2018 Annual Report #17134 Testing Fault-based Forecasts

Principal Investigator: David D. Jackson June 13, 2018

Abstract ThegoalsofthisprojecthavebeentoexaminetheimplicationsofthehiatusofpaleoseismiceventsinCaliforniasince1918,tostudymethodstoassociateandtestobjectivedescriptionsoffiniterupturesonfaults,andtotesttheagreementofpastforecastsbytheWorkingGrouponCaliforniaEarthquakeProbability(WGCEP)withsubsequentearthquakes.ThepaleoseismichiatusinCaliforniaisstatisticallyinconsistentwiththerateofeventsestimatedfrompaleoseismicdata,almostallofwhichwereestimatedbeforeseismicnetworksbeganoperatingintheearly20thcentury.Hypothesesforthediscrepancywereextremeluck,anunknown(andcounter-intuitive)physicalmechanismforcoordinatingrupturestatewide,andover-estimationofearthquakeratebyincludingnon-earthquakeeventsbeforeseismicnetworkscouldverifytrueseismicevents.Eachofthesehypotheseshasquitedifferentimplicationsforuseofpaleoseismicdatainearthquakeforecasts.We’vediscoveredthatasimilarhiatusoccurredinNewZealand.ExplainingthejointobservationsforCaliforniaandNewZealandprettymuchrulesoutthefirsttwohypotheses,implicatingover-estimation.We’vetestedboththe1988and1995WGCEPforecastsretrospectivelyandprospectivelyusingmodificationsoftheCSEPN-testandS-test.The1988forecastisforlargeearthquakesonthemajorfaults,assumingquasi-periodiccharacteristicsonprescribedfaultsegments.Since1988onlytheParkfieldearthquakeof2004matchestheprescribedsegmentdescription,andtheforecastfailstheN-testat95%confidence.Ifthesegmentboundariesareretrospectivelyadjustedtomatchtheruptureofthe1989LomaPrietaEarthquake,theN-testisbarelymetat95%.TheforecastisconsistentwiththeS-testofearthquakelocation.The1995WGCEPassociateseachearthquakewithoneof65seismotectoniczones,andincludesbothatimedependentandtimeindependentforecast.Bothsignificantlyover-predictthenumberofexpectedeventsafter1995,andthetime-dependentforecastsignificantlyover-predictsthenumberfrom1932tillnow.

Intellectual Merit Ourprojectexplorestheextenttowhichpriorearthquakeforecasts,basedoncommonassumptions,agreewithearthquakeoccurrencebeforeandaftertheforecasts.Thusiscontributestotheunderstandingofearthquakeprocesses,amajorgoaloftheEarthquakeForecastingandPredictionprojectinSCEC.Wealsodevelopedmethodsforassociatingfiniterupturesonfaultswithprescribeddescriptionsofsuchevents,anecessarystepinevaluationforecastsofsuchevents,andwetestedsomeoftheseideasonforecastsoftheWorkingGrouponCaliforniaEarthquakeProbabilities(WGCEP).AssociationandtestingoffiniteruptureshavebeenmajorgoalsoftheCollaboratoryforStudyofEarthquakePredictability(CSEP).Broader Impacts. Ourresearchaffectsseismichazardestimation,whichhasimplicationsforpublicsafety,riskmanagement,andpublicpolicy.We’vepresentedourresultstostudentsandfacultyatuniversities,includingUCLA,UCR,ETHZurich,VictoriaUniverityWellington,andatthe2017SCECAnnualmeeting,showinghowPhysics,Geology,Geodesy,andStatisticscanbecombinedtosolverealworldproblems.We’vediscussedouranalysiswithstudentsandfacultyinStatistics,showingthemapplicationsandencouragingtheirparticipationinseismologicalresearch. Technical Report Paleoseismic Tests WecontinuedanalysisofpaleoseismicdatesandratesinCaliforniaandNewZealand.The32paleositesusedintheUCERF3reportand29sitesinNewZealandshowednoeventsinthelastcenturyexceptforonesecondarydisplacementinNewZealand.StatisticalaspectsofthehiatusinbothlocationswerereportedthisyearatStatSei10,(10thStatisticalSeismologyconferenceinNewZealand),theSSAmeetinginDenver,andaEuropeanProbabilisticSeismicHazardmeetingatLenzberg,Switzerland.NumericalresultsforCaliforniawerepresentedinlastyear’sreport,sowe’llfocushereonresultsforNewZealandasshowninTable1.

Table1.PaleoseismiceventdataforNewZealand,compiledbyNicoletal., withratecalculationsprovidedbyusfromdataintheNicoletal. paper. Theprobabilityforhiatussince1918inCalifornia,givenpasteventrates,islessthan1.4percent.ForNewZealand,thecomparableprobabilityisabout4percent.Theprobabilityofbothoccurringatrandominthelastcenturyislessthan0.1percent.WesearchedforreasonswhysitesthroughoutCaliforniaorNewZealandmightbeinternallydependent,withnosatisfactoryanswers.Inter-dependencebetweenCaliforniaandNewZealandpaleoeventsisevenmoredifficulttoexplain.ForCaliforniasiteswelistedthreepossiblealternatives:extremeluck,unknownphysicalinteractionsoverlargedistances,oroverestimationofeventratescausedbylackofseismologicalconfirmationbeforeseismicnetworkswereinstalledintheearly20’thcentury.WithNewZealanddataincludedinthepuzzle,thesolutionappearsmoreandmoretobeoverestimationofratesbeforeseismicnetworks.Testingofthe1988and1995ForecastModelsoftheWorkingGrouponCaliforniaEarthquakeProbabilities(WGCEP) Wesetupcriteriaforprospectivefault-ruptureforecasts,andshowedexamplesusingretrospectivetestsonearthquakes.Weusedonlyinstrumentallyrecordedeventsafter

1932.Acentralproblemforfiniterupturesis“association”:findingobjectivemeasurestodescribeearthquakerupturesinordertoassignaprobabilitytoeachfutureevent.TheWorkingGrouponCaliforniaEarthquakeProbabilitieshaspublishedlong-termearthquakerateforecastsforCaliforniasince1988.Wefocusedfirstonprospectiveandretrospectivetestingofthe1988and1995versions,becausethosecontaininformationbywhichprobabilitiescanbeassociatedwithanyfutureearthquakeovermagnitude6.0.The1988WGCEPforecastwasbasedonthecharacteristicearthquakeassumptionandwasspecificallytimedependent.Fortestingpurposesweconstructedatime-independentmodelassumingPoissonrecurrencewiththesameratesonthesamesegmentsadoptedbyWGCEP.The1995forecast,forSouthernCaliforniaonly,attributedearthquakesto65“seismotectonic”zonesthatcoveredtheentirestudyarea.Weparticipatedinbi-weeklyphoneconferencewithotherCSEPmembers,andacommonthemeisbroadeningCSEPtestingbeyondepicenterratestofinitesurfacerupture.Forthemtheassociationproblemischallenging,becausefiniteruptureshavemoredimensionsthanhypocentersdo,theyneedsomequantitativedescriptions,andthereisnoauthoritativecompilationasthereisforepicenters.Webeganwithretrospectivetestsofpublishedforecastslikethe1995WGCEPreport(Jacksonetal.,1995)whichprovidesfault-centeredpolygonswhichcancapturefiniterupturesonornearknownmajorfaults.“Near”isimportant,becauselargeearthquakesmaygetoffthetracks.WGCEPforecastsgenerallycover30-yeartimeintervals.Revisionscomemorefrequentlythanthatbutthefundamentalideasandassumptionsintheforecastsareolderthan30years.Bothretrospectiveandprospectivetestsarestillimportantforseveralreasons.Testing“outdated”modelscanstillrevealimportantinformationaboutwhatworkedwellordidn’t,andformulatingthetestscanbeveryhelpfulinunderstandingwhattheforecastmodelisreallyabout.IsitmeanttoillustratetheimpactofcertainassumptionslikeCoulombstresstriggeringoron-faultmagnitudedistribution,orrathertoprovidearobustsourcemodelforhazardestimates?Isthetargetlargestructureslikebridges,infrastructure,orprivatehomes?Eachsuggestsdifferenttestmeasures.Ideallyforecastsandtestplansshouldbedevelopedtogether.Ourpurposeistoexpandthemenuoftestingoptionsbytestingpastforecasts.Figure1belowshowsthegeometryof65“seismotectonic”zonesusedinthe1995WGCEPreportandthelocationsofmagnitude6+earthquakessince1932.TheN-test,whetherthetotalnumberofeventsisconsistentwiththeforecast,issimplest.Theforecastedannualratesofearthquakesinthe1995reportareshowninthethirdcolumnofTable2.Thefourthcolumn,lamda,istheexpectedratefor86yearintervals(theintervalbetween1932and2018),Nistheobservednumberinthelast86years,andProb(0-N)istheprobabilitythattheobservednumberorfewerwouldoccur,basedonaPoissonmodel.ThePoissonmodelisappropriateforthe“TI”orTimeIndependentalternativemodel.The“TD”orTimeDependentmodel,describedas“preferred”inthe1995model)assumeslognormalrecurrence.The2010ElMayor-Cucupaheventwasnotcountedbecauseitsepicenterwasoutsidethestudyarea.Includingitwouldnotchangetheresults.Evenretrospectively,both1995modelsover-estimatedtheratesof

magnitude6+and7+earthquakes.Thediscrepancyat6+fortheTimeDependentmodelishighlysignificantand,well,dramatic.ThebasicassumptionsoftheTDmodelarestillusedinmodifiedforminUCERF3andotherforecasts,sofurtherexaminationisappropriate.

Figure1.Earthquakesafter1932occurringwithinthe65"seismo-tectonic"zonesusedinthe1995WGCEPreport.Thereare28eventsovermagnitude6,butonlythree(HectorMine1999,SanSimeon2003,andParkfield2004)occurredfrom1995to2018.The1995WGCEPtime-dependentmodelforecastover13suchevents,andthetimeindependentmodelforecastover9suchevents.

Table2.RetrospectiveperformanceoftheTimeDependentandTimeIndependentversionsofthe1995WGCEPforecastonearthquakessince1932.LastcolumniscumulativePoissonprobabilityofNgivenexpectednumberlamda.Redhighlightindicatesstatistically significantdiscrepancy.Twentyfouryearshaveelapsedsincethe“30year”1995reportwasfinalized,butwehavedoneanearlyprospectivetestwithjustafewadjustments.Inthiscasewe“rolledtheclockback”tocalculatetheTimeDependentprobabilitiesfora23yearinterval.Amapoftheearthquakessince1995isshowninFigure2,andresultsinTable3.Onlythreem6+andonem7+eventshaveoccurredsince1995.The1995reportcontainedspecificdescriptionsof“characteristic”earthquakes,includedmagnitudesandpossiblecombinationsofinvolvedfaultsegments.Nosucheventsoccurredafter1995.Again,boththeTDandTIforecastsover-predictedtheearthquakerate,significantlysoforthoseinredfontinTable3.TheTIandTDmodelsbothpassthemagnitudedistributiontests(M-test)andconditionallocationstests(S-test).Clearlythe1995forecasts,especiallythepreferredmodel,over-estimatedthefutureearthquakeratesignificantly.The1995reportacknowledgedthattheestimatedratesexceededthehistoricrateof0.35/a,buttheassumptionwasmadethatanearthquakedeficitwouldhavetoberepaid.Thelesson:instrumentallydeterminedearthquakedataafter1932donotsupportthetime-dependentmodelforsouthernCalifornia.AsecondlessonvisibleinFigure1isthattheearthquakesoccurrednearbutnotpreciselyonpreviouslymappedfaults.Futureforecastscoulduseaftershockstomeasurefinitefaultextentinprospectivetests,butassociatingearthquakesbyzoneismorerobust.

Table3.EarlyprospectiveN-testof1995WGCEPforecast.

Figure2.Mapofseismo-tectoniczoneusedinthe1995WGCEPforecast,andm=3+aftershockswithin30daysofthefoursubsequentm=6+earthquakes.Probabilitieswereassignedtoeachofthezonesin1995.Aftershockzonesareadequatetoassigneachm=6+eventtoazone,butnotnecessarilytoanindividualfault.Publications Jackson,D.D.2018.TestingtheClassic1988Forecast.SeismologicalResearchLettersdoi:https://doi.org/10.1785/0220180039WeiserD,PortoN,JacksonD.D.2018,FaultDimensionsandMaximumMagnitude.Proceedingsofthe11thNationalConferenceinEarthquakeEngineering,EarthquakeEngineeringResearchInstitute,LosAngeles,CA.2018Presentations Jackson, D.D., 2018. Testing Fault-Based Rupture Models. Presentation at Seismological Society of America, Miami FL, 5/16/2018.

Jackson, D.D. , 2018. The Collaboratory for the Study of Earthquake Predictability: Achievements and Priorities. Presentation at Seismological Society of America, Miami FL, 5/17/2018 Jackson, D.D., 2017. Large earthquake rates from geologic, geodetic, and seismological perspectives, Presentation and Annual Meeting of the American Geophysical Union, New Orleans, LA Jackson, D.D. , 2017. Mind the Gap, Keynote Presentation at 10th International Workshop on Statistical Seismology, 21 February 2017, Wellington, New Zealand. doi: 10.21420/G23S33 Jackson, D.D. , D. Weiser, and N. Porto, 2017. Resistance to Rupture, Keynote Presentation at PSHA Workshop “Future directions for probabilistic seismic hazard assessment at a local, national, and transnational scale”, Lensburg, Switzerland, Sep 5 – 7, 2017