摘要
Anovelhybridapproachforearthquakelocationisproposedwhichusesacombinedcoarseglobalsearchandfinelocalinversionwithaminimumsearchroutine,plusanexaminationoftherootmeansquares(RMS)errordistribution.ThemethodexploitstheadvantagesofnetworkraytracingandrobustformulationoftheFréchetderivativestosimultaneouslyupdateallpossibleinitialsourceparametersaroundmostlocalminima(includingtheglobalminimum)inthesolutionspace,andfinallytodeterminethelikelyglobalsolution.Severalsyntheticexamplesinvolvinga3-Dcomplexvelocitymodelandachallengingsource-receiverlayoutareusedtodemonstratethecapabilityofthenewly-developedmethod.Thisnewglobal-localhybridsolutiontechniquenotonlyincorporatesthesignificantbenefitsofourrecentlypublishedhypocenterdeterminationprocedureformultipleearthquakeparameters,butalsoofferstheattractivefeaturesofglobaloptimalsearchingintheRMStraveltimeerrordistribution.Unlikethetraditionalglobalsearchmethod,forexample,theMonteCarloapproach,wheremillionsoftestshavetobedonetofindthefinalglobalsolution,thenewmethodonlyconductsamatrixinversiontypelocalsearchbutdoesitmultipletimessimultaneouslythroughoutthemodelvolumetoseekaglobalsolution.ThesearchisaidedbyinspectionoftheRMSerrordistribution.Benchmarktestsagainsttwopopularapproaches,thedirectgridsearchmethodandtheoct-treeimportantsamplingmethod,indicatethatthehybridglobal-localinversionyieldscomparablelocationaccuracyandisnotsensitivetomodestlevelofnoisedata,butmoreimportantlyitofferstwo-orderofmagnitudespeed-upincomputationaleffort.Suchanimprovement,combinedwithhighaccuracy,makeitapromisinghypocenterdeterminationschemeinearthquakeearlywarning,tsunamiearlywarning,rapidhazardassessmentandemergencyresponseafterstrongearthquakeoccurrence.
出版日期
2009年05月15日(中国期刊网平台首次上网日期,不代表论文的发表时间)