automatic-earthquake-recognition-and-timing-from-s

BulletinoftheSeismologicalSocietyofAmerica,Vol.68,No.5,pp.1521-1532,October1978AUTOMATICEARTHQUAKERECOGNITIONANDTIMINGFROMSINGLETRACESByREXV.ALLENABSTRACTAcomputerprogramhasbeendevelopedfortheautomaticdetectionandtimingofearthquakesonasingleseismictrace.Theprogramoperatesonlineandissufficientlysimplethatitisexpectedtoworkininexpensivelow-powermicroprocessorsinfieldapplications.Intestswithanalogtapesofearthquakes,theprogramcorrectlyidentifiedandtimedtowithin0.05secabout70percentoftheeventswhichwouldnormallybetimedinoperationofanetwork.Theprogramevaluatestheaccuracyofitspicks,anditsestimatesappeartobequitereliable.Thealgorithmisworkingatpresentina16-bitminicomputerandappearstobecompatiblewithpresentlyavailablemicroprocessors.INTRODUCTIONAnewcomputerprogramfortheautomaticprocessingofseismicdatahasbeendevelopedandtestedbytheU.S.GeologicalSurvey.ThegoalhasbeentoproduceacomputerprogramthatcanduplicatetheperformanceofahumanoperatorinprocessingParrivalsofsingleseismometertraces.Itmustrecognizeearthquakes,timethemtoanaccuracyofbetterthan0.05sec,determineamplitudes,directionoffirstmotionandafewotherdescriptiveparameters,andformulateanestimateofhowreliablethepicktimeistobeconsidered.Tobeatalluseful,theprogrammustbecapableofdistinguishingearthquakesfromnoisesourcessuchaswindorvehiculartraffic,anditmustbeabletooperatewithsmallweakearthquakessuchasarecommonlyusedinroutineoperationofalocationnetworksuchastheUSGSCentralCaliforniaNetwork.Contraintsontheprogramaresimple,butratherstringent.Itmustbesuitableforeventualreal-timeoperationinaninexpensivelow-powermicroprocessoratremotefieldlocations.Thisconstraintmeansthattheprogrammaynotlookbackintime,anditmayuseonlyalimitedamountofmemorybothforprograminstructionsandforanyscratch-padstorageusedduringoperation.Bynotlookingback,Imeanthateachsampleisprocessedsequentially,andtheprogramisnotallowedtoaccessearlierpointsunlessithasstoredtheminaseparatescratchpadarray.Thisisanextremelyrigoroussetofrequirements,andinmanycasesnotalltheconstraintswillberequired.Insomecasesforinstance,thealgorithmwillmerelyberequiredtorecognizeanearthquakeforrecordingbyaneventrecorder.Inthiscaseprecisetimingandmostmeasurementswillbemadelaterbyahumanoperator.Inanotherpossibleapplicationanumberofprocessorsmightbelinkedtogetherinprocessingtelemetereddataatacentralcollectionpointforareal-timeearthquakelocator.Inthisuseitwouldbepossibletocross-referenceresultsbetweenstationsandthussimplifytherecognitionproblem.Thealgorithmhasbeendevelopedandtestedunderthemostrigorousconstraintsandifsomecanberelaxeditsperformanceshouldbeevenbetter.Microprocessorshavecomeintouseonlyinthepastfewyears,andarestillevolvingsorapidlythatitisdifficulttostayabreastofwhatisavailable.SoucekAnyuseoftradenamesortrademarksinthispublicationisfordescriptivepurposesonlyanddoesnotconstituteendorsementbytheU.S.GeologicalSurvey.15211522REXV.ALLEN(1976)reviewstheircapabilitiesandusesanddiscussessomeoftherepresentativeprocessorscurrentlyavailable.ProgramdevelopmentforthesystemdescribedherewascarriedoutpartlyonCDC6600and7600computersandpartlyontheUSGSCDC170016-bitcomputerintheMenloParkoffice.WhenoperatingontheCDC6600andCDC7600machines,theprogramprocessespreviouslydigitizeddatasetsupto45secinlength,digitizedat200samples/sec.Inthisprocessing,therestrictionofnotlookingbackisstrictlyobserved:Inthisfashion,arathergoodsimulationofon-lineoperationcanbeachieved.Mostoftheillustrationsinthispaperareofthesedigitaldatasets,becauseitispossibletostoreandlaterplotdetailedtimehistoriesoftheprogramoperation,includingoperationalparameters.Mostoftheprogramdevelopmentandtesting,however,wascarriedoutontheCDC1700computer.Sincethismachinecandigitizeandoperatewithanaloginputsignals,itispossibletotestaprogramwithlongrunsofanalogtapedatasetsortooperateinrealtimebymonitoringselectedtelemetrychannelsoftheUSGSCentralCaliforniaNetwork.Thisabilitytouseanalogdatasetsisabsolutelynecessaryindevelopinganalgorithmthatisresistanttonoisesources.Digitalrecordsareunsatisfactorybecauseseismicreportlibrariescontainearthquakerecords,notnoise,andtherealtestofanautomaticpickingalgorithmisitsabilitytorejectstrongnoiseburstswhilestillrecognizingweakearthquakes.Anillustrationofthisprogram'seffectivenessisprovidedinFigure1,whichplotstenarrivalsfromamagnitude2.8earthquakeatdistancesrangingfrom50to90kmillustratingtheaccuracywhichmaybeexpected.PROGRAMOPERATIONInthefollowingdiscussionofprogramoperation,Ishallusethetermeventtomeanasectionofaseismicrecordwhichisnoticeablydifferentincharacterfromtheimmediatelyprecedingrecord.Anearthquakewillbeaneventinthissense,aswillwindorculturallyproducednoisebursts.Thecharacterofatraceisdescribedbyaspecificfunctionoftraceamplitudeandfirsttimedifference.Iftheseismictraceisregardedasatimeseriesfit)withfirstdifferencef'(t),thenthecharacteristicfunctionE(t)isdefinedasE(t)=f(t)2+(22+f'(t)2.Ceisaweightingconstantwhosepurposeistovarytherelativeweightassignedtheamplitudeandfirstdifferencedependingondigitalsamplerateandnoisecharacter-isticsoftheindividualseismicstation.Thisparticu