adaptive meshes on the sphere: cubed-spheres versus latitude-longitude grids

36
Adaptive Meshes on the Sphere: Cubed-Spheres versus Latitude-Longitude Grids Christiane Jablonowski University of Michigan Dec/8/2006

Upload: isaac-rogers

Post on 30-Dec-2015

22 views

Category:

Documents


1 download

DESCRIPTION

Adaptive Meshes on the Sphere: Cubed-Spheres versus Latitude-Longitude Grids. Christiane Jablonowski University of Michigan Dec/8/2006. Acknowledgments. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Adaptive Meshes on the Sphere: Cubed-Spheres versus

Latitude-Longitude Grids

Christiane JablonowskiUniversity of Michigan

Dec/8/2006

Page 2: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

AcknowledgmentsAcknowledgments The AMR comparison is based on a joint paper with The AMR comparison is based on a joint paper with

Amik St-Cyr and collaborators from NCAR, submitted to Amik St-Cyr and collaborators from NCAR, submitted to Monthly Weather Review in November 2006Monthly Weather Review in November 2006

The AMR Spectral Element Model was mainly developed by The AMR Spectral Element Model was mainly developed by Amik St-Cyr, John Dennis & Steve Thomas (NCAR)Amik St-Cyr, John Dennis & Steve Thomas (NCAR)

The AMR FV model is documented inThe AMR FV model is documented inJablonowski (2004), Jablonowski et al. (2004, 2006)Jablonowski (2004), Jablonowski et al. (2004, 2006)

Contributors to the AMR FV model areContributors to the AMR FV model areMichael Herzog (GFDL) & Joyce Penner (UM)Michael Herzog (GFDL) & Joyce Penner (UM)Robert Oehmke (NCAR) & Quentin Stout (UM)Robert Oehmke (NCAR) & Quentin Stout (UM)Bram van Leer (UM) & Ken Powell (UM)Bram van Leer (UM) & Ken Powell (UM)

The AMR comparison is based on a joint paper with The AMR comparison is based on a joint paper with Amik St-Cyr and collaborators from NCAR, submitted to Amik St-Cyr and collaborators from NCAR, submitted to Monthly Weather Review in November 2006Monthly Weather Review in November 2006

The AMR Spectral Element Model was mainly developed by The AMR Spectral Element Model was mainly developed by Amik St-Cyr, John Dennis & Steve Thomas (NCAR)Amik St-Cyr, John Dennis & Steve Thomas (NCAR)

The AMR FV model is documented inThe AMR FV model is documented inJablonowski (2004), Jablonowski et al. (2004, 2006)Jablonowski (2004), Jablonowski et al. (2004, 2006)

Contributors to the AMR FV model areContributors to the AMR FV model areMichael Herzog (GFDL) & Joyce Penner (UM)Michael Herzog (GFDL) & Joyce Penner (UM)Robert Oehmke (NCAR) & Quentin Stout (UM)Robert Oehmke (NCAR) & Quentin Stout (UM)Bram van Leer (UM) & Ken Powell (UM)Bram van Leer (UM) & Ken Powell (UM)

Page 3: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

OverviewOverview

Computational Grids on the SphereComputational Grids on the Sphere

Adaptive mesh refinement (AMR) Adaptive mesh refinement (AMR)

techniquestechniques Why are we interested in variable Why are we interested in variable resolutions / resolutions / multi-scales?multi-scales?

Overview of two AMR shallow water modelsOverview of two AMR shallow water models

Finite volume (FV) modelFinite volume (FV) model

Spectral element model (SEM)Spectral element model (SEM)

Results: Static and dynamic adaptationsResults: Static and dynamic adaptations

2D shallow water experiments 2D shallow water experiments

Conclusions and OutlookConclusions and Outlook

Computational Grids on the SphereComputational Grids on the Sphere

Adaptive mesh refinement (AMR) Adaptive mesh refinement (AMR)

techniquestechniques Why are we interested in variable Why are we interested in variable resolutions / resolutions / multi-scales?multi-scales?

Overview of two AMR shallow water modelsOverview of two AMR shallow water models

Finite volume (FV) modelFinite volume (FV) model

Spectral element model (SEM)Spectral element model (SEM)

Results: Static and dynamic adaptationsResults: Static and dynamic adaptations

2D shallow water experiments 2D shallow water experiments

Conclusions and OutlookConclusions and Outlook

Page 4: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Latitude-Longitude GridLatitude-Longitude Grid

Popular choice Meridians converge:polar filters or/andtime steps

Orthogonal

Popular choice Meridians converge:polar filters or/andtime steps

Orthogonal

Page 5: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Platonic solids - Regular grid structures

Platonic solids - Regular grid structures

Platonic solids can be enclosed in a sphere

Platonic solids can be enclosed in a sphere

Page 6: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Cubed Sphere Geometry

Courtesy of Ram Nair (NCAR)

Advection of a cosine bell around the sphere (12 days)at a 45o angle

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.

Page 7: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Adaptive Mesh Refinements (AMR)

Latitude-Longitude grid:Model FV

Cubed-sphere grid:Model SEM

Page 8: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

SEM: Grid Points within Spectral Elements

Circles: Gauss-Lobatto-Legendre (GLL) points for vectors

Squares:Gauss-Lobatto (GL) points forscalars

Elements are split in case of refinements

Page 9: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

FV: Block-Structured Adaptive Mesh Refinement Strategy

Self-similar blocks with 3 ghost cells in x & y direction

Page 10: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Other AMR Grids

Source: DWD

Model ICON

Icosahedral grid with nested high-resolution regions

under development at the German WeatherService (DWD) and MPI, Hamburg

Page 11: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Features of Interest in a Multi-Scale Regime

Hurricane Ivan

Hurricane Frances

September/5/2004

Page 12: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

High Resolution: Multi-Scale Interactions

W. Ohfuchi, The Earth Simulator Center, Japan

10 km resolution

Page 13: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

AMR Transport of a Slotted Cylinder

AMR Transport of a Slotted Cylinder

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.

Model FV

Page 14: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Transport of a Slotted CylinderTransport of a Slotted Cylinder

Page 15: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Transport of a Slotted CylinderTransport of a Slotted Cylinder

Page 16: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Transport of a Slotted CylinderTransport of a Slotted Cylinder

Page 17: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Transport of a Slotted CylinderTransport of a Slotted Cylinder

Page 18: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Transport of a Slotted CylinderTransport of a Slotted Cylinder

• Slotted cylinder is reliably detected and trackedSlotted cylinder is reliably detected and tracked

Page 19: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Shallow Water Equations

∂ r

v h

∂t+ (ζ + f )

r k ×

r v h +

r ∇(K −νD + g(h + hs)) = 0

∂h

∂t+

r ∇ • h

r v ( ) = 0

Momentum equation in vector-invariant form

Continuity equation

vh horizontal velocity vector relative vorticityf Coriolis parameterK= 0.5*(u2 + v2) kinetic energyD horizontal divergence, damping coefficienth free surface height, hs height of the orographyg gravitational acceleration

Page 20: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Developed by Lin and Rood (1996), Lin and Rood (1997) 3D version available (Lin 2004), built upon the SW model:

hydrostatic dynamical core used for climate and weather predictions

Currently part of NCAR’s, NASA’s and GFDL’s General Circulation Models

Numerics: Finite volume approach– conservative and monotonic transport scheme

– upwind biased 1D fluxes, operator splitting

– van Leer second order scheme for time-averaged numerical fluxes

– PPM third order scheme (piecewise parabolic method)for prognostic variables

– Staggered grid (Arakawa D-grid)

– Orthogonal Latitude-Longitude computational grid

Finite Volume (FV) Shallow Water Model

Page 21: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Documented in Thomas and Loft (2002), St-Cyr and Thomas (2005), St-Cyr et al. (2006) 3D version available Experimental tests within NCAR’s Climate Modeling

Software Framework

Numerics: Spectral Elements– Non-conservative and non-monotonic

– Allows high-order numerical method

– Spectral convergence for smooth flows

– GLL and GL collocation points

– Non-orthogonal cubed-sphere computational grid

Spectral Element (SEM) Shallow Water Model

Page 22: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Overview of the AMR comparisonOverview of the AMR comparison

2D shallow water tests: (Williamson et al., JCP 2D shallow water tests: (Williamson et al., JCP

1992)1992)

Dynamic refinements for pure advection Dynamic refinements for pure advection

experimentsexperiments

Cosine bell advection test (test case 1)Cosine bell advection test (test case 1)

Static refinements in regions of interest Static refinements in regions of interest

(test case 2)(test case 2)

Dynamic refinements and refinement criteria: Dynamic refinements and refinement criteria:

Flow over a mountain (test case 5) Flow over a mountain (test case 5)

Rossby-Haurwitz wave with static refinements Rossby-Haurwitz wave with static refinements

(test case 6) (test case 6)

2D shallow water tests: (Williamson et al., JCP 2D shallow water tests: (Williamson et al., JCP

1992)1992)

Dynamic refinements for pure advection Dynamic refinements for pure advection

experimentsexperiments

Cosine bell advection test (test case 1)Cosine bell advection test (test case 1)

Static refinements in regions of interest Static refinements in regions of interest

(test case 2)(test case 2)

Dynamic refinements and refinement criteria: Dynamic refinements and refinement criteria:

Flow over a mountain (test case 5) Flow over a mountain (test case 5)

Rossby-Haurwitz wave with static refinements Rossby-Haurwitz wave with static refinements

(test case 6) (test case 6)

Page 23: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Snapshots: Advection of a Cosine Bell

Snapshots: Advection of a Cosine Bell

Page 24: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Snapshots: Advection of a Cosine Bell

Snapshots: Advection of a Cosine Bell

Page 25: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Error norms: Cosine Bell Advection

Error norms: Cosine Bell Advection

Days Days

Rotation angle = 45:Errors in SEM are lower than in FV

Page 26: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Error norms after 12 daysError norms after 12 days

Rotation angle = 0

SEM produces undershoots

Errors arecomparable

Page 27: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Snapshots: Cosine Bell at day 3Snapshots: Cosine Bell at day 3

North-polar stereographic projection at day 3 for a = 90

Convergence of blocks in FV

Page 28: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

2D Static adaptations2D Static adaptations

• Smooth flow in regimes with strong gradientsSmooth flow in regimes with strong gradients

FV model:

Test case 2, = 45

Page 29: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Error norms: Test case 2Error norms: Test case 2

Days Days

Rotation angle = 45:Errors in FV partly due to errors at AMR interfaces

Page 30: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

2D Dynamic adaptations in FV

Vorticity-basedadaptation criterion

QuickTime™ and aYUV420 codec decompressor

are needed to see this picture.

2D shallowwater test #5:15-day run

Page 31: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Snapshots: Flow over a mountainSnapshots: Flow over a mountain

Longitude Longitude

Geopotential height field (test case 5)

Page 32: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Snapshots: Flow over a mountainSnapshots: Flow over a mountainGeopotential height field (test case 5)

SEM FV

Page 33: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Error norms: Test case 5Error norms: Test case 5

Hours Hours

Errors in SEM converge quicker to the NCAR reference solution

Page 34: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Snapshots: Rossby-Haurwitz WaveSnapshots: Rossby-Haurwitz Wave

Geopotential height field (test case 6) at day 7

Smooth flow through static refinement regions

Page 35: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Alternative AMR: Unstructured Triangular Grid

Alternative AMR: Unstructured Triangular Grid

QuickTime™ and aBMP decompressor

are needed to see this picture.

Hurricane Floyd(1999)

Colors indicate the wind speed

OMEGA model

Courtesy ofA. Sarma (SAIC, NC, USA)

Page 36: Adaptive Meshes on the Sphere:  Cubed-Spheres versus  Latitude-Longitude Grids

Conclusions & OutlookConclusions & Outlook

Both grids, cubed-sphere meshes and latitude-Both grids, cubed-sphere meshes and latitude-

longitude grids, are options for AMR techniqueslongitude grids, are options for AMR techniques

SEM model shows lower error norms in comparison to SEM model shows lower error norms in comparison to

FV:FV:

Mainly due to high-order numerical methodMainly due to high-order numerical method

Partly due to different AMR approach that does Partly due to different AMR approach that does

not need interpolations of ‘ghost cells’ in not need interpolations of ‘ghost cells’ in

blocksblocks

But: SEM is non-monotonic and non-conservativeBut: SEM is non-monotonic and non-conservative Cubed-sphere grid has clear advantages:Cubed-sphere grid has clear advantages:

No convergence of the meridians, no polar filtersNo convergence of the meridians, no polar filters But, GLL and GL points for numerical method in SEM But, GLL and GL points for numerical method in SEM

are clustered along boundaries of spectral elementsare clustered along boundaries of spectral elements

Future interests: Finite-volume AMR method on a Future interests: Finite-volume AMR method on a

cubed-sphere gridcubed-sphere grid

Both grids, cubed-sphere meshes and latitude-Both grids, cubed-sphere meshes and latitude-

longitude grids, are options for AMR techniqueslongitude grids, are options for AMR techniques

SEM model shows lower error norms in comparison to SEM model shows lower error norms in comparison to

FV:FV:

Mainly due to high-order numerical methodMainly due to high-order numerical method

Partly due to different AMR approach that does Partly due to different AMR approach that does

not need interpolations of ‘ghost cells’ in not need interpolations of ‘ghost cells’ in

blocksblocks

But: SEM is non-monotonic and non-conservativeBut: SEM is non-monotonic and non-conservative Cubed-sphere grid has clear advantages:Cubed-sphere grid has clear advantages:

No convergence of the meridians, no polar filtersNo convergence of the meridians, no polar filters But, GLL and GL points for numerical method in SEM But, GLL and GL points for numerical method in SEM

are clustered along boundaries of spectral elementsare clustered along boundaries of spectral elements

Future interests: Finite-volume AMR method on a Future interests: Finite-volume AMR method on a

cubed-sphere gridcubed-sphere grid