基于k-均值算法的无监督模型降雨预报.(IJMSC-V1-N1-2)

I.J.MathematicalSciencesandComputing,2015,1,11-20PublishedOnlineJuly2015inMECS()DOI:10.5815/ijmsc.2015.01.02Availableonlineat:Rainfallprediction,Gaussianmixturemodel,K-Meansalgorithm,rainfallestimation,PSNR,MSE.©2015PublishedbyMECSPublisher.Selectionand/orpeerreviewunderresponsibilityoftheResearchAssociationofModernEducationandComputerScience1.IntroductionWeatherforecastingisconsideredtobeoneofthemostrecurrentchallengingissues.Theprosperityandeconomicscenarioofacountrycompletelydependsontherainfall.Duetotheincreaseinglobalheats,itisinevitabletopredicttherainfallaccurately.Thepredictionofrainfallisgenerallybasedonthecloudformulation.Thecloudsareclassifiedintothreegroups,namely;nimbostratus,cumulonimbusandcumulus.Amongtheseclouds,thefirsttwotypesofcloudsproducerainfallandtherainfallassociatedwiththethirdtypeofcloudisrare.Thereforepredictivemethodologiesaretobeimplementedfortheidentificationofcloudandtherebyestimatingtheprobabilityofrainfall.Manymodelshavebeendiscussedintheliteratureaboutthecloudformationandpredictionoftherainfall.Ingeneral,theonsitepredictionofrainfalliscarriedoutintwostages,stage-1discussesaboutthedatasetandextractionofthefeaturesfromthedataset.Inthestage-2,detectionofthecloudisestimated.Inordertoportraythepresentmethodology,thedatasetisconsideredfrom*Correspondingauthor.Tel.:E-mailaddress:[e-mail]12PredictionofRainfallUsingUnsupervisedModelbasedApproachUsingK-MeansAlgorithmsatelliteimagesfromthemeteorologicaldepartment,Visakhapatnam.Thestudyisbasedontheinfraredvisibleregionsandwatervaporregions.TheseimagesareextractedfromINSAT-3AandKALPANA-1.Fortheclouddetection,thesatelliteimagesarepreprocessedandbasingonthehighreflectanceofthecloudinthevisiblespectrumandlowtemperaturesininfraredspectrum,theanalysisiscarriedout.Fromthesatelliteimages,thecloudtoptemperatureisextractedusingtheinfraredraysandbasingontheprecipitation,onecananalyze,whetherthecloudisacold-cloudorahot-cloud.Intheabsenceofcloud,thesatelliteimagesprocessthedataandrecordthedataasinfraredintensity.Clouddensityhelpstounderstandthethicknessofthecloudwhichindirectlydecidesintheformationoftherain.ItisgenerallyconsideredasaBooleanwhichreturnsazero,whencloudisnotpresentandintensityvaluewhichisequaltoone,ifcloudispresent.Watervaporisanothersignificantparameterthatdecidesthestatusoftherainfall.ThearchitectureinvolvedispresentedbelowinFig.1Fig.1.ArchitectureTheresearchworkinthisareahasbeencarriedoutbymanymeteorologistsandmostofthemodelsdevelopedbytheeminentresearcherscouldnotabletopredicttherainfallefficiently,becauseofthecompressionratiowhichisassociatedwiththeretrievedimages.Theotherfactorsassociatedwiththesatelliteimagesinclude;snow,sandandicyregions,theseassociatedfactorsaffecttheretrievalaccuracy.Thereforeeffectivemethodologiesarestilltobeinvolvedforeffectiveidentificationofcloudandformationofcloud.Therefore,thispaperhighlightsamodelbasedonfeatureextractiontogetherwithclusteringtechniquesVisibleImageInfraredImageWaterVaporImageCloudDensityCloudTopTemperatureSeaSurfaceTemperatureHumidityEstimationFusedAND/ORRainfallPredictionDatabasePredictionofRainfallUsingUnsupervisedModelbasedApproachUsingK-MeansAlgorithm13basedonk-meansalgorithm.Theclassificationofcloudsisbasedontheparameterswhichincludewatervapor,humidityandtheotherparametersestimatedfromGaussianmixturemodel.Therestofthepaperisorganizedasfollowssection-2ofthepaperdealswithrelatedworkinthearea;section-3presentsthefeatureextractionmethodology.TheclusteringtechniqueforsegmentingtheimagesbasedonK-Meansalgorithmispresentedinsection-4,inthesection-5ofthepaperhighlightsabouttheclassificationalgorithmbasedonGaussianmixturemodel,section-6ofthepaperdiscussesaboutthefusionandthepredictionoftherainfalliscarriedoutbyusingKL-divergencewhichishighlightsinsection-7ofthepaper.Insection-8thedesignandimplementationisshown.Insection-9evaluationishighlightedandsection-10ofthepapershowsexperimentsandresultsandthenthepaperconcludes.2.RelatedWorkManyauthorshavepresentedthereviewofthepredictionofrainfall.D.K.RichardsandG.D.Sullivanhaspresentedamethodologyinwhichthepredictionoftherainfallisbasedonthediscriminationofthecloudforwhichtheclassifi