Page 92 - 水利学报2025年第56卷第3期
P. 92

darvariablesintoadeep - learningmodel[J].GeophysicalResearchLetters,2021,48(21):e2021GL095302.
                [24] 刘昱辰,刘佳,刘 录 三,等.基 于 LSTM 实 时 校 正 的 WRF?WRF - Hydro耦 合 径 流 预 报 [J].水 利 学 报,
                      2023,54(11):1334 - 1346.
                [25] 胡春歧,赵才.陆气耦合洪水预报技术应用研究[J].水力发电,2019,45(9):48 - 51.
                [26] 严昌盛,朱德华,马燮铫,等.基于雷达短时临 近 降 雨 预 报 的 王 家 坝 洪 水 预 报 研 究 [J].水 利 水 电 技 术,
                      2020,51(9):13 - 23.
                [27] WUZ,WUJ,LUG.Aone - waycoupledatmospheric - hydrologicalmodelingsystem withcombinationofhigh -
                       resolutionandensembleprecipitationforecasting[J].FrontiersofEarthScience,2016,10(3):432 - 443.
                [28] 糜佳伟,田济扬,楚志刚,等.耦合天气雷达定量降水估计与定量降水预报的中小流域洪水预报研究[J].
                       华北水利水电大学学报(自然科学版),2022,43(3):11 - 18.
                [29] 陈豫英,苏洋,杨银,等.贺兰山东麓极端暴雨的中尺度特征[J].高原气象,2021,40(1):47 - 60.
                [30] FRANCHG,MAGGIOV,COVIELLOL,etal.TAASRAD19,ahigh - resolutionweatherradarreflectivitydataset
                       forprecipitationnowcasting [J].ScientificData,2020,7(1):234.
                [31] XUQ,LIUS,XUEM.BackgrounderrorcovariancefunctionsforvectorwindanalysesusingDoppler - radarradial -
                       velocityobservations[J].QuarterlyJournaloftheRoyalMeteorologicalSociety,2006,132(621C):2887 - 2904.
                [32] 李娟芳,王文川,车沛沛,等.HEC - HMS模型和 TOPMODEL模型在东庄流域山洪预报的应用研究 [J].
                       水电能源科学,2019,37(3):50 - 53.
                [33] PULKKINENS,NERINID,HORTALAA P,etal.Pysteps:anopen - sourcePythonlibraryforprobabilistic
                       precipitationnowcasting(v1.0)[J].GeoscientificModelDevelopment,2019,12(10):4185 - 4219.



                          Coupledforecastingofrainstormsandfloodsinsmallwatershedbasedon
                                          DeepLearningandHEC - HMSModel

                                                                         2
                                                                                       4
                                                      3
                                                                                                     4
                                        2
                           1,2
                   LIUWan ,XIEShuai,ZHONGDeyu,WANGYongqiang,BAOShuping,ZHUXudong
                     (1.SchoolofCivilandHydraulicEngineering,HuazhongUniversityofScienceandTechnology,Wuhan 430074,China;
                                    2.ChangjiangRiverScientificResearchInstitute,Wuhan 430010,China;
                           3.StateKeyLaboratoryofHydroscienceandEngineering,TsinghuaUniversity,Beijing 100084,China;
                        4.NingxiaHydrologicalandWaterResourcesMonitoringandEarlyWarningCenter,Yinchuan 750001,China)
                  Abstract:Therainfallrunoffprocessinsmallwatershedissoquickthattheforecastleadtimeofconventionalflood
                  monitoringandforecastingisshort,whichisdifficulttoprovideeffectivesupportforfloodpreventionanddisaster
                  relief.Inthisstudy,acoupledforecastingframeworkforrainstorm andfloodforecastingisestablished.Inthis
                  framework ,thehydrologicalmodelisdrivenbytherainfallnowcastingtoprolongtheforecastleadtimeofflooding,
                  whichisofgreatimportanceforfloodpreventioninsmallwatershed.Intherainfallnowcastingmethod,thispaper
                  proposesaRainfallNowcastingMethodwithCombinedReflectanceandWindFieldasInputs(RNMCW),the
                  combinedreflectionandradar - retrievedwindfieldthatcaneffectivelyreflectthewatervaporcontentandwaterva
                  portransportationareselectedasinputs ,andthedeeplearningmodelisproposedtoextractthetemporalandspa
                  tialcharacteristicsofinputstoforecasttherainfallinthebasin.Theresultsshowthatthecorrelationcoefficientof
                  therainfallnowcastingmethodishigherthan0.75ateachstation,whichisabout5% higherthanthatofthecon
                  ventionalopticalflowmethod.Then,theparametersoftheHEC - HMSmodelarecalibratedwiththerainfallpre
                  dictedbyRNMCW,andtherainfallforecastresultsareusedastheinputofthehydrologicalmodeltobuildastorm
                  andfloodcouplingforecastmodel.ComparedwiththeHEC - HMSmodel,thecoupledmodelreducestheflood
                  peakpredictionerrorof2016floodby2.17%,andincreasestheNashcoefficientof20180722floodpredictionby
                  0.002.Themodelcaneffectivelyforecastthefloodprocessandextendtheeffectivepredictionperiodbyupto2
                  hours,thussignificantlyimprovingthefloodpredictioneffectandprovidingbettersupportforpracticalapplication.
                  Keywords:rainfallnowcasting;deeplearning;radar - retrievedwindfield; coupledforecastingofrainstorms
                  andfloods

                                                                                    (责任编辑:耿庆斋)

                —  3 7  —
                     4
   87   88   89   90   91   92   93   94   95   96   97