motivations, successes, and problems, with the lagrangian
TRANSCRIPT
Motivations, successes, and problems, with the Lagrangian-remap dynamical core in MOM6
AlistairAdcroftPrincetonUniversity/NOAA-GFDL
• Wintonetal.,1998:z-coordinatemodelsexcessmixinginoverflows– Numerouscoordinate/mixingstudies:Griffies etal.,2000;Chassignetetal.,2003;Leggetal.,2006;Burchard&Rennau 2008;Megann etal.,2010;…
• Dunneetal.,2012:comparedESM2MandESM2G,both1° resolution– ESM2M:z-coordinate,shallowAMOC– ESM2G:isopycnal layermodel,deepAMOCwithoverflows
2
Z-coordinatesv.isopycnal coordinates
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
Dunneetal.,2012
ESM2M
ESM2G
Talley,2003Ganachaud &Wunsch, 2003
• Delworth etal.,2012,coupledmodelseries(CM2.1,CM2.5,CM2.6):– 50kmatmosphere– 1°,¼° and0.1° ocean– 1° alonehasSGSeddyparameterization(Gent-McWilliams)
• Griffies etal.,2015,diagnosedhowtransienteddiesina0.1° oceantransportheatupwards– CM2.6leastheatuptakeofCM2.xseries– ArguethatCM2.5hasmostheatuptakeduetoweakeddiesandnoGM
Roleofmesoscaleeddies
Evolutionofhorizontallyaveragedpotential temperature,θ (°C).
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 3
CM2.5 CM2.6¼° ocean 1/10° oceanCM2-1deg
1.1°C 1.3°C 0.8°C
1° ocean
• Measureworkdonebymixinginspindownexperiments– Turningoffexplicitmixingleavesonlynumericalmixing*
• Ilicak etal.,2012,argued¼° z-coordinatemodelhadnumericalmixingaslargeas“real”mixing
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 4
Measuringspuriousmixing
[Wm-2]
Ilicak etal.,2012
• ReducingspuriousmixingmotivatedMOM6development– Generalcoordinatetoavoidlimitationsofpotentialdensity(usedbyisopycnal layermodels)
• MOM6broadlyreproducedpriorsolutionsusingz-coordinates
• Adoptionofhybridz-densitycoordinatesreducedheatuptake– success?
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 5
MOM6andOM4
OM41/4° z-coordinates
OM41/4° hybrid-coordinates
Adcroftetal.,2019
• HydrostaticPrimitiveEquations(forocean)ingeneralcoordinates𝑟 = 𝑟(𝑥, 𝑦, 𝑧, 𝑡) :
Generalcoordinates(Boussinesq)
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
𝑧+ =𝜕𝑧𝜕𝑟
Starr,1945;Kasahara,1974;…
6
• Integratingbetweensurfacesofconstant𝑟 converts𝑧+ → ℎ = ∫ 𝑧+𝑑𝑟 and𝜕+ → 𝛿+
• Choosing�̇� = 0 followstheverticalmotion→VerticallyLagrangian– removesexplicitverticaltransports→noverticalCFL
7
Layerintegratedgeneralcoordinateequations
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
Lagrangianremapalgorithm
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 8
Reconstruction
𝜃†
𝑧5
𝜃5
Dynamics+“physics”
ℎ†ℎ5
�̇� = 𝟎
Average
𝑧589
𝜃589
Gridgeneration
Eulerian,ALEandLRMalgorithmsside-by-side
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 9
𝑣;589 = 𝑣;5 + Δ𝑡 − 9?@𝛻B𝑝 +⋯
𝜕E𝑤 = −𝛻 G 𝑣;589
𝜕E𝑝 = −𝑔𝜌 𝑧, 𝑆5, 𝜃5
𝜃589 = 𝜃5 − Δ𝑡 𝛻 G 𝑣;589𝜃5 +𝜕E 𝑤𝜃5 + ⋯
𝑣;589 = 𝑣;5 + Δ𝑡 − 9?@𝛻B𝑝 +⋯
𝛿K(𝑤∗ + 𝑤M) = −𝛻 G ℎ5𝑣;589
𝜕E𝑝 = −𝑔𝜌 𝑧, 𝑆5, 𝜃5
ℎ589𝜃589 = ℎ5𝜃5
−Δ𝑡 𝛻 G ℎ5𝑣;589𝜃5
+𝛿K 𝑤∗𝜃5 + ⋯
ℎ† = ℎ5 − Δ𝑡𝛻 G (ℎ5𝑣;†)
𝜕E𝑝 = −𝑔𝜌 𝑧, 𝑆5, 𝜃5
ℎ†𝜃† = ℎ5𝜃5 − Δ𝑡 𝛻 G ℎ5𝑣;†𝜃5+⋯
ℎ589 ← 𝛿K𝑍 𝑧†
𝜃589 = 𝜃† 𝑍(𝑧†)
𝑣;† = 𝑣;5 + Δ𝑡 − 9?@𝛻B𝑝 +⋯
Eulerian A.L.E. Langrangian-remap
wΔ𝑡Δ𝑧 < 1
𝑤∗Δ𝑡Δ𝑧 < 1
ℎ589 = ℎ5 + Δ𝑡𝛿K(𝑤M)
𝑤∗ = 𝑤− 𝑤M
introduces grid motion 𝑤M
Grid generation
Remap
SeeGriffies etal.,JAMES2020
Assumingexplicit-in-timetransport
Assumingexplicit-in-timetransport
Sub-cyclingwithLagrangianverticaldynamics
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
𝑣;S89 = 𝑣;S + 9TUV?@−𝛻W𝑝 − 𝜌𝛻WΦ +⋯
ℎS89 = ℎS − 9TΔ𝑡𝛻W G (ℎ
S𝑣;S89)
𝛿K𝑝 = −𝜌 𝑧, 𝑆5, 𝜃5 𝛿KΦ
ℎ∗𝐶∗ = ℎ5𝐶5 −𝑀Δ𝑡 𝛻 G [ ℎS𝑣;S89𝐶5T
S\9
ℎ589 ← 𝛿K𝑍 ℎ∗ ; 𝐶589 = 𝐶∗ 𝑍(ℎ∗) ; …
×𝑀
×𝑁
Internal gravity waves
𝚫𝒕𝒄𝒊𝒈𝚫𝒙 < 𝟏
𝐌𝚫𝒕𝒖𝒉𝚫𝒙 < 𝟏
𝑈l89 = 𝑈l + 9mΔ𝑡 −𝛻𝜂
l + ⋯𝜂l89 = 𝜂S − 9
mΔ𝑡𝛻W G (𝐻𝑈l89)
Barotropic gravity waves
×𝐿 𝚫𝒕√𝒈𝑯𝑳𝚫𝒙 < 𝟏
Tracers
Vertical remap
10
Interpolationforgridgeneration• Accurate• Continuous (singlevalued)– resultingprofilesaregenerallynotconservative
Reconstructionforremapping• Accurate• Conservative– profilesoftenbecomediscontinuoustobeconservative
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 11
Consistentgrid-generationandremapping
Areas not equal
Discontinuity between layers
Second order interpolation (i.e. linear)
Third order reconstruction
Piecewise Parabolic (PPM)
Discrepancies reduce with higher order
Remappingimplementations
𝑧589
𝜃K589
𝜃†
𝑧† 𝑧589
𝜃K589
𝜃†
𝑧†
Remappingvia“advectiveflux” Remappingby“projection”
ℎK589𝜃K589 = ℎKt𝜃K
t + u 𝜃t𝑑𝑧Evwxy
zwx
Evwxy
{−u 𝜃t𝑑𝑧
Ev|xy
zwx
Ev|xy
{ℎK589𝜃K589 = u 𝜃t𝑑𝑧
Ev|xy
zwx
Evwxy
zwx
AccurateconservationbutlocallyinaccurateforCFL>1 Inaccuratetotalconservation butaccuratelocally
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 12
Remapby“sublayers”• Dividesupersetofold/newgridsintosublayers(2N-1)
• Integrateprofileforeachsub-layer– Bydefinitionusesonlyonesourcelayerprofile
• Replacelargest sub-layerwith
• Sumsub-layerstonewlayer
Accurateconservationwhenremapping
𝑧589
𝜃K589
𝜃†
𝑧†
ℎ}~𝜃}~ = u 𝜃t𝑑𝑧E�|xy
�
E�wxy
�ℎK589𝜃K589 = [ℎ}~𝜃}~
S
}\l
ℎ}~𝜃}~ = ℎKt𝜃K
t − [ 1− 𝛿}�} ℎ}�~ 𝜃}�~S
}�\l(insight fromHallberg)
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 13
• PLM:twodegreesoffreedom– Cellmean+slope
• PPM:threedegreesoffreedom– Verywidelyused– Cellmean+twoedgevalues
• PQM:fivedegreesoffreedom– Cellmean+twoedgevalues+twoedgeslopes
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 14
Piecewise*Methods (*=C,L,PorQ)PLM
PPM
PQM
InspiredbyDaru &Tenaud, JCP2004– introduced OSi i=1..7
Successiveschemesprovidemoreflexibilitytorepresentstructures→moreaccurate
(White&Adcroft, JCP2008)
IllustratingmethodsandspuriousdiffusionLayeredIsopycnal
PCM
PPMih4
PQMih6
zPLM-PCM
PPMih4-PPMih4
PQMih4-PQMih4
ρ
Inappropriate interpolator leads to collapse of stratification
Diffusive behavior reduces with increasing order of remapping but
does not vanishMinimal smoothing
of interface
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 15
• Denseflowdownaslope– Hydrostaticandadiabatic
• Layeredisopycnal solutionistrulyadiabatic
• z*andterrain-followingbothhavemodifiedwatermasses
• LRM+densitycoordinates(foundbyinterpolation)behavesverysimilarlytolayeredmodel
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 16
Overflows
• Lateralprocesses(incl.transport)actingoninclinedisopycnals
• Alongisopycnal processesinnon-isopycnal coordinates
• Dia-surfacetransportinvertical– Minimizingdia-surfacemotionreducesnumericaldiffusion
• Choosingacoordinateclosetoisopycnal isdesirabletominimizespuriousdiffusion
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 17
Sourcesofspuriousmixing
• Alreadyknowthatpureisopycnal coordinateshavelimitations
• Bleck,2002,introducedhybridz-rhocoordinateinHYCOM
• z-tilde:Leclair &Madec,2011;Petersenetal.,2015
• adaptive:Hofmeisteretal.,2010;Gräwe etal.,2015;Gibson(thesis)2019
• OM4used“HYCOM1”whichisasimpleinterpretationofactualHYCOMhybridcoordinate– OngoingworkwithWallcraft andChassignet toimprovegridgeneration(c.f.AMOCresults)
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 18
Hybridandothercoordinates
• Comparingz*tohybridcoordinateprovidesmagnitudeofspuriousnumericalmixing
• Reducingresolutionfrom¼° to½° (eddyingtonon-eddying)inhibitsresolvedre-stratification– ½° requireseddyparameterization
• Refiningresolutionfrom¼° to⅛°reducesheatuptake:– moreefficienteddies– and/orlessnumericalmixing?
• Thisreallytellsuswe(inadvertently)managedtoperfectlycompensateforweakeddiesat¼° resolution
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 19
Balancingheattransferbyeddiesandmixing
More efficient eddies?
Parameterized eddies
Spurious mixing
Zanna et al. 2019
• MainconcernisOM4withhybridcoordinatestillhasashallowAMOC– Isthismixinginoverflows?
– ItturnsoutwedidhavetoomuchparameterizedmixingbutreducingthathashadnoaffectondepthofAMOC.Infactnothingwe’vetriedsofarseemsto!
20
OM4AMOC
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
• Motivatedbyresultswithisopycnal layeredmodelswebuiltMOM6,capableofusingarbitrarilygeneralcoordinatesfollowingHYCOM’spioneeringhybridcoordinate
• LagrangianRemapAlgorithmhassignificantadvantagesforEarthSystemModelswithmanyconstituents
• High-ordernumericalmethods+LRMdeliverlowlevelsofspuriousmixing
• InOM4,puttingeverythingtogether,amysteryremainsastowhyAMOCisshallowerthaninESM2G
21
Wrapup
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
• Potentialenergy
• Availablepotentialenergy(APE)
• ρ*istheadiabaticallyre-arrangedstatewithminimalpotentialenergy
• RPEcanonlybechangedbydiapycnalmixing– Mixingraisescenterofmass
Quantifyingspuriousmixingusingenergetics
Wintersetal.,JFM1995Ilicak etal.,OM2012
ρ+
ρ-
ρ+
ρ-
½(ρ- +ρ+)
APE
½(ρ- +ρ+)ρ+
ρ-½(ρ- +ρ+)
APE=PE−RPE
RPE=g�ρ∗zdV
PE=g�ρzdV
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft
• OM4istheice-oceancomponentofGFDL’slatestcoupledmodelCM4(Heldetal.,2019)
• OM4configuration– Identicalsetup/parameterstoCM4– Developedalmostexclusivelyincoupledmode• UncoupledOM4
– Nominallyeddy-permitting¼°horizontalresolution• Non-eddying½° witheddyparameterization(GM+EKEscheme)
Ingredients:• MOM6
– usinghybridz-ρ verticalcoordinates• ePBL
– Reichl andHallberg,2019• Scale-awareMLErestratification
– Fox-Kemperetal.,2011• Shear-dependentmixing
– Jacksonetal.,2008• Internal-wavedrivenmixing
– Melet etal.,2012• BBL
– Leggetal.,2006
Adcroftetal.,JAMES 2019
ECMWFAnnualSeminarLagrangian-remapdynamiccoreofMOM6- Adcroft 23
OM4
c.f.HYCOM,Bleck 2002