Thispaperteststheideaofsubstitutingtheatmosphericobservationswithatmosphericreanalysiswhensettingupacoupleddataassimilationsystem.Thepaperfocusesonthequantificationoftheeffectsontheoceanicanalysisresultedfromthissubstitutionanddesignsfourdifferentassimilationschemesforsuchasubstitution.AcoupledLorenz96systemisconstructedandanensembleKalmanfilterisadopted.Theatmosphericreanalysisandoceanicobservationsareassimilatedintothesystemandtheanalysisqualityiscomparedtoabenchmarkexperimentwherebothatmosphericandoceanicobservationsareassimilated.Fourschemesaredesignedforassimilatingthereanalysisandtheydifferinthegenerationoftheperturbedobservationensembleandtherepresentationoftheerrorcovariancematrix.Theresultsshowthatwhenthereanalysisisassimilateddirectlyasindependentobservations,theroot-mean-squareerrorincreaseofoceanicanalysisrelativetothebenchmarkislessthan16%intheperfectmodelframework;inthebiasedmodelcase,theincreaseislessthan22%.Thisresultisrobustwithsufficientensemblesizeandreasonableatmosphericobservationquality(e.g.,frequency,noisiness,anddensity).Iftheobservationisoverlynoisy,infrequent,sparse,ortheensemblesizeisinsufficientlysmall,theanalysisdeteriorationcausedbythesubstitutionislessseveresincetheanalysisqualityofthebenchmarkalsodeterioratessignificantlyduetoworseobservationsandundersampling.Theresultsfromdifferentassimilationschemeshighlighttheimportanceoftwofactors:accuraterepresentationoftheerrorcovarianceofthereanalysisandthetemporalcoherencealongeachensemblemember,whicharecrucialfortheanalysisqualityofthesubstitutionexperiment.
