You are on page 1of 18

Problems

Snow/ice&cloudsarebothbright
Cloudsoftenmadeofice
Cloudscanbewarmerorcolderthansnow/icesurface
Visualinspectionrequiresexperiencetodiscriminate
cloudsfromicesheet
Thincloudscanbeproblematic

Butthincloudsdontobscuresurface

ACCA
Twopass:
Pass#1:only1of8filters(NDSI)meaningfulforice/snow
Pass#2:thermalchannelanalysisisambiguousoverice/snow

Somegoodnews
Sunisrarelydirectlyoverhead
Thincloudsarelessproblematic
Surfacenotobscured

ACCAcontains1(of8)filtersthathasvalue
NormalizedDifferenceSnowIndex
band2 band5
NDSI=
band2+band5
Similarband2reflectance,butolder
snowdarkerthancloudsinband5

OurApproach
Separatingbrightthingsfromdarkthingsis
easy
Shadows&rocks&ocean
vs.
clouds&ice/snow
Everythinghasanedge
Shadowedgesmatchcloudedges
Especiallywhensundirectionknown
Alsogetcloudelevationaboveground

Phase1:Classifications
UseNDSI(initialthresholdof0.7)
definesbinaryimageofpossiblecloud &notcloud

Simplifymorphologyofpossiblecloud regions
fillsmallholesandclosingnarrowgaps
Appliesnumerictagtoeachpossiblecloud region

Edgedetectionofpossiblecloud regions
Thinnedto1pixelwidth

Detectcloudshadowsandwater
Cloudshadowsarebrighterthanoceananddarkerthaneither
snoworclouds
Bands3and4areausedwiththresholdsdependentonsun
elevation
2classesidentified
possiblecloudshadow
Water

Phase2:MatchingandScoring
Uses:
edgesofpossibleclouds(1pixelwide)
Pixelsofpossiblecloudshadows
Pixelsofwater

Moveedgesdownsun
Whenanedgemeetsashadoworwaterpixel
aratioiscalculated:
Ratio=#edgepixelsthatcoincidewithashadow
pixel/#ofalledgepixels

Phase3:Ratiointerpretation
Idealratiovalues:
=1.0,ifclouddoesntoverlapshadow
=0.5,ifcloudoverlapsshadow

Empiricallysetthresholdsbasedon8imagedataset:
Shadowratio>0.2 possiblecloudiscloud
Waterratio>0.25 possiblecloudiscloud
Imageedgeratio>0.2 possiblecloudiscloud

Resultsisanimageseparatedinto5classes:

Cloud
Water
Detectedshadows
Rejectedclouds
Snow(icesheet)

Identifiedclouds(lightgray),detectedshadows(darkgray),detected
waterpixels(grid)andtherejectednoncloudpixels(black).

Identifiedclouds(lightblue),
detectedshadows(red),rejected
noncloudpixels(darkblue).And
icesheet(black)

NDSIthresholdof0.7doesntalwayswork
Bestvaluesrangedfrom0.56to0.79
Iterationofthresholdvalueintroducedalongwith
cloudscore tomeasuresuccessofclouddetection
foreachiteration
CloudScore=S1 *Sratio 0.5*R
S1 =#edgepixelsmatchingcloudshadow
Sratio =S1 /total#ofedgepixels
R=#edgepixelsnotmatchingcloudshadow

Sratio biasedtheresultsawayfromlargeNDSI
thresholdsthatdetectedmorepossiblecloudregions
andgreatlyincreasedthenumberofedgepixels

Normalized Cloud Score

0 .5
2 2 7 ,1 1 7

0
0 .5

0 .5 5

0 .6

0 .6 5

0 .7

0 .7 5

0 .8

0 .8 5

3 4 ,1 1 9
1 2 ,1 1 5

- 0 .5

7 ,1 2 1
2 2 9 ,1 1 8

-1

2 2 9 ,1 1 9
5 3 ,1 1 5

- 1 .5

2 9 ,1 1 7

-2

- 2 .5
NDS I t h r e s h o ld

Wearewisernow
LIMAtaughtus
thatsnowisnota
diffusereflector
atlowsun
elevationsanda
betterreflectance
modelisrequired
(Bindschadleretal.,2008)

Minutes

Imagesubsamplingtoreduceruntimes

Millionsofpixels

Summary
Useofcloudshadowstodetectcloudsover
snowandiceworks
NDSIisausefulpredetectionstepandcan
probablybeimprovedbycorrectconversion
ofsnowradiancetoreflectance
Useofreferenceimagesmaybemore
effectiveforicesheets.

The results of Automated NDSI threshold decision


for LANDSAT 7 7/121(path/row), 29/117, 229/119, 229/118

Normalized Cloud Score

0.5
0
-0.5

0.6

0.65

0.7

0.75

0.8

0.85
p7r121.img

-1

p29r117.img
p229r119.img

-1.5

p229r118.img
-2
-2.5
-3
-3.5
NDSI threshold

The results of Shadow / Cloud Ratio


for LANDSAT 7 7/121(path/row), 29/117, 229/119, 229/118

Normalized Cloud Score

2.5

2
p7r121.img

1.5

p29r117.img
p229r119.img
1

p229r118.img

0.5

0
0.6

0.65

0.7

0.75

NDSI threshold

0.8

0.85

ProceduresforAutomatedNDSIthresholddecision
StartCDSEwith
initialNDSI
threshold(0.6)

IncreasetheNDSI
thresholdby0.01

Calculatetheclouddetectionscore
foreachcloudclusterduring
CDSEprocedures
CloudScore=S1*S_Ratio 0.5*(#of
rejectedcloudedgepixels)

Findthemaximumscore
anddecideNDSI
thresholdwiththe
maximumscore

CalculateTotal_Score
addingthecloud
scoresforallcloud
clusters

Stoptheiterationwhen
theshadowcloudratio
(totalshadowpixels/total
cloudpixels)islessthan
0.15

Theclouddetectionscorewilldecrease
1.

WhenthedetectedcloudclustersgrowtoomuchduetotoohighNDSI
value.(S_Ratio)

2.

Whentheimagehasmorerejectedcloudclusters.

ThepurposeofmultiplyingS1(S1*SRatio)isgivingaweightofthecloudcluster
sizetoavoidthattheimagehavingsmallcloudclustershashigherscore.

Thescorevaluecouldbeanynumberandrelative.

CloudDetectionusingShadowMatching(CDSE)
Clouddetectionusing
NDSIthreshold
cloudmaskimage

Cloudshadow
detectionusing
band3,4,5

Detectcloudclusters
Labelregion
algorithm

Searchfordirectionalcloud
shadowexistencefor
eachcloudedgecluster

Cloudedgedetection
Morphological
operations

Calculateashadowratio
(S_Ratio)=S1/(total#of
cloudedgepixels)
S1=#ofcloudedgepixelshaving
detectedshadow

You might also like