SPATIALCHARACTERIZATIONOFAHYDROGEOCHEMICALLY HETEROGENEOUSAQUIFERUSINGATHREE-DIMENSIONAL DISTRIBUTEDPARAMETEREXTENDEDKALMANFILTER By YANZHANG ADISSERTATIONPRESENTEDTOTHEGRADUATESCHOOL OFTHEUNIVERSITYOFFLORIDAINPARTIALFULFILLMENT OFTHEREQUIREMENTSFORTHEDEGREEOF DOCTOROFPHILOSOPHY UNIVERSITYOFFLORIDA 1997 ACKNOWLEDGEMENTS Thecompletionofthisworkwouldnothavebeenpossibleifitwerenotfor the help and support ofmany individuals; their assistance is therefore cheerfully acknowledged. First,IwouldliketothankDr. P.S.C.Raoforhisinvaluablehelpandguidance throughoutthecourseofthisresearch. Dr. Raohasprovidedmewithopportunities toparticipateininterestingscientificandengineeringprojects. Iappreciatethemany intellectualandchallengingdiscussionswehadovertheyears. SpecialthanksalsogotoDr. KirkHatfield,Dr. MichaelD.Annable,andDr. KennethL.Campbellforservingasmycommitteemembers,andfortheirinsightful commentsprovidedduringmanyresearchmeetings. IwouldalsoliketothankDr. AndyI.JamesandDr. AshieAkpojifortheir valuableinputtothisresearch. Inaddition,mygratitudegoestomyfellowgraduate studentsintheHydrologicalSciencesAcademicClusterfortheirhelpandfriendship; inrandomorder,GeorgeG.Demmy,LiyongLi,XavierFoussereau,RandyK.Sillan, JamesW.Jawitz,DongpingDai,andHeonkiKim. Iwouldliketothankmyfamilyandfriends,whoseloveandsupporthavebeen essential. Finally,Iamdeeplyindebtedtomyadvisor,Dr. WendyD.Graham,forher guidance,support,encouragement,andconstantenthusiasm. Iamgratefulforthe patienceandconfidenceshehasshowninme. Ifeelmyselfveryfortunatenotonly ii tohavebenefitedfromheracademicguidancebutalsotohaveenjoyedthefreedom shehasgivenmeduringmyPh.D. FinancialsupportforthisresearchwasprovidedinpartbytheAirForceOffice ofScientificResearchandtheUniversityofFloridaCollegeofAgriculture-Hydrologic SciencesAcademicClusterResearchAssistantshipProgram. iii TABLEOFCONTENTS ACKNOWLEDGEMENTS ii LISTOFTABLES vi LISTOFFIGURES viii ABSTRACT xiv CHAPTERS 1 INTRODUCTION 1 1.1 ResearchBackground 1 1.2 StochasticSubsurfaceHydrology,InverseProblems 8 1.2.1 Flowandtransportprocesses 8 1.2.2 Stochasticapproachtoflowandtransportmodeling ... 10 1.3 OutlineoftheDissertation 16 2 INFERENCEOFUNCONDITIONALMOMENTS 17 2.1 IntroductionandLiteratureReview 17 2.2 Theory,UnconditionalMomentEquations 29 2.3 SolutionMethods 38 2.4 CaseStudies 38 2.4.1 Definitionoftheproblem 39 2.4.2 Unconditionalsimulationresults 44 2.5 Summary 54 3 INFERENCEOFCONDITIONALMOMENTS 69 3.1 IntroductionandLiteratureReview 69 3.2 Theory,KalmanFilteringEquations 78 3.2.1 Problemformulation 81 3.2.2 Momentpropagationequations 85 3.2.3 Momentupdateequations 92 3.2.4 Filterinitialization 94 3.2.5 Solutionmethods 96 3.3 SyntheticCaseStudies 97 3.3.1 Problemdefinition,boundaryconditions 100 3.3.2 Conditionalsimulationresults 101 3.4 Summary 113 iv 4 ANALYSISOFFIELDDATA 137 4.1 Introduction 137 4.2 OU-1SiteBackgroundandInstallation 141 4.3 OU-1TracerTests 145 4.4 ProblemFormulation 146 4.4.1 Descriptionofthestateandmeasurement 146 4.4.2 Initialization 150 4.5 DiscussionoftheResults 153 4.5.1 Estimatedparameterdistributions 153 4.5.2 NAPLdistributionbasedonsoilcoreandmomentanalysis 156 4.6 Summary 180 5 CONCLUSIONS 182 APPENDIXES A DERIVATIONOFFLUXCOVARIANCES 187 B THESQUARE-ROOTDECOMPOSITIONMETHOD 193 C DEFINITIONOFPARAMETERSINTABLES3.3-3.6 200 D CHOLESKYDECOMPOSITION 201 E FORMULATIONOFTHEITERATIVELINE-SORSCHEME 202 REFERENCES 209 BIOGRAPHICALSKETCH 219 v 423 LISTOFTABLES 2.1 Inputparametersfortheunconditionalsimulations 42 2.2 Inputparametersforunconditionalcasestudies 44 3.1 Inputparametersforconditionalsimulations 100 3.2 Inputparametersfortheconditionalcasestudies 103 3.3 SummaryofthePriorandPosteriorstatisticsforsyntheticcaseNo.l (p=0.0)conditionedwithconcentrationmeasurementsatfrequencies 0.3days,0.6daysand0.9days 109 3.4 SummaryofthePriorandPosteriorstatisticsforsyntheticcaseNo. (p=0.25)conditionedwithconcentrationmeasurementsatfrequencies 0.3days,0.6daysand0.9days 110 3.5 SummaryofthePriorandPosteriorstatisticsforsyntheticcaseNo. (p=—0.25)conditionedwithconcentrationmeasurementsatfrequen- cies0.3days,0.6daysand0.9days Ill 3.6 SummaryofthePriorandPosteriorstatisticsforsyntheticcaseNo.l s((yppnt—=he0—t,1i.cs0a,cmappsleeirnfNegoc.tnl5eyt(wnpoerg=katis0,vheoluwysnecoi4rn2r0eFlicagotunercdeen\3t.nr4Kabt),iaonsndynmltehnae#stniucrfieecmladessne),tNsaon.adt time=0.5and1.0days) 112 4.1 Tracers used inInterwell PartitioningTracerTest (IWPT) at OU-1 testcell, includingpartitioningcoefficients (KN) andinjectedtracer concentration ' 146 4.2 InputparametersforstochasticsimulationusingOU-1IWPTdata . 151 4.3 Locationsofmultilevelsamplersinexperimentalandsimulationdomain152 4.4 Colorcodescorrespondingtoelevationsaboveclayformulti-levelsam- plermeasurementpointsinexperimentalandsimulationdomain. . . 153 4.5 TargetanalyteconcentrationinanLNAPLsamplecollectedfromOU-1 testcell 157 4.6 ComparisonofNAPLsaturationspredictedbythesoilcoreanalysis andpartitioningtracers 157 4.7 Massrecovery,wellsweptvolume,andaverageNAPLsaturation(after Annableetal. [3]) 158 vi C.l DefinitionofthePriorandPosteriorstatisticsusedinTable3.3through Table3.6 200 vii LISTOFFIGURES 2.1 Schematic diagramofthe three-dimensional simulationdomain and boundaryconditions 40 2.2 Finitedifferencemeshfortheexampleproblem 41 2.3 Unconditionalmeanconcentration isshown asdashed contour lines superimposedonthesyntheticallysimulatedsinglereplicateconcen- tlCarayaseterioNZno4.fil(e0ld(.p7s6=2hom0w)nabaosvseoltihdecboansteo)uratlitniesm.e0S.h1,ow0n.5,ab1.o0v,eainsdho1r.i5zdoanytsa.l 46 2.4 U(0n.c7o6n2dimtiaobnoavlectohnecebnatsrea)tiaotntismtean0d.a1,rd0.d5e,v1i.a0t,ioanndat1.h5o.riCzaosnetaNlol.alye(rpZ=40) 47 2.5 Plotofunconditionalconcentrationstandarddeviationversusxalong c(opn=tro0l)lineY4Z4attime0.2, 0.5, 1.0, 1.5,and2.0days. CaseNo.l 48 2.6 PatloltocoaftiKonNse05.6-3|5[,Py1c.(3x9.,7,x,2£.)1]59v,erasnuds2t.i9m2e1amlofnrgomthtehecoinntfrloolwlbioneunYd4aZr4y onthecontrollineY4Z4. CaseNo.l (p=0) 49 2.7 Plotof£ [P9lC(x,x,t)]versusxalongcontrollineY4Z4attime0.2, 0.5,1.0,1.5,and2.0days. CaseNo.l (p=0) 49 2.8 sUpneccotndtiotiroenfaerlenccoerreploaitnitonxsp(2?/j./1(5x9,x'm),,1p.Cj7/7(x8,xm',),0a.7n6d2pmcc)(xa,rxe')shwiotwhnraes- sSohloidwncolaobrovceonitsouhrorliiznoenst,aslolliadyecronZt4ou(r0.l7i6ne2s,manadbodvaeshtehdebcaosnet)ouartltiniemse. 0.1,0.2,0.5,1.0,1.5,and2.0days. CaseNo.l(p=0) 57 2.9 Uernecnocnediptoiionntalxco(r2r.e1l5a9timo,ns1p.y7y7(8x,mx,')0a.7n6d2pmcq)i(xa,rex')shwoiwtnhraesspseocltidtocorleofr- cZNoo4n.tl(o0u.(r7p6=2linm0e)saabnodvesotlhiedbcaosnet)ouartltiinemse.0S.1h,ow0.n5,a1b.o0v,eanisdh1o.r5izdoanytsa.llCaayseer 58 2.10 Unconditionalmeanconcentrationisshownassolidcontourlinessu- aaapsbenorddviaeCmsaphtsoehesdeeNdcbooa.osn3net)o(tupahrte=ltisn-iye0mns.te.2he50St.)h5i,coaw1l.nl0y,asbaionmvduel1ai.st5ehddoarycisoz.nocnetCanaltsrelaatyNieoor.n2Zf4i(epl(d0=.s7h06.2o2w5mn) 59 viii 2.11 Unconditionalcorrelationspyy(x,x'),pcy(x,x'),andpcc(x,x')withre- specttoreferencepointx(2.159m, 1.778m,0.762m)areshownas sS0.oh2l,oidw0n.c5o,laob1r.o0vc,eoanitnsoduhro1r.l5iizndoeansty,asslo.lliCadyaescroenZNt4oou.(r20.l(7ip6ne2=s,m0a.n2a5db)odvaeshtehdecbaosnet)ouartltiniemse. 60 2.12 Unconditionalcorrelationspyy(x,x'),pcy(x,x'),andpcc(x,x')withre- specttoreferencepointx(2.159m, 1.778m,0.762m)areshownas s0S.oh2l,oidw0n.c5o,laob1r.o0vc,eonaitnsoduhro1r.l5iizndoeansy,tsa.slollCiaadyseceronNZto4o.u3(r0{.lp7i6n=e2s,m-0a.na2db5o)dvaeshtehdecbaosnet)ouartltiniemse. 61 2.13 Unconditionalcorrelationspyy(x,x'),pcy(x,x'),andpcc(x,x')withre- specttoreferencepointx(2.159m, 1.778m,0.762m)areshownas s0S.oh2l,oidw0n.c5o,laob1r.o0vc,eoanitnsoduhro1r.l5iizndoeansy,tsa.slollCiaadyseceronNZto4o.u4(r0(.lp7i6n=e2s,m1)anadbodvaeshtehdecbaosnet)ouartltiniemse. 62 2.14 Unconditionalcorrelationspyy(x,x'),pcy(x,x'),andpcc(x,x')withre- specttoreferencepointx(2.159m, 1.778m,0.762m)areshownas sS0.oh2l,oidw0n.c5o,laob1r.o0vc,eoanitnsoduhro1r.l5iizndoeansyt,sa.slollCiaadyseceronNZto4o.u5(r0(.lp7i6n=e2s,m-a1)nadbodvaeshtehdecbaosnet)ouartltiniemse. 63 2.15 ComparisonofUnconditionalconcentrationstandarddeviationversus xalongcontrollineY4Z4attime0.5days 64 2.16 vCeormspuasrxisaolnonogfucnocnotrnodlitliionneaYl4lZon4giattudtiinmael0m.a5crdoadyisspersivefluxP9iC(x,x) 65 2.17 ComparisonofunconditionalPyc(x,x)at(1.397m,1.778m,0.762m) versustime 66 2.18 [A] ComparisonofUnconditionalpyc(x,x') withrespecttoreference pointx'(2.159m,1.778m,0.762m)versusxalongcontrollineY4Z4at time0.5days. [B]ComparisonofUnconditionalp9lC(x,x')withrespect toreferencepointx(2.159m,1.778m,0.762m)versusxalongcontrol lineY4Z4attime0.5days 67 2.19 Unconditionalcorrelationspyy(x,x'),pcy(x,x'),andpcc(x,x')withre- specttoreferencepointx(2.159m, 1.778m,0.762m)areshownas sSohloidwncolaobrovceonitsouhrorliiznoens,taslolliadyecronZt4ou(r0.l7i6ne2s,manadbodvaeshtehdecbaosnet)ouartltiniemse. 0.2,0.5,1.0,and1.5days. Hydraulicconductivityisspatiallyinvariant. 68 3.1 Estimatepropagation. ConcentrationmeasurementsC*_jandC*be- comeavailableattimet^xandtimeU,respectively 80 3.2 StructureofthesequentialKalmanfiltering 98 3.3 Schematicdiagramoftheunconditionalsimulationdomainandbound- aryconditions. Twelvemultilevelsamplers (MLSs) generate60con- centrationmeasurementsateachmeasuringtime 102 ix 3.4 Horizontalplanviewofthespatialplacement ofmultilevelsamplers (MLSs). (a) 12 MLSsevenlydistributed; (b) 12-MLSconfiguration; (c)84-MLSsnap-shotconfiguration 117 3.5 NaturalloghydraulicconductivityInKgeneratedbyaturning-band a0.l5g4o4r,it0h.7m62i,sasnhdow0n.9a8t0mhoraibzoovntealthelaybearsse,Zr2e,spZe3c,tivZe4l,y)and Z5 (0.327, 118 3.6 NaturalloghydraulicconductivityInifgeneratedbyaturning-band a2.l7g9o4rimthmfriosmshtohewnoraitgivneritnicya-ldisrleiccetsioYn2,,rYes4p,ecatnivdelYy6.)(0.762,1.778,and 119 3.7 Vhoercitzoorntpallotlayoefrsx-Zy2,coZ3m,poZn4e,natnd(q.Zx5y)(0o.f32t7h,e0s.5i4m4u,la0t.e7d62,Daarncdy0f.l9u8x0amt abovethebase,respectively) 120 3.8 VveercttiocralpslloitcesofYx2-,zZc4o,mpaonndenYt6((q0x.z7)62,of1t.h7e78,siamnuldat2e.d794Damrcyfrfolmuxthaet origininy-direction,respectively) 121 3.9 tVaelctloaryeprlsotZ2o,fZx3-,ycZ4o,mpaonndenZt5((q0x.y32)7o,f0e.s5t4i4m,at0.e7d62D,aracnydf0l.ux98a0tmhoraibzoovne- thebase,respectively). SyntheticcaseNo.l: p=0 122 3.10 VsleicctesorY2pl,otZ4o,fxa-nzdcoYm6po(n0.e7n6t2,(q1.xz77)8o,faensdtim2a.t7e94dDmarfcryomflutxheatorviegritnicailn y-direction,respectively). SyntheticcaseNo.l: p=0 123 3.11 Conditionalandunconditionalmeanconcentrationsareshownassolid anddashedcontourlines,respectively,superimposedonthesyntheti- ctSaayllnltlyhaesyteiirmcuZlc4aats(ee0d.Nc7o6o.2nlc:menptar=baotv0ieontfhieelbdass(ei)nactolotri)m.esSh0.o5w,n1.a0b,oavendis1h.o5rdiazyosn.- 124 3.12 Ct1.oo0un,rdialtniindoesn1a.al5tdchaooynrsci.zeonSntytranaltthileoatnyiecsrtcZaa4nsde(a0rN.do7.6ld2:evmipaat=bioo0vneisthpelobtatseed)aaststoilmiedsc0o.n5-, 125 3.13 tEhsetiomraitgiinoanlloyfslinm9unlaitsedshNoAwnPLasrsaonldidomcofnitelodur(inlinceoslors)u.peSrhiomwponseadbooven 0ar.e98f0oumrhaobroivzeontthaelblaasyee,rsreZs2p,ecZt3i,veZl4y,).anSdyntZh5et(i0c.3c27a,se0.N5o4.4l,:0p.7=62,0.an.d. 126 3.14 Estimationofln0„isshownassolidcontourlinessuperimposedonthe originallysimulatedNAPLrandomfield(incolor). Shownaboveare threeverticalslicesY2,Y4,andY6 (0.762, 1.778,and2.794mfrom theorigininy-direction,respectively). SyntheticcaseNo.l: p=0. . . 127 3.15 Conditionalstandarddeviationofln9nisshownassolidcontourlines a0t.9f8o0urmhaorbiozvoenttahlelbaayseer,srZe2s,peZc3t,iveZl4y,).anSdynZt5het(0i.c32c7a,se0.N5o4.4l,:0p.7=62,0.an.d. 128 3.16 aCtontdhirteieonvearltisctaalnsdlaircedsdYe2v,iaYt4i,onanofdlYn60n(0i.s7s62h,ow1n.77a8s,saolnidd2c.o7n9t4oumrflirnoems theorigininy-direction,respectively). SyntheticcaseNo.l: p=0. . . 129 x