上半年工作總結(jié)和下半年工作計劃課件

Section12

AirQualityForecastingTools1工作報告BackgroundForecastingtoolsprovideinformationtohelpguidetheforecastingprocess.Forecastersuseavarietyofdataproducts,information,tools,andexperiencetopredictairquality.Forecastingtoolsarebuiltuponanunderstandingoftheprocessesthatcontrolairquality.Forecastingtools:SubjectiveObjectiveMoreforecastingtools=betterresults.2工作報告BackgroundPersistenceClimatologyCriteriaStatisticalClassificationand

RegressionTree(CART)RegressionNeuralnetworksNumericalmodelingPhenomenologicaland

experiencePredictorvariablesFewerresources,loweraccuracy

Moreresources,potentialforhigheraccuracy3工作報告SelectingPredictorVariables(2of3)Selectobservedandforecastedvariables.Predictorvariablescanconsistofobservedvariables(e.g.,yesterday’sozoneorPM2.5concentration)andforecastedvariables(e.g.,

tomorrow’smaximumtemperature).Makesurethatpredictorvariablesareeasilyobtainablefromreliablesource(s)andcanbeforecast.Consideruncertaintyinmeasurements,particularlymeasurementsofPM.5工作報告SelectingPredictorVariables(3of3)Beginwithasmanyas50to100predictorvariables.Usestatisticalanalysistechniquestoidentifythemostimportantvariables.Clusteranalysisisusedtopartitiondataintosimilaranddissimilarsubsets.Unique(i.e.,dissimilar)variablesshouldbeusedtoavoidredundancy.Correlationanalysisisusedtoevaluatetherelationshipbetweenthepredictand(i.e.,

pollutantlevels)andvariouspredictorvariables.Step-wiseregressionisanautomaticprocedurethatallowsthestatisticalsoftware(SAS,

Statgraphics,Systat,etc.)toselectthemostimportantvariablesandgeneratethebestregressionequation.Humanselection

isanothermeansofselectingthemostimportantpredictorvariables.6工作報告CommonOzonePredictorVariablesVariableUsefulnessConditionforHighOzoneMaximumtemperatureHighlycorrelatedwithozoneandozoneformationHighMorningwindspeedAssociatedwithdispersionanddilutionofozoneprecursorpollutantsLowAfternoonwindspeedAssociatedwithtransportofozone-CloudcoverControlssolarradiation,whichinfluencesphotochemistryFewRelativehumiditySurrogateforcloudcoverLow500-mbheightIndicatorofthesynoptic-scaleweatherpatternHigh850-mbtemperatureSurrogateforverticalmixingHighPressuregradientsCauseswinds/ventilationLowLengthofdayAmountofsolarradiationLongerDayofweekEmissionsdifferences-MorningNOxconcentrationOzoneprecursorlevelsHighPreviousday’speakozoneconcentrationPersistence,carry-overHighAloftwindspeedanddirectionTransportfromupwindregion-7工作報告AssemblingadatasetDeterminewhatdatatouseWhatdatatypesareneededandavailableWhatsitesarerepresentativeWhatairqualitymonitoringnetwork(s)touse(forexample,continuousversuspassiveorfilter)Whattypeofmeteorologicaldataareavailable(surface,upper-air,satellite,etc.)Howmuchdataisavailable(years)9工作報告AssemblingadatasetAcquirehistoricaldataincludingHourlypollutantdataDailymaximumpollutantmetrics,suchasPeak1-hrozonePeak8-hraverageozone24-hraveragePM2.5orPM10HourlymeteorologicaldataRadiosondedataModeldataMeteorologicaloutputsMM5/TAPMOtherSurfaceandupper-airweatherchartsHYSPLITtrajectories10工作報告AssemblingadatasetQualitycontroldataCheckforoutliersLookattheminimumandmaximumvaluesforeachfield;aretheyreasonable?Checkrateofchangebetweenrecordsateachextreme.TimestampsHasalldatabeenproperlymatchedbytime?TimeseriesplotscanhelpidentifyproblemsshiftingfromUTCtoLST.MissingdataIsthesameidentifierusedforeachfield?I.e.,–999.UnitsAreunitsconsistentamongdifferentdatasets?I.e.,m/sorknotsforwindspeeds.ValidationcodesArevalidationcodesconsistentamongdifferentdatasets?Dothevalidationcodesmatchthedatavalues?I.e.,aredatavalues

of–999flaggedasmissing?11工作報告ForecastingToolsandMethods

(2of2)ForeachtoolWhatisit?Howdoesitwork?ExampleHowtodevelopit?StrengthsLimitationsOzone=WS*10.2+…13工作報告Persistence(1of2)Persistencemeanstocontinuesteadilyinsomestate.Tomorrow’spollutantconcentrationwillbethesameasToday’s.Bestusedasastartingpointandtohelpguideotherforecastingmethods.Itshouldnotbeusedastheonlyforecastingmethod.Modifyingapersistenceforecastwithforecastingexperiencecanhelpimproveforecastaccuracy.MondayTuesdayWednesdayUnhealthyUnhealthyUnhealthyPersistenceforecast14工作報告Persistence(2of2)Sevenhighozonedays(red)Fiveofthesedaysoccurredafterahighday(*)Probabilityofhighozoneoccurringonthedayafterahighozonedayis5outof7daysProbabilityofalowozonedayoccurringafteralowozonedayare20outof22daysPersistencemethodwouldbeaccurate25outof29days,or86%ofthetime

DayOzone(ppb)DayOzone(ppb)18016120*25017110*350188047019805802070610021607110*2250890*23509802470108025801180268012702770138028801490296015110*3070Peak8-hrozoneconcentrationsforasamplecity15工作報告Persistence–LimitationsPersistenceforecastingcannotPredictthestartandendofapollutionepisodeWorkwellunderchangingweatherconditionswhenaccurateairqualitypredictionscanbemostcritical17工作報告ClimatologyClimatologyisthestudyofaverageandextremeweatherorairqualityconditionsatagivenlocation.Climatologycanhelpforecastersboundandguidetheirairqualitypredictions.18工作報告Climatology–ExampleAveragenumberofdayspermonthwithozoneineachAQIcategoryforSacramento,California19工作報告DevelopingClimatologyConsideremissionschangesForexample,fuelreformulationItmaybeusefultodividetheclimatetablesorchartsinto“before”and“after”periodsformajoremissionschanges.McCarthyetal.,200521工作報告ClimatologyHighpollutiondaysoccurmostoftenonWednesdayHighpollutiondaysoccurleastoftenatthebeginningoftheweekDayofweekdistributionofhighPM2.5days22工作報告ClimatologyDistributionofhighozonedaysbyupper-airweatherpattern23工作報告Climatology–LimitationsClimatologyIsnotastand-aloneforecastingmethodbutatooltocomplementotherforecastmethodsDoesnotaccountforabruptchangesinemissionspatternssuchasthoseassociatedwiththeuseofreformulatedfuel,alargechangeinpopulation,forestfires,etc.Requiresenoughdata(years)toestablishrealistictrends25工作報告CriteriaUsesthresholdvalues(criteria)ofmeteorologicalorairqualityvariablestoforecastpollutantconcentrationsForexample,iftemperature>27°Cand

wind<2m/sthenozonewillbeintheUnhealthyAQIcategorySometimescalled“rulesofthumb”CommonlyusedinmanyforecastingprogramsasaprimaryforecastingmethodorcombinedwithothermethodsBestsuitedtohelpforecasthighpollutionorlowpollutionevents,orpollutioninaparticularairqualityindexcategoryrangeratherthananexactconcentration26工作報告DevelopingCriteria(2of2)

Determinethethresholdvalueforeachparameterthatdistinguisheshighandlowpollutantconcentrations.Forexample,createscatterplotsofpollutantvs.weatherparameters.Useanindependentdataset(i.e.,adatasetnotusedfordevelopment)toevaluatetheselectedcriteria.29工作報告Criteria–StrengthsEasytooperateandmodifyAnobjectivemethodthatalleviatespotentialhumanbiasesComplementsotherforecastingmethods30工作報告Criteria–LimitationsSelectionofthevariablesandtheirassociatedthresholdsissubjective.Itisnotwellsuitedforpredictingexactpollutantconcentrations.31工作報告Classificationand

RegressionTree(CART)CARTisastatisticalproceduredesignedtoclassifydataintodissimilargroups.Similartocriteriamethod;however,itisobjectivelydeveloped.CARTenablesaforecastertodevelopadecisiontreetopredictpollutantconcentrationsbasedonpredictorvariables(usuallyweather)thatarewellcorrelatedwithpollutantconcentrations.AnexampleCarttreeformaximumozonepredictionforthegreaterAthensarea.32工作報告Classificationand

RegressionTree(CART)Ozone(Low–High)

ModeratetoHighModeratetoLowTemplowTemphighWS-strongModerateModerateHighLowWS-calmWS-calmWS-light33工作報告CART–HowItWorks

(1of2)ThestatisticalsoftwaredeterminesthepredictorvariablesandthethresholdcutoffvaluesbyReadingalargedatasetwithmanypossiblepredictorvariablesIdentifyingthevariableswiththehighestcorrelationwiththepollutantContinuingtheprocessofsplittingthedatasetandgrowingthetreeuntilthedataineachgrouparesufficientlyuniform34工作報告CART–HowItWorks(2of2)ToforecastpollutantconcentrationsusingCARTsStepthroughthetreestartingatthefirstsplitanddeterminewhichofthetwogroupsthedatapointbelongsin,basedonthecut-pointforthatvariable;Continuethroughthetreeinthismanneruntilanendnodeisreached.Themeanconcentrationshownintheendnodeistheforecastedconcentration.Note,slightdifferencesinthevaluesofpredictedvariablescanproducesignificantchangesinpredictedpollutantlevelswhenthevalueisnearthethreshold.35工作報告

Variables:T850-12Z850MBtempDELTAP-thepressuredifferencebetweenthebaseand

topoftheinversionMI0-Synopticweatherpotential(scalefrom1-lowto5-high).FAVGTMP-24-houraveragetemperatureatLaPazFAVGRH-24-houraveragerelativehumidityatLaPaz.CARTclassificationPM10inSantiago,Chile

NodexVariableandcriteriaSTD=StandarddeviationAvg=AveragePM10(ug/m3)N=numberofcasesin nodeCassmassi,1999Istheforecastedtemperatureat850mb10.5°C?YesNoCART–Example36工作報告DevelopingCARTDeterminetheimportantprocessesthatinfluencepollution.Selectvariablesthatproperlyrepresenttheimportantprocesses.Createamulti-yeardatasetoftheselectedvariables.ChooserecentyearsthatarerepresentativeofthecurrentemissionprofileReserveasubsetofthedataforindependentevaluation,butensureitrepresentsallconditionsBesurevariablesareforecastedUsestatisticalsoftwaretocreateadecisiontree.Evaluatethedecisiontreeusingtheindependentdataset.37工作報告CART–StrengthsRequireslittleexpertisetooperateonadailybasis;runsquickly.Complementsothersubjectiveforecastingmethods.Allowsdifferentiationbetweendayswithsimilarpollutantconcentrationsifthepollutantconcentrationsarearesultofdifferentprocesses.SincePMcanformthroughmultiplepathways,thisadvantageofCARTcanbeparticularlyimportanttoPMforecasting.38工作報告CART–LimitationsRequiresamodestamountofexpertiseandefforttodevelop.Slightchangesinpredictedvariablesmayproducelargechangesinthepredictedconcentrations.CARTmaynotpredictpollutantconcentrationsduringperiodsofunusualemissionspatternsduetoholidaysorotherevents.CARTcriteriaandstatisticalapproachesmayrequireperiodicupdatesasemissionsourcesandlandusechanges.39工作報告RegressionEquations–

HowTheyWork(1of5)RegressionequationsaredevelopedtodescribetherelationshipbetweenpollutantconcentrationandotherpredictorvariablesForlinearregression,thecommonformis

y=mx+bAtright,maximumtemperature(Tmax)isagoodpredictorforpeakozone[O3]=1.92*Tmax–86.8r=0.77r2=0.5940工作報告Morepredictorscanbeadded(“stepwiseregression”)sothattheequationlookslikethis: y=m1x1+m2x2+m3x3+……mnxn+bEachpredictor(xn)hasitsown“weight”(mn)andthecombinationmayleadtobetterforecastaccuracy.Themixofpredictorsvariesfromplacetoplace.RegressionEquations–

HowTheyWork(2of5)41工作報告OzoneRegressionEquationforColumbus,Ohio8hrO3=exp(2.421+0.024*Tmax+0.003*Trange-0.006*WS1to6+0.007*00ZV925-0.004*RHSfc00-0.002*00ZWS500)VariableDescriptionTmaxMaximumtemperatureinoFTrangeDailytemperaturerangeWS1to6Averagewindspeedfrom1p.m.to

6p.m.inknots00ZV925Vcomponentofthe925-mbwindat00ZRHSfc00Relativehumidityatthesurfaceat00Z00ZWS500Windspeedat500mbat00ZRegressionEquations–

HowTheyWork(3of5)42工作報告Thevariouspredictorsarenotequallyweighted,somearemoreimportantthanothers.Itisessentialtoidentifythestrongestpredictorsandworkhardestongettingthosepredictionsright.Tmaxvs.O3PreviousdayO3vs.O3WindSpeedvs.O3RegressionEquations–

HowTheyWork(4of5)43工作報告Inanexamplecase,mostofthevarianceinO3isexplainedbyTmax(60%),withtheadditionalpredictorsadding~15%.Overall,75%ofthevarianceinobservedO3

isexplainedbytheforecastmodel.Ourjobasforecastersistofillintheadditional25%usingothertools.AccumulatedexplainedvarianceRegressionEquations–

HowTheyWork(5of5)44工作報告DevelopingRegression(1of2)Determinetheimportantprocessesthatinfluencepollutant

concentrations.Selectvariablesthatrepresenttheimportantprocessesthatinfluencepollutantconcentrations.Createamulti-yeardatasetoftheselectedvariables.Chooserecentyearsthatarerepresentativeofthecurrentemissionprofile.Reserveasubsetofthedataforindependentevaluation,butensureitrepresentsallconditions.Besurevariablesareforecasted.Usestatisticalsoftwaretocalculatethecoefficientsandaconstantfortheregressionequation.Performanindependentevaluationoftheregressionmodel.45工作報告DevelopingRegression(2of2)Usingthenaturallogofpollutantconcentrationsasthepredictandmayimproveperformance.Donotto“overfit”themodelbyusingtoomanypredictionvariables.An“over-fit”modelwilldecreasetheforecastaccuracy.Areasonablenumberofvariablestouseis5to10.Uniquevariablesshouldbeusedtoavoidredundancyandco?linearity.Stratifyingthedatasetmayimproveregressionperformance.SeasonsWeekendvs.weekday46工作報告Regression–StrengthsItiswelldocumentedandwidelyusedinavarietyofdisciplines.Softwareiswidelyavailable.Itisanobjectiveforecastingmethodthatreducespotentialbiasesarisingfromhumansubjectivity.Itcanproperlyweightrelationshipsthataredifficulttosubjectivelyquantify.Itcanbeusedincombinationwithotherforecastingmethods,oritcanbeusedastheprimarymethod.47工作報告Regression–LimitationsRegressionequationsrequireamodestamountofexpertiseandefforttodevelop.Regressionequationstendtopredictthemeanbetterthanthetails(i.e.,thehighestpollutantconcentrations)ofthedistribution.Theywilllikelyunderpredictthehighconcentrationsandoverpredictthelowconcentrations.Regressioncriteriaandstatisticalapproachesmayrequireperiodicupdatesasemissionsourcesandlandusechanges.Regressionequationsrequire3-5yearsofmeasurementdataintheregionofapplication,includingmanyinstancesofairpollutionevents,todevelop.48工作報告NeuralNetworksArtificialneuralnetworksarecomputeralgorithmsdesignedtosimulatethehumanbrainintermsofpatternrecognition.Artificialneuralnetworkscanbe“trained”toidentifypatternsincomplicatednon-lineardata.Becausepollutantformationprocessesarecomplex,neuralnetworksarewellsuitedforforecasting.However,neuralnetworksrequireabout50%moreefforttodevelopthanregressionequationsandprovideonlyamodestimprovementinforecastaccuracy(Comrie,1997).49工作報告NeuralNetworks–HowItWorksNeuralnetworksuseweightsandfunctionstoconvertinputvariablesintoaprediction.Aforecastersuppliestheneuralnetworkwithmeteorologicalandairqualitydata.Thesoftwarethenweightseachdatumandsumsthesevalueswithotherweighteddatumateachhiddennode.Thesoftwarethenmodifiesthenodedatabyanon-linearequation(transferfunction).Themodifieddataareweightedandsummedastheypasstotheoutputnode.Attheoutputnode,thesoftwaremodifiesthesummeddatausinganothertransferfunctionandthenoutputsaprediction.Comrie,199750工作報告DevelopingNeuralNetworksDeterminetheimportantprocessesthatinfluencepollutant

concentrations.Selectvariablesthatrepresenttheimportantprocesses.Createamulti-yeardatasetoftheselectedvariables.Chooserecentyearsthatarerepresentativeofthecurrentemissionprofile.Reserveasubsetofthedataforindependentevaluation,butensureitrepresentsallconditions.Besurevariablesareforecasted.Trainthedatausingneuralnetworksoftware.SeeGardnerandDorling(1998)fordetails.Testthetrainednetworkonatestdatasettoevaluatetheperformance.Iftheresultsaresatisfactory,thenetworkisreadytouseforforecasting.51工作報告NeuralNetworks–StrengthsCanweightrelationshipsthataredifficulttosubjectivelyquantifyAllowsfornon-linearrelationshipsbetweenvariablesPredictsextremevaluesmoreeffectivelythanregressionequations,providedthatthenetworkdevelopmentalsetcontainssuchoutliersOncedeveloped,aforecasterdoesnotneedspecificexpertisetooperateitCanbeusedincombinationwithotherforecastingmethods,oritcanbeusedastheprimaryforecastingmethod52工作報告NeuralNetworks–LimitationsComplexandnotcommonlyunderstood;thus,themethodcanbeinappropriatelyappliedanddifficulttodevelopDonotextrapolatedatawell;thus,extremepollutantconcentrationsnotincludedinthedevelopmentaldatasetwillnotbetakenintoconsiderationintheformulationoftheneuralnetworkpredictionRequire3-5yearsofmeasurementdataintheregionofapplication,includingmanyinstancesofairpollutionevents,todevelop.53工作報告NumericalModelingMathematicallyrepresentstheimportantprocessesthataffectpollutionRequiresasystemofmodelstosimulatetheemission,transport,diffusion,transformation,andremovalofairpollutionMeteorologicalforecastmodelsEmissionsmodelsAirqualitymodels54工作報告NumericalModeling–HowItWorks55工作報告ProcessesTreatedinGridModelsEmissionsSurfaceemittedsources(on-roadandnon-roadmobile,area,

low-levelpoint,biogenic,fires)Pointsources(electricalgeneration,industrial,other,fires)Advection(Transport)Dispersion(Diffusion)ChemicalTransformationVOCandNOxchemistry,radicalcycleForPMaerosolthermodynamicsandaqueous-phasechemistryDepositionDrydeposition(gasandparticles)Wetdeposition(rainoutandwashout,gasandparticles)BoundaryconditionsHorizontalboundaryconditionsTopboundaryconditions56工作報告PhotochemicalGridModelConcept57工作報告EulerianGridCellProcesses58工作報告CouplingBetweenGridCells59工作報告NumericalModeling–Example60工作報告DevelopingaNumericalModelDesignandplanthesystemIdentifyandallocatetheresourcesAcquirerequiredgeophysicaldataImplementthedataacquisitionandprocessingtools,componentmodels(emissions,meteorological,andairquality),andanalysisprograms.DeveloptheemissioninventoryTesttheoperationofalldataacquisitionprograms,preprocessorprograms,componentmodels,andanalysisprogramsasasystemIntegratedataacquisitionandprocessingtools,componentmodels,andanalysisprogramsintoanoperationalsystemTest,evaluate,andimprovetheintegratedsystem61工作報告DevelopingaNumericalModel(1of7)DesignandplanthesystemDecideonwhichpollutantstoforecast.Definemodelingdomainsconsideringgeographyandemissionssources.Selectcomponentmodelsconsideringforecastpollutants,domains,componentmodelcompatibility,availabilityofinterfaceprograms,andavailableresources.Determinehardwareandsoftwarerequirements.Identifysourcesofmeteorological,emissions,andairqualitydata.Prepareadetailedplanforacquiringandintegratingdataacquisition,modeling,andanalysissoftware.Planforcontinuousreal-timeevaluationofthemodelingsystem.

62工作報告IdentifyandallocatetheresourcesStaffforsystemimplementationandoperationsComputingandstorageconsistentwiththeselectionofdomainsandmodelsCommunicationsfordatatransferintoandoutofthemodelingsystemAcquirerequiredgeophysicaldataTopographicaldataLandusedataDevelopingaNumericalModel

(2of7)63工作報告Implementthedataacquisitionandprocessingtools,componentmodels(emissions,meteorological,andairquality),andanalysisprograms.Implementeachprogramindividually.Usestandardtestcasestoverifycorrectimplementation.DevelopingaNumericalModel

(3of7)64工作報告DeveloptheemissioninventoryAcquireneededemissioninventoryrelateddata.Reviewtheemissionsdataforaccuracy.Besurethattheemissioninventoryincludesthemostrecentemissionsdataavailable.Updatethebaseemissioninventoryannually.DevelopingaNumericalModel

(4of7)65工作報告TestandEvaluateTesttheoperationofalldataacquisitionprograms,preprocessorprograms,componentmodels,andanalysisprogramsasasystem.Reviewtheprognosticmeteorologicalforecastdataforaccuracyoverseveralweeksundervariousweatherpatterns.Runthecombinedmeteorological/emissions/airqualitymodelingsysteminaprognosticmodeusingavarietyofmeteorologicalandairqualityconditions.Evaluatetheperformanceofthemodelingsystembycomparingitwithobservations.Refinethemodelapplicationprocedures(i.e.,themethodsofselectingboundaryconditionsorinitialconcentrationfields,thenumberofspin-updays,thegridboundaries,etc.)toimproveperformance.DevelopingaNumericalModel

(5of7)66工作報告Integratedataacquisitionandprocessingtools,componentmodels,andanalysisprogramsintoanoperationalsystemImplementautomatedprocessesfordataacquisition,thedailydataexchangefromtheprognosticmeteorologicalmodelandtheemissionsmodeltothe3-Dairqualitymodelandanalysisprograms,andforecastproductproduction.Implementautomatedprocessesbyusingscriptingandschedulingtools.Verifythattheforecastproductsreflecttheactualmodelpredictions.DevelopingaNumericalModel

(6of7)67工作報告Test,evaluate,andimprovetheintegratedsystemRunthemodelinreal-timetestmodeforanextendedperiod.Compareoutputtoobserveddataandnotewhentherearemodelfailures.Afterobtainingsatisfactoryresultsonaconsistentbasis,usethemodelingsystemtoforecastpollutantconcentrations.Documentthemodelingsystem.Continuouslyevaluatethesystem’sperformancebycomparingobservationsandpredictions.Implementimprovementsasneededbasedonperformanceevaluationsandnewinformation.DevelopingaNumericalModel

(7of7)68工作報告NumericalModeling–StrengthsTheyarephenomenologicalbased,simulatingthephysicalandchemicalprocessesthatresultintheformationanddestructionofairpollutants.Theycanforecastforalargegeographicarea.Theycanpredictairpollutioninareaswheretherearenoairqualitymeasurements.Themodelforecastscanbepresentedasmapsofairqualitytoshowhowpredictedairqualityvariesoveraregionhourbyhour.Themodelscanbeusedtofurtherunderstandtheprocessesthatcontrolairpollutioninaspecificarea.Forexample,theycanbeusedtoassesstheimportanceoflocalemissionssourcesorlong-rangetransport.69工作報告NumericalModeling–LimitationsInaccuraciesintheprognosticmodelforecastsofwindspeeds,winddirections,extentofverticalmixing,andsolarinsulationmaylimit3-Dairqualitymodelperformance.Emissioninventoriesusedincurrentmodelsareoftenoutofdateandbasedonuncertainemissionfactorsandactivitylevels.Site-by-siteozoneconcentrationspredictedby3-Dairqualityforecastmodelsmaynotbeaccurateduetosmall-scaleweatherandemissionfeaturesthatarenotcapturedinthemodel.Substantialstaffandcomputerresourcesareneededtoestablishascientificallysoundandautomatedairqualityforecastsystembasedona3-Dairqualitymodel.70工作報告AustralianAirQualityForecastingSystemPeterManinsCSIROMarineandAtmosphericResearchAustraliaWMOGURMESAGmemberDemonstrationProject71工作報告Phenomenological–HowItWorksReliesonforecasterexperienceandcapabilitiesForecasterneedsgoodunderstandingoftheprocessesthatinfluencepollutionsuchasthesynoptic,regional,andlocalmeteorologicalconditions,plusairqualitycharacteristicsintheforecastarea.Forecastersynthesizestheinformationbyanalyzin