Amathematicalmodelisdefinedbyaseriesofequations,inputvariablesandparametersaimedatcharacterizingsomeprocessunderinvestigation.Someexamplesmightbeaclimatemodel,aneconomicmodel,orafiniteelementmodelinengineering.Increasingly,suchmodelsarehighlycomplex,andasaresulttheirinput/outputrelationshipsmaybepoorlyunderstood.Insuchcases,themodelcanbeviewedasablackbox,i.e.theoutputisanopaquefunctionofitsinputs.

　　Quiteoften,someorallofthemodelinputsaresubjecttosourcesofuncertainty,includingerrorsofmeasurement,absenceofinformationandpoororpartialunderstandingofthedrivingforcesandmechanisms.Thisuncertaintyimposesalimitonourconfidenceintheresponseoroutputofthemodel.Further,modelsmayhavetocopewiththenaturalintrinsicvariabilityofthesystem(aleatory),suchastheoccurrenceofstochasticevents[3]

　　Goodmodelingpracticerequiresthatthemodelerprovidesanevaluationoftheconfidenceinthemodel.Thisrequires,first,aquantificationoftheuncertaintyinanymodelresults(uncertaintyanalysis);andsecond,anevaluationofhowmucheachinputiscontributingtotheoutputuncertainty.Sensitivityanalysisaddressesthesecondoftheseissues(althoughuncertaintyanalysisisusuallyanecessaryprecursor),performingtheroleoforderingbyimportancethestrengthandrelevanceoftheinputsindeterminingthevariationintheoutput.[1]

　　Inmodelsinvolvingmanyinputvariables,sensitivityanalysisisanessentialingredientofmodelbuildingandqualityassurance.Nationalandinternationalagenciesinvolvedinimpactassessmentstudieshaveincludedsectionsdevotedtosensitivityanalysisintheirguidelines.ExamplesaretheEuropeanCommission(seee.g.theguidelinesforimpactassessment),theWhiteHouseOfficeofManagementandBudget,theIntergovernmentalPanelonClimateChangeandUSEnvironmentalProtectionAgency'smodellingguidelines.

　　维基给的