ChinesescientistsusemachinelearningforpreciseAntarcticseaiceprediction_GuangmingOnlineBEIJING,March26(Xinhua)--ChinesescientistsmadeaccuratepredictionsregardingAntarcticseaiceforDecember2023toFebruary2024usingdeeplearningmethods.TheresearchteamutilizedaConvolutionalLongShort-TermMemory(ConvLSTM)neuralnetworktoconstructaseasonal-scaleAntarcticseaicepredictionmodel.TheirforecastindicatedthatAntarcticseaicewouldremainclosetohistoricallowsinFebruary2024,buttherewaslessindicationofitreachinganewrecordlow.Thepredictedseaicearea(SIA)andseaiceextent(SIE)forFebruary2024were1.441millionsquarekilometersand2.105millionsquarekilometers,respectively,slightlyhigherthanthehistoriclowsobservedin2023.Theteam,ledbyresearchersfromSunYat-senUniversityandtheSouthernMarineScienceandEngineeringGuangdongLaboratory(Zhuhai),submittedtheirpredictionresultsinDecember.TheresultswerepublishedinthejournalAdvancesinAtmosphericSciencesinearlyFebruary.TheirpredictionwasthenvalidatedbythelatestsatelliteobservationsforFebruary.TheobservedSIAandSIevalsuesforFebruary2024are1.510millionsquarekilometersand2.142millionsquarekilometers,respectively.Accordingtotheresearchers,thecomparisonbetweenthepredictionsandobservationsindicatesaremarkablyclosealignment.Furthermore,theseaiceareaandextentfromDecembertoFebruaryfallwithinonestandarddeviationofthepredictedvalues,underscoringthereliabilityoftheforecastingsystem.ThesuccessfulcomparisonbetweenthepredictionandobservationdatavalidatestheaccuracyoftheConvLSTMmodelanditspotentialforreliableAntarcticseaiceforecasting,saidtheresearchers."OursuccessfulpredictionnotonlyunderscoresthesignificanceofstrengtheningAntarcticseaicepredictionresearchbutalsodemonstratesthesubstantialapplicationpotentialofdeeplearningmethodsinthiscriticalarea,"saidYangQinghua,aprofessorofSunYat-senUniversity.Editor:DisclaimerTheviewsandopinionsexpressedinthisarticlearethoseoftheauthor's,GMW.cnmakesnorepresentationsastoaccuracy,suitability,orvalidityofanyinformationonthissiteandwillnotbeliableforanyerrors,omissions,ordelaysinthisinformation.