环境评价的模糊方法在大气质量评价上的运用.docx
《环境评价的模糊方法在大气质量评价上的运用.docx》由会员分享,可在线阅读,更多相关《环境评价的模糊方法在大气质量评价上的运用.docx(24页珍藏版)》请在冰豆网上搜索。
环境评价的模糊方法在大气质量评价上的运用
英文参考文献
Fuzzyapproachestoenvironmentaldecisions:
applicationtoairquality
BernardE.A.Fisher
abstract
Thispaperconsidersflexibleapproachestodecisionsdesignedtoimproveenvironmentalqualityhavingregardtouncertainty.Theperformanceofsimpleandcomplexmodels,forforecastingairqualityarereviewed,andbothtypesareshowntoinvolveconsiderableuncertaintyregardedastypicalofenvironmentalsystems.Thismeansthatdecisionsusuallydependoncombiningtwoormorequiteuncertainenvironmentalcriteria,anditisshownthatthiscanbeapproachedsystematicallyifafuzzylogicframeworkisadopted.Fuzzysetaggregationincludes,asspecialcases,otherdecision-makingframeworks,suchasmulti-criteriaanalysisandconventionalprobabilitybasedmethods.Examplesarepresentedofhowitcanbeappliedtosituationsinvolvingmodelsandusedtoincorporatebroaderfactorsinvolvingrisk,andsocio-economicconsiderations.
1.Theproblem
Ideallyenvironmentalmodelsshouldcontainthebestknownscience,betestedagainstmeasurements,andthenusedforpredictionanddecision-makingiftheyaresupportedbythenecessaryinputdataandperformwellagainstmeasurements.Thishasledinrecentyearstothedevelopmentofmorecomplex,fundamentalmodels,inwhicheverypartoftheenvironmentalsystemisdescribedinasmuchdetailaspossible,theso-called‘reductionist’approach.Ifthedevelopmentinmodellingproducedbetterpredictionsthiswouldbethewaytoproceed.Fromexamplesinthefieldofairpollution,itisarguedinthispaperthatlimitationsinprocessdescription,andthelackofdetaileddataonconcentrations,depositionoremissionsetc.,meanthatthisapproachhasnotproducedusefullybetterpredictions.Eventhemostadvancedmodelsarestillassociatedwithlargeuncertainties(Hunt,2000;FuntowiczandRavetz,2005;Saloranta,2001).
Thealternativeapproachtoforecastingistoconsiderthedecisionthatislikelytobringenvironmentalimprovementsandtoconsidermodelswhichallowdecisionstobemade,so-called‘fitforpurpose’models.Thesemodelsmayinvolvegreateruncertaintythanmorecomplexmodels,buttheyfacilitatemorereadilythetreatmentofuncertainty,throughsensitivityanalysis.Moreover,thedatarequirementstorunthemarelessstringent.Theymayalsoallowbroaderfactorstobeincludedintothedecision-makingprocess,incorporatingrisk,andoptimisingsocialandeconomicfactors,etc.
Anenvironmentaldecisionoughttodependonacriterionmeetinganumericalobjectivewithuncertaintytreatedexplicitly(RoyalCommissiononEnvironmentalPollution,1998).Onewayofincludingtheuncertaintyistoassumethatthecriterioninvolvesthemembershipofafuzzyset.Thisisanobviousapplicationoffuzzylogictoenvironmentaldecision-making.Insomecases,theenvironmentalcriterionisvagueorimprecise,suchaswhendealingwiththe‘qualityoflife’(MendesandMotizuki,2001),therankingofecosystemsintermsofenvironmentalconditionsandimpacts(Tranetal.,2002),orenvironmentalimpactassessment(EneaandSalemi,2001).
Afuzzysetisageneralisationofanormalsetforwhichthereisnotasharpboundarybetweenbelongingandnotbelongingtotheset.AfuzzysetisdefinedbyitsmembershipfunctionandtwoexamplesareshowninFig.1,inwhichtwohypotheticalenvironmentalqualitycriteria(relatingtohealthandnoise)alongacross-sectionthoughacityareshown,wherexisthedistancefromthecitycentre.Inbothcasesnearthecitycentre(atsmallvaluesofx),theenvironmentalqualityispoorer,andthereforethereisagreaterpossibilitythatenvironmentalqualityobjectivesareexceeded.Thisisshownbythehighermembershipfunctionsatshortdistancesfromthecitycentreforthefuzzysetsdescribingunsatisfactoryhealthandnoise.Iftheenvironmentalqualityweredescribedbyanormalset,themembershipfunctionwouldequal1outtothedistanceatwhichtheenvironmentalqualitywasjudgedsatisfactoryandthemembershipfunctionwouldbesetequalto0atgreaterdistances.
Qualitativefactors,involvingexpertjudgementofthepedigreeofamodel,shouldbepartofanassessmentofmodelperformance,goingbeyondtraditionalscientificapproaches,suchassensitivityanalysisandmodelvalidation(VanderSluijsetal.,2003).Widerjudgementsarerequiredwhenreviewingtheformulationofamodelordecidingwhetheritisappropriatetouseanestablishedmodelinanovelapplication.Becauseofitshistoryoftestingmodelperformance,andtheuseofobjectivesinairqualitymanagement,airqualityprovidesagoodtestcaseoffuzzydecision-making.However,fuzzydecision-makingiscloselyrelatedtootherenvironmentaldecision-makingframeworks,suchasmulti-criteriadecisionanalysisforflooddefence(Tung,2002),ortheweightedutilityapproachforabruptclimatechange(Perrings,2003).Itisnotproposedthatfuzzydecision-makingissuperiortotheseothermethods,butitdoesprovideastructuredframeworkwithinwhichothermethodsmaybeincorporated.
Inthispaper,weconsiderhowconventionalapproachestoairqualitymodelscanbeextendedwithinafuzzylogicframeworktoincludewiderconsiderations,notjustthoseconcernedwiththeaccuracyofthemodel.Thesewider,generallypolicyrelated,considerationsarerelevanttohowthemodellingresultswillbeused,andhenceshouldalsobeanintegralpartofthedecision.
2.Performanceofairqualitymodels:
simpleandcomplex
2.1.Short-rangemodels
Anumberofassessmentsofdispersionmodels’performancetopredictconcentrationsovershortdistancesoutto30kmhavebeenmade.SomeofthesehavebeensummarisedbyIreland(2003),andforurbanareasbyFisher(2005a).Theyshowthatvarioussummarystatisticscanbeusedtojudgemodelperformance.Modelperformancealsovariesaccordingtothewayresultsareinterpreted.Forexample,forpointsourcesmuchoftheerrorcanarisefromerrorsinthepredictionofthedirectionoftheplumecentreline.Theerrorismuchsmallerifpredictedandmeasuredmaximumgroundlevelconcentrationsarecomparedwithoutregardtothelocationofthemaxima.Generally,studiessuggestthatmostpredictionsarelikelytobewithinafactoroftwoofthepredictions,asimple,usefulmeasureofperformance,butperformanceisnotconsistentlybetterthanafactoroftwo.Theuncertaintyinmodelpredictionscouldbebasedonotherestimators,suchastheresidualsumofsquares.Thefactoroftwoiseasytoexplaintodecisionmakersandwouldbeinagreementwiththeuncertaintyassociatedwithsettingtheboundaryofairqualitymanagementareas,recommendedininformalguidanceproducedbyNationalSocietyforCleanAir(2003).
Theuncertaintyassociatedwitherrorsinmodelformulation,orwhetherthemodelisappropriatetoanapplicationaremoredifficulttodealwith.Inpart,thismaybeaddressedbyusingmorethanoneairqualitydispersionmodelinpredictions(Fisher,2003;Fisheretal.,2002).Theoveralluncertaintycanonlybeaddressedbyrecoursetocomparisonswithmeasurementsofsuitablequalityandquantity.Thepedigreeofthemodelcanhelp(VanderSluijsetal.,2003)andwouldbringinconsiderationsastowhetherthemodelhadbeenusedforsimilarapplicationsandiftheformulationofthemodelisreadilyavailableforreviewbypotentialusers.Whenthestructureoftheenvironmentalmodelisverycomplex,whichisoftenthecaseinregionalairqualitymodels,sensitivityanalysistochoosethebestvaluesforparameters,isnotusuallyundertaken,becauseeithertheruntimesaretoolong,orthedatasetsofobservationsaretoolimitedtoevaluateeachcomponentindetail.
2.2.Regionalmodels
Thepredictionofconcentrationsanddepositionoverlongandshortdistancesusingairqualitymodelsdependsinpartontheuncertaintyintheinputvaluesfortheparametersinthemodel.Thiscanbeassessedinarelativelystraightforwardway,ifpossibleerrorsinmodelformulationareignored.Tworecentstudieshavedeterminedtheuncertaintyinmodelpredictionsforaciddeposition(Abbottetal.,2003)anddispersioncalculations(Halletal.,2000a,b)basedonsensitivitystudies,Monte-Carlosimulationandscenarioanalysis.Thedegreeofuncertaintyisagainbroadlywithinafactoroftwoi.e.changinginputvalueswithintheirboundsofuncertaintygenerallyleadstopredictionswhichliewithinarangeoftwoofacentralvalue,butpredictionsbetterthanafactoroftwocannotbeobtained.Abbottetal.(2003)havealsoshownthatthecommonlyappliedmethodsofderivingcriticalloadshaveuncertaintiesofcomparablemagnitude.
Anexampleofacomplexintegratedatmosphericmodeltocalculatelong-termannualaverageatmosphericconcentrationsanddepositionsoveraregionsuchasEurope,istheuseofModels-3.Recentresearch(Cocksetal.,2004)hasdemonstratedthatmeso-scalepollutiontransportmodels,suchasModels-3,canberuntoproducehourlytime-seriesofpollutionconcentrationanddepositionfieldsoverlongperiods,suchasayear,providingdirectcomparisonswithpercentileexceedencesofshort-termairqualityobjectives.Long-termaveragesderivedfromtime-seriesalsoprovideabenchmarkforcomparisonwitholderestablishedstatisticalaciddepositionmodels,whichmayonlybeusedtocalculatelong-termaverages.IssuesregardingthepracticaluseofModels-3forregulatoryassessmentpurposesremain,butthedemonstrationsupportstheviabilityoftheconceptofamodularassessmentframeworkincorporatingcomplexinteractions.Asanintegrated,‘reductionist’model,theresultsprovideextrainformationonquantities,suchassecondaryparticulatematter.
T