Semi-lagrangian advection methods and their applications in geoscience 1st edition steven j. fletche

Page 1


Semi-LagrangianAdvectionMethodsandTheir ApplicationsinGeoscience1stEditionStevenJ. Fletcher

https://ebookmass.com/product/semi-lagrangian-advectionmethods-and-their-applications-in-geoscience-1st-editionsteven-j-fletcher/

Instant digital products (PDF, ePub, MOBI) ready for you

Download now and discover formats that fit your needs...

Data Assimilation for the Geosciences. From Theory to Application 1st Edition Edition Steven J. Fletcher (Auth.)

https://ebookmass.com/product/data-assimilation-for-the-geosciencesfrom-theory-to-application-1st-edition-edition-steven-j-fletcher-auth/

ebookmass.com

Data Assimilation for the Geosciences: From Theory to Application 2nd Edition Steven J. Fletcher

https://ebookmass.com/product/data-assimilation-for-the-geosciencesfrom-theory-to-application-2nd-edition-steven-j-fletcher/

ebookmass.com

Tribology of Graphene: Simulation Methods, Preparation Methods, and Their Applications Oleksiy V. Penkov

https://ebookmass.com/product/tribology-of-graphene-simulationmethods-preparation-methods-and-their-applications-oleksiy-v-penkov/

ebookmass.com

Ophthalmology Secrets Janice A. Gault

https://ebookmass.com/product/ophthalmology-secrets-janice-a-gault/

ebookmass.com

https://ebookmass.com/product/beguiled-by-her-mainely-books-clubbook-7-chelsea-m-cameron/

ebookmass.com

Greek and Roman Technology: A Sourcebook of Translated Greek and Roman Texts 2nd Edition Andrew N. Sherwood

https://ebookmass.com/product/greek-and-roman-technology-a-sourcebookof-translated-greek-and-roman-texts-2nd-edition-andrew-n-sherwood/

ebookmass.com

Medieval Arms and Armour: A Sourcebook, Volume I: The Fourteenth Century Ralph Moffat

https://ebookmass.com/product/medieval-arms-and-armour-a-sourcebookvolume-i-the-fourteenth-century-ralph-moffat/

ebookmass.com

Feinde John Grisham

https://ebookmass.com/product/feinde-john-grisham/

ebookmass.com

GOVT 10: Principles of American Government 10th Edition

https://ebookmass.com/product/govt-10-principles-of-americangovernment-10th-edition-edward-i-sidlow/

ebookmass.com

Mary, Countess of Derby, and the Politics of Victorian Britain Jennifer Davey

https://ebookmass.com/product/mary-countess-of-derby-and-the-politicsof-victorian-britain-jennifer-davey/

ebookmass.com

SEMI-LAGRANGIANADVECTION METHODSANDTHEIRAPPLICATIONS

INGEOSCIENCE

SEMI-LAGRANGIAN ADVECTION METHODSAND THEIR APPLICATIONS

INGEOSCIENCE

ResearchScientistIII

CooperativeInstituteforResearchintheAtmosphere(CIRA) ColoradoStateUniversity FortCollins,CO,UnitedStates

Elsevier

Radarweg29,POBox211,1000AEAmsterdam,Netherlands

TheBoulevard,LangfordLane,Kidlington,OxfordOX51GB,UnitedKingdom

50HampshireStreet,5thFloor,Cambridge,MA02139,UnitedStates

Copyright©2020ElsevierInc.Allrightsreserved.

MATLAB® isatrademarkofTheMathWorks,Inc.andisusedwithpermission.

TheMathWorksdoesnotwarranttheaccuracyofthetextorexercisesinthisbook.

Thisbook’suseordiscussionofMATLAB® softwareorrelatedproductsdoesnotconstituteendorsementorsponsorshipby TheMathWorksofaparticularpedagogicalapproachorparticularuseoftheMATLAB® software.

Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans,electronicormechanical, includingphotocopying,recording,oranyinformationstorageandretrievalsystem,withoutpermissioninwritingfromthe publisher.Detailsonhowtoseekpermission,furtherinformationaboutthePublisher’spermissionspoliciesandour arrangementswithorganizationssuchastheCopyrightClearanceCenterandtheCopyrightLicensingAgency,canbefound atourwebsite: www.elsevier.com/permissions.

ThisbookandtheindividualcontributionscontainedinitareprotectedundercopyrightbythePublisher(otherthanasmay benotedherein).

Notices

Knowledgeandbestpracticeinthisfieldareconstantlychanging.Asnewresearchandexperiencebroadenour understanding,changesinresearchmethods,professionalpractices,ormedicaltreatmentmaybecomenecessary.

Practitionersandresearchersmustalwaysrelyontheirownexperienceandknowledgeinevaluatingandusingany information,methods,compounds,orexperimentsdescribedherein.Inusingsuchinformationormethodstheyshouldbe mindfuloftheirownsafetyandthesafetyofothers,includingpartiesforwhomtheyhaveaprofessionalresponsibility.

Tothefullestextentofthelaw,neitherthePublishernortheauthors,contributors,oreditors,assumeanyliabilityforany injuryand/ordamagetopersonsorpropertyasamatterofproductsliability,negligenceorotherwise,orfromanyuseor operationofanymethods,products,instructions,orideascontainedinthematerialherein.

LibraryofCongressCataloging-in-PublicationData

AcatalogrecordforthisbookisavailablefromtheLibraryofCongress

BritishLibraryCataloguing-in-PublicationData

AcataloguerecordforthisbookisavailablefromtheBritishLibrary

ISBN:978-0-12-817222-3

ForinformationonallElsevierpublications visitourwebsiteat https://www.elsevier.com/books-and-journals

Publisher: CandiceJanco

AcquisitionEditor: AmyShapiro

EditorialProjectManager: RubySmith

ProductionProjectManager: VigneshTamil

Designer: MatthewLimbert

TypesetbyVTeX

1.Introduction

Contents

2.Eulerianmodelingofadvectionproblems

2.1Continuousformoftheadvection equation7

2.2Finitedifferenceapproximationtothe Eulerianformulationoftheadvection equation14

2.3Implicitschemes49

2.4Predictor–correctormethods65

2.5Summary66

3.Stability,consistency,andconvergence ofEulerianadvectionbased numericalmethods

3.1Truncationerror69

3.2Dispersionanddissipationerrors77

3.3Amplitudeandphaseerrors78

3.4Stability81

3.5Quantifyingthepropertiesoftheexplicit finitedifferenceschemes87

3.6Linearmultistepmethods96

3.7Consistencyandstabilityofexplicit Runge–Kuttamethods100

3.8Implicitschemes102

3.9Predictor–correctormethods107

3.10Summary108

4.Historyofsemi-Lagrangianmethods

4.1Fjørtoft(1952)paper111

4.2Welander(1955)paper115

4.3Wiin-Nielsen(1959)paper120

4.4Robert’s(1981)paper124

4.5Summary127

5.Semi-Lagrangianmethodsforlinear advectionproblems

5.1DerivationoftheLagrangianformfor advection129

5.2Derivationofthesemi-Lagrangian approach131

5.3Semi-Lagrangianadvectionofthebell curve135

5.4Semi-Lagrangianadvectionofthestep function140

5.5Summary145

6.Interpolationmethods

6.1Lagrangeinterpolationpolynomials148

6.2Newtondivideddifferenceinterpolation polynomials158

6.3Hermiteinterpolatingpolynomials163

6.4Cubicsplineinterpolation polynomials166

6.5Summary172

7.Stabilityandconsistencyanalysis ofsemi-Lagrangianmethodsforthe linearproblem

7.1Stabilityofsemi-Lagrangian schemes175

7.2StabilityanalysisofLagrangeinterpolation polynomials177

7.3StabilityanalysisofthecubicHermite semi-Lagrangianinterpolation scheme190

7.4Stabilityanalysisofthecubicspline semi-Lagrangianinterpolation scheme195

7.5Consistencyanalysisofsemi-Lagrangian schemes199

7.6Summary203

8.Advectionwithnonconstantvelocities

8.1Semi-Lagrangianapproachesforlinear nonconstantadvectionvelocity205

8.2Twoandthreetimelevelschemes209

8.3Semi-Lagrangianapproximationsto nonlinearadvection219

8.4NonlinearinstabilityI224

8.5NonlinearinstabilityII226

8.6Boundaryconditionsforlimitarea models230

8.7Summary234

9.Nonzeroforcings

9.1Methodsofcharacteristics approach236

9.2Semi-implicitintegration241

9.3Semi-implicitsemi-Lagrangian (SISL)243

9.4Spatialaveraging246

9.5Optimalaccuracyassociatedwith uncenteringtimeaverages247

9.6Semi-Lagrangiantrajectoriesanddiscrete modes252

9.7Time-splitting258

9.8Boundaryconditionsforthe advection-adjustmentequation261

9.9Summary266

10.Semi-Lagrangianmethodsfor two-dimensionalproblems

10.1Bivariateinterpolationmethods269

10.2Gridconfigurations276

10.3Semi-implicitsemi-Lagrangianfinite differencesintwodimensions280

10.4Nonlinearshallowwaterequations284

10.5Finiteelementbasedsemi-Lagrangian method287

10.6Semi-Lagrangianintegrationinflux form294

10.7Semi-Lagrangianintegrationwithfinite volumes302

10.8Semi-Lagrangianadvectioninflowswith rotationanddeformation307

10.9Eliminatingtheinterpolation311

10.10Semi-Lagrangianapproachwithocean circulationmodels318

10.11Transparentboundaryconditions320

10.12Testcasesfortwo-dimensional semi-Lagrangianmethods331

10.13Semi-Lagrangianmethodswiththe2D quasi-geostrophicpotentialvorticity(Eady model)335

10.14Summary346

11.Semi-Lagrangianmethodsfor three-dimensionalproblems

11.1Trivariateinterpolationmethods351

11.2Semi-Lagrangianadvectionintheprimitive equations355

11.33Dfluxformsemi-Lagrangian360

11.4Three-dimensionalfullyelasticEuler equationswithsemi-Lagrangian361

11.5Sensitivitytodeparturepoint calculations368

11.6Consistencyofsemi-Lagrangiantrajectory calculations372

11.7Semi-implicitEulerianLagrangianfinite elements(SELFE)374

11.8Summary380

12.Semi-Lagrangianmethodsonasphere

12.1Vectoroperatorsinspherical coordinates381

12.2Griddevelopmentforasphere383

12.3Grid-pointrepresentationsofthe sphere388

12.4Spectralmodeling394

12.5Semi-Lagrangianandalternatingdirection implicit(SLADI)scheme413

12.6Globalsemi-Lagrangianmodelingofthe shallowwaterequations416

12.7Spectralmodelingoftheshallowwater equations435

12.8Semi-implicitsemi-Lagrangianschemeon thesphere442

12.9RemovingtheHelmholtzequation447

12.10Stableextrapolationtwo-time-levelscheme (SETTLS)449

12.11Fluxformonasphere456

12.12Numericaltestcasesforthesphere464

12.13Summary468

13.Shape-preservingandmass-conserving semi-Lagrangianapproaches

13.1Shape-preservingsemi-Lagrangian advection471

13.2Cascadeinterpolation482

13.3Semi-Lagrangianinherentlyconservingand efficientscheme(SLICE)493

13.4Flux-formsemi-Lagrangianspectralelement approach521

13.5Conservativesemi-LagrangianHWENO methodfortheVlasovequations527

13.6Summary540

14.Tangentlinearmodelingandadjointsof semi-Lagrangianmethods

14.1Derivationofthelinearizedmodel541

14.2Adjoints542

14.3Testofthetangentlinearandadjoint models545

14.4Differentiatingthecodetoderivethe adjoint546

14.5Tangentlinearapproximationsto semi-Lagrangianschemes548

14.6Perturbationforecastmodeling559

14.7Sensitivityofadjointofsemi-Lagrangian integrationtodeparturepoint iterations562

14.8Summary566

15.Applicationsofsemi-Lagrangian methodsinthegeosciences

15.1Atmosphericsciences569

15.2Atmosphericchemistry576

15.3Hydrologicalandocean applications580

15.4Earth’smantleandinterior586

15.5Otherapplications591

15.6Summary592

16.Solutionstoselectexercises

Bibliography 597 Index 605

1 Introduction

Advectionplaysavitalpartinmanydifferentformsofgeophysicalmodelingthatcanaffect everybody,everydayinsomeform.Advection,initslinearornonlinearform,affectsthe weatherandoceanforecasts,riversedimentaswellaschemicaltransport,alongwiththe solarwindforecast,modelingofmagmaflows,andhydrothermaltransport.Aninaccurate forecastinanyofthesesituationscouldleadtoquitecatastrophic,andlifethreatening,results.

Moisturetransportforanonshoreflow,ifmiscalculated,couldleadtoamisleadingforecastofanextremerainevent,whichinturncouldleadtoaflashfloodwarningnotbeing issued.TheonsetofthetornadoseasonintheUnitedStates’Midwestispartiallydependent ontheadvectionofthewarmmoistairfromtheGulfofMexicointothecontinent.Anover-, orunder-,predictionofthisadvectioncouldleadtoverydifferentoutcomeswhichcould indicatesevere,ornotsevere,weather.

Aswesawin2010,theeruptionofEyjafjallajökullinIcelandledtotheshutdownofthe NorthAtlanticairspaceforoveraweekbecausethenumericalweatherpredictioncenters didnothavegoodenoughmodelsforthetransportoftheash.Sincethentherehavebeen majordevelopmentsinthisarea,particularlyintransportasaformofadvection,wheresemiLagrangianapproacheshavebeendeveloped.

Thestatementaboveindicatesthatadvectionisalsoreferredtoasatransportproblemin atmosphericchemistry,sedimenttransportinriversandoceanmodeling.Ifwegobelowthe Earth’ssurface,whenweareconcernedwithtemperature,advectionisoftenreferredtoas convectionmodeling.Advectionalsooccursinspacethroughtheionosphereaswellassolar windinteractionswiththemagnetosphere.

ApproximationsofadvectionintheearlynumericalweatherpredictionswereconsideredintheEulerianformwhichisassociatedwithmodelingtheflowasitpassesapoint. Ontheotherhand,theLagrangianformulationdeals withtheflow.Bothapproacheshave advantagesanddisadvantages.Intheearlynumericalweatherprediction,themodelingof advectionintheEulerianformwaseitherdoneusingafinitedifferenceorfiniteelementformulation.Theproblemsthatwereobservedinthoseearlyattemptswereassociatedwiththe Courant–Friedrichs–Lewy,or CFLcondition,whichisassociatedwiththe stability ofthe numericalscheme.Effectively,thisconditionsplacesarestrictiononthesizeofthetimestep, aswellasthegridorelementsize.

Itwasshownearlyoninnumericalweatherpredictionthatthisconditioncouldleadto veryunstableforecastsifitwasnotmet,butthisrestrictedthesizeofthetimestep,whichin turnpreventedrunninglongerforecastsasthecomputationalresourceswerenotavailableto

runatsuchafinetemporalresolution.Therewasalsotheproblemofdampingwithcertain Eulerian-basedfinitedifferenceschemes.

IfweweretoconsidertheLagrangianframework,wewouldstartwithasetofparticles, followthemintime,andthenapplythenumericalapproximationtothedifferentialoperators there.Theproblemwiththisapproachisthattheparticlescouldbecometoofarapart,which isnotpracticaltoachieveviableapproximations.WeshowacopyofFig.2fromWelander [198]inFig. 1.1,whichillustratesthisproblem.

FIGURE1.1 CopyoftheLagrangiandeformationplot,Fig.2fromWelander,1955:Studiesonthegeneraldevelopmentofmotioninatwo-dimensionalidealfluid, Tellus, 17,141–156. https://www.tandfonline.com/doi/abs/10. 3402/tellusa.v7i2.8797 https://creativecommons.org/licenses/by/4.0/

TheadvantageoftheLagrangianapproachisthatthevalueoftheparticlefollowingthe trajectoryisconstant.However,wecannotkeeptrackofalltheseparticles,butwewould likesomeformofatechniquethatutilizesthis.Herecomesemi-Lagrangianapproaches.The basisofsemi-LagrangianapproachesisthatwehaveafixedEuleriangridwithknownvalues ofthefieldatthesegridpointsattime t n andweknowthelocationofthegridpointsat t n+1 , butwedonotknowwhatisthevalueofthefieldthere.

TheLagrangianformoftheadvectionequation,whichyouwillseethroughoutthebook, isgivenby

Dψ Dt = 0, subjecttothekinematicequation

D x Dt = u, where u isreferredtoastheadvectionvelocity.

Thesemi-Lagrangianapproachamountstosayingthatthesecondequationabovetellsus howfartheparticlehastraveledinoneortwotimestepswithoutforcing.Whereasthefirst equation,whenintegratedwithrespecttotime,tellsusthatthevalueofthefieldat t n 1 , or t n ,dependingonwhetheroneusesatwoorthreetimeleveldiscretizationforthetime derivative,isthevalueofthetraceratthe arrivalpoint at t n+1 .Theproblemwehaveisthat wearenotguaranteedthatthe departurepoint willbeatagridpointat t n or t n 1 .Thevalue of ψ canbefoundthroughinterpolationbutwehavetobecarefulwhichorder,andwhat properties,wewishtheinterpolationpolynomialtosatisfy.Thisisthefinitedifferencepoint ofview;thereisalsoasimilarviewwiththefiniteelementapproachwhereitnotapointbut anelementtodealwith.Finally,therearethefluxformandfinitevolumeapproachesthat conservepropertiesofthefieldbetweenthetimelevels.

Giventhismotivation,wenowmoveontobrieflysummarizewhatwehaveineachchaptertoaddressthesituationdescribedabove.Wefinishwithapplicationsofsemi-Lagrangian approachesinthegeosciences,followedbysolutionstoselectexercises.

InChapter 2 westartbyshowingderivationsoftheadvection/transportequationsand showthecharacteristicsbasedapproachwhichwillenableustomoveontothesemiLagrangianapproaches.WewillpresentdifferentexplicitandimplicitEulerian-basedfinite differenceschemesanddemostratetheperformanceoftheseschemeswithasmoothGaussianbelladvection,aswellaswithadiscontinuousstepfunction.

Chapter 3 coversthedifferentpropertiesoftheEulerian-basedfinitedifferencescheme, whereweshallintroducedifferentformsofstability,convergence,andconsistency.Wewill lookatdispersionanddiffusionpropertiesofthedifferentschemessothatweknowtheir likelyperformance.

GiventhebehavioroftheEulerianscheme,wemoveontothehistoryoftheideastowardsadoptingasemi-LagrangianapproachinChapter 4;weshouldnoteherethatthe semi-Lagrangianapproachwasreferredtoinitiallyasquasi-Lagrangian,buttheschemesare alsoreferredtoas Eulerian–Lagrangian approach,andinsomedisciplinesasa particletrajectorytracking.

InChapter 5 wewillintroducethesemi-Lagrangianapproachforthelinearonedimensionalscalaradvectionandshowitsperformancewithdifferentorderofinterpolation fortheadvectionoftheGaussianbellcurveandthestepfunction.

Giventhatwerequireinterpolationpolynomials,weshallintroducedifferentformsof interpolationpolynomialsinChapter 6,andshowtheirperformancesinreconstructingthe bellcurveandstepfunctionwithdifferentnumberofinterpolationpointstoillustratehow welltheapproximationsimprovewiththenumberofpoints.

Thenextstageoftheintroductionofthesemi-Lagrangianapproachesistoconsiderhow todetermineiftherearerestrictionsonthetimeandspacestepsofthedifferentschemes;this isaccomplishedinChapter 7.

Thepropertiesandtechniquestoperformthesemi-Lagrangianintegrationsofarhave beenintendedfortheconstantvelocityandzeroforcingone-dimensionalscalarproblem. InChapter 8 wewillintroducethetechniquesthatarerequiredtobeabletosolvethecase wherethetrajectoryvelocityisnolongerconstant,whichimpliesthatwehavetoestimate thevelocityalongthetrajectory,butalsotoevaluatethevelocityatthedeparturepointinan implicititerativesolution.Wewillconsidereithertwoorthreetimelevelschemesandlook atthepropertiesofbothoftheseapproaches.

InChapter 9 wewillintroducethesituationwherewehaveaforcingtermontherighthandsideoftheadvectionequations.Therearemanydifferentformsofadvectionproblems thathavenonzeroforcingterms,andgiventhedynamicalsituationthatisbeingconsidered, theapproachtodealwiththesesituationscouldleadtononphysicalmodesbeingexcitedin thenumericalsolution.Awidelyusedapproachtodealwiththissituationisreferredtoas thesemi-implicitsemi-Lagrangian(SISL)method,wherenonlineartermsaretreatedasexplicit,whilethelineartermsaretreatedimplicitly.Weshallintroducethistechniquealong withothertechniquesthathavebeendevelopedovertheyearstostabilizeandincreasethe efficientlyofthesemi-Lagrangianintegrationwhenaforcingtermispresent.

Wemoveontothetwo-dimensionalprobleminChapter 10.Inthischapterweshallintroducethetechniquesofbivariateinterpolationtoestimatethevalueofthefieldatthe departurepoint,alongwiththestabilityanalysisintwodimensions.Weshallalsointroducedifferentstaggeredhorizontalandverticalgridsthatarequiteoftenusedfornumerical modeling.Weshallalsolookattwo-dimensionalapproximationsfornonconstantvelocities, aswellasfornonzeroforcingtermswherethedeparturepointnowisinanarea,element,or cellforthefinitedifference,finiteelements,orfinitevolume/fluxformapproaches.Weshall alsopresentsometestcasesthatarefrequentlyusedtoanalyzetheperformanceofnewdevelopmentsinsemi-Lagrangiantheory.Finally,weshallshowanexampleofsemi-Lagrangian advectionwiththeEadymodelwhichdescribesbaroclinicinstabilityinan x –z planeapproximationoftheatmosphere.

Thenextstepistointroducethethirddimensionfortheadvectionproblem,whichwe doinChapter 11.Inthischapterweshallextendtheideasfromthepreviouschaptersto thethree-dimensionalproblem.Weconsiderthecasethatweneedtointerpolatetoadeparturepointinthreedimensions.Weshallintroducesomedifferentverticalcoordinatesystems thatareusedinoperationalnumericalweatherpredictionsystems,aswellasinoceanmodeling.Weshallconsiderfinitedifference,finiteelement,andfinitevolumeapproaches,all asanextensionoftheflux-formfromthelastchapteronchemicaltransport.Werevisitthe assumptionsofonlyneedingtwoiterationstofindthedeparturepoint,aswellasbeinginconsistentintheorderofthevelocityinterpolation,comparedtothefields’interpolationto thedeparturepoint.

Aswedonotliveonacube,weneedtolookathowwecanextendthetechniqueto sphericalcoordinates,andwedothisinChapter 12.Hereweshallintroducesemi-Lagrangian developmentonthesphere.Thischaptercomprisesoftwoparts:thefirstintroduceshow thevectorcalculustransferstosphericalcoordinates,projections,numericalgrids,aswellas introducesthetheoryofspectralmethods.Thesecondpartdescribeshowsemi-Lagrangian

theoryisappliedtodifferentmodelsinsphericalcoordinatesandmultipledimensions,where wewillpresentfinitedifferenceandfinitevolumeapproaches,aswellassemi-Lagrangian methodswithspectralmethods.

Oneoftheproblemsthathasbeennoticedinthestandardfinitedifferenceapproachto thesemi-Lagrangianmethodsisthattheymaynotpreservetheshapeoftheobjectbeing advected;itmayalsoberequiredthatthe mass isconserved.InChapter 13 wewillintroduce differenttechniquesthathavebeendevelopedtoensurethatthesemi-Lagrangianscheme canpreservetheshapeofthebodybeingadvected,toavoidtheGibbsphenomenaofunderandovershoots.WeshallintroducethecascadeinterpolationmethodsbetweenEulerianand Lagrangiangrids,alongwithdifferentfinitevolumeapproachesthatenabletheconservation ofmasseitherlocallyofglobally.WewillalsointroduceSLICEwhichisafinitevolumebased schemeutilizingthecascadeinterpolation.ThechapterfinishesbyintroducingtheVlasov familyofnonlinearPDEsandtheHWENOapproachtosolvethem.

Inthelasttheoreticalchapter,Chapter 14,weshallintroducetheconceptofthetangent linearmodel,alongwithadjoints.Weshallalsointroducethenotionofdifferentiatingthe codetoobtaintheadjointratherthanderivingitanalytically.Wewillalsointroducetheperturbationforecastmodelwhichisseenasanalternativetothetangentlinearapproximation. Afterwardswewillapplyallthesetechniquestosemi-Lagrangianmethods,wherewewill seethatthereisanimportantpropertyinthederivationthatputsalimitonthesizeofthe variationinthedeparturepointandtheinterpolation.

Thepenultimatechapterofthebook,Chapter 15,willpresetdifferentapplicationsofsemiLagrangian,Eulerian–Lagrangian,andquasi-Lagrangianapproachesforsolvingadvection, convection,transport,Navier–Stokes,andVlasov–Maxwellequations.Wewillshowapplicationsinnumericalweatherandoceanprediction,hydrologyglaciermovement,volcanicash transport,airpollution,oceanridgehydrothermalmodels,tonameafew,toshowhowfar fetchingsemi-Lagrangianmethodsareused.Thefinalchapeterissolutionstoselectexercises.

Thatbeingsaid,wenowmoveontolearnaboutsemi-Lagrangianadvectionmethodsand theirapplicationsingeosciences.

Eulerianmodelingofadvection problems

Asweshallmentionagainsoon,therearedifferentwaysofdescribingthemovementofa particle.Itcanbewithrespecttotheparticleitself,oritcanbewithrespecttotheparticles goingpastacertainfixedpoint.Itisthelatterdescriptionthatthenexttwochaptersare concernedwith.Inthischapterweshallpresentdifferentnumericalapproximationstothe one-dimensionalconstant-velocityadvectionequation.Forsomeofthenumericalschemes, weshallpresentplotsoftheirperformancewith advecting aGaussianbellcurveandadiscontinuousstepfunctionaroundaperiodicdomain.Weshallseesomebehaviorsoftheschemes thatweshallquantifyandverifyinthenextchapter.Westartthischapternowwithabrief summaryoftheone-,two-,andthree-dimensionalgeneralEulerianformsoftheadvection equation,followedbyacoupleofbriefderivationsoftheadvectionequation.

2.1Continuousformoftheadvectionequation

Therearetwoapproachesthatcanbeusedtodescribetheprocessofadvection:Eulerian, wherewehaveavolumeelementthatisfixedinspaceinasetframeofreference,orLagrangian,wherethesurfaceofthevolumeelementis co-moving withthefluid,inthefluid’s frameofreference.ThusinEulerianmodelingofadvectionweareconsideringtheproblem oftheflowpastapoint,andassuchthecontinuousone-dimensionalgenerallinearadvection modelisgivenbythepartialdifferentialequation

(x,t )

+ u (x,t )

where ψ (x,t ) isthedependentvariable, t istime, x isthespatialcoordinate, u (x,t ) isthe advectionvelocity,whichcanbeafunctionofbothspaceandtime,andfinally, f (x,t ) is referredtoasthe forcingterm.Notethattheforcingtermcanbezero,linear,ornonlinear. Theadvectionvelocitycanalsobeconstant,linear,ornonlinear.

Ifwenowthinkthatthereisafieldthatistwo-dimensional,meaningthatitcanmovein bothhorizontaldirections,orinonehorizontalandtheverticaldirections,thenthereisan associatedversionoftheadvectionequationintwodimensions.Forexample,ifweconsider thehorizontalcase,thentheassociatedtwo-dimensionalpartialdifferentialequationforthis

situationisgivenby ∂ψ (x,y,t )

+ u (x,y,t ) ∂ψ (x,y,t )

(x,y,t )

)

)

where v (x,y,t ) istheadvectionvelocityinthe y direction,butitcanbeafunctionofthe otherspatialcoordinate.Finally,ifweareconsideringathree-dimensionalfieldthenitis possiblethatitcanbeadvectedwithrespecttoallthreespatialcoordinatesinaCartesian formulation,oranyothercoordinatesystem,andtheassociatedpartialdifferentialequation forthissituationisgivenby ∂ψ (x,y,z,t )

+ u (x,y,z,t )

(x,y,z,t )

w (x,y,z,t )

(x,y,z,t )

+ v (x,y,z,t )

(x,y,z,t )

where w (x,y,z,t ) istheadvectionvelocityintheverticaldirection.Notethatallthreeadvectingvelocitiescanbefunctionsofallthreespatialdimensions,alongwithtime.

Forthemultidimensionalequationsabove,(2.2)and(2.3),itispossibletosimplifythese equationsusingvector-differentialnotationas

isthespatial

gradientoperatorgivenby

Wenowconsiderhowtoderivetheone-dimensionaladvectionequationthatwepresented above.Weshallpresenttwodifferentapproaches:onebasedonadynamicaldescriptionof thesituation,andonethatusesamoremathematicalargument.

2.1.1Derivationoftheone-dimensionalEulerianadvectionequation

Asjustmentioned,inthissectionweshallpresenttwodifferentapproachestoderivethe one-dimensionaladvectionequation.Weshouldnoteherethattheadvectionequationhas manydifferentnames,itisquiteoftenreferredtoasthetransportequation,butitisalso referredtoastheconvectionequation,alongwithbeingassociatedwiththewaveequation. Thefirstderivationthatweconsiderhereisadynamics-basedapproachemployingtheideas ofmassconservation,continuity,andincompressibility.

Massconservationderivationoftheadvectionequation

Atthebeginningofthischapterwereferredtoavolumeelement,whichwewillnow denoteas dV asshowninFig. 2.1,where P (x,y,z) isthecentroidofthevolumeelement.As weareconsideringtheEulerianformulation,wehavethatthesidesofthevolumeelement

arefixedinspace.Therefore,wewishtoconsiderthesituationwherewehaveaflowinand outofthevolumeelementthroughthesides.

FIGURE2.1 Schematicofagenericvolumeelement, dV

Ifwedenotethemassdensityatthepoint P (x,y,z) as ρ (x,y,z),whichrepresentsthe massdividedbythevolume,thenweassumethatitisanaverage,andnearlyuniformmass densityinallof dV .Thusthetotalmass, M ,thatiswithin dv isgivenbytheintegralquantity

Thenextassumptionthatwemakeisthattherearenosourcesorsinksofmasswithin dV .As suchwehavethat dM dt representstherateatwhichmassentersorexitsthevolumethrough thesurface dS .

Whenweareconsideringthethree-dimensionalvolume,therearemultiplesurfacesofthe volume,assuchwehaveasideofthevolumedenotedby d S,andtheunitnormalvectordenotedby n,whereitisassumedthatthenormalvectorpoints“out”ofthevolumeelement. Thenextimportantquantitythatwehavetointroduceisthe flux ofthemass,whichrepresentsmassperunitarea,perunittime,thatwehavepassingthroughasurface,andwhichis givenby ρ v,where v isthefluidvelocity.

Wenowrequirethatthemassperunittimethatisflowingthroughthesideofthevolume, d S,begivenby ρ v d S ≡ ρ v ndS ,andthusthetotalrateofflowofmassoutofthe volumeelement dV isgivenbythesumofthemassperunittimeflowingthrougheachof thesurfaces/facesofthevolumeelement.Thusincontinuousformthisisgivenbythesurface integrals

Asmentionedbefore,wehave dM dt representingtherateatwhichmassisenteringor exitingthevolumeelementthroughthesurface,andassuchwerequire(2.6)tobeequalto

2.Eulerianmodelingofadvectionproblems

dM dt ,whichimpliesthat

AsweareconsideringafixedEuleriansurface,weareabletobringthetotaltimederivative d dt insidethevolumeintegralasapartialderivative,whichthengivesus

Whiletheexpressionin(2.8)maylookquiteintimidating,wenowintroduceatheoremthat willhelpushere.Thistheoremisreferredtoas Gauss’Theorem andisstatedbelow.

Theorem2.1 (Gauss’Theorem). Let V bearegioninspacewithboundary ∂V .Thenthevolume integralofthedivergence ∇· F of F over V andthesurfaceintegralof F overtheboundary ∂V of V arerelatedby

Gauss’Theoremisquiteoftenreferredtoasthe divergencetheorem,wherethetheoremis themathematicalexpressionforthephysicalsituationthat,intheabsenceofthecreationor destructionofmatter,thedensitywithinaregionofspacecanchangeonlybyhavingitflow inoroutfromtheregionthroughitsboundary,whichisthesituationthatwehavehere.

UponapplyingGauss’theoremtothesituationabove,weobtain

Therefore,wenowhave

Werequiretheexpressionin(2.11)toholdforallarbitraryshapedvolumes,andthustheonly waythattheintegralequationin(2.11)canbesatisfiedisiftheintegrandisequaltozero.This thenimpliesthat

Eq.(2.12)isreferredtoasthe continuityequation,whereitisdescribingthe conservation ofmass withintheEulerianframeofreference.Wecanseethatthisisalmosttheadvection equationthatwepresentedearlier,butwehavetomakeonemoreassumption,namelythat thefluidisincompressible,whichimpliesthatthedensitywithinthevolumeisconstant. Anotherwayofinterpretingthisassumptionisthatthedivergenceoftheflowvelocityis

zero.Mathematicallythenon-divergenceofthevelocityfieldisequivalentto ∇· v = 0.Ifwe nowexpandtheterm ∇· (ρ v) wehave

ρ, wherewecanseethatwerequirethedivergenceofthevelocity.Usingtheincompressibility assumptionresultsinthecontinuityequationforanincompressiblefluidas

whichisthevectorialformofthethree-dimensionaladvectionequationthatwepresented in(2.4),butwherethereisnoexternalforcingterm.

Taylorseriesexpansionderivationoftheone-dimensionaladvectionequation

Thisderivationoftheadvectionequationisfortheone-dimensionsituation,whichstarts byconsideringtheone-dimensionaldriftofanincompressiblefluid.Therefore,let ψ (x,t ) representaparticledensitythatonlychangesduetoadvectionprocessessothat

Ifwenowassumethat t issufficientlysmallsothatwecanapplyaTaylorseriesexpansion toeithersideof(2.14),thenweobtain,tofirstorder,

whichsimplifiesto

where u isanonzeroconstantvelocity.

Asanasidehere,weshouldnotethataswearedealingwithapartialdifferentialequation.Therearethreedifferenttypesthatapartialdifferentialequationcanbe:hyperbolic, parabolic,orelliptical,whichrepresentmanydifferentphysicalandgeophysicalprocesses. Thepartialdifferentialequationin(2.16)ishyperbolic.Thustheuniquesolutionof(2.16)is determinedbyaninitialcondition,denotedby

,suchthat

and

Oneapproachtosolvetheproblemaboveisbythe MethodofCharacteristics.

2.1.2Methodsofcharacteristics

Thefirstthingtonoticehereisthattheadvectionequationin(2.16)canbefactorizedas

Thenextstepistotransform x , t tonewvariables r , s suchthatthederivativetransformsas

Ifitispossibletointroducethischangeofvariable,thenwewouldhavethat ∂ψ ∂r = 0,where thisequationiseasiertosolve.Therefore,ifwelet x = x (r,s ) and t = t (r,s ),then,applying thechainrule,weobtain

Ifwenowcompare(2.20)with(2.18),weseethat ∂x ∂r = u and ∂t ∂r = 1.Integratingthese twoequationsresultsin x = ur + q (s ) and t = r + p (s ).Wenowusethe s dependenceby applyingthepropertythattheinitialconditionsspecifythesolutionalongthe x -axis.Tomake theapplicationoftheinitialconditionsinthetransformedspace,weintroducethecondition thatthe x -axisat t = 0 transformsontothe s -axis (r = 0);specifically,weneed r = 0,implying that t = 0 and x = s .Thusbysetting r = 0 and t = 0,wehave q (s ) = s and p (s ) = 0,implying that

(2.21)

Thenextstepistoinvertthetransformationabove,whichresultsin r = t and s = x ut , whichthenenableustowrite(2.18)intheform ∂ψ

= 0,implyingthat ψ = ψ (s ) = ψ (x ut ).

Giventheinitialconditions,wethereforecanconcludethatthesolutionoftheadvection problemis

Therefore,theexpressionin(2.22)definesatravelingwavethatmoveswithspeed u,andit isalsoseenthatthesolutionisconstantalongthelinesoftheform x ut = β where β is aconstant.Theselinesarereferredtoasthe characteristics forthepartialdifferentialequation,andthemethodthatwasappliedheretofindthesolutionisknownasthe methodof characteristics

InFig. 2.2 wehaveanillustrationoftheappearanceofthecharacteristicsfortheadvection equation,wereweseethatthecharacteristicsaredefinedbytheirvalueastheycrossthe x -axis,butthenhavethatvaluealongthewholepath.

Tohelpillustratethisapproach,weshallconsidertheexampleofastepfunction,which weshallrevisitmanytimesthroughoutthebookindifferentforms.Theinitialconditionsthat definethestepfunctionshowninFig. 2.3 aregivenby

Thusfrom(2.22)thesolutionis

FIGURE2.2 Exampleoftheconstantcharacteristicsfortheone-dimensionallinear,constantvelocity,advection partialdifferentialequation,withinitialconditions ψ (x, 0) = (x )

FIGURE2.3 Schematicofadvectingastepfunctionstartingatinitialtimeinthetopplot,andthenafteratimeof 1inthelowerplot.

2.Eulerianmodelingofadvectionproblems orequivalently,

0 (x ) = 1,ut ≤ x ≤ 1 + ut, 0,x<ut,x> 1 + ut. (2.25)

InFig. 2.3 wehavepresentedboththeinitialstepfunction,asmentionedabove,andthe stepfunctionatafuturetime t ,wherethesolutionistheoriginalstepfunctionthathasbeen movedtotheinterval ut ≤ x ≤ 1 + ut

Thereareafewpropertiesofthelinearconstantvelocityequationthatareusefulwhen assessingtheperformanceofthenumericalapproximationtoit:

1. Theinitialprofileispreserved;thisistrueforjumpsanddiscontinuities.

2. Theinformationoftheinitialprofiletravelsatspeed u,andtheassociateddirectionofthe travelisone-wayonly.

3. Thesolutionatagivenspatiallocation x andattime t isentirelydeterminedbythevalues oftheinitialconditionsat x = x ut .Wehaveprovidedaschematicofthissituationin Fig. 2.4.

FIGURE2.4 Schematicofback-tracingofthecharacteristicstotheinitialconditionfromthepoint x, t .

GiventhecontinuousEulerianapproachforadvection,wenowconsiderseveraldifferent numericalapproximationtotheone-dimensionallinearadvectionequationwithconstant velocity.Theapproachesthatweshallshowinthenextsectionarereferredtoa finitedifferencing scheme,wherewewillpresentexplicitandimplicitschemes,andacombinationof thetwo,whichcanbemultistepalgorithmsreferredtoaspredictor–correctormethods.

2.2FinitedifferenceapproximationstotheEulerianformulationofthe advectionequation

Inthissectionweshallintroducemanydifferentapproximationstothelinear,zero-forcing versionoftheone-dimensionaladvectionequation.Thefirstthingthatwehavetointroduce hereistheconceptofthenumericalgridforbothtimeandspace.

InFig. 2.5 wehavepresentedauniformdistributedsetofgridpoints,whereforthissetup wehaveequalspacingintimeandspace,althoughweshouldnotethatspacinginthetemporaldirectionisnotalwaysthesamesizeasthatinthespatialdirection.Wecanseethat

thereisadistancebetweenthepointsinbothtimeandspace.Thedistancebetweenthegrid pointsinthe x -directionisdenotedby x ,whereasthedistancebetweenthetimegridpoints isdenotedby t ,andtheyarereferredtoasthegridspacingandtimestep,respectively.

Illustrationofanumericalgrid.

Ifwestartwiththespatialdirection,thenweintroducetheindex i forthe x direction,while forthetemporaldirectionwithhavetheindex n.Thusatpoint (i,n) wehaveanumerical approximationtothecontinuoustracer, ψ (x,t ),as ψ (xi ,t n ),whichisusuallydenotedby ψ n i ,for i = 1, 2,...,I and n = 0, 1,...,N .Pleasenotethatthesuperscript n heredoesnot representapowerofthefunction,itissimplythetemporalindex.Therefore,ifweconsider the x directionderivative,thenwecouldusethefollowingapproximation:

wheretheapproximationin(2.26)isreferredtoasan upwind scheme. Itisalsopossibletodefineanupwindschemeforthetemporalderivativeforthenumerical approximationtothetimecomponentaswell,whichisgivenby

Theapproachespresentedhereareexamplesofwhatisreferredtoas finitedifference approximationstothegradientsofthefieldinthepartialdifferentialequation.Giventhe conceptofthegridpointsintroducedabove,weshallnowpresentaseriesofdifferentexplicitnumericalandimplicitnumericalschemes,andtheschemesthatcombineexplicitand implicitfeaturesfortheEulerianformulationoftheone-dimensionaladvectionequation.

2.2.1UpwindforwardEuler

Dependingonthesignoftheadvectivevelocity,wecandefinetwodifferentversionsof thisscheme.If u> 0,thentheversionofthe forwardEuler approximationisbaseduponthe upwindapproximationtothespatialderivativepresentedabovecombinedwiththetemporal

FIGURE2.5

2.Eulerianmodelingofadvectionproblems approximationaboveaswell,whichresultsin

Toimplementthisnumericalschemepresentedin(2.28),werearrange(2.28)as

where C ≡ u x t isreferredtoasthe Courantnumber,whichwillbecomeimportantinthe nextchapter.

InFig. 2.6 wehavepresentedthestencil,whichisthesetofpointsthatareusedinthis approximation,fortheforwardEuler,usinganupwindapproach,scheme.Itshouldbenoted thatforwardEulerreferstotheforwardapproximationofthetemporalcomponent.Forall ofthestencilplotsthatweshowinthischapter,thegreencirclesarethosethatareusedto createthatfinitedifferencescheme.

FIGURE2.6 IllustrationofthestencilfortheupwindforwardEulerfinitedifferencescheme.

Iftheadvectivevelocityis u< 0,thenwecandefineadownwindapproximationtothe spatialderivative,andifweusetheforwardEulerapproximationforthetimecomponent, thenweobtain

Toimplementthisnumericalschemepresentedin(2.30),werearrangeitas

Notes:(1)Thereasonwhytheschemein(2.30)isnotreferredtoasthebackwardEuleris becausethatnamebelongstotheimplicitversionsofthetimederivativethatweshallpresent later.(2)Eqs.(2.28)and(2.30)arereferredtoasthe differenceequations forthecontinuous partialdifferentialequation.

Throughoutthischapterweshallassesstheperformanceofsomeoftheadvectionschemes withtheadvectionofastepfunction,aswellasaGaussianbellcurve.Beforethatweinvestigatesomepropertiesofeachschemeinthenextchapter.Aswesawearlier,thecharacteristics

2.2.FinitedifferenceapproximationstotheEulerianformulationoftheadvectionequation 17

tellusthattheshapesshouldsimplybemovingalong,withoutanydistortions,withadvectionspeed u,withoutslowingdownorspeedingup.

UpwindforwardEulerwiththebellcurve

Forthefirstsetofplotsforthistestcase,weshallconsideranadvectionspeedof u = 1, andspatialandtemporalgridspacingof x = t = 0 1 units.Thebellcurveisdefinedby

wheretheparameter μ determineswherethebellisinitiallycentered,and σ determinesthe spreadofthecurve.WeshouldnotethatthisissimilartothedefinitionoftheGaussian distribution,exceptthatwedonothavethenormalizingfactorof 1

2πσ

Forthedemonstrationsthatwepresentinthischapter,weuse μ = 5,implyingthatthebell willinitiallybecenteredat x = 5,and σ = 0.5.TheinitialprofileispresentedinFig. 2.7.

FIGURE2.7 Plotofthetruebellcurveforthesituationwhen μ = 5 and σ = 0 5 at t = 0

GiventheinformationpresentedhereweknowthattheCourantnumberforthisconfigurationis C ≡ u t x = 1.Weareusingaspatialdomainof x ∈ [0, 10],andatemporaldomainof 0 ≤ t ≤ 10.Thismeansthatthereare101 x gridpointsand101 t pointsforthisconfiguration. Weareusingaperiodicityconditionastheboundaryconditionsforeitherendofthedomain,

2.Eulerianmodelingofadvectionproblems andwiththeadvectionspeedof 1,thisthenmeansthatthebellcurvecantakeonecomplete circuitofthedomaininthistime.

Weareabletoplotthetruesolutionateverytimestepasweknowthatitisgivenby

where n istheindexforthespecifictimestepthatthenumericalmodeisat.Thiswillenable ustoquantifythe error associatedwitheachnumericalscheme.

InFig. 2.8 wehaveplottedboththetrueandnumericalsolutionsfortheconfiguration shownatevery10timesteps.Weshouldpointoutthatthereisnoerrorinthisplot,simply since,when C = 1,theupwindforwardEulerapproximatesthecontinuousbellfunctionvery well.

FIGURE2.8 Plotofthetruebellcurve,inred,andthenumericalsolutionfromtheupwindforwardEulerscheme for tn = 10 t, 20 t,..., 90 t steps.

Wewillnowlookattheeffectsofchangingthespatialstepsize.InFig. 2.9 wehavepresentedasimilarplottothatinFig. 2.8,butnowwehavethesituationwhere C = 1 25,which ariseskeeping t = 0 1 buthavingincreasedtheresolutionofthe x directionto x = 0 08 Whilethischangemaynotseemlikeabigdeal,wecanseefromFig. 2.9 thatweareno longermaintainingthestructureofthebellcurves.Wecanclearlyseesomedampingofthe curvewithin20timesteps,andthat,bythetimewearriveat70timessteps(panel7),wecan clearlyseethatthesolutionisstartingtodisplaysomefeaturebehindthebellcurve,andby 90timesstepswecanseedistortionsbothbehindthewaveandalsotravelingfurtherbehind thecurve.Thiscouldindicatearelationshipbetween C , t ,and x .

FIGURE2.9 PlotoftheupwindforwardEulerschemeforthebellcurvewhentheCourantnumberis C = 1 25.

Thenextsetofplotsthatwepresentforthissituationwiththisnumericalschemearefor thecasewhenwehave C = 0 5.Herewearetryingtosavesomestoragespaceandalsospeed upthenumericalintegration.Wehavekept t = 0.1,butnow x = 0.2.Thefirstfeature tonoticeinFig. 2.10 isthattheupwindforwardEulerschemeshowwhatisreferredtoas damping ofthebellcurveastimegoeson;onapositivenote,wecanseethattheschemeisnot outofphase withthetruesolution.Therefore,byreducingthenumberofspatialgridpoints, wehaveactivatedanotherfeatureofthenumericalschemethatwasnotpresentwhenwe hada higherspatialresolution.

Asmentionedearlier,thisupwindforwardEulerissupposedtobeusedwhen u> 0,and wenowpresentaplotoftheaffectsofapplyingthisschemewhen u< 0.InFig. 2.11 wehave plottedtheresultsat10and20timesstepsofapplyingtheupwindforwardEulerscheme when u< 0.Weseethatafter10timessteps,thenumericalsolutionsisstartingtogooutof phasewiththetruesolution,andisalsointroducingan amplitudeerror,aswellasovershoots. However,only10timestepslater,weseethatthenumericalsolutionisnothinglikethetrue solutionandisstartingtoblowup.Therewasnopointinproducingoneofthesetsof9plots forthiscaseastheschemebecamecompletelyunstableafterafewtimesteps.

Wenowmoveontothediscontinuouscaseofthestepfunctionwiththesamesetupused heretoseetheimpactthatdiscontinuitieshaveontheperformanceoftheupwindforward Eulerscheme.

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.