active and passive microwave observations of snowfall from
TRANSCRIPT
Activeandpassivemicrowaveobservationsofsnowfallfromspace.
DanieleCasella1,GiuliaPanegrossi1,PaoloSanò1,AnnaCinziaMarra1,StefanoDietrich1,1 Istitute of Atmospheric Sciences andClimate - CNR,Rome,Italy
BenjaminT.Johnson2,2 AtmosphericandEnvironmentalResearch@NOAAJointCenterforSatelliteDataAssimilation(JCSDA),CollegePark,
MD,USA
MarkS.Kulie3.
3 SpaceScienceandEngineeringCenter,UniversityofWisconsin-Madison,Madison,Wisconsin,USA
Motivation• OneofthemaingoalsoftheGPMmissionistoimprove
snowfallretrievalaccuracy
Objectives:• Dual Precipitaion Radar
– AssesstheGPMDPR-basedprecipitationproductsabilitytoestimate andidentify snowfall
– AnalyzetheDPRKu/Karadarsensitivitytosnowfall– ProposeasimplealgorithmtoincreasetheDPR’ssnowfalldetection
capabilityusingmatchedDPRKu/Karadarreflectivityvalues
• MWradiometers– Surface backgroundcharacterization– Analysis of snowfall related signal
CloudSat CPRis used as areference inthis study
8thIPWG5thIWSSMJointWorkshop,Bologna,3-7October 2016
CPRvsDPRGPM DPR GPM DPR CloudSat CPR
Ku Ka
Frequency 13.6GHz 35.55GHz 94.05 GHz
Orbit Drifting 65° Sunsynchronous13:30asc.
Altitude 407km 705km
Sensitivity 12-13dBZ 16.32dBZ -28dBZ
Scanning electronic None
Swath size 245km 125km 1.5km
HorizontalResolution 5km 1.5km
VerticalResolution 250m 250mMS
500mHS 480m
VerticalSpacing 125 m 125mMS
250mHS 240m
DPRKu
DPRKa
CPR
8thIPWG5thIWSSMJointWorkshop,Bologna,3-7October 2016
We collected 74750coincidentDPR-CPRsnowfall observationsfrom 2B-CSATGPMproductdataset was enriched with
various CludSat-GPM productsperiodMarch2014toMay2015Selectedcoincidenceswithin:
5minutesand2.5km
Casestudy:extensive frontal snowfallWidespreadfrontalsnowfalleventoccurredoverEasternRussianorthoftheSeaofOkhotskon30April2014
CPR:• TypicalmaximumCPRZ:10-15dBZ• maximumcloudtopheightsbetween~5-8km• shallowercloudstructureswithcloudtopheightslessthan~2km
DPRMeasuredReflectivity(Z)• KuandKa-HSuncorrectedZsomestructurebelow~4kminthedeepersnowfallsegment
• mostoftheeventnorthof60o latitudeandathigheraltitudesismissed
• significantrandomnoisearound12dBZ (KuandKaHS)or18dBZ (KaMS)
• SidelobecluttersignalintheKu
DPRCorrectedReflectivity(Zc)(2A-DPR)• Completesuppressionofrandomnoiseandsidelobe clutter
• Attenuationofcorrectionbelowthefree-clutterlevel
• Partoftheweaksignalrelatedtosnowfallisalsoeliminated
8thIPWG5thIWSSMJointWorkshop,Bologna,3-7October 2016
MorecasestudiesShallowconvectivesnowfall
OregonCascademountainRangeIntenseFrontalSnowfall
InteriorAlaska
8thIPWG5thIWSSMJointWorkshop,Bologna,3-7October 2016
2A-DPR:Ku andKa Reflectivity values2B-DPRCMB:DPRandGMIcombined
Assesment of snowfall rates inDPRL2products2B-DPR-
CMB Ku
2B-DPR-CMB
Ka MS
2A-DPRKu
2A-DPRKaMS
2A-DPRKa HS
ME*[mm h-1] -0.338 -0.372 -0.1136 -0.167 -0.450
RMSE*[mm h-1] 0.731 0.820 0.674 0.734 0.637
ARMSE [mm h-1] 0.648 0.731 0.665 0.714 0.451
POD 0.0675 0.0757 0.0605 0.0676 0.0486
FAR 0.507 0.542 0.538 0.580 0.636
6.3%
87.5%
6.2% 7.1%
86.9%
6.0% 5.8%
89.3%
4.9% 6.4%
88.9%
4.7% 4.7%
92.6%
2.7%
HIT MISS FALSE
Ku Ka
2A-DPR
KuMSHS
DPR-
CPRfreeClutter
binhe
ight
diffe
rence
Ku Rayindex
DPRProducts Snowfall DetectionCapabilities
2B-DPR-CMB
Ku
2B-DPR-CMB
Ka MS
2A-DPRKu
2A-DPRKa MS
2A-DPRKa HS
Percentage of snowfall
mass detected28.43% 34.09% 27.74% 32.08% 30.33%
DPR-CPRLowLevel Reflectivity Comparison
Mean Reflectivity inthefirst500mover DPRclutter-free level
Proposed algorithm for DPRsnowfall detectionincrease
1. Assuming Ka and Ku coherence -> -0.5 < DFR < 5
2. Assuming cloud structure continuity -> Z > 8 dBZ for 3 or more of the 4 Ku reflectivity
3. Assuming a minimum ->Z dBZ10>Z
DRF: dual frequency ratio (DFR = dBZKu-dBZKa):
DPR Algorithmsversion 4
2B-DPR-CMB
Ku
2A-DPRKu
2B-DPR-CMB
Ka MS
2A-DPRKa MS
2A-DPRKa HS
% snowfall mass
detected
28.43
%
27.74
%
34.09
%
32.08
%
30.33
%Proposed
algorthmKu Ka
% snowfall mass
detected57.16% 52.86%
Reddots:detected by thenew algorithm
Crosses :detected by2A-DPR
Section 2:MWradiometers
D.Casella,G.Panegrossi,P.Sanò,A.C.Marra,S.Dietrich,B.T.Johnson,M.S.Kulie,EvaluationoftheGPM-DPRSnowfallDetectionCapability:ComparisonwithCloudSat-CPR,JGR- Atmosphere,submitted
ATMSradiometer
• Somepreliminary results from theanalisysof ATMSradiometr compared to CPRwillbe shown
• ATMSobserve highlatitudes with 5channels inthe183GHz WVabsorptionband
• General approach for Passivemicrowave– Lowfreq.channels used for surface
classification– Highfreq.for snowfall-related signal
• Dataset of coincident CPR-ATMSobservations size:2.300.000ATMSpixels
Centralfrequency(GHz)
Bandwidth(MHz)
Quasi-polarisation NEΔT
23.8 270 QV 0.90K31.4 180 QV 0.90K50.3 180 QH 1.20K51.76 400 QH 0.75K52.8 400 QH 0.75K
53.596± 0.115 170 QH 0.75K54.4 400 QH 0.75K54.94 400 QH 0.75K55.5 330 QH 0.75Kf0=
57.290344 330 QH 0.75Kf0± 0.217 78 QH 1.20K
f0± 0.3222 ±0.048 36 QH 1.20K
f0± 0.3222 ±0.022 16 QH 1.50K
f0± 0.3222 ±0.010 8 QH 2.40K
f0± 0.3222 ±0.0045 3 QH 3.60K89.5 5000 QV 0.50K165.5 3000 QH 0.60K
183.31± 7.0 2000 QH 0.80K183.31± 4.5 2000 QH 0.80K183.31± 3.0 1000 QH 0.80K183.31± 1.8 1000 QH 0.80K183.31± 1.0 500 QH 0.90K
Surface Classification
Casestudy Ice Class Test
ATMSclassifierOceanisBlue,newiceiscyanandyellowisfirstyear/multilayerice(inbothpanels).
OSISAF(productOSI-403-a)
MODISObservationovertheoceanbetweenGreenlandandSvalbardIslandson01/05/2015at10:05UTC,reflectancefromchannel2(Visible841-876nm)isshown.AlargestripeofbrokeniceisclearlyvisibleinthissectionofoceanfromMODIS.
ATMSWVchannelsNoCloud
Columnar WaterVapor[kgm-2]
183.31GHz
183GHz ΔTbCloud Free
Columnar WaterVapor[kgm-2]
Columnar WaterVapor[kgm-2]
183GHz ΔTbSurface Snowfall >0
ConclusionsDPR
• DPRprecipitationproductsmissaverylargefractionofsnowfallprecipitationevents (92.5- 95.2%ofmisses)
• fractionofsnowfallmasscorrectlydetectedbyDPRisnotnegligible(27-34%)• ThemostimportantissuesoftheDPRinsensingsnowfallproducingcloudsare
relatedtothelimitedsensitivityandtothepresence(intheKuprofiles)ofsidelobeclutterechoes.
• thecluttermitigationtechniqueseliminatealarge fractionoftheweaksignalrelatedtosnowfall
• AsimpleprototypealgorithmhasbeendevelopedandtestedforimprovingDPRsnowfalldetectioncapability:thepercentageoftheestimatedsnowfallmassdetectedincreasessignificantlyto57.16%(52.86%)fortheKu(Ka)radar
ATMS• Only two lowfrequencychannels(23.8GHz31.4GHz),togetherwithsome
ancillarydata,canbesuccessfullyusedtoclassifythebackgroundsurfaceeveninpresenceofprecipitatingclouds.
• The183GHz channeldifferencesaresensibletodeep/intensesnowclouds.• The characterizationofthebackgroundsurfaceisessential,oversomesurface
class(SnowA-B,Coast)cloudsnowdetectionisreallydifficult,overice-freeoceanandsnow-freelandrelativelyeasier
• Columnarwater vaporsignalisstrongerthancloud-snowsignal
Thank you
Thanks to Joe Turk, Norm Wood, Guosheng Liu and Grant Petty for the valuable discussions and suggestions.
This study has been carried out within:
EUMETSAT H-SAF/UW-SSEC/UMBC-JTEC Federated Activity “Cooperation on the use ofcombined spaceborne active and passive MW observations for precipitation retrieval”
Programma Nazionale di Ricerche in Antartide (PNRA) through the project “Characterization ofprecipitation in the Antarctic region based on satellite observations”
DPRKu Sidelobe clutter
Furukawa et al.2014
Kubota et al.2014
8thIPWG5thIWSSMJointWorkshop,Bologna,3-7October 2016
Kubota, et al. (2014), Evaluation of precipitation estimates by at-launch codes of GPM/DPR algorithms using synthetic data from TRMM/PR Observations, IEEE J. Sel.Top. Appl. Earth Obs. Remote Sens., 7(9), 3931–3944
Furukawa,Kinji,et al."Theorbital checkoutstatusof thedual-frequency precipitation radarontheglobalprecipitation measurement core spacecraft."2014IEEEGeoscienceandRemoteSensing Symposium.IEEE,2014.
Detectionof Ice over Ocean
Partitioningoftheoveroceanobservationsintoocean-iceandwarm-ocean.Reddotsrepresenticeoveroceanandno-iceoveroceanasrepresentedintotheSurface_type variableintothe2C-PRECIP-COLUMNCloudsat product.Theblackcurverepresentsthediscriminant functionappliedbythealgorithm.ThiscategorizationshowsaPODof94%(93%)andaFARof0.3%(1%)consideringthecloudfree(precipitation)dataset.
Detectionof Snow over Land
Snowoverlandidentification.ThecolorscalerepresentsthesnowdepthliquidwaterequivalentascalculatedbytheECMWFERAInterimreanalysis.Theblackcurverepresentsthediscriminant functionappliedbythealgorithm.ThiscategorizationshowsaPODof97%(93%)andaFARof7%(10%)consideringthecloudfree(precipitation)datasetandathresholdforsnowcoveredsurfaceof0.1mofLWEsnowdepth.
Ice over ocean classificationexample of classification for Ice over ocean. The 2 dimensional spaceof the 23 GHz channel divided by the 2 meter temperature (em 23GHz in the x axis) and 31 GHz channel divided by the 2 metertemperature (em 31 GHz in the y axis) is divided into several classeswith a k means clustering approach (the circles represents thecluster centroids) then the various clusters are grouped into 3 mainclasses: blue for New ice characterized by high emissivity almostconstant with the frequency. Red for multilayer Ice, with highlyvariable emissivity stronger at 23 GHz than at 31 GHz; and lightgreen for broken ice with a variable emissivity constant with thefrequency. Figure 5 show the results of the airborne retrieval ofemissivity fron Hewison and English 1999 in supporting theclassification operated from the ATMS surface classifier.
NadirEmissivityasreportedfromtheAirborneretrival ofHewison andEnglish1999
Statistics for theIce ClassesNEWICE MULTILAYERICE BrokenICE
NEWICE MULTILAYERICE BrokenICE
Peaksbeforewinterseason
Peaksinsummer
Noclearpeaks
NEWICE
MULTILAYERICE
BrokenICE
Snow over Land ClassificationTbrat =Tb23 /Tb31
ECMWFERA-ISnow Depth
Cloud Free With Clouds
Tbrat >1.1
Tbrat <1.1
NoSnow NoSnow
NoSnow NoSnow
Statistics for theSnow Classes
SnowAandBonlyoverAntarcticaandcontinentalGreenland,TypeBinparticularispresentonlyovertheAntarcticaPlateu (andoverinnerGreenland)duringwinter.
Snow A Snow B
Snow C
Globaltrends
Summary of DPR-CPRComparison• DPRprecipitationproductsmissaverylargefractionofsnowfall
precipitationevents (92.5- 95.2%ofmisses)• fractionofsnowfallmasscorrectlydetectedbyDPRisnotnegligible(27-
34%)• snowfallratesestimatedbytheDPRlevel2productssufferfrom
underestimation• allfalsealarmsinourdatasetareduetodifferencesintheestimated
phaseofprecipitation,thatisrelatedprimarilytothedifferentnearsurfacebinconsideredintheCPRandDPRproducts
• ThemostimportantissuesoftheDPRinsensingsnowfallproducingcloudsarerelatedtothelimitedsensitivityandtothepresence(intheKuprofiles)ofsidelobeclutterechoes.
• thecluttermitigationtechniqueseliminatealargefractionoftheweaksignalrelatedtosnowfall
• AsimpleprototypealgorithmhasbeendevelopedandtestedforimprovingDPRsnowfalldetectioncapability:thepercentageoftheestimatedsnowfallmassdetectedincreasesdramaticallyto57.16%(52.86%)fortheKu(Ka)radar