Systemwide Intersection Safety Prioritization Development and

Click here to load reader

  • date post

    03-Jan-2017
  • Category

    Documents

  • view

    215
  • download

    0

Embed Size (px)

Transcript of Systemwide Intersection Safety Prioritization Development and

  • SystemwideIntersectionSafetyPrioritizationDevelopmentandAssessment

    XiaoQin(Correspondingauthor)

    AssistantScientist,DepartmentofCivilandEnvironmentalEngineeringUniversityofWisconsinMadison

    B239EngineeringHall,1415EngineeringDriveMadison,WI53706

    Phone:(608)2623649Fax:(608)2625199Email:xqin@engr.wisc.edu

    LuisLaracuanteGraduateStudent,DepartmentofCivilandEnvironmentalEngineering

    UniversityofWisconsinMadison1261EngineeringHall,1415EngineeringDrive

    Madison,WI53706

    DavidA.NoyceAssistantScientist,DepartmentofCivilandEnvironmentalEngineering

    UniversityofWisconsinMadison1204EngineeringHall,1415EngineeringDrive

    Madison,WI53706Phone:(608)2651882Fax:(608)2625199Email:noyce@engr.wisc.edu

    MadhavV.Chitturi

    ResearchAssociate,DepartmentofCivilandEnvironmentalEngineeringUniversityofWisconsinMadison

    B243EngineeringHall,1415EngineeringDriveMadison,WI53706

    Phone:(608)2623649Fax:(608)2625199Email:mchitturi@wisc.edu

    TotalWords:5401forText+7Figures+1Table=7401

    SubmittedforPresentationatthe88thTRBAnnualMeetingandPublicationintheTransportationResearchRecord

    TRB 2009 Annual Meeting CD-ROM Original paper submittal - not revised by author.

    mailto:xqin@engr.wisc.edumailto:noyce@engr.wisc.edumailto:mchitturi@wisc.edu

  • ABSTRACT

    Thispaperpresentsthedevelopmentandassessmentofasystemwideintersectionsafetyprioritizationtool.Generationofasystemwideintersectioncrashlistandrankingofintersectionsusinganewintersectionsafetyrankingmethodologyarethetwomaincomponents.Anautomated,comprehensiveandsustainableprocessofgeneratingtheintersectioncrashlistwasdesignedanditaddressesissuessuchascrashlocationanddistinguishingbetweenintersectionandcorridorsafetyproblems.Thenewrankingmethodologyusesascorethatisacompositeofcrashfrequency,crashseverityandcrashtype.Sensitivityanalysesindicatethattherankingsgeneratedfrom1yeardataaresignificantlydifferentfromthe5yeardatawithPercentageDifferent(PD)intersectionsrangingfrom31to47andcorrelationbetweentherankingsvaryingfromlessthan0toamaximumof0.42.For3yeardatathePDvariedfrom14to21andcorrelationrangedfrom0.21to0.76.Therefore1yeardatashouldbeusedwithgreatcautionforprioritizingintersectionsafetyanditisstronglyrecommendedthataminimumof3yeardataisused.Itwasalsofoundthatweightsusedinthecompositerankingwerenotassignificantinalteringtheprioritizationofsafetyasthenumberofyearsofdataused.UsingMedianscorewasexploredasanalternativetousingtotalscorefortheintersectionprioritization.ItwasfoundthatalthoughthePDwaslow(around14)thecorrelationvariedsignificantlyfrom0.73to0.21.Medianrankingshouldbeusedwithcautionforgeneratinglargerlistsofintersectionsforsafetyprioritization.INTRODUCTION

    Intersectionsafetyisacrucialcomponentofhighwaysafety;intersectionfatalitiesconstituteapproximately22%oftotalhighwayfatalitiesandintersectioncrashescomposeofover45%ofallreportedcrasheseveryyear(1).Methodstoidentifyandrankintersectionswithsafetyconcerns,whichmaybeaddressedthroughengineeringimprovements,arecommonlyusedbytransportationagencies.Thesemethodologiesallowthedevelopmentofarankedlistofintersectionsbasedonreportedcrashdata.Eventhoughmanymethodssharethepurposeofidentifyingintersectionswiththemostsafetyconcerns,themethodscanhavesignificantlydifferentoutcomes.Themethodsavailabletoidentifyunsafeintersectionscouldbepresentedinfourcategories:Counts,Rates,Composite,andEmpiricalBayes(24).Thereareafewvariationsineachcategorysuchasfrequency,equivalentpropertydamageandvaluelossincounts.Veryoften,counts,rates,andcrashseveritymeasuresarecombinedtoformaseverityindexaspartofthecompositemethods.AstandaloneapproachtomeasuresafetyistheEmpiricalBayes(EB)method(57).Inthispaperthedevelopmentandassessmentofasystemwideintersectionsafetyprioritizationtoolispresented.FirstthedevelopmentoftheintersectionCrashListispresentedfollowedbythenewrankingmethodology.Alsopresentedaretheresultsofsensitivityanalysesperformedonthenewmethodology.Finallyconclusionsandrecommendationsarepresented.

    SYSTEMWIDEINTERSECTIONCRASHLIST

    Thefirststepintherankingofintersectionsforidentifyingsafetyimprovementlocationsistogenerateareliableintersectioncrashlist.Althoughthisappearstoberatherstraightforward,severalissuesneedtobeaddressed.Theseissuesareimproperrecordingofcrashlocations(inthecrashreports)atanintersectionperhapsduetoaliasorspellingerrors,anddistinguishingcorridorsafetyissuesfromintersectionsafetyissues.

    TRB 2009 Annual Meeting CD-ROM Original paper submittal - not revised by author.

  • Inthisresearchanautomated,comprehensiveandsustainableprocessofgeneratingtheintersectioncrashlistwasdesigned.Thelistgenerationshouldberapidandrequireminimummanualoperations,hencetheneedforautomation.Comprehensiveindicatesthatintersectionsafetyhasmultifacetcausesandshouldnotsolelybeevaluatedbycrashcounts.Sincethecrashdataissubjecttochangessuchasadding/removing/correctingcrashreportsorotherinformation,theprocessshouldbedesignedforrapidregeneration,maintenanceandupdates.

    CrashDataCollectionandCompilation

    CrashdataiscontinuallycollectedandarchivedbytheWisconsinDepartmentofTransportation(WisDOT).Crashlocationinformationisenteredmanuallybylawenforcementagenciesonthecrashreportformthenverifiedmanuallyandassignedareferencepoint(RP)andanoffsetbyWisDOTstaffifitisastatehighwayrelatedcrash.ThepostprocessensuresthatcrashesthatoccurredonthestatehighwaysystemcouldbespatiallylocatedusingtheWisDOTlinearreferencingsystem(LRS).Anycrasheslocatedmorethan0.02mileawayontheintersectingnonSTHroadwayareconsideredasanoffstateroutecrashandrarelygeocoded.

    InformationaboutIntersectionsontheStateTrunkNetwork(STN)iskeptinageospatialdatabasethatincludesallintersectinghighways.Atotalof22,863spatiallyuniqueintersectionsareidentifiedandupdatedregularlyintheAccessPointTable,includingdrivewaysandvariousaccesspointstotheSTNSystem.Classificationofacrashasanintersectioncrashismadebyalawenforcementofficialandenteredasintersectionrelated.

    MethodologyofGeocodingIntersectionCrashes

    Crashreferencepoints(RPs)areemployedtorelatecrashestotherelevantintersectionsspatially.TheprocessisperformedusingWisDOTLocationControlManger(LCM)builtupononthedepartmentlegendaryLRS.Theprocessbeginsbycollectingtheintersectionrelatedcrashdatafromthecrashdatabase.Eachintersectionrelatedcrashislinkedtothenearestintersectionbasedonthespatialdistance,resultinginanaggregationofallcrashestoaparticularintersection.AstatehighwayintersectionscrashcountmapisillustratedinFigure1withdatafrom2002through2005.Crashrecordsarefurthercategorizedbycrashseverity,crashtypeormannerofcollision.Crashseverityincludespropertydamageonly(PDO)crashes,andTypeC,B,AandKcrashes.Theselettersindicatetheseverityofthecrashinjuriesstartingfromminor(typeC)tofatalinjuries(typeK).Thepostprocesseddataincludesintersectionswithdatacategorizedby:totalnumberofcrashes,numberofcrashesbyseverity,andnumberofcrashesbymannerofcollisionorcrashtype.

    AlternativeGeocodingProcedureUsingGoogleMapsAPI

    Relatingcrashestothenearestintersectionmayresultingroupingcrashestoanintersectionwheretheydidnotoccur.Inthisstudy,theGoogleMapsAPIwasresearchedasanalternativecrashlocationtool(8).Inaddition,GooglecanbeusedtocreateaGoogleMaplinktoanintersectionsothatengineerscanopenthelinkandperformpreliminaryscansofintersectionconfigurationsusingGoogleimagesoraerialmaps.

    TRB 2009 Annual Meeting CD-ROM Original paper submittal - not revised by author.

  • FIGURE1Wisconsinstatewideintersectioncrashcountmap.

    TRB

    2009 Annual M

    eeting CD

    -RO

    MO

    riginal paper submittal - not revised by author.

  • ThelocationinformationsuppliedtoGoogleMapsforgeocodingwasthelistofcrasheswithlocationdata.Googlefoundmatchesforabout80percentofthecrashes.DespitethepowerfulcapabilityprovidedthroughtheGooglemapapplication,theresultsreturnedfromtheGoogleMapsAPIshouldbeanalyzedwithcaution.SpecificallythefollowingareissuesinusingGoogleMapsAPI:

    1. Googleappliesacomplicatedalgorithmtoprovideabestguessiftheinputlocationdoesnothavetheexactmatchinitsdatabase;Thusfurtherlocationqualityassurancemightberequiredtoensurethattheidentifiedlocationsarecorrect.

    2. Googlecannotrecognizesomehighwayfacilitiesincludingfrom,to,on,off,median,ramp,exit,bound,eb,wb,sb,nb,etc;therefore,thehighwaynamesinthecrashreportsneedtobecleaned.

    3. GooglehasadifferentwayofspellingUShighways,StatehighwaysandCountyhighways,soaconversionisneededbeforemappinginGoogle.

    4. Google'sdatabaseforgeocodingmaybedifferentfromtheoneforGoogleMapsprovidedbyNAVTAQTM.Thediscrepanciesintworesourcesmaycausesomeissues,suchasdifferentaddressesinGooglegeocodingandaGooglemaplink;

    QualityControlUsingGoogleMapsAPI

    GoogleMapsAPIhasalsobeenexperimentedasapossibletoolforQA/QCforthefollowing:

    1) ToStandardizeintersectionnamebyselectingtheintersectionaddressthatappearedthemostinpolicereports;and

    2) Todistinguishbetweenanintersectionsafetyissueandacorridorsafetyissue.

    Foreachintersection,allofthecrasheswereexaminedfortheintersectionname.Thestandardizednamewasdeterminedbasedonthenamethatappearedmostinthecrashesthatwereassignedtothatintersection.Forexample,inFigure2a,141crasheswereidentifiedaroundintersection44984duringthefiveyearperiodandabout98%werecalledWJohnsonSt&NPioneerRd,FondduLac,WIinpolicecrashreports.Sincethemajorityofcrashes(98%)werecodedtothesameintersection,thenamewasstandardizedasWJohnsonSt&NPioneerRd,FondduLac,WIwithahighlevelofconfidence.

    Itispossiblethatthecrashesareassignedtoanearbyintersectionratherthantotheactualintersectionbecausethecrasheswereaggregatedtoanintersectionbydistancefromtheintersection.Thissituationusuallyoccurredinurbanareaswhereintersectionswerecloselyspaced.Consequentlyacorridorsafetyissuemaybecategorizedasanintersectionsafetyissue.Forexample,intersection10201(showninFigure2b)wasgeocodedwith68crashes.About35%ofthemwererecordedinpolicereportsasSHowellAve&WLaytonAve,Milwaukee,WIwitht