Vegetationbasedclassificationtreesforrapidassessme

VegetationbasedclassificationtreesforrapidassessmentofisolatedwetlandconditionMatthewJ.Cohen*,CharlesR.Lane1,KellyC.Reiss,JamesA.Surdick,ElianaBardi,MarkT.BrownH.T.OdumCenterforWetlands,POBox116350,DepartmentofEnvironmentalEngineeringSciences,UniversityofFlorida,Gainesville,FL32611-6350,USAAccepted24January2005AbstractBiologicalmetricsareincreasinglyemployedbyregulatoryagenciestoassessecosystemconditionforthepurposesofconservation,restoration,mitigationandcreation.Whileindicesofbioticintegrity(IBI)havebeenshowntocorrespondstronglywithanthropogenicdisturbance,theyaretimeconsumingandexpensivetoimplementandrequireadvancedtaxonomicknowledge.Regulatoryimplementationfrequentlynecessitatessimplifiedindicatorsofcommunitycompositionthatdonotexcessivelysacrificeecologicalspecificity.Detailedobservationsofvegetativecommunitycompositionandstructureinisolateddepressionalherbaceous(n=75)andforested(n=118)wetlandswereundertakentoidentifyintegrativebioticmetricsofcommunitycondition.Habitat-specificwetlandconditionindices(WCI)composedoffivetosixrelativizedbioticmetrics(%non-nativeplantspecies,%tolerantand%sensitiveplantspecies,perennial-to-annual/biennialratio,meanfloristicquality,specieshydricstatus)werepreviouslydeveloped.Fromthesequantitativeindicesofcondition,classification(CT)andregressiontree(RT)modelsweredevelopedforcategorical/continuouspredictionbasedonpresence/absencedataforspeciesandecologicallyappropriategroupsofspecies.Optimalcategoricalresolutionwithrespecttomodelvalidationaccuracywasachievedforthreeimpactcategories(minimally,moderately,andseverelydisturbed).Overallcross-validationefficiencieswere77%and74%forherbaceousandforestedwetlands,respectively.Errorwasprimarilyobservedinthemoderatedegradationclassforbothsystemsandmisclassifiedsitestendedtofallnearcategoricalthresholdvalues.Highvalidationspecificitywasobservedforseverelydegradedsites(89%forbothsystems)andminimallyimpactedsites(90%forbothsystems).TheadvantagesofCTmodelsforregulatoryapplicationincludeinterpretivesimplicity,ecologicalspecificity,standardizedrepeatableassessmentbetweenassessors,andreducedbotanicalexpertiserequirements.#2005ElsevierLtd.Allrightsreserved.Keywords:Wetlandcondition;Depressionalwetlands;Rapidassessment;Biologicalmetrics;Classificationandregressiontrees;Florida;UniformmitigationassessmentmethodThisarticleisalsoavailableonlineat:(2005)189–206*Correspondingauthor.Tel.:[phone];fax:[phone].E-mailaddress:mjc@ufl.edu(M.J.Cohen).1Presentaddress:EPACincinnati,OH,USA.1470-160X/$–seefrontmatter#2005ElsevierLtd.Allrightsreserved.doi:10.1016/j.ecolind.2005.01.0021.IntroductionQuantitativeinferenceofecosystemhealthisanincreasinglycentralobjectiveofregulatoryagencieschargedwithmanagement,restoration,assessmentandmitigation.Themonitoringandassessmentofwetlandresourceshasbeenachievedthroughacombinationofquantitativemethodsandbestprofes-sionaljudgment(e.g.,Hydrogeomorphic(HGM)assessment–Brinson,1993;Nobleetal.,2002;wetlandevaluationtechnique(WET)–Adamusetal.,1991;WetlandRapidAssessmentProcedure(WRAP)–MillerandGunsalus,1997).Whilethesewhole-systemapproacheshavedominatedregulatoryassess-ment,developmentandimplementationofmethodsbasedmoreexplicitlyonsystembiota(Karr,1981,1993;Danielson,1998)havebeengainingfavorasaresultofincreasedecologicalunderstandingandorganismalspecificityandsensitivitytomodesandintensityofhumandisturbances.Thisisparticularlytruewheretemporallyintegratedresponsestocumu-lativeimpactsarerequired.Typically,developmentofindicesofbioticintegrity(IBI)beginsbyidentifyingsitesrepresentativeofreferenceconditionsagainstwhichothersitescanbecomparedusingcommunitycompositionmetricsbasedoninsects(e.g.,Florida’sstreamconditionindex(SCI)–Barbouretal.,1996;Reynoldsonetal.,1997),diatoms(PanandStevenson,1996),birds(O’Connelletal.,1998),fishesandamphibians(Simonetal.,2000),and/orplants(Mack,2001;DeKeyseretal.,2003).Whiletheseapproacheshavesignificantlyimprovedindicatorspecificityandsensitivity,theyhavelimitedregulatoryrapidassessmentapplicationprimarilybecausemanyrequireextensivesamplingeffortandbiologicalexpertiseforcommunitycharacterization.InFlorida,qualitativewhole-systemapproacheshavedominatedregulatoryassessmentduetotimeconstraintsforquantifyingwetlandconditionimposedonassessors.Effortstostandardizemethodsamongassessorsandcommunitytypeshaveresultedinthedevelopmentoftheuniformmitigationassessmentmethod(UMAM)rule(Chapter62-345,F.A.C.),whichbecameeffectiveinFebruary2004.WithintheUMAMframework,sitesofallhabitattypes(e.g.,forested,emergent,riverine,coastalmarshes,coralreefs)areevaluatedinthreecategoriesandscorednumericallyonascalefrom0to10(where10indicatesaminimallyimpairedsystem).Thefirstcategory,‘‘locationandlandscapesupport,’’examinestheecologicalcontextwithinwhichthesystemoperates.Thesecondcategoryexaminesthe‘‘waterenvironment,’’includingrapidinferenceofhydrologicalterationandwaterqualityimpairmentwhereasthethirdfocuseson‘‘communitystructure’’andmorespecifically‘‘vegetationandstructuralhabitat’’.Whilet