EMPIRICAL POLARIZATION DISTRIBUTION MODELS: UPDATE• Considered PDMs at least 2 std. dev. away from...
Transcript of EMPIRICAL POLARIZATION DISTRIBUTION MODELS: UPDATE• Considered PDMs at least 2 std. dev. away from...
EMPIRICALPOLARIZATIONDISTRIBUTIONMODELS:UPDATE
DANIEL GOLDIN (SSAI/NASA-LARC), CONSTANTINE LUKASHIN (NASA-LARC), WENBO SUN (SSAI/NASA-LARC)
DanielGoldin POLARIZATION:RECAP
• PolarizationstatefullyspecifiedbydegreeofpolarizationP,angleofpolarizationc andtotalintensityI (alternatively,maybespecifiedbyStokesparametersI,Q,U)
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
3
Imager’ssensitivitytopolarization
Totalreflectance
Polarizedradiance
Totalradiance where
CLARREO’sownaccuracy UncertaintiesinmandP
• (Mean)reflectanceneedstobecorrectedforpolarizationeffects:
• Uncertaintyduetopolarizationcontributestouncertaintyinreflectance:
• m isafunctionofc.SunandXiong[2007]haveshowncyclical dependenceforMODIS.TheexactdependencewouldbeestablishedbyCLARREOinstrument.Fornow,thisdependenceisfoldedintodm(usemeandm)
DanielGoldin EMPIRICALPDMSFROMPARASOL
• PARASOLwastheonlyinstrumentthatprovidedpolarizationmeasurementsonorbit
• 3wavelengthsavailablefromPARASOL:470,670and865nm
• Goal:constructP andc PDMsforthe3bandsforvariousscenetypes(IGBPs,clear-sky,cloudy,aerosols)
• UseinterpolationtoconstructP andc PDMsbetweenthe3bands
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
4
DanielGoldin RECENTRESULTS
• Finished2Dfitsforthehighestpolarizationscenetypes
• PDMfits,advantages:
• Aswithanyfits,PDMfitsusefultosmoothoutstatisticalfluctuationsandfillgapsindata
• Compact(only7-8parameters)anduniversal(appliedtoanyscenetype)• Robust,evenforlow-statisticsPDMs
• Usefits(meansandfiterror)toempiricalP andc PDMsinthefinalmodule
• Implementedaworkingversion(C++/C)ofthePDMmoduletoretrievedegreeofpolarizationP andangleofpolarizationc,withcorrespondingstd.dev., based on PARASOL data
• Workingonlowerpolarizationscenes(cloudyscenetypes)
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
5
DanielGoldin PDMCLASSIFICATION
• Computedtotalmeansandstd.devs forPPDMs
• Initiallyconsideredhighestpolarizationscenarios:
• Pickedshortestavailablewavelength(l =490nm)
• SZA=40(closetothetypicalrangeofBrewster’sangles)
• ConsideredPDMsatleast2std.dev.awayfrom0
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
6
IGBP Surface Type Pmean
P std. dev.
1 Evergreen needle-leaf forest 0.19 0.11
2 Evergreen broad-leaf forest 0.26 0.07
3 Deciduous needle-leaf forest 0.14 0.11
4 Deciduous broad-leaf forest 0.20 0.11
5 Mixed forest 0.16 0.12
6 Closed shrubland 0.18 0.12
7 Open shrubland 0.17 0.10
8 Woody savannas 0.17 0.12
9 Savannas 0.23 0.06
10 Grasslands 0.18 0.09
11 Permanent wetlands 0.16 0.13
12 Croplands 0.20 0.08
13 Urban and Built-up 0.24 0.09
14 Cropland Mosaics 0.21 0.09
15 Permanent snow and ice
16 Bare soil and rocks 0.16 0.06
17 Water Bodies 0.31 0.08
18 Tundra
19 Fresh Snow
20 Sea Ice
DanielGoldin PPDMFITFUNCTION
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
7
Gaussiansdescribing(mostly)glintregion
basedon1st orderscatteringapproximation(awayfromglint)
• Performac2 fitonP PDM:
DanielGoldin PPDMFITFORIGBP=9(l =670nm):STD.DEVS
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
9
70%C.L.(≈1s)uncertainty
DanielGoldin PPDMS:SUMMARY
• 2Dfitsprovidegood(DP ≈+/-0.1)approximationtoPPDMsmeansandstd.devs.
• FinalempiricalPPDMsrecorded(TBD)intheformofthefitcoefficientstof(f,q)orbinnedvaluesoff(f,q) andper-binfituncertainties
11/28/16
Sprin
g20
16CLARR
EOSDT
Meetin
g,Univ.ofM
ichigan
14
DanielGoldin c:SINGLESCATTERINGAPPROXIMATION
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
15
• Performac2 fitonc PDM:
DanielGoldin PPDMFITFORIGBP=14(l =865nm,50o <SZA<60o )
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
16
Second/third orderscatteringEffectswhenSZA≈VZA
DanielGoldin c PDMS:SUMMARY
• Singlescatteringisagoodapproximationtowithin+/-4o forempiricalc inthe50o <RAZ<310o region
• IntheregionRAZ=0o/360o,SZA=VZA:
• PARASOLhaslowresolutioninc• Singlescatteringnotagoodapproximationthere:higherorderscatteringmodel
needed• Singlescatteringfityieldsreasonablemeanvaluesbutdoesn’tyieldreasonablefit
uncertainties,soplan(TBD)tousehybridapproachforfinalrecordedc values:
• UsePARASOLforbinswithdata,css approximationforbinswithnodata
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
17
DanielGoldin TOBEDONE
• Workonlowerpolarizations(clouds)andotherIGBPs
• Implementice&watercloudsPDMs
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
18
DanielGoldin HIGHcUNCERTAINTYREGIONS
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
20
c PDMhigh-uncertaintyregion:RAZ>310,VZA>30
c PDMcentralregion:50<RAZ<300
• RegionsRAZ=0o/360o,SZA=VZAhaveStokesparamer≈0o,closetoPARASOL’sresolution
• ForsinglescatteringQ=0isundefined,needhigherorderscattering:
DanielGoldin PUNCERTAINTIES
11/28/16
Fall20
16CLARR
EOSDT
Meetin
g,NIA
21
• Plotshowsintercal.uncertaintyinreflectancevs.degreeofpolarizationassumingtheintercalibratedimager’ssensitivitytopolarizationm=0.03anddmsettothreedifferentvalues
• ConsideringCLARREO’sowntargetuncertaintyduetopolarizationdr0 =0.15%,onecanconcludethat:
• valuesofP <0.1,maybeconsideredbelownoisethreshold.Thus:
• duetosharpdrop-offinP(e.g.,nextslide),somePDM’sfor670and865nmcanbeneglected
• snow-coveredsurfacescanbeneglected(IGBP=15,19and20)
• Lowslopeindicatesintercal.P isinsensitivetostd.devs<≈0.1.
• PDMuncertaintieshaveroughlythenecessaryprecision
dm =100%dm =20%dm =10%
Clear sky ocean (IGBP = 17)
dm =100%dm =20%dm =10%
Evergreen Broadleaf Forests (IGBP = 2)