OptimizingtheProphetmodelusingBayesianOptimizationforlandsubsidencepredictionincentralwardsandcommunesofCaMau

Abstract

LandsubsidenceintheCaMauregionhasbecomeincreasinglycomplex,posingasignificantthreattothesustainabledevelopmentofthiscoastaldeltaarea.Inthiscontext,thetemporalmonitoringandforecastingoflandsubsidenceareessentialforearlywarningsystemsandforsupportingdisasterresponseandspatialplanning.Thisstudyproposestheapplicationofanadvancedmachinelearningmodel-Prophet-integratedwithBayesianOptimization(BO)toimprovetheaccuracyoftemporalsubsidenceprediction.Thedatasetcomprisestimeseriesofgrounddeformationat1817points,extractedusingPersistentScattererInterferometricSyntheticApertureRadar(PS-InSAR)technology.ThesedatawerecollectedcontinuouslyfromJanuary2015toJanuary2019,withatotalof178temporalacquisitions.ExperimentalresultsdemonstratethattheProphetmodelcombinedwithBOachieveshighpredictiveperformance,withanaverageRootMeanSquareError(RMSE)of3,4mmandaMeanAbsoluteError(MAE)of2,6mm.Notably,atthereferencedateofJanuary31,2019,thepredictedvaluesexhibitedastrongcorrelationwithPS-InSARobservations(R²=0,96).Giventhislevelofaccuracy,theproposedmodelshowsgreatpotentialforlong-termsubsidencetrendmonitoringandriskmapping,particularlyinareaswithslowandstablesubsidencesuchastheCaMaucoastalplai