Multi-Resolution Climate Modeling: Principle Investigators

Multi-Resolution Climate
Modeling
Principle Investigators
F. Baer (UMCP)
J. Tribbia(NCAR)
A. Fournier (UMCP)
Collaborators
Mark Taylor (Los Alamos)
Steve Thomas (NCAR)
Rich Loft (NCAR)
M. Fox-Rabinovitz (Fac. Assoc.)
Goal of the Project
• To develop a climate model

methodology which allows for
seamless concurrent integration of the
planetary scales together with regional
scales;
• To optimize the capabilities of
currently available computer hardware
(parallel processors) with this model;
• To incorporate the best features of
current state-of the-art climate models.
2
Rationale
• Variable resolution in a model helps
define climate;
• Nonlinear effects introduced by
regional scales must be incorporated
into a climate;
• Smaller scale effects often grow on
shorter time scales;
• Identification and prediction of
regional climate should help in
understanding the evolution of the
global climate.
• Integrations must be sped up to
perform all computations needed for
solving the climate modeling problem.
3
The Method and Model
SEAM: Spectral Element

Atmospheric Model
• A global model offering great
flexibility and advantages in:
– Using geometric properties of finite
element methods;
– Incorporating local mesh refinement and
regional detail;
– Utilization of parallel processing;
– Maintaining the accuracy of spectral
models;
– Computational efficiency;
– Having no pole problem.
4
Plans for the Model
• Use as the dynamical core with

a forced model like the CCM4;
• Consider the finite-volume
technique;
• Use semi-implicit integration
(currently under study);
• Consider semi-Lagrange
integration;
• Create an option for linking to
an OGCM.
5
Model Domain
• Tile spherical surface with arbitrary

number and size of rectangular
elements;
– Inscribe a polyhedron with rectangular
faces inside sphere,
– Map surface of polyhedron to surface of
sphere with a gnomonic projection,
– Use the cube (most elementary
polyhedron),
– Subdivide each of the six faces of the
cube as desired.
– Can use Local Mesh Refinement (LMR)
as desired.
6
Cube Subdivision 1
Uniform Resolution Rectangles

An example of Local Mesh-refinement
on the globe.
Application of the method
Use the shallow water equations as an example:

V 1 
   f k  V -  V  V +gH 
t 2 
h
   hV
t
H = h +hs   k    V f  2sin 
 Generate an Integral form of the equations;

 Multiply by a global test function () and
integrate over the entire domain (spherical
surface).
V  1 
A t dA   A   f k  V 2 V  V +gH  dA
h
A t dA   A   hVdA
spectral element discretization
• Within each rectangle, use Legendre cardinal
functions as basis functions ( k, j ) in each direction.
 k, j  k (x) j (y)
h(x, y)   hk , j k (x) j (y)
k, j
V(x,y)   Vk, j k (x) j (y)

k, j
 k (x i )  Legendre Cardinal functions = k ,i
xi  Gauss - Lobatto Points

spectral element discretization
•The Gauss- Lobatto quadrature points

distribute as follows on the local rectangles:
•Within each rectangle, estimate the integral

equations by Gauss- Lobatto quadrature .
Global Test Functions
•Simple combinations of Legendre cardinal functions;

•One global test function for each grid point.
Element interior points
Element boundary points

Example of the computational system
The system of integral equations,

h
A t dA     hVdA
A
Reduces to:
A set for element interior points:

h
J (x i )  J  hV
t
And a set for element boundary points:
h
J  J  t xi    hVJ   hVJ 
   
where J is the area integration weight.

In Summary
• Tile the sphere with rectangles of

arbitrary number and size;
• Represent the prediction equations
in integral form;
• Use Gauss-Lobatto quadrature for
integration;
• Use Legendre cardinal functions for
the basis functions;
• Use test functions based on the
Legendre cardinal functions.
These choices result in an extremely

simple finite element method with a
diagonal mass matrix.
14
Does the model work?
•Tests with the shallow water

equations:
–With test suite (Williamson et al.,
1992);
–Efficiency;
–With local mesh refinement;
–With semi-implicit integration;
–With high resolution turbulence
problem;
–With vortex studies.
•Tests with a 3-D PE dynamical
core and Held-Suarez forcing.
15
Shallow water test case 7
•ICs taken from 500 mb observed data;
•Compare spectral element model to NCAR
spectral, CSU twisted icosohedral grid, and A-L
grid point models.
•L2 is normalized error between computed and
true height field at end of run.
Efficiency
•Curves show increasing number of elements
(M) for fixed spectral degree (N);
•Increasing N is more costly than increasing M;
•N = 8 appears to be the most efficient index.
•Select M for desired resolution.
488x8x8
Mesh
Refine
1784x8x8
Global
6936x8x8
•Test case 5
•Ht. field 150x8x8
210x8x8
Local
310x8x8
Mesh Refine 488x8x8
Zoom
1784x8x8
Global
6936x8x8
150x8x8
210x8x8
Local
310x8x8
•Errors comparing the explicit and semi-
implicit integrations of SEAM.
Semi-implicit
(Thomas and Loft)
Explicit
Linf
(Taylor)
•Ratio of semi-implicit to explicit

integrations of SEAM with the shallow
water equations for various truncations:
C refers to number of elements;
N refers to spectral degree.
•Note that semi-implicit is somewhat faster as
well as more accurate.
Thomas and Loft

Shallow water equations / Jupiter
• A study of decaying turbulence with high

resolution using SEAM;
• Equivalent depth on Jupiter = 20000 m;
• Use very weak dissipation;
• T170, T360, T533, T1033 runs on a CRAY T3E
with 128 processors;
• T1033 has 60000 elements (8x8) ~ 3000
equatorial pts.;
• Potential vorticity at 276 Jupiter days, T1033.
Shallow water equations / Jupiter
•Zonal wind at 700 Jovian days (left column);

•Equatorial jet strength in time (right column).
•Both demonstrate the effect of resolution.
Zonal Wind Equatorial Jet Strength

0
100
T170
200
100
T1033
200
0 500 1000
Latitude Jovian days

Dynamical Core/SEAM/LMR
Local mesh-refine Andes topography
Global
Zoom
Dynamical Core/SEAM
•Held-Suarez forcing; 384x8x8 ~ T85/L20 SEAM;
• Lat. vs Ht. zonal mean wind and eddy variance of T;
•Compare T63, G72 and SEAM.
T*2 (K2) U (m/s)
T63
T*2 (K2)
G72
Latitude Latitude
SEAM
Dynamical Core/SEAM
•Held-Suarez forcing;
•SEAM with uniform grid;
•Scaling results for various resolutions; almost
insensitive to processor number change.
HP Exemplar SPP2000
320km/L20
160km/L20
80km/L20
320km/L20 (dotted)
160km/L20 (solid)
80km/L20 (dashed)
Dynamical Core/SEAM
•Parallel scaling on various computers

•Triangles denote SEAM
•Horizontal resolution-T181, (g)seaborg- T533
•Other symbols for other models
•Log-log plot, flops vs processors.
Mflops per processor Gflops
# of processors
The Research Plan
• First and foremost, incorporate

SEAM as a dynamical core into
CCM4.
• Experiments to be undertaken
concurrently with this development:
– Apply the H-S forced version of
SEAM (already running) to
meaningful simulations such as
turbulence decay;
– Develop algorithmic refinements of
the 3-D simple forced model to
include semi-implicit integration.
This is currently under study by our
collaborators Thomas and Loft and a
test version is running.
28
Research Plan, cont.
• Mesh refinement studies:
– Test effectiveness of LMR using 3-D
PE version of SEAM with H-S test case
forcing model including real
topography;
– Test a suite of long integrations at
various uniform resolutions;
– Test high resolution over the Andes and
Himalayas with lower uniform
resolution elsewhere;
– Undertake comparison studies of LMR
with the stretched grid model of Fox-
Rabinovitz (already begun);
– Undertake the same comparison with a
global model which uses an imbedded
fine-scale model (say MM5) if
available.
29
• Vertical representation:
– Currently we use -coordinates; we will consider
the use of the finite volume method together with
spectral elements in the horizontal.
• Time extrapolation:
– The model currently uses explicit integration;
– We will study the feasibility of using semi-
Lagrange integration;
– The semi-implicit scheme is currently under study;
it appears to be between 2 to 3 times faster than
the explicit scheme for the SWE.
• Time filtering;
– We have developed a time filtering scheme which
we will test with SEAM.
• Storm tracks with LMR:
– We will put LMR into SEAM in the vicinity of
quasi-stationary storm tracks to test sensitivity.
30
Studies with SEAM and full physics

(CCM4)
• Using this model, repeat tests described
above with the simple forcing model and
compare to those results;
• Space resolution:
– Test quality of predictions against the
standard CCM4 with various resolutions on
a uniform grid.
• Apply LMR:
– One region at a time;
– Multiple regions;
– Consider LMR use in tropics for tropical
wave evolution;
– Compare LMR integrations with other
models using LMR, such as Fox-Rabinovitz’
stretched grid.
31
• Time resolution:
– We will determine the effect of SEAM
methodology on the climate period
selected; i.e., seasonal vs. decadal, etc.
– We will consider this effect using LMR.
• Computational efficiency:
– We will attempt to run the various
experiments on machines with largest
number of processors to determine
scaling.
32
• Interactions:
– NCAR
• CCM4 staff
• SCD staff
• Scientists associated with the project
• Computers
– NERSC
• Computers
• Support staff
– UMCP
• Scientists and students associated
with the project
• Stretched-grid development group.
• Staff:
– PIs: Baer, Tribbia, Fournier
– Co-Investigator: Taylor
– Faculty Affiliate: Fox-Rabinovitz
– Collaborators: Thomas, Loft
33
The End
34

Multi-Resolution Climate Modeling: Principle Investigators

Uploaded by

Copyright:

Available Formats

Multi-Resolution Climate Modeling: Principle Investigators

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Multi-Resolution Climate Modeling: Principle Investigators

Uploaded by

Copyright:

Available Formats

Multi-Resolution Climate

• To develop a climate model

SEAM: Spectral Element

• Use as the dynamical core with

• Tile spherical surface with arbitrary

Uniform Resolution Rectangles

Use the shallow water equations as an example:

 Generate an Integral form of the equations;

V(x,y)   Vk, j k (x) j (y)

 k (x i )  Legendre Cardinal functions = k ,i

xi  Gauss - Lobatto Points

•The Gauss- Lobatto quadrature points

•Within each rectangle, estimate the integral

•Simple combinations of Legendre cardinal functions;

Element interior points

Element boundary points

The system of integral equations,

A set for element interior points:

where J is the area integration weight.

• Tile the sphere with rectangles of

These choices result in an extremely

•Tests with the shallow water

(Thomas and Loft)

•Ratio of semi-implicit to explicit

Thomas and Loft

• A study of decaying turbulence with high

•Zonal wind at 700 Jovian days (left column);

Zonal Wind Equatorial Jet Strength

Latitude Jovian days

Local mesh-refine Andes topography

•Parallel scaling on various computers

Mflops per processor Gflops

• First and foremost, incorporate

Studies with SEAM and full physics

You might also like