基于历史排位降雨阀值的粤港澳大湾区滑坡危险性预警

打开文本图片集

（1.SchoolofCivilEngeing，SnYatsenUniversityZuhai9o82，Cna;2.CenterfrWaterResourceandEnvirotu Yat-senUniversity，Guangzhou51O275，China）

Abstract：ThisstudyfocusedontheGuangdong-HongKong-MacaoGreaterBayAreaandconstructedagrid-basedlandslidehazard assessment model toenhanceregionaldisasterpreventionandmitigationcapabilities.Asemi-supervisedlearing methodwasused to optimizetheproporioalselectionoflandsldepointsandnn-landslidepoints toreducetheuncertaintyofsusceptibilitymodelingA historicalrankingainfallthreshold-basedmethodwasproposedtoclasifydailyrainfall，3-daycumulativerainfall，and7-day cumulativerainfalldata.Thespatialsusceptibilityoflandslidesandrainfall-inducedprobabilitywerequantitativelycoupledto establishadynamiclandslidehazardwarningsystem.Theresultsindicatethatwhena1.5-meterevaluationunitscaleisusedwithin theGuangdong-HongKong-MacaoGreaterBayArea，theoptialratiooflandslidepoints tonon-landslidepointsis1：4.Furthore， theareaundercurve（AUC）valueof tesusceptibiltymodelreachesashighasO.973.InpracticalapplicationduringJune2O18，the warning system accurately predicted 25 rainfall-induced landslide events，with 72% occurring in extremely high-risk warning zones and 28% inhigh-rsk warning zones，validatingthemodel'seffectiveness.Thissystemachieves fine-scalelandslidehazard warnings in the Greater Bay Area，providing scientific support for regional landslide risk management.

Keywords：semi-supervised machinelearning;non-lndslidesample;rainfallthreshold;landslidehazard;Guangdong-Hong Kong Macao Greater Bay Area

滑坡是斜坡岩土体沿着一定的软弱面或软弱带顺坡向下滑移的自然现象，不仅发生频率高，造成的损失也极为严重。（剩余10481字）