Coupled FiniteVolume Simulations at OneDegree Resolution

1
Coupled Finite-Volume Simulations at One-Degree
Resolution

• Art Mirin and Govindasamy Bala
• Lawrence Livermore National Laboratory

2
Outline of presentation

• Introduction to finite-volume dynamical core
• Parallelization, unification and simulation
  efforts
• Diagnosis of seasonal ice buildup
• Acknowledgments
• Dani Bundy, Brian Eaton
• Phil Rasch, Mariana Vertenstein
• Work performed under the auspices of the U.S.
  Department of Energy by University of California
  Lawrence Livermore National Laboratory under
  contract No. W-7405-Eng-48. This is LLNL Report
  UCRL-PRES-219702.

3
Attributes of finite volume dynamical core

• Developed by S.J. Lin and R. Rood of NASA GSFC
• Terrain-following floating Lagrangian
  control-volume vertical coordinate
• Two-dimensional conservative semi-Lagrangian
  transport within a control-volume
• Monotonicity-preserving mass-, momentum-, and
  total energy-conserving mapping algorithm to
  Eulerian reference coordinate

4
Terrain-following vertical coordinate
Lagrangian coordinate evolves according to
vertical transport
5
Attributes of finite volume dycore, cont.

• Dynamics subcycled with respect to remapping
• Multiple, staggered horizontal Eulerian grids
  invoked
• Semi-Lagrangian transport conserves key physical
  quantities
• Semi-Lagrangian algorithm circumvents polar
  singularity
• Fast timescale contains geopotential calculation
  that couples vertical levels through indefinite
  integrals

6
The FV dycore uses a hybrid parallel model

• Multi-two-dimensional domain decomposition
• latitude-vertical (yz) 2-D domain decomposition
  for most of dynamics
• longitude-latitude (xy) 2-D domain decomposition
  for remapping and geopotential calculation
• Shared memory parallelism (OpenMP) largely in
  vertical, but also in latitude
• Decompositions connected by transposes using
  Pilgrim and Mod_comm libraries (NASA/GSFC)
• MPI derived types
• MPI-2 one-sided communication

7
FV unification effort

• FV originally implemented in NASA FVGCM (GEOS4)
• FV then implemented in CAM
• NASA GEOS5 contains FV as ESMF module
• FV on cubed sphere under development
• Unification
• unification of CAM and GEOS5 (ESMF) versions
  nearly complete outside world will see only
  lat-lon decomposition
• S-J Lins FVGCM improvements will be merged into
  CAM/GEOS5 version
• common repository at GFDL
• eventual merge with cubed sphere version

8
Coupled FV usage and study at LLNL

• CCSM3.0 ported to LLNL thunder IA64 Linux cluster
• platform-relevant configuration script files
• perturbation growth test for CAM
• T85 validation test
• Instituted support for 1x1.25_gx1v3 mesh
• created land surface file
• coupler mapping files
• historical run
• 10 simulated years/day using 118 (4-processor)
  nodes
• Detection and Attribution of Regional Climate
  Change
• targeting western United States
• coupled FV at 1x1.25 1000 year spin-up
• Investigating known sea ice problem with coupled
  FV

9
The ice problem in FV 2x2.5
Excessive ice south of Greenland
10
Sensitivity tests at 2x2.5 to diagnose sea ice
issue

• We have performed the following sensitivity
  tests
• albedo reduction
• increase/decrease wind stress over ocean
• increase/decrease wind drag on ice
• eliminate ice dynamics
• upwind ice advection algorithm
• Removing ice dynamics largely eliminates the ice
  problem
• Other variations have little effect

11
Sea ice T85, FV 2x2.5, FV 2x2.5 without ice
dynamics
Eulerian top left FV 2x2.5 top right FV 2x2.5
without ice dynamics bottom left
12
Discussion of FV 1x1.25 simulation

• Modified the following parameters based on CAM
  stand-alone tuning exercise at 1x1.25 (Bala in
  correspondence with Hack)
• low cloud threshold (rhminl) 0.88 gt 0.87
• cold ice autoconversion (icritc) 9.5e-6 gt
  18.0e-6
• Initiated 21-year CCSM run that showed SST drift
  of -1.57
• Increased rhminl from 0.87 to 0.91 and initiated
  new 20-year run
• ice buildup was as bad as with 2x2.5 case

13
Discussion of FV 1x1.25 simulation, cont.

• Lowered ice and snow albedos
• albicev0.68 (vs 0.73)
• albicei0.30 (vs 0.33)
• albsnowv0.90 (vs 0.96)
• albsnowi0.62 (vs 0.68)
• Lower albedos improve ice in winter but worsen
  summer ice
• seasonal cycle has larger amplitude with FV
• Note all runs use improved remap topography

14
Winter ice fraction FV 1x1.25 vs T85 (left),
1x1.25 vs 2x2.5 (right)
15
Winter ice thickness FV 1x1.25 vs T85 (left),
1x1.25 vs 2x2.5 (right)
16
Summer ice fraction FV 1x1.25 vs T85 (left),
1x1.25 vs 2x2.5 (right)
Missing plot
17
Summer ice thickness FV 1x1.25 vs T85 (left),
1x1.25 vs 2x2.5 (right)
18
Seasonal cycle FV 1x1.25 vs T85
19
Discussion of FV ice issue

• Ice buildup near Greenland is seasonal
• Decreasing ice/snow albedos improves winter ice
  at expense of summer ice seasonal amplitude
  insensitive to albedos
• Ice issue with coupled FV is not necessarily the
  fault of the FV dycore
• nonlinear coupling among components
• some cite errors in surface wind stress
• some cite errors in ocean surface current
• Case at 1x1.25 with original albedos being
  continued further
• Next steps???