NO.1156

JULY2025

LossesfromNaturalDisasters:County-LevelDataonDamages,Injuries,andFatalities

MatteoCrosignani|MartinHiti

LossesfromNaturalDisasters:County-LevelDataonDamages,Injuries,andFatalities

MatteoCrosignaniandMartinHiti

FederalReserveBankofNewYorkStaffReports,no.1156

July2025

/10.59576/sr.1156

Abstract

Weintroducethefirstcomprehensivepubliclyavailabledatasetoncounty-leveldamages,injuries,andfatalitiesfromnaturaldisastersintheU.S.andpresentafewfactsontheeconomicandhumancostsofextremeclimateevents.OursourceistheNationalOceanicandAtmosphericAdministration’sStormEventsDatabase,whichreportslossesforgeographicareaslargelydefinedbasedonmeteorological

science.Wemaptheseareastocountiesusinggeographictoolstogetherwiththespatialdistributionofpopulation,housingstock,andeconomicactivity.Ourestimatesareparticularlyaccurateforsevere

disasters.TheLossesfromNaturalDisastersdatasetisregularlyupdatedat

/research/policy/natural-disaster-losses

.

JELclassification:H12,H71,Q54

Keywords:naturaldisasters,physicalrisk

Crosignani(correspondingauthor),Hiti:FederalReserveBankofNewYork(email:

matteo.crosignani@).TheauthorsthankseminarparticipantsattheNewYorkFedfortheircomments.

Thispaperpresentspreliminaryfindingsandisbeingdistributedtoeconomistsandotherinterested

readerssolelytostimulatediscussionandelicitcomments.Theviewsexpressedinthispaperarethoseoftheauthor(s)anddonotnecessarilyreflectthepositionoftheFederalReserveBankofNewYorkortheFederalReserveSystem.Anyerrorsoromissionsaretheresponsibilityoftheauthor(s).

Toviewtheauthors’disclosurestatements,visit

/research/staff_reports/sr1156.html

.

2

1Introduction

Naturaldisastersgeneratesubstantialcostsforhouseholds,firms,andgovernments.Busi-nessesmonitortheirexposuretophysicalrisks,homeownersfacerisinginsurancepremiums,andthemountingcostsofdisasterresponseoftenstrainpublicbudgets.However,somewhatsurprisingly,empiricalanalysesoftheeconomicandhumancostsofnaturaldisastersarecurrentlyhinderedbydatalimitations.TheNationalOceanicandAtmosphericAdminis-tration(NOAA)collectsofficialdataondamages,injuries,andfatalities,butnearly40%oftheseestimatesarereportedformeteorological“zones”—geographicunitsthatmightspanmanycounties(e.g.,avalleyintersectingthreecounties)orincludeonlypartsofone(e.g.,thecoastalportionofonecounty),makingitdifficulttomergelossesfromNOAAwithstandardadministrativedatasets.

Inthispaper,weintroduceLossesfromNaturalDisasters—newandcomprehensivedataoncounty-leveldamages,injuries,andfatalitiesfromnaturaldisastersintheU.S.from1996tothepresent.Usinggeographictoolstogetherwithweightingschemesbasedonthespatialdistributionofpopulation,housingstock,andeconomicactivity,weapportionzone-leveldamagestocounties(datauserscanchoosetheweightingschemethatbestsuitstheirempiricalanalysis).LossesfromNaturalDisastersprovidesnominalandinflation-adjusteddamages,injuries,andfatalitiesforeachnaturaldisasterreportedbyNOAA.Ourcounty-levelestimatesareparticularlyaccurateforseveredisastersandcanbeeasilyaggregatedtodifferentlevels,suchascounty-month,county-month-disastertype,state-year,andsoon.Finally,weuseourdatatodiscussthreefactsontheeconomicandhumancostsofnaturaldisasters.

LossesfromNaturalDisastersispubliclyavailableandregularlyupdatedat

/research/policy/natural-disaster-losses

.Thiswebpagealsoincludesinterac-tivemaps,FAQs,datadictionaries,andotherdetailsaboutthedataset.

Figure1

showsaheatmapoftotalcounty-leveldamagesfromfloods,hurricanes/tropicalstorms,andcoastaldisasters(toppanel)andwildfires(bottompanel)between1996and2023usingourdata.

1

1Coastaldisastersareastronomicallowtide,coastalflooding,highsurf,ripcurrent,andstormsurge/tide.

3

DamagesfromFloods,Hurricanes/TropicalStorms,andCoastalDisasters

DamagesfromWildfires

Figure1:Totaldamagesfromflooding,hurricanes/tropicalstorms,coastaldisasters,and

wildfiresfrom1996to2023.Thisfigureshowsaheatmapoftotalcumulativecounty-leveldamagesfromflooding,hurricanes/tropicalstorms,andcoastaldisasters(toppanel)andwildfires(bottompanel)between1996and2023.

TableOA.3

providesdefinitionsofeachdisastertype.DamagesareinflationadjustedtoDecember2023USDusingtheCPI.

FigureOA.2

showsdamagesfromalldisastertypesanddroughts.Sources:NOAA,U.S.CensusBureau.

4

Thesemaps,obtainedusingpopulation-basedweights,documentthatFlorida,theSoutheast,andpartsoftheEastCoasthavesufferedthemostfromflooding,hurricanes,andcoastaldisasters,whiletheWesthasbeenparticularlyaffectedbywildfires.

OurmaindatasourceistheStormEventsDatabase(SED),anofficialNOAApublication.

TheSEDrecordssignificantorunusualweatherphenomenathatarecollectedinanationaldatabasebytheNationalWeatherService(NWS)andclassifiedintoover55typesofweathereventssince1996(wefurthergroupdisastertypesinto13broadercategories).TheSEDreports(i)directpropertydamages,(ii)cropdamages,and(iii)directandindirectfatalitiesandinjuries.Asanticipated,thelocationofeacheventiseitheracountyorazone,wheretheNWSuseszonestoallowformoreaccurateforecastsdueto“suchthingsaselevationorproximitytolargebodiesofwater.”

Ourmethodologytomapalleventstocountiesisbasedonthreesteps.First,weusehistoricalmapstoidentifywhichcounty(orcounties)eachzoneintersects.Second,forzonesthatoverlapmultiplecounties,weuseCensusblockgroupdatatoconstructasetofweightsthatallocatedamagestocountiesbasedonthespatialdistributionofpopulation,housingstock,andeconomicactivity(employmentandincome).Forexample,if75%ofazone’spopulationlivesincountyAandtheremainderincountyB,thepopulationweightsassign75%ofthereporteddamagesinthatzonetocountyAand25%tocountyB.Themappingfromzonestocountiesandthesetofweightsleadtoacrosswalkfromzonestocounties.WebuildthismappingforthecurrentandhistoricalvintagesofNOAA’sgeographicunitssothatthecrosswalkisbasedonboundariesthatwereineffectatthetimeofthedisaster.

Ourmethodologyleadstocounty-levelestimatesthatareparticularlyaccurateforseveredisasters.Westartbydocumentingthatourweightingschemesleadtosubstantiallydifferentestimatescomparedwithna¨?ve“equalweighting”schemes(commonlyusedinexistingwork)inthecontextofthe2017–2023WesternWildfiresandHurricaneKatrina,twodisasterslargelystudiedintheliterature(e.g.,

Issleretal.

,

2021

;

Deryuginaetal.

,

2018

).Wethenanalyzethefullsampleofdisastersfrom1996and2023anddocumentthatourestimatessystematicallydeviatefromana¨?veequal-weightingscheme,especiallyformoredamagingdisasters.Thisempiricalregularityisconsistentwiththehigherlikelihoodofhigherdamagesinurbanareasandtheobservationthatseveraloftheespeciallycostlydisastertypes(e.g.,

5

hurricanesandwildfires)areonlyreportedatthezonelevelbyNOAA.

Wealsodocumentthreefactsabouttheeconomicandhumancostsofnaturaldisasters.

First,theseverityofdisastersisveryskewedwithasmallshareofdisastersbeingresponsibleforthemajorityofdamages,injuries,andfatalities.Forexample,whilethemediandamageforacounty-yearhitbyadisasteris§4.6perperson,the99thpercentileisastaggering§4,355.8.Thisskewnessisparticularlypronouncedforhurricanes/tropicalstorms,coastaldisasters,anddroughts—andisalsopresentwhenlookingatinjuriesandfatalities.Second,thefrequencyandseverityofdamagesarenegativelycorrelatedacrossdisastertypes.Inotherwords,verydamagingdisasterstendtoberelativelyinfrequentwhilelessdamagingonestendtobeparticularlyfrequent.Forexample,droughts,hurricanes/tropicalstorms,andcoastaldisastersareextremelydamagingandinfrequent,whilewindandfloodsarefrequentandlessdamaging.Third,sometypesofdisasters,suchashurricanes/tropicalstorms,floods,andcoastaldisasters,havebecomemoresevereinthelast25years,whileotherdisastertypes,suchasdroughtandwildfires,havebecomelesssevere.

Relatedliterature.Ournew,comprehensive,publiclyavailabledatasetondamages,injuries,andfatalitiesfromnaturaldisasterscontributestothelargeandgrowingliteraturethatusescounty-leveldataondamagesfromnaturaldisastersineconomics(e.g.,

Barrotand

Sauvagnat

,

2016

;

Fried

,

2022

)andfinance(e.g.,

Bernileetal.

,

2017

;

Heimeretal.

,

2019

;

Ge

,

2022

;

Aloketal.

,

2020

;

Lietal.

,

2024

;

Morse

,

2011

;

Cort′esandStrahan

,

2017

;

Dessaintand

Matray

,

2017

;

GeandWeisbach

,

2021

;

Diamondetal.

,

2024

;

Ersahinetal.

,

2024

;

Aretzet

al.

,

2019

;

Sch¨uweretal.

,

2019

;

Choietal.

,

2023

;

HuynhandXia

,

2023

;

Dlugoszetal.

,

2024

).

Outline.Theremainderofthepaperisstructuredasfollows.

Section2

presentstheSEDdata.

Section3

discussesourmethodologytoconstructcounty-leveldamages,injuries,andfatalities.

Section4

showsthatourweightingschemeleadstodifferentestimatescomparedwithna¨?veequalweighting,especiallyforseveredisasters.

Section5

presentsthreefactsaboutdirectdamages,injuries,andfatalitiesfromnaturaldisasters.

Section6

concludes.

6

2Theofficialdatasource

Wenowprovideabriefdescriptionofourmaindatasource.ThesourceofourestimatesistheStormEventsDatabase(SED),anofficialpublicationoftheNationalOceanicandAtmosphericAdministration(NOAA).Thereaderisreferredto

NWSDirective10-1605

foradetaileddiscussionoftheSEDdata.

TheunitofobservationintheSEDisan“event.”Aneventmustmeetoneofthefollowingcriteria:(i)theoccurrenceofstormsandothersignificantweatherphenomenahavingsufficientintensitytocauselossoflife,injuries,significantpropertydamage,and/ordisruptiontocommerce;(ii)unusualweatherphenomenathatgeneratemediaattention(e.g.,snowflurriesinSouthFlorida);or(iii)othersignificantmeteorologicalevents,suchasrecordmaximumorminimumtemperatures.

Theunderlyingdatacomesfromavarietyofsources,including,amongothers,NationalWeatherService(NWS)trainedspotters,county,state,andfederalemergencymanagementofficials,locallawenforcement,andmediareports.

2

Thedisparaterecordsarecollected,compared,andverifiedby“stormdatapreparers”atthe123NWSWeatherForecastOffices(WFOs).Afterthisverificationprocedure,theWFOssendthedatafilestotheNWSheadquarters,whichstandardizestheformatandupdatesthenationaldatabase.TheSEDisupdatedwithapproximatelyathree-monthlag(e.g.,thedataforJanuaryisusuallymadeavailableinApril).From1996tothepresent,NOAAhasincludedover55typesofweathereventsinthedatabase,rangingfromhurricanestoduststormsandflashfloods.

TableOA.3

providesacompletelistoftheeventtypesintheSEDdata.Thetablealsoprovidesacategorizationofeacheventtypeintoabroader“disastertype”category.

WeprimarilyrelyontwosetsofvariablesfromtheSED—thosethatquantifylossesfromeventsandthosethatdocumenttheirlocation.

2

TableOA.1

and

TableOA.2

reporttheshareofeventsintheNOAAdatabysource.Approximately65%oftheeventscomefromNOAA/NWSorothergovernment-affiliatedofficials.

7

Lossesfromnaturaldisasters.TheSEDreportsestimateddirectpropertyandcropdamages,alongwithdirectandindirectfatalitiesandinjuries.

Propertydamagereferstodamageinflictedtoprivatepropertyaswellaspublicinfrastruc-tureandfacilities.Damageisreportedif“areasonablyaccurateestimate”isavailablefromaninsurancecompanyorotherqualifiedentitiesincluding“emergencymanagers,U.S.GeologicalSurvey,U.S.ArmyCorpsofEngineers,utilitycompanies,andnewspaperarticles.”Cropdamagesareobtainedfromthe“U.S.DepartmentofAgriculture(USDA),thecounty/parishagriculturalextensionagent,thestatedepartmentofagriculture,cropinsuranceagencies,oranyotherreliableauthority.”

TheSEDonlyreportsdirectdamages,i.e.,monetizedvaluesofphysicaldestruction.

Indirectdamages,thosethathappenaftertheeventandareaconsequenceofdirectlosses,arenotincludedinthedamageestimates.Forexample,arupturedpowerlinethatfallsonahomeduringatornadoresultsindirectdamages.Subsequent(indirect)lossesmayoccuriflocalbusinessesareforcedtocloseduetoalackofpowerfromthedownedline.NOAAreferstotheseindirectdamagesas“otherrelatedcosts,”andStormDataPreparersareinstructednottoincludethemintheSED.

TheSEDalsorecordsdirectandindirectfatalitiesandinjuries.Stormdatapreparersareresponsiblefordeterminingthenumberoffatalitiesandinjuriesandwhethertheyareadirectorindirectresultoftheevent.Directfatalitiesandinjuriesarethoseinwhichthe“active”agentistheweathereventitselforthedebriscreatedbytheevent.Indirectfatalitiesandinjuriesoccurnearanevent,butarenot“directlycausedbyimpactordebrisfromtheevent.”Forexample,“fatalities/injuriescausedbywind-drivendebrisduringahurricane”aredirect,and“avehicleaccidentcausedbyahurricane-relatedmissingtrafficsignal”isconsideredanindirectinjuryorfatality.

Locationofnaturaldisasters.TheSEDconsistentlydocumentsthelocationofdisastersusinggeographicareas,afieldpopulatedforallobservationsintheSED.ThetypeofgeographicareaintheSEDdependsontheeventtype.Asshownin

TableOA.3

,10typesofeventsarereportedatthecountyleveland34typesofeventsarereportedatthezonelevel.Theremaining11eventtypesaredesignatedformarinezonesandcategorizeeventsthatdo

8

notoccuroverland.ZoneswereestablishedbytheNWStoallowformoreaccurateforecastsbecauseofthe“differencesinweatherwithinacountyduetosuchthingsaselevationorproximitytolargebodiesofwater.”

3

Forexample,onecountymaybesplitintotwozones,acoastalandinlandportion,orazonemaytracetheboundariesofavalleywhichoverlapsmultiplecounties.Generally,zonesaresmallerthancounties,butanindividualzonecanintersectwithmorethanonecounty.

Importantly,eventsaredefinedtobezone-specificorcounty-specific.Hence,ifthesamenaturaldisaster(orstormsystem)affectsmultiplezonesorcounties,thedatafeaturesaseparateevent(i.e.,aseparateobservation)foreachaffectedzoneorcounty.

4

Toconnecteventsthatarepartofthesamenaturaldisasterbutoccurindifferentareas,theSEDassignseacheventtoabroader“episode.”Wepreservethisevent-to-episodemappingtoallowresearcherstotrackthespatialdistributionoflossesfromagivendisasterorstormsystem.

3Mappinglossestocounties

Wenowdiscusshowweconstructourcounty-levelestimatesoflossesfromnaturaldisasters.Specifically,wedescribehowwebuildcrosswalksfromhistoricalmapstoallocatedamagesintheSEDtothecountylevelusingweightsthatreflectthespatialdistributionofpopulation,housingstock,andeconomicactivity.

ThedatachallengewithPublicForecastZones.Whileallowingformoreaccurateweatherforecasts,zonespreventtheSEDfrombeingreadilyusedforcounty-levelanalyses.Indeed,about40%ofalleventsarereportedatthezonelevelintheSED.AlthoughNOAAprovides“zone-countycorrelationfiles”thatmapeachzonetooneormorecounties,twolimitationspreventthesecrosswalksfrombeingsufficienttoaccuratelycalculatecounty-leveldamages.First,NOAAregularlychangesthezoneboundariesandidentifiers(e.g.,splitting

3See

/gis/PublicZones

formoredetailsaboutzones.

4Weusetheterm“stormsystem”torefertobroaderdisasters/weatherphenomenasuchasHurricaneHarveyorthe2023HawaiiFirestorms.

9

onezoneintotwo),butthecorrelationfilesonlycontainamappingforthemostrecentvintagesofthezones.Hence,merginghistoricaleventswiththecurrentcrosswalkleadstounmatchedeventsandpotentiallyinaccuratemapping.Second,thecrosswalksdonotprovidearuleforapportioningzone-leveldamagestocountieswhenthezonespansmultiplecounties,whichoccursforapproximately22%ofzonesasof2023.

Giventheseissues,weconstructourowncrosswalksusingzoneandcountymaps.Thecrosswalkslinkeachzonetoallcountiesthatthezoneoverlaps.Forzonesthatoverlapmultiplecounties,wecalculateasetofweightsthatcanbeusedtoallocatedamagestocountiesbasedontheshareofazone’spopulation,housingstock,oreconomicactivitylocatedineachoverlappingcounty.Crucially,weconstructthesecrosswalksusingcurrentandhistoricalvintagesoftheNOAAzones,ensuringthatourallocationofdamagestocountiesisconsistentwiththezoneboundariesatthetimetheeventoccurred.

Constructingcrosswalks.Tobuildthecrosswalks,weoverlayshapefilesofNOAAzonesandU.S.countiesfromthesameyearasthevintageoftheNOAAzones.

5

Shapefilesareadataformatthatcontainacollectionofpolygonsorientedinspace.Theintersectionofzonesandcountiesgeneratesanewmapconsistingofsmallerandmutuallyexclusivespatialsubunits.Eachofthesesubunitsisauniquezone-countypairwithanon-zerogeographicintersection.UsingCensusblockgroup-leveldata,wenextcalculatethearea,population,employedpopulation,aggregateincome,andthenumberofhousingunitsineachzone-countypair.Inthecaseofpopulation,wethencalculatetheweightsforeachzone-countypairbydividingthepopulationofthezone-countypairbythetotalpopulationofthezone.Weusethesamemethodologyforweightsbasedonarea,employedpopulation,aggregateincome,andnumberofhousingunits.

Forexample,if80%ofazone’spopulationislocatedincountyAand20%incountyB,ourpopulation-basedweightsallocate80%ofdamagesfromweathereventsinthezonetocountyAandtheremaining20%tocountyB.Azonelocatedentirelywithinonecounty

5Weconstructmultiplecrosswalks(oneperyear)forvintagesthatwereineffectovermultipleyears.

10

hasaweightof1,soalldamagesareassignedtothatcounty.Weconstructsimilarweightsusingdifferentweightingvariables,namelygeographicarea,employedpopulation,aggregateincome,andthenumberofhousingunits.Webenchmarkourmethodologyagainstanaequally-weightedweightingscheme(used,tothebestofourknowledge,bycommerciallyavailabledatasets)thatdistributesdamagesequallyamongallcountiesthatintersecttheaffectedzone.

Example:Zone504inWashingtonstate.Tomakethisproceduremoreconcrete,

Figure2

providesanexampleofthecrosswalkconstructionusingzone504inWashingtonstatefromthe2017vintageofzoneboundaries.PanelAshowsthezoneoverlaidontotheWashingtoncountymap,andPanelBzoomsinonthethreecountiesthezoneoverlaps,namelyMason,Thurston,andLewiscounties.PanelCillustratesthefirststepoftheprocedure,i.e.,thespatialintersectionofthezoneandcountyboundaries.Theresultisadatasetwherethereisonlyoneobservationforeachzone-countypair.ThethreecoloredpolygonsinPanelCrepresenteachobservationinthespatialintersection.

Next,weconstructasetofweightsthatapportiondisasterdamagesfromzoneztocountiesc1,...,cn(inthisexample,n=3).Formally,theweightforeachzone-countyintersectionz∩ciis

wheref(z∩ci)isthevalueofoneofourweightingvariables(area,population,employment,numberofhousingunits,orincome)forthezone-countyintersection.Intuitively,theweightsrepresenttheshareofthezone’stotalpopulation(oranyotherweightingvariable)thatislocatedinaparticularzone-countysubunit.Thena-weightedweightingschemeforeachzone-countypairissimplycalculatedas1/n.

Weconstructtheweightsbycalculatingthevalueoftheunderlyingweightingvariableforeachzone-countypair.Fortheareaweightingvariable,weperformthiscalculationusingtheGeographicInformationSystem(GIS)software.Valuesoftheeconomicanddemographicvariablescannotbedirectlyobtainedbecausethesesubunitsdonotconformtostandardadministrativeboundaries.Hence,weestimatethesevaluesusingAmericanCommunitySurvey(ACS)5-yeardataattheblockgrouplevel.Theprocedureforobtainingourestimates

11

PanelA:Zone504inWAPanelB:Zone504

PanelC:Zone-CountyIntersectionsPanelD:OverlayingCensusBlockGroups

Figure2:Exampleofcrosswalkconstruction.Thisfigureshowsanexampleofcrosswalkconstruction

forzone504inWashingtonstate.PanelAshowsthezoneoverlaidontothecountymap.PanelBzoomsinonMason,Thurston,andLewiscounties—thecountiesthatoverlapwithzone504.PanelCandPanelDillustratethefirstandsecondstepoftheprocedure,respectively.Sources:NOAA,U.S.CensusBureau.

isshowninPanelDof

Figure2

,whichoverlayscensusblockgroupsontothezone-countysubunits.Wecalculatethevalueoftheweightingvariablebysummingacrossallblockgroupslocatedinthezone-countypair.Ifablockgroupislocatedinmultiplezone-countypairs,beforeaddingtheweightingvariabletothezonecounty-total,wemultiplyitbytheshareoftheblockgroup’sgeographicareathatislocatedinthezone-countypair.

Inourexampleofzone504inWashingtonstate,thepopulationweight(0.799)andeconomicweights(0.820,0.842,0.793foremployment,income,andhousingstock,respectively)forThurstonCountyaremorethandoublethanitsequalweight(0.333)andarea-basedweight(0.404).ThisdiscrepancyisduetothisportionofThurstonCountycontainingthecityofOlympia.Interestingly,thenacountywhich,asshowninthefigure,onlysharesaverytinyintersectionwithzone504.

12

Finalsteps:50states(plusD.C.andPuertoRico)and46vintages.Theprocessusedinthisexampleisexecutedforeachzoneinall50statesplusD.C.andPuertoRico.ThecrosswalksforeachstatearethenappendedtoformasinglecrosswalkforeachNOAAmap.Werepeatthisprocessforthe46vintagesofNOAAmapsthatwereeffectivebetween2006and2023.Forvintagesthatwereineffectacrossmultipleyears,weconstructmultiplecrosswalks(oneforeachyear)usingthecorrespondingcountyboundariesandCensusblockgroupdata.Hence,forthisperiod,weareabletomergetheeventswiththecrosswalkconstructedfromthezoneboundariesthatwereineffectatthetimeoftheevent.Foreventsthatoccurredfrom1996to2006,wemergeeventswithacrosswalkcreatedfromtheoldestvintageofzonesforwhichweobtainedashapefilemap,namely2006.

Asafinalstep,wemergeallzone-leveleventswiththecrosswalkfromthezoneboundariesthatwereineffectatthetimeoftheevent.WeappendthisdatatotheeventsintheSEDthatarereportedatthecountylevel.Theseeventsareassignedaweightofoneforeachofthesixmethodologiessincetheyoccuronlyinonecounty.

Thefinaldataset.Ourfinaldatasetisanevent-leveldatasetthatcanbecollapsedtothecountylevelbyselectingaweighttypeandsummingtheweighteddamagesacrossalleventsinthecounty.Wegroupeventsin13broaddisastertypes:coastaldisaster,winterweather,wind,flood,drought,wildfire,tornado,hail,heat,hurricane/tropicalstorm,thunderstorm/rain,tsunami/seiche,andother.Thedatacanbeeasilyaggregatedtodifferentlevelssuchascounty-month,county-month-disastertype,state-year,andsoon.

4Precisedamageestimates

Wenowshowthattheweightingschemesinourmethodologyleadtoestimatesthatareparticularlyaccurateforseveredisasters.First,wedocumentsubstantialdifferencesinestimatescomparedtona¨?veequalweightsinthecontextofthe2017–2023WesternWildfiresandHurricaneKatrina.Wethenshow,moreformally,thattheaccuracyofourdataincreasesindisasterseverity.

13

WesternWildfires(2017–2023)

HurricaneKatrina(2005)

Figure3:Population-weightedrelativetoequal-weighteddamages:2017—2023WesternWildfiresandHurricaneKatrina.Thisfigureplotstherelativedifferencebetweenpopulation-weightedandequal-weighteddamagesforthe2017-2023WesternWildfiresinWashington,Oregon,andCalifornia(leftpanel)andHurricaneKatrinainMississippiandLouisiana(rightpanel).WedefineHurricaneKatrinatoincludealleventswithadisastertypeofhurricane/tropicalstorm,coastaldisaster,orfloodwithastartdatebetweenAugust25,2005,andAugust30,2005.See

TableOA.3

formoredetailsonwhich

eventsareincludedinthesedisastertypes.Therelativedistancemetricisdefinedasd|Equal=

(PopulationWeightedDamagesc-EqualWeightedDamagesc)/EqualWeightedDamagesc.Sources:NOAA,U.S.CensusBureau.

HurricaneKatrinaandthe2017—2023WesternWildfires.Wenowdocumenttheimportanceoftheweightingschemetoaccuratelycomputedamagesfromnaturaldisasters.Westartbyfocusingonthe2017–2023WesternWildfiresandHurricaneKatrina,twodisastertypes(wildfiresandhurricanes)reportedatthezone-levelbyNOAA.Theseepisodeshavebeenwidelystudiedintheliterature.Forexample,

Issleretal.

(

2021

),

McConnelletal.

(

2021

),and

Biswasetal.

(

2023

)investigatetheimpactofwildfiresonmigrationandthehousingmarket,and

Vigdor

(

2008

),

GallagherandHartley

(

2017

),

Deryuginaetal.

(

2018

),and

Groen

etal.

(

2020

)examinetheeconomicandfinancialconsequencesofHurricaneKatrina.

14

Figure3

showsthatdamagesestimatedusingpopulation-basedweights