Progress in the Development of Quasi-3D Multiscale … › events › wg-meetings › 2019 › presentations › ...(CRM developed at CSU/UCLA) + GCM component: dynamical core based

Progress in the Development of Quasi-3D Multiscale Modeling Frameworkas a Physics Option in CAM-SE Joon-Hee Jung, Celal S. Konor, David A. Randall,

Department of Atmospheric Sciences, Colorado State University, USA

Peter Lauritzen, and Steve GoldhaberNational Center for Atmospheric Research, Boulder, Colorado, USA

CESM: Atmosphere Model Working Group Meeting NCAR, Mesa Lab, Boulder, CO, February 19-21, 2019

This research has been supported by NSF AGS-1500187 (and partly by DOE ACME DE-SC0016273 and DOE CMDV DE-SC0016305).

Quasi-3D Multiscale Modeling Framework

z

x

y

z

x

y

GCM grid CRM grid GCM grid cell CRM ghost grid

Q3D MMF MMF

CRMs in GCM grid columns are seamlessly connected;

CRMs are three-dimensional, although covering only channel-like domains.

Two segments of CRMs perpendicularly pass each other (but don’t intersect) at the center of each GCM cell;

Jung and Arakawa (2010, 2014), Jung (2016)

Quasi-3D Multiscale Modeling Framework

z

x

y

z

x

y

GCM grid CRM grid GCM grid cell CRM ghost grid

Q3D MMF MMF

Surface orography can be resolved by the CRMs;Cloud systems travel freely along the CRM channels;

Subgrid-scale vertical momentum transport can be directly simulated.

CRMs in GCM grid columns are seamlessly connected;

CRMs are three-dimensional, although covering only channel-like domains.

Two segments of CRMs perpendicularly pass each other (but don’t intersect) at the center of each GCM cell;

Jung and Arakawa (2010, 2014), Jung (2016)

Development of a Global Q3D MMF

GCM component: dynamical core based on the cubed sphere grid

Quadrilateral geometry allows a straightforward extension to the sphere of our limited-area model based on rectangular Cartesian coordinate. Cubed sphere grid has relatively uniform horizontal grid spacings almost everywhere, allowing the CRM channels to be almost uniformly distributed.

CRM component: the CRM used in the limited-area Q3D MMF

A global version of Q3D MMF has been created.

Development of a Global Q3D MMF

Group 1

Group 2

Group 3 CAM-SE CSLAM dynamical core

Vector Vorticity Model(CRM developed at CSU/UCLA)

+

GCM component: dynamical core based on the cubed sphere grid

Quadrilateral geometry allows a straightforward extension to the sphere of our limited-area model based on rectangular Cartesian coordinate. Cubed sphere grid has relatively uniform horizontal grid spacings almost everywhere, allowing the CRM channels to be almost uniformly distributed.

CRM component: the CRM used in the limited-area Q3D MMF

A global version of Q3D MMF has been created.

Vector Vorticity Modelas a CRM Component of the Q3D MMF

3D nonhydrostatic anelastic model;

Pressure gradient force is eliminated from the governing Eqs;

Vertical velocity is a solution of a 3D elliptic equation;

CRM-type physics parameterizations are included.

The prognostic dynamical variables are vorticity components;

Vector Vorticity Modelas a CRM Component of the Q3D MMF

3D nonhydrostatic anelastic model;

Pressure gradient force is eliminated from the governing Eqs;

Vertical velocity is a solution of a 3D elliptic equation;

CRM-type physics parameterizations are included.

Suitable for the Q3D MMF It is convenient to formulate the lateral boundary conditions required by the narrow CRM channels and the lower boundary condition over steep topography.

The prognostic dynamical variables are vorticity components;

Implementation of the Vector Vorticity Dynamical Core on Cubed Sphere

for Use in the Q3D MMFJoon-Hee Jung, Celal S. Konor, and David Randall

J. Adv. Model. Earth Syst. (in press)

Implementation of the Vector Vorticity Dynamical Core on Cubed Sphere

for Use in the Q3D MMFJoon-Hee Jung, Celal S. Konor, and David Randall

J. Adv. Model. Earth Syst. (in press)

100

10-1

10-2

10-3

10-4

10-5

100 2550 6.2512.5

Nor

mal

ized

Erro

rs

Grid Size (km)

day 12

Rotation of a cosine bell along the equatorFollowing Williamson et al. (1992)

90 180 270 0 90135 225 315 45Longitude (degrees)

3525

45

155

-35-25

-45

-15-5

Latit

ude

(deg

rees

)

VVM: Barotropic Instability Test (Case 1)Initial Vorticity

Similar to Galewsky et al. (2004)


3525

45

155

-35-25

-45

-15-5

Latit

ude

(deg

rees

)


3525

45

155

-35-25

-45

-15-5

90 180 270 0 90135 225 315 45

3525

45

155

-35-25

-45

-15-5


Latit

ude

(deg

rees

) Longitude-latitude grids

Longitude (degrees)

Cubed-sphere grids

3525

45

155

-35-25

-45

-15-5

90 180 270 0 90135 225 315 45

3525

45

155

-35-25

-45

-15-5


Latit

ude

(deg

rees


Longitude (degrees)

Cubed-sphere gridst = 168 h

dx = dy ~ 100 km

dx = dy ~ 6 km

-8 -7 -6 -5 -4 -3 -2 -1 1 2 3 4 5 6 7 8 10-5 (s-1)

Similar to Galewsky et al. (2004)

-45

3525

45

155

-35-25-15-5

90 180 270 0 90135 225 315 45-45

3525

45

155

-35-25-15-5


Latit

ude

(deg

rees


Longitude (degrees)

Cubed-sphere grids


3525

45

155

-35-25

-45

-15-5

Latit

ude

(deg

rees

)


t = 120 hdx = dy ~ 100 km

-16 -14 -12 -10 -8 -6 -4 -2 2 4 6 8 10 12 14 16 10-5 (s-1)

-45

3525

45

155

-35-25-15-5

90 180 270 0 90135 225 315 45-45

3525

45

155

-35-25-15-5


Latit

ude

(deg

rees


Longitude (degrees)

Cubed-sphere grids


3525

45

155

-35-25

-45

-15-5

Latit

ude

(deg

rees

)


t = 120 hdx = dy ~ 100 km

-45

3525

45

155

-35-25-15-5

90 180 270 0 90135 225 315 45-45

3525

45

155

-35-25-15-5


Latit

ude

(deg

rees


Longitude (degrees)

Cubed-sphere gridsdx = dy ~ 6 km

-16 -14 -12 -10 -8 -6 -4 -2 2 4 6 8 10 12 14 16 10-5 (s-1)

VVM: Baroclinic Instability TestIdealized setting: f = 2Ωsin ϕ +π 4( )

Zonally Uniform Initial State

252015105 30 35 40

(m/s)

Latitude (deg)

45 15 -45-15

20

15

10

5

0

25

He

igh

t (k

m)

320

720

290

300

270

280

3525

45

155

-35-25

-45

-15-5

-180 0 180-90 90

Latit

ude

(deg

)

Longitude (deg)

302 300

266

Initial Perturbation on Potential Temperature

VVM: Baroclinic Instability TestIdealized setting: f = 2Ωsin ϕ +π 4( )

Zonally Uniform Initial State

252015105 30 35 40

(m/s)

Latitude (deg)

45 15 -45-15

20

15

10

5

0

25

He

igh

t (k

m)

320

720

290

300

270

280

3525

45

155

-35-25

-45

-15-5

-180 0 180-90 90

Latit

ude

(deg

)

Longitude (deg)

302 300

266

Initial Perturbation on Potential Temperature

3525

45

155

-35-25

-45

-15-5

-180 -90 0 90 180-135 -45 45 135

3525

45

155

-35-25

-45

-15-5

-180 -90 0 90 180-135 -45 45 135

3525

45

155

-35-25

-45

-15-5

-180 -90 0 90 180-135 -45 45 135

3525

45

155

-35-25

-45

-15-5

-180 -90 0 90 180-135 -45 45 135

Longitude (degrees) Longitude (degrees)

Latit

ude

(deg

rees

)

Longitude-latitude grids Cubed-sphere grids

dx = dy ~ 100 km

dx = dy ~ 12 km

t = 288 h

Horizontal Grid Structure of the Q3D MMF

GLL grid & physics grid (CSLAM configuration)

x

y

x

y

Low-resolution GCM grid High-resolution CRM grid

The CRMs are coupled with the physics grids.

Two Grid Systems

For communication, horizontal interpolation and cell-average are needed.

Example of Horizontal Grid Distribution

CRM grid channels

Group 1

Group 2 Group 3

Group 1

Group 2

Group 3

Nphy = 3, Ne = 5, Nc = 25 (dxGCM ~ 660 km, dxCRM ~ 26 km)

# of CRM-grid columns on the sphere = (# of physics-grid columns) x Nc x 2 = 67,500# of physics-grid columns on the sphere = (Nphy x Ne) x 6 = 1,3502

Example of Horizontal Grid Distribution

CRM grid channels

Group 1

Group 2 Group 3

Group 1

Group 2

Group 3

Nphy = 3, Ne = 5, Nc = 25 (dxGCM ~ 660 km, dxCRM ~ 26 km)

# of CRM-grid columns on the sphere = (# of physics-grid columns) x Nc x 2 = 67,500# of physics-grid columns on the sphere = (Nphy x Ne) x 6 = 1,3502

(dxGCM ~ 100 km, dxCRM ~ 4 km) 48,600 / 2,430,000

vGCM

k=nLevelk=nLevel+1

k=1 (layer)k=1 (interface)

....

....

32 la

yers

(~2.3 mb)u v Τ

ω

ω

ω

ω

ω

ω

u v Τ

u v Τ

u v Τ

pressure vertical coordinate (hybrid)

height vertical coordinate

Vertical Grid Structure of the Q3D MMF

VVM

....

u v θζη wξ

η wξ

u v θζη wξ

η wξ

u v θζη wξ

η wξ

....

k=2 (layer)

k=nk2

k=1 (interface)

k=nk2

k=2

k=nk2-1

k=nk2_extk=nk2_ext

....

acti

ve d

omai

n(d

ynam

ics

& a

ll ph

ysic

s)ex

tra

dom

ain

(onl

y ra

diat

ion)

(30 km)

(46 km)

30 la

yers

6 la

yers

θ

For communication, vertical interpolation is needed.

Ge

ne

ratio

n o

f d

ata

on

ph

ysic

s g

rid

s

vG

CM

co

mm

un

ica

tio

ns (m

ovin

g da

ta to

cha

nnel

dec

ompo

sitio

n)

Inte

rpo

latio

n o

f vG

CM

da

ta t

o C

RM

po

ints

CR

M C

OM

PU

TE

CR

M C

OM

PU

TE

Ca

lcu

latio

n o

f m

ea

n C

RM

fe

ed

ba

cks (p

hysi

cs te

nden

cies

)

POST-CRM

ProceduresCRM PREDICTION

PRE-CRM

Procedures

DY

CO

RE

CO

MP

UT

E

with

ph

ysic

s t

en

de

ncie

s

PREDICTION

CR

M C

OM

PU

TE

Dis

trib

utio

n o

f C

RM

fe

ed

ba

cks t

o D

yco

re p

oin

ts

vG

CM

co

mm

un

ica

tio

ns (m

ovin

g da

ta to

phy

sics

dec

ompo

sitio

n)

CR

M C

OM

PU

TE

t

Q3D MMF Computation Algorithm

t0

“physics”

Similar to the coupling approach of CAM-SE

t1

Hybrid Dynamics/Physics

coupling

x

y

GCM GLL grid

CRM grid

Input to CRM: GCM-scale solutions

interpolated to CRM grids

Output from CRM: Eddy transport

& diabatic effects(cell averages)

Communication between Dycore & CRMs

x

y

x

y

GCM physics grid

“Interface”

Two model components are coupledat every GCM time step.

Test of Q3D MMF InterfacesThe GCM-scale input data are distributed to CRM grids through horizontal and vertical interpolations and gathered back to physics grids without performing the CRM-prediction.

Qv_in Qv_out

Actual Output Actual Output

Plotted on uniform grid Plotted on uniform grid

Longitude (deg)

Latit

ude

(deg

)La

titud

e (d

eg)

Latit

ude

(deg

)La

titud

e (d

eg)

Longitude (deg)

k = 15

Actual Output Actual Output

Plotted on uniform grid Plotted on uniform grid

Longitude (deg)

Latit

ude

(deg

)La

titud

e (d

eg)

Latit

ude

(deg

)La

titud

e (d

eg)

Longitude (deg)

k = 32 (Lowest model layer)

Short-Term Simulation of the Q3D MMF

Latit

ude

(deg

)

t = 0 t = 24 hWithout CRM Feedback

Surface Pressure

Longitude (deg) Longitude (deg)La

titud

e (d

eg)

Longitude (deg) Longitude (deg)

Latit

ude

(deg

)La

titud

e (d

eg)

T

Qv

Qr

(Lowest model layer)

995

1006

1017

1028

1039

1050

(mb)

Latit

ude

(deg

)

t = 0 t = 24 hWith CRM Feedback (Microphysical effect)

Surface Pressure


titud

e (d

eg)


Latit

ude

(deg

)La

titud

e (d

eg)

T

Qv

Qr


995

1006

1017

1028

1039

1050

(mb)


Latit

ude

(deg

)

t = 0 t = 24 hWith CRM Feedback (Microphysical effect)

Surface Pressure


titud

e (d

eg)


Latit

ude

(deg

)La

titud

e (d

eg)

T

Qv

Qr


995

1006

1017

1028

1039

1050

(mb)


We are making progress in testing and debugging the new global Q3D MMF with short-term simulations.

SummaryA global version of Q3D MMF has been created by coupling the VVM on cubed-sphere grids with the CAM-SE-CSLAM dynamical core (https://svn-ccsm-models.cgd.ucar.edu/cam1/branches/Q3D).

https://svn-ccsm-models.cgd.ucar.edu/cam1/branches/Q3D

We are making progress in testing and debugging the new global Q3D MMF with short-term simulations.

SummaryA global version of Q3D MMF has been created by coupling the VVM on cubed-sphere grids with the CAM-SE-CSLAM dynamical core (https://svn-ccsm-models.cgd.ucar.edu/cam1/branches/Q3D).

- Finish up the code development;- Make improvement to numerical methods and parameterizations;- Speed up the computing time (hybrid MPI/OpenMP)

We will continue to evaluate and improve the model for its application to long-term simulations.

https://svn-ccsm-models.cgd.ucar.edu/cam1/branches/Q3D

Documents

Progress in the Development of Quasi-3D Multiscale … › events › wg-meetings › 2019 › presentations › ...(CRM developed at CSU/UCLA) + GCM component: dynamical core based