变分数据同化方法中的误差问题分析-error analysis in variational data assimilation method.docxVIP

AbstractInvariationaldataassimilation,theerrorcovariancematricesareveryimportantforassimilationsystem.Tosomeextent,theydeterminethequalityofassimilativeanalysis.Errorcovariancesnotonlyreflecttherelationshipbetweendifferentvariables,butalsoensuretheuniquenessofvariationalsolutionandsmoothnessintheprocessofanalysis.Therefore,itisvitalforassimilationsystemtoconstructthereasonableerrorcovariancematrix.Atpresent,theresearchfortheestimationandtreatmentoftheerrorcovarianceinvariationaldataassimilationhasalatestartinourcountry.Andcomparedwiththesequentialdataassimilationmethod,ithasitsownuniqueproperties.Hence,onthebasisofabsorptionofexistingscientificachievement,predecessors’achievementsandexperienceofothers,weshoulddeeplyexploretheproductionmechanism,estimationanddisposalmethodoferrorcovariance,combiningwiththecharacteristicsofcontinuousdataassimilation.Ultimately,thatwillpromotevariationaldataassimilationtobetterbusinessandpracticaluse.Inthispaper,wewillstudythestatisticalcharacteristicoferrorcovarianceinvariationaldataassimilationanditsinfluenceonthree-dimensionalvariationalassimilationsystem.Variouskindsofsolutionmethodsandbackgrounderrorcovarianceofvariationaldataassimilationarecombedanddiscussedinminutedetailinthepaper.Ontheonehand,itstrengthensthetheoreticalbasisofthearticle,ontheotherhand,itmeetstherequirementsoftheguidelinesforthenumericalexperiments.BasedontheLorenz-1963modelforforecastingmodel,three-dimensionalvariationalassimilationmethodischoosenforassimilationalgorithm.Wehavestudiedthebackgrounderrorcovariance,theanalysiserrorcovariance,theerrorcovariancebetweenanalysisandforecast,theerrorcovariancebetweenanalysisandtruevalue.Afterthat,weexplorethestatisticaldistributioncharacteristicsofallkindsoferrorcovariances,andmakeacrosscomparisonanddiscussionaboutthem.Forthevariabledimensionandwidthofassimilationwindow,thesensitivenumericalexperimentsaresetupinordertounderstandtheeffectofthoseparametersintheanalysisofassimilationsystem.Inaddition,abouttheNMCmethod,then