcing on climate (IPCC 2007 – ar4 2.4.1). Changes in land use can cu
s
result in an anthropogenic influence on the atmospheric burden of desert dust. Dust s
io
also has indirect radiative effects by acting as cloud condensation nuclei (CCN) and n
P
5 modifying precipitation. Iron transported by desert dust and deposited into the sea ap
affectsphytoplankton(Jickellsetal.,2005). e
r
Satellites can provide global measurements of desert dust and have particular im-
|
portanceinremoteareaswherethereisalackofinsitumeasurements. Desertdust
D
sourcesareofteninjustsuchpoorlyinstrumentedremoteareas. Satelliteaerosolre- is
c
10 trievalshaveimprovedconsiderablyinthelastdecadeandthenumberofrelatedpub- us
lications has correspondingly increased. However intercomparison exercises (Myhre sio
et al., 2005) have revealed that discrepancies between satellite measurements are n
P
particularly large during events of heavy aerosol loading. In the past, aerosol re- ap
e
trievals for satellite radiometers have typically made use of visible and near-infrared r
15 measurements,theinterpretationofwhichbecomesdifficultoverbrightsurfacessuch |
as deserts. To overcome these difficulties more recent algorithms make use of addi-
D
tionalinformationavailablefromcertaininstruments,forexamplemulti-angleobserva- is
c
tions,shorter(ultraviolet)wavelengths,thermalinfraredwavelengths,andpolarization. us
s
All algorithms must also make prior assumptions about aerosol composition and the io
n
20 propertiesoftheunderlyingsurfaces,astheretrievalofaerosolpropertiesisaninher- P
a
ently under-constrained optimisation problem. Instrumental observing capability and p
e
algorithm implementation (such as the use and formulation of prior information) give r
retrievalsthataresensitivetodifferentaspectsofthedustaerosolloading. Usingmea- |
surementsfromthesamesensor,largevariationsinAODcanbefoundbetweendiffer- D
25 entalgorithms(Kokhanovskyetal.,2007)evenintheidealisedcaseofanon-reflecting isc
u
surface (Kokhanovsky et al., 2010), or when only the assumed aerosol microphysical s
s
propertiesinaretrievalalgorithmarechanged(e.g.Bulginetal.,2011). io
n
Inthispaper,wereportresultsfromtheDesertdustRetrievalIntercomparison(DRI) P
a
project,whichperformedacomparisonofretrievalsforaSaharandesertdustepisode pe
r
694 |
in March, 2006 using data from a wide range of state-of-the-art schemes. This com- D
is
parison reveals the dependence of the results on approximations made by retrieval cu
s
algorithms, the accuracy of the aerosol model assumed, and the importance of good s
io
quality control. The aim of the study was to highlight and understand differences be- n
P
5 tweentheschemesinordertorecognizethestrengthsofparticularschemesandiden- ap
tify areas for improvement. The comparisons were performed separately over ocean e
r
(where satellite retrievals are less affected by problems in modelling the surface con-
|
tributiontothetopofatmospheresignal)andoverland(wheretheretrievalproblemis
D
morechallenging). Theprojectalsocompiledadatabaseofretrievalresultswhichcan is
c
10 beusedinfutureworktotestalgorithmimprovements. us
s
io
n
2 Datasets Pa
p
e
r
Most of the instrument radiance measurements considered here gave rise to more
than a single estimate of AOD through the application of different retrieval algo- |
rithms. Apixel-by-pixelanalysisofdifferentalgorithmsappliedtothesameinstrument D
is
or datasets from collocated pixels from different instruments, as MODIS and MISR c
15 u
(Mishchenko et al., 2010; Kahn et al., 2011), could lead to interesting results, but we ss
io
leave this to future work. Here the comparison of satellite datasets is limited to the n
P
spatiallyandtemporallyaverageddata. a
p
The different spatial and temporal sampling available from different satellite instru- e
r
ments means that a direct comparison at the individual pixel (field-of-view) level is
20
|
generally not possible. These differences in sampling can give differences in the re-
D
trieved AOD, particularly when the aerosol loading is spatially and/or temporally het-
is
erogeneous, orwherecloudfieldsmovebetweentheoverpassesofdifferentsensors cu
s
suchthatthecloud-freeareaimagedisnotthesame(e.g.Levyetal.,2009; Sayeret s
io
al., 2010b). Thequalitycontrolineachindividualretrievalalgorithmmaybedifferent, n
25 P
soevenretrievalsfromthesameinstrumentwillnotnecessarilyhavethesamecover- a
p
age. Tominimisetheeffectsofsamplingdifferencesontheintercomparison,dataare er
695 |
aggregated to daily temporal resolution and a relatively fine spatial grid. To facilitate D
is
comparisonseachdataproviderhasgiventwosetsofresults: cu
s
s
1. A daily AOD field formed by averaging individual retrievals onto a common spa- io
n
tialgrid,namelyahalfdegreeregularlyspacedgridinlatitudeandlongitude. The P
a
meanAODineachgridcellisprovided,alongwiththestandarddeviationofallin- p
5 e
r
dividualretrievedAODvaluesinthecellandthenumberofthesesamples. These
aggregated daily fields are directly compared (neglecting the fact that satellites |
may sample at different times of day). Most algorithms provide AOD at 550nm D
withtheexceptionofAIRS(900cm−1)andOMI-NASA(440nm). isc
u
s
s
10 2. For comparison with ground-based direct-Sun observations from the Aerosol io
n
RoboticNetwork(AERONET;Holbenetal.,1998),weusethemeanandstandard P
a
deviationofallindividualretrievalswithinaradiusof50kmofselectedAERONET p
e
sites,togetherwiththenumberofindividualretrievalsandthemeanobservation r
timeofthosesamples. |
D
15 Inthefollowingtexttheterm“dataset”isusedtorefertotheAODproducedbytheappli- is
c
cationofanamedretrievalalgorithmtomeasurementsbyaspecificinstrument. Com- u
s
parisons are restricted to a region enclosed by latitudes 0 and 45◦ N and longitudes sio
50◦Wand50◦E.Table1summarisesthedatasetsincludedandabriefdescriptionof n
P
eachdatasetisprovidedbelow,organisedbyinstrument. a
p
e
20 The differences in using desert dust optical properties computed from spherical or r
non-spherical model can lead to significant effects (Mishchenko et al., 2003). In this
|
comparison,theAATSR-ORAC,MISR,MODIS,POLDER-ocean,OMI-KNMIdatasets
D
include non-spherical optical models, the other datasets use only spherical models. is
c
This can lead to differences in retrieved AOD, especially for the datasets that make u
s
s
25 useofultravioletandvisiblewavelengths. Moreoverifdustismodeledassumingnon- io
n
spherical particles, then additional decisions need to be made as to the specific dis- P
a
tribution of particle shape(s) to use. These differences, which can be significant, will p
e
alsoaffectthecalculatedphasefunctionandsoretrievedAOD(e.g.Kalashnikovaand r
696 |
Sokolik,2002;Kalashnikovaetal.,2005). Theseeffectsarenotanalyzedinthepresent D
is
paper (due to the complexity added by multi-angle retrievals), but for future research cu
s
we suggest that an analysis of AOD differences between datasets as a function of s
io
scatteringanglecouldhelpisolatephasefunctioneffects. n
P
a
p
5 2.1 AATSR er
2.1.1 Globaerosol |
D
The GlobAEROSOL project (http://www.atm.ox.ac.uk/project/Globaerosol/) was car- isc
u
ried out as part of the European Space Agency’s Data User Element programme. s
s
All the products are on a common 10km sinusoidal grid and together provide almost io
n
continuouscoveragefor1995–2007. Theinstrumentsusedbytheprojectarethesec- P
10 a
ond Along Track Scanning Radiometer (ATSR-2), the Advanced ATSR (AATSR), the pe
r
MediumResolutionImagingSpectrometer(MERIS),andtheSpinningEnhancedVisi-
bleInfraredImager(SEVIRI). |
TheDRIintercomparisonhasmadeuseoftheAATSRGlobAEROSOLproductde- D
is
rivedusingtheOxford-RutherfordAppletonLaboratory(RAL)RetrievalofAerosoland c
15 u
Cloud(ORAC)optimalestimationscheme. Afulldescriptionoftheretrievalisgivenby ss
io
Thomasetal.(2009). ORACmakesuseoftheATSRnadirandforwardviewchannels n
P
centred at 0.55, 0.67, 0.87 and 1.6µm. The forward model includes a bidirectional a
p
reflection distribution function (BRDF) description of the surface reflectance. By con- e
r
straining the relative strengths of the direct, hemispherical and bi-hemispherical sur-
20
|
facereflectancetheaerosolopticaldepth,effectiveradiusandbi-hemisphericalsurface
D
albedoineachchannelareretrieved. Theapriorisurfacereflectanceisdeterminedby is
the MODIS BRDF product (MCD43B1 Collection 5.0) over land (Schaaf et al., 2002) cu
s
andbyanoceansurfacereflectancemodel(Sayeretal.,2010a)overtheocean. s
io
Retrievalsareperformedforeachoffivepredefinedaerosoltypes: desertdust,mar- n
25 P
itime clean, continental clean, urban (all using component optical properties from the a
p
OpticalPropertiesofAerosolsandClouds(OPAC)databaseofHessetal.,1998)and er
697 |
biomass-burning (Dubovik, 2006). From these five results a best match is selected D
is
basedonthequalityofthefittothemeasurementsandaprioriconstraints,providing cu
s
acrudespeciationoftheaerosol. s
io
n
2.1.2 ORAC P
a
p
e
r
The AATSR-ORAC dataset represents an updated version of the aerosol retrieval al-
5
gorithmusedinAATSR-GlobAerosol. Threemajoraspectsofthealgorithmhavebeen |
improved,describedindetailbyLean(2009)andSayeretal.(2012): D
is
c
– Improved surface reflectance treatment. The error budget of the MODIS BRDF u
s
products used to generate the a priori surface albedo has been improved, and sio
n
a correction algorithm applied to account for the differences between the visible
10 P
channelspectralresponsefunctionsoftheMODISandAATSRinstruments. ap
e
r
– Implementation of aerosol type flags (volcanic ash, biomass burning over land,
|
desert dust over sea) to identify aerosol pixels misclassified as cloudy by the
suppliedcloudflag. D
is
c
u
15 – Development of a new aerosol microphysical model for desert dust. This uses ss
the same refractive indices (derived from OPAC components) as in the AATSR- io
n
GlobAerosol retrieval, but treats the particles as spheroids using T-matrix code P
a
(Mishchenkoetal.,1997,1998)ratherthanspheres. Amodifiedlognormaldistri- p
e
r
bution of spheroid aspect ratios (the ratio between major and minor axis length)
asgivenbySect.3.4ofKandler(2007)isusedwithequalnumbersofoblateand |
20
prolatespheroids. D
is
c
u
2.1.3 Swansea s
s
io
n
TheSwanseaUniversityretrievalalgorithmhasbeendesignedtoretrievetheaerosol P
a
optical thickness and type, and surface reflectance over both land and ocean. The p
e
treatment of atmospheric radiative transfer is by look-up table (LUT) using the scalar r
25
698 |
version of 6S code (Vermote et al., 1997; Grey et al., 2006a). Five aerosol models D
is
are represented: two coarse mode (oceanic, desert) and three fine mode (biomass cu
s
burning, continental and urban). The optimum value of AOD and aerosol model is s
io
selected by iterative inversion based on fit to a model of surface reflectance. Over n
P
5 ocean the algorithm uses the low spectral reflectivity at near and mid infra-red chan- ap
nelstoconstrainaerosolretrieval(Greyetal., 2006b; Bevanetal., 2011). Overland, e
r
the algorithm uses the AATSR dual-view capability to estimate aerosol without prior
|
assumptions of land surface spectral properties, based on inversion of a simple pa-
D
rameterized model of surface anisotropy (North et al., 1999; North, 2002; Davies et is
c
10 al., 2010). This model defines spectral variation of reflectance anisotropy accounting us
forvariationindiffuselightfromtheatmosphereandmultiplescatteringatthesurface. sio
Cloud clearing is based on instrument flags enhanced by the cloud detection system n
P
developedbyPlummer(2008). ap
e
The retrieval procedure was implemented within ESA’s Grid Processing on De- r
15 mand(GPOD)high-performancecomputingfacilityforglobalretrievalsofAODandbi- |
directional reflectance from ATSR-2 and AATSR, at 10km resolution, and these data
D
are used in the current study. Global validation with AERONET and other satellite is
c
sensorswaspresentedbyGrey(2006b)andBevanetal.(2011). Bevanetal.(2009) us
s
performedvalidationofGPODATSR-2andexploredtheimpactofatmosphericaerosol io
n
20 frombiomassburningintheAmazonregionoverthefull13-yrATSR-2/AATSRdataset. P
a
FurtherdetailsonthealgorithmaregiveninGreyandNorth(2009). p
e
r
2.2 AIRS |
2.2.1 JCET D
is
c
Operational since September 2002, the Atmospheric Infrared Sounder (AIRS) instru- u
s
25 ment (Aumann et al., 2003) on NASA’s Aqua satellite provides data for temperature sio
n
andhumidityprofiles,usedinnumericalweatherprediction. AIRShas2378channels,
P
coveringthespectralrange649–1136,1217–1613,2181–2665cm−1. Eachcrosstrack ap
e
swathconsistsof90pixels,withafootprintof15kmatnadir. r
699 |
Upwellingradiancesinthe8–12µmthermalinfrared(TIR)atmosphericwindoware D
is
measured with a large number of high-resolution, low noise channels, making it pos- cu
s
sibletodetectsilicatebasedaerosols(DeSouza-Machadoetal., 2006, 2010)dayor s
io
night,overoceanorland. Adustdetectionalgorithmhasbeendevelopedusingbright- n
P
5 ness temperature differences (BTDs) for a set of 5 AIRS channels in the TIR region. ap
The algorithm is based on simulations of dust-contaminated radiances for numerous e
r
atmospheric profiles over ocean using dust refractive indices from Volz (1973). This
|
flagwasdesignedtodetectdustovertropicalandmidlatitudeoceans(whichhaveno
D
cloudcoveroverthedust),andcanbemodifiedtoworkoverlandsurfaces. is
c
10 TheretrievalusesamodifiedversionoftheAIRSRadiativeTransferAlgorithm(AIRS- us
RTA) which computes radiative transfer through a dusty atmosphere (Chou et al., sio
1999). The AIRS-RTA assumes a plane parallel atmosphere divided into 100 layers, n
P
withthedustprofileoccupyingoneormoreconsecutivepressurelayers. ap
e
Thealgorithmassumesknowledgeoftheeffectiveparticlesizeanddusttop/bottom r
15 height. Here an effective particle diameter of 4µm for a lognormal distribution is |
adopted. The dust height comes from GOCART climatology (Ginoux et al., 2001).
D
Giventhis,alinearizedNewton-RaphsonmethodfitsforthecolumndustloadingΓ(in is
c
gm2)tominimizeaχ2 leastsquarefitofbrightnesstemperatures(BTs)inthewindow us
s
regions. ThedustloadingisrelatedtotheTIRdustAODτ by io
n
P
τ(v)=σ (v,r )Γ (1) a
20 dustmodel mode p
e
Here σdustmodel(v,rmode) is the mass extinction efficiency in m2g−1. It is important r
to note that the retrieved TIR AOD depends critically on the assumed particle height. |
Comparisons show that when the heights are correct, AIRS AODs have a very high D
is
correlationagainstMODISandPOLDERAODs,especiallyovertheocean(DeSouza- c
u
s
25 Machadoetal.,2010). Thiscorrelationdropsnoticeablywhentheheightsareincorrect. sio
n
P
a
p
e
r
700 |
2.3 OMI D
is
c
u
2.3.1 NASA-GSFC ss
io
n
The first step in the OMAERUV algorithm is the calculation of the UV Aerosol Index P
a
(UVAI)asdescribedinTorresetal.(2007). TheinformationcontentoftheOMIUVAIis p
e
r
turnedintoquantitativeestimatesofaerosolextinctionopticaldepthandsinglescatter-
5
ingalbedoat388nmbyapplicationofaninversionalgorithmtoOMInear-UVobserva- |
tions at 354 and 388nm (Torres et al., 2007). These aerosol parameters are derived D
is
by an inversion algorithm that uses pre-calculated reflectances for a set of assumed c
u
aerosol models. A climatological data-set of near-UV surface albedo derived from ss
io
10 long-termTOMS(TotalOzoneMappingSpectrometer)observationsisusedtocharac- n
terizesurfacereflectiveproperties. Threemajoraerosoltypesareconsidered: desert Pa
p
dust,carbonaceousaerosolsassociatedwithbiomassburning,andweaklyabsorbing e
r
sulfate-basedaerosols. Theselectionofanaerosoltypemakesuseofacombination
|
of spectral and geographic considerations (Torres et al., 2007). The aerosol models
particle size distributions were derived from long term AERONET statistics (Torres et D
15 is
al.,2007). cu
s
Sincetheretrievalprocedureissensitivetoaerosolverticaldistribution,theaerosol s
io
layerheightisassumedbasedonaerosoltypeandgeographiclocation. Carbonaceous n
P
aerosollayersbetweenwithin30◦ oftheEquatorareassumedtohavemaximumcon- ap
20 centration at 3km above ground level, whereas mid and high-latitude (pole wards of er
±45◦)smokelayersareassumedtopeakat6km. Theheightofsmokelayersbetween |
30◦ and45◦ latitudeinbothhemispheresisinterpolatedwithlatitudebetween3and6
D
km. The location of desert dust aerosol layers varies between 1.5 and 10km, and is is
c
givenbyamulti-yearclimatologicalaverageofChemicalModelTransportcalculations u
s
25 usingtheGOCARTmodelatalat-lonresolutionof2◦ 2.5◦ (Ginouxetal.,2001). For sio
× n
sulfate-based aerosols the assumed vertical distribution is largest at the surface and
P
decreasesexponentiallywithheight. ap
e
r
701 |
For a chosen aerosol type and assumed aerosol layer height, the extinction optical D
is
depth and single scattering albedo at 388nm are retrieved and aerosol absorption c
u
s
optical depth is calculated. Results are also reported at 354 and 500nm to facilitate s
io
comparisonswithmeasurementsfromotherspace-borneandgroundbasedsensors. n
P
5 Aerosol parameters over land are retrieved for all cloud-free scenes as determined ap
byaninternalcloudmask. Retrievalsovertheocean,however,arelimitedtocloud-free e
r
scenescontainingabsorbingaerosols(i.e.smokeordesertdust)asindicatedbyUVAI
|
values larger than unity. Since the current representation of ocean surface effects in
D
theOMAERUValgorithmdoesnotexplicitlycorrectforoceancolorsignal,theretrieval is
c
10 ofaccuratebackgroundmaritimeaerosolisnotcurrentlypossible. us
Algorithmqualityflagsareassignedtoeachpixel. MostreliableOMAERUVretrievals sio
haveaqualityflag0. Qualityflag1indicatessub-pixelcloudcontamination. Forquanti- n
P
tativeapplicationsusingOMAERUVderivedaerosolopticaldepthandsinglescattering ap
e
albedoonlydataofqualityflag0isrecommended. Forthiscomparisonflag0datahave r
15 beenextrapolateto440nm. |
D
2.3.2 KNMI is
c
u
s
The OMI multi-wavelength algorithm OMAERO (Torres et al., 2002) is used to de- s
io
riveaerosolcharacteristicsfromOMIspectralreflectancemeasurementsofcloud-free n
P
scenes. Undercloud-freeconditions, OMIreflectancemeasurementsaresensitiveto a
p
20 the aerosol optical depth, the single-scattering albedo, the size distribution, altitude er
of the aerosol layer, and the reflective properties of the surface. However, from a
|
principal-componentanalysisappliedtosyntheticreflectancedata(Veihelmannetal.,
D
2007), it was shown that OMI spectra contain only two to four degrees of freedom is
c
of signal. Hence, OMI spectral reflectance measurements do not contain sufficient u
s
25 information to retrieve all aerosol parameters independently. The OMAERO level-2 sio
n
data product reports aerosol characteristics such as the AOD, aerosol type, aerosol
P
absorption indices as well as ancillary information. The AOD is retrieved from OMI ap
e
spectralreflectancemeasurementsandabest-fittingaerosoltypeisdetermined. The r
702 |
single-scatteringalbedo,thelayeraltitudeandthesizedistributionassociatedwiththe D
is
best-fittingaerosoltypearereported. c
u
s
Cloudy scenes are excluded from the retrieval using three tests. The first test is s
io
based on reflectance data in combination with the UV absorbing aerosol index. The n
P
5 second test uses cloud fraction data from the OMI O2-O2 cloud product OMCLDO2 ap
(Acarreta and Hann, 2002; Acarreta et al., 2004; Sneep et al., 2008). The third test e
r
isbasedonthespatialhomogeneityofthescene. Thelattertestisthemoststrictfor
|
screeningcloudsinthecurrentimplementationofthealgorithm.
D
The OMAERO algorithm evaluates the OMI reflectance spectrum in a set of fifteen is
c
10 wavelength bands in the spectral range between 330 and 500nm. The wavelength us
bands are about 1nm wide and were chosen such that they are essentially free from sio
gas absorption and strong Raman scattering features, except for a band at 477nm, n
P
which comprises an O2-O2 absorption feature. The sensitivity to the layer altitude ap
e
andsingle-scatteringalbedoisrelatedtotherelativelystrongcontributionofRayleigh r
15 scatteringtothemeasuredreflectanceintheUV(Torresetal.,1998). Theabsorption |
band of the O2-O2 collision complex at 477nm is used in OMAERO to enhance the
D
sensitivitytotheaerosollayeraltitude(Veihelmannetal.,2007). is
c
The multi-wavelength algorithm uses forward calculations for a number of micro- us
s
physical aerosol models that are defined by the size distribution and the complex re- io
n
20 fractiveindex,aswellastheAODandtheaerosollayeraltitude. Themodelsarerep- P
a
resentative for the main aerosol types of desert dust, biomass burning, volcanic and p
e
weaklyabsorbingaerosol. Severalsub-typesormodelsrepresenteachofthesemain r
types. Syntheticreflectancedatahavebeenpre-computedforeachaerosolmodelus- |
ingtheDoubling-AddingKNMIprogram(DeHaanetal.,1987; Stammesetal.,1989; D
25 Stammes,2001),assumingaplane-parallelatmosphereandtakingintoaccountmul- isc
u
tiple scattering as well as polarization. For land scenes the surface albedo spectrum s
s
is taken from a global climatology that has been constructed using Multi-angle Imag- io
n
ing Spectroradiometer (MISR) data measured in four bands (at 446, 558, 672, and P
a
866nm)thatare extrapolatedtotheUV. Foroceansurfaces thespectralbidirectional pe
r
703 |
reflectancedistributionfunctioniscomputedusingamodelthataccountsforthechloro- D
is
phyllconcentrationoftheoceanwaterandthenear-surfacewindspeed(Veefkindand cu
s
deLeeuw,1998). s
For each aerosol model, an AOD is determined by minimizing the χ2 merit func- ion
P
5 tion obtained with the spectra of measured reflectances, the computed reflectances ap
(functionoftheAOD),andtheerrorinthemeasuredreflectances. e
r
The aerosol model with the smallest value of χ2 is selected and the corresponding
|
AODatfourteendifferentwavelengthsisreportedastheretrievedAOD.Otherreported
D
parameters are the single-scattering albedo, the size distribution and the aerosol alti- is
c
10 tudethatareassociatedwiththeselectedaerosolmodel. us
Aerosol models are post-selected based on a climatology of geographical aerosol sio
distribution(Curieretal., 2008). TheaccuracyoftheAODretrievedbytheOMAERO n
P
algorithm is estimated to be the larger of 0.1 or 30% of the AOD value. This is an ap
e
errorestimatethatwasalsousedfortheTOMSaerosolalgorithm(Torresetal.,2005). r
15 MoreinformationontheOMAEROalgorithmanddataproductmaybefoundinTorres |
etal.(2007).
D
is
c
2.4 MISR u
s
s
io
2.4.1 JPL/GSFC n
P
a
p
The Multi-angle Imaging SpectroRadiometer (MISR) was launched into a sun- e
r
synchronous polar orbit in December 1999, aboard the NASA Earth Observing Sys-
20
|
tem’sTerrasatellite. MISRmeasuresupwellingshort-waveradiancefromEarthinfour
D
spectral bands centred at 446, 558, 672, and 866nm, at each of nine view angles
is
spreadoutintheforwardandaftdirectionsalongtheflightpath,at70.5◦,60.0◦,45.6◦, cu
s
26.1◦, and nadir (Diner et al., 1998). Over a period of seven minutes, as the space- sio
craft flies overhead, a 380-km-wide swath of Earth is successively viewed by each n
25 P
of MISR’s nine cameras. As a result, the instrument samples a very large range of a
p
scatteringangles(betweenabout60◦ and160◦ atmidlatitudes),providinginformation er
704 |
about aerosol microphysical properties. These views also capture air-mass factors D
is
rangingfromonetothree,offeringsensitivitytoopticallythinaerosollayers,andallow- cu
s
ingaerosolretrievalalgorithmstodistinguishsurfacefromatmosphericcontributionsto s
io
thetop-of-atmosphere(TOA)radiance. Globalcoverage(to 82◦ latitude)isobtained n
± P
5 aboutonceperweek. ap
The MISR Standard aerosol retrieval algorithm reports AOD and aerosol type at e
r
17.6km resolution, by analyzing data from 16 16 pixel regions of 1.1km-resolution,
× |
MISRtop-of-atmosphereradiances(Kahnetal.,2009a). Overdarkwater,operational
D
retrievalsareperformedusingthe672and867nmspectralbands,assumingaFresnel- is
c
10 reflecting surface and standard, wind-dependent glint and whitecap ocean surface us
models. Coupled surface-atmosphere retrievals are performed using all four spectral sio
bands over most land, including bright desert surfaces (Martonchik et al., 2009), but n
P
notoversnowandice. ap
e
MISRAODhasbeenvalidatedandusedovermanydesertsurfaces(Martonchiket r
15 al., 2004; Christopher et al., 2008, 2009; Kahn et al., 2009b; Koven and Fung, 2008; |
Xia et al., 2008, 2009), as well as other, less challenging environments. Sensitivity
D
to AOD and particle properties varies with conditions; at least over dark water, un- is
c
der good retrieval conditions and mid-visible AOD larger than about 0.15, MISR can us
s
distinguishaboutthree-to-fivegroupingsbasedonparticlesize,two-to-fourgroupings io
n
20 in single-scattering albedo (SSA), and spherical vs. non-spherical particles (Chen et P
a
al., 2008; Kalashnikova and Kahn, 2006). The algorithm identifies all mixtures that p
e
meet the acceptance criteria from a table of mixtures, each composed of up to three r
aerosol components; the same mixture table is applied for all seasons and locations, |
overbothlandandwater. Version22oftheMISRStandardAerosolProduct, usedin D
25 this study, contains 74 mixtures and eight components (Kahn et al., 2010), including isc
u
amedium-mode,non-sphericaldustopticalanaloguedevelopedfromaggregated,an- s
s
gularshapesandacoarse-modedustanaloguecomposedofellipsoids(Kalashnikova io
n
etal.,2005). P
a
p
e
r
705 |
2.5 MERISandSEAWIFS D
is
c
u
2.5.1 LOV ss
io
n
The MERISalgorithm (Antoine and Morel, 1999) isa full multiple scatteringinversion P
a
scheme using aerosol models and pre-computed look-up tables (LUTs). It uses the p
e
r
path reflectances in the near infrared, where the contribution of the ocean is null, as
5
well as visible reflectances, where the marine contribution is significant and varying |
withthechlorophyllcontentofoceanicwater. AtechniquewasproposedbyNobileau D
is
andAntoine(2005)toovercomethedifficultyindiscriminatingbetweenabsorbingand c
u
non-absorbing aerosols. In the present regional application, a climatology is used ss
io
10 for water reflectance and error as described in Antoine and Nobileau (2006). After n
absorption has been detected, the atmospheric correction is restarted using specific Pa
p
setsofabsorbingaerosolmodels(i.e.specificLUTs). Theaerosolopticalthicknessat e
r
allwavelengthsandtheA˚ngstro¨mexponentarethenderived.
|
Fornon-absorbingaerosol,asetoftwelveaerosolmodelsisusedfromShettleand
Fenn(1979)andGordonandWang(1994). Thissetincludesfourmaritimeaerosols, D
15 is
fourruralaerosolsthataremadeofsmallerparticles,andfourcoastalaerosolsthatare cu
s
amixingbetweenthemaritimeandtheruralaerosols. Themeanparticlesizesofthese s
io
aerosols, and thus their optical properties, vary as a function of the relative humidity, n
P
whichissetto50,70,90and99%(hencethe3timesfourmodels). Inadditiontothese a
p
20 boundary-layeraerosols,constantbackgroundsareintroducedinthefreetroposphere er
(2–12km),withacontinentalaerosolAODof0.025at550nm(WCRP,1986)andinthe
|
stratosphere(12–30km),withH SO aerosolAODof0.005at550nm(WCRP,1986).
2 4 D
Fortheabsorbingcase,thelook-uptablesusethesixdustmodelsandthethreever- is
c
ticaldistributionsproposedbyMoulinetal.(2001),whichwerederivedasthemostap- u
s
25 propriatetoreproducetheTOAtotalradiancesrecordedbySeaWiFSabovethickdust sio
plumesoffwesternAfrica. ThemeanA˚ngstro¨mexponentofthesemodelsisabout0.4 n
P
whencomputedbetween443and865nm. Whentheseaerosolsarepresent,aback- ap
e
groundofmaritimeaerosolismaintained,usingtheShettleandFenn(1979)maritime r
706 |
modelforarelativehumidityof90%andanopticalthicknessof0.05at550nm(Kauf- D
is
manetal., 2001). Thebackgroundsinthefreetroposphereandthestratosphereare cu
s
unchanged. s
io
Aspecifictestusingthebandat412nmwasdevelopedinordertoeliminateclouds n
P
5 without eliminating thick dust plumes (see also Nobileau and Antoine, 2005), which ap
are quite bright in the near infrared and therefore are eliminated when using a low e
r
thresholdinthiswavelengthdomain(asdoneforinstanceinthestandardprocessing
|
of the SeaWiFS observations). The same algorithm is applied to SeaWiFS. In this
D
case,specificlook-uptablesareusedthatcorrespondtotheSeaWiFSbandset. This is
c
10 istheonlydifferenceascomparedtotheMERISversion. us
s
io
2.6 MODIS n
P
a
p
2.6.1 NASA-GSFC e
r
InthisintercomparisontheDeepBlueretrievalhavebeenconsideredandthisdataset |
include only data over land. The principal concept behind the Deep Blue algorithm’s D
is
retrieval of aerosol properties over surfaces such as arid and semi-arid takes advan- c
15 u
tageofthefactthat,overtheseregions,thesurfacereflectanceisusuallyverybrightin ss
io
theredpartofthevisiblespectrumandinthenearinfrared,butismuchdarkerinthe n
P
blue spectral region (i.e. wavelength less then 500nm). In order to infer atmospheric a
p
properties from these data, a global surface reflectance database of 0.1◦ latitude by er
20 0.1◦ longitude resolution was constructed over land surfaces for visible wavelengths
|
usingtheminimum reflectivitytechnique(forexample, findingtheclearestscenedur-
D
ingeachseasonforagivenlocation). ForMODIScollection5.1DeepBlueproducts,
is
thesurfaceBRDFeffectsaretakenintoaccountbybinningthereflectivityvaluesinto cu
s
variousviewinggeometries. s
io
CloudmasksusedintheDeepBluealgorithmaredifferentfromthestandardMODIS n
25 P
cloudmasks. Theyaregeneratedinternallyandconsistof3steps: (1)determiningthe a
p
spatial variance of the 412nm reflectance; (2) using the visible aerosol index (412– er
707 |
470nm) to distinguish heavy dust from clouds; and (3) detecting thin cirrus based on D
is
the 1.38 micron channel reflectance. After cloud screening, the aerosol optical depth cu
s
and aerosol type are then determined simultaneously in the algorithm using look-up s
io
tables to match the satellite observed spectral radiances. The final products include n
5 spectralaerosolopticaldepth,A˚ngstro¨mexponent,aswellassinglescatteringalbedo Pap
fordust. MoreinformationonthedeepbluealgorithmcanbefoundatHsuetal.(2004) e
r
andHsuetal.(2006).
|
2.7 SEVIRI D
is
c
u
2.7.1 Globaerosol ss
io
n
TheDRIintercomparisonhasmadeuseoftheSEVIRIGlobAEROSOLproduct. Thisis P
10 a
derivedusingtheOxford-RALAerosolandCloudoptimalestimationretrievalscheme. pe
r
AfulldescriptionoftheretrievalisgivenbyThomas(2009). Itmakesuseofthe0.64,
0.81and1.64micronchannelsofSEVIRIat10:00,13:00and16:00UTC.Theforward |
modelincludesaBRDFdescriptionofthesurfacereflectanceandbyconstrainingthe D
is
relativestrengthsofthedirect,hemisphericalandbi-hemisphericalsurfacereflectance c
15 u
toapriorivalues,theaerosolopticaldepth,effectiveradiusandthesurfacereflectance ss
io
ineachchannelcanberetrieved. Theapriorisurfacereflectanceisdeterminedbythe n
P
16-day MODIS BRDF product over land and by an ocean surface reflectance model a
p
overtheocean. e
r
Retrievalsaredoneforeachoffivepredefinedaerosoltypes: desertdust,maritime
20
|
clean,continentalclean,andurban,allusingcomponentspropertiesfromtheOptical
D
PropertiesofAerosolsandClouds(OPAC)database(Hessetal.,1998)andbiomass-
is
burning (from Dubovik et al., 2006). From these five results a best match is selected cu
s
basedonthequalityofthefittothemeasurementsandaprioriconstraints,providing s
io
acrudespeciationoftheaerosol. n
25 P
a
p
e
r
708 |
2.7.2 ORAC D
is
c
u
The SEVIRI-ORAC dataset represents an updated version of the aerosol retrieval al- ss
gorithm used in SEVIRI-GlobAEROSOL. The main difference is the addition of the ion
infrared channels that allow the SEVIRI retrieval over bright surface. A simultane- Pa
p
5 ousaerosolretrievalthatconsidersthevisible,near-infraredandmid-infaredchannels er
(0.64, 0.81, 1.64, 10.78, 11.94 micron) is used, as described in detail by Carboni et
|
al.(2007).
ThemaindifferencefromtheGlobaerosolalgorithmistheadditionof2IRchannels D
is
around 11 and 12 microns, (assuming surface emissivity equal to ocean emissivity). cu
s
For these infrared channels ECMWF profile and skin temperature are used to define s
10 io
theclearskyatmosphericcontributiontothesignal. TogetherwithAODat550nmand n
P
effective radius, the altitude of the aerosol layer and surface temperature are part of a
p
the state vector of the retrieved parameters. The retrieval is run only with the desert er
dustaerosolclass(sphericalparticles): thiswillproduceerrorsinnondustconditions,
|
andinparticularwillproduceanoverestimationofAODincleanconditionsandmore
15
D
scatteringaerosoltype(nondust). InthisdatasetonlytheSEVIRIscenesacquiredat is
c
12:12 UT are analysed, to allow the maximum thermal contrast and with no need for u
s
interpolationofECMWFdata. sio
n
A data cut is then performed, excluding pixels that result in AOD greater than 4.9,
P
20 AODlessthe0.01,effectiveradiusgreaterthe3µm,brightnesstemperaturedifference ap
e
at 11–12 microns greater than 1.2K, cost function greater than 15 and pixels where r
theretrievalisnotconverging. |
D
2.7.3 IMPERIAL is
c
u
s
TwodifferentretrievalschemesareemployeddependingonwhethertheSEVIRIobser- s
io
vationsaretakenoverlandoroverocean. Inbothcasesretrievalsareonlyperformed n
25 P
if the scene is designated non-cloudy. Over ocean, the cloud detection scheme de- a
p
scribedbyIpeetal.(2004)isutilisedinconjunctionwithasubsequenttesttorestore er
709 |
anydustypointsincorrectlyflaggedascloud(BrindleyandRussell,2006). Overland, D
is
theschemeofIpeetal.(2004)issupplementedbytheclouddetectionduetoDerrien cu
s
and Le Gleau (2005). Again, dusty points incorrectly flagged as cloud are restored s
io
basedonthethresholdtestsdevelopedundertheauspicesoftheSatelliteApplication n
P
5 FacilityforNowcasting(Meteofrance,2005). ap
e
r
2.7.4 ImperialVIS
|
Over ocean, optical depths at 0.6, 0.8 and 1.6 microns are obtained independently D
is
fromtherelevantchannelreflectancesaccordingtothealgorithmdescribedinBrindley c
u
andIgnatov(2006). Briefly,thisschemeinvolvestheuseofreflectancelook-uptables ss
io
10 (LUTs)derivedasafunctionofsolar/viewinggeometryandaerosolopticaldepth. For n
a given sun-satellite geometry and channel, the retrieved optical depth is that which Pa
p
minimises the residual between the observed and simulated reflectance. One fixed e
r
“semi-empirical” aerosol model is used in the construction of the LUTs, matching the
|
representationoriginallyemployedintheretrievalschemedevelopedfortheAdvanced
Very High Resolution Radiometer (AVHRR) (Ignatov and Stowe, 2002). Using the D
15 is
optical depths derived from the different channels, one can also obtain estimates of cu
A˚ngstro¨m coefficients: these can subsequently be used to scale the retrievals to al- ss
io
ternative wavelengths as required. De Paepe et al. (2008) show that retrievals using n
P
this method exhibit RMS differences with co-located MODIS optical depths that are a
p
20 typicallylessthan0.1. er
|
2.7.5 ImperialIR
D
is
Over land, the lack of contrast between aerosol and surface reflectance in the solar c
u
bandsmakesitdifficulttousethesealonetoobtainaquantitativemeasureofaerosol ss
io
loading. Instead a relatively simple method is used which relates dust-induced vari- n
P
25 ations in SEVIRI 10.8 and 13.4 micron brightness temperatures to the visible optical a
p
depth (Brindley and Russell, 2009). This technique essentially builds on the method e
r
710 |