towards multiscale simulations of cloud...
TRANSCRIPT
![Page 1: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/1.jpg)
Towards Multiscale Simulations of Cloud Turbulence
Ryo Onishi*1,2 and Keiko Takahashi*1
*1Earth Simulator Center/JAMSTEC *2visiting scientist at Imperial College, London
(Prof. Christos Vassilicos)
International MetStorm Conference 9 June, 2011
***Some figures and slides are removed from original talk***
![Page 2: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/2.jpg)
MSSG (Multi-Scale Simulator for the Geoenvironment)
High-resolution cloud simulations (LES) with MSSG-Bin microphysical scheme Turbulent collisions of droplets RICO
DNS for colliding droplets in high-Re flows Our numerical schemes validation of turbulent collision models
![Page 3: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/3.jpg)
3
Earth Simulator Mar. 2002 – Sept. 2008
Processors : 5,120 (640 nodes)
Peak performance:40 TFLOPS
Main memory:10 TB
Earth Simulator 2 Mar. 2009 -
Processors : 1,280 (160 nodes)
Peak performance:131 TFLOPS
Main memory:20 TB
![Page 4: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/4.jpg)
4
urban scale
mesoscale
global scale
Applicable to global, regional and local scales seamlessly
Ying-Yang grid for globe Consists of 3 modes; atmos. /
ocean / coupled Highly optimized for the Earth
Simulator (ES)
![Page 5: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/5.jpg)
![Page 6: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/6.jpg)
Observation (CMAP, 1979-2001JJA)
MSSG-A (ΔH=40 km, 32 levels, 5yrs ave-JJA)
precipitation at JJA
**skip**
![Page 7: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/7.jpg)
7
observation (TRMM 3B42)
NICAM-7km (Miura et al., 2008)
MSSG-10km
5[m/s]
5[m/s]
5[m/s]
30 days from Dec 15, 2006
![Page 8: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/8.jpg)
ΔH=2.7 km, 72 layers
MSSG result is operationally
provided to
![Page 9: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/9.jpg)
turbulence (large-eddy) resolving mesoscale simulation
![Page 10: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/10.jpg)
lat-long system for regional simulations
Yin-Yang system for global simulations Ref. Kageyama & Sato (2004)GGG Baba et al. (2010)MWR
![Page 11: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/11.jpg)
MSSG-Bulk model (Reisner et al. (1998) with modification by Thompson (2004))
![Page 12: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/12.jpg)
12 MSSG-Bin model (Onishi & Takahashi, submitted)
purely spectral-bin scheme for warm clouds
![Page 13: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/13.jpg)
Gravitational (Hydrodynamic) Collision Kernel (No turbulent collisions)
Turbulent Collision Kernel
( R : collision radius (=r1+r2), V∞:settling velocity )
|wr| : radial relative velocity at contact (Wang et al. 2000)
g(R) : radial distribution function at contact (Onishi 2005, Onishi et al. 2007 etc…)
)()(),( 112
21 rVrVRrrKc ∞∞ −= π
∫∫∞
−=
∂
∂00
')'()()',( ')'()''()',''( 21
)(drrnrnrrKdrrnrnrrK
trn
ppc
r
ppccol
p
Stochastic Collision Equation (SCE)
Collision kernel
![Page 14: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/14.jpg)
( ) 32SCs
SGS
∆=
≈ εε
MSSG Dynamics
Microphysics(Bin method)
・compressive N-S eq. (static Smagorinsky model) ・etc…
・collision growth ・condensation growth ・etc…
'15 ulεν
λ =
2/' Su ∆≈ν
λλ
lu'Re =
( ) 4/13 / ενη =lFlow turbulent statistics local isotropy
![Page 15: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/15.jpg)
Field campaign for shallow cumulus developing over the ocean near the Barbuda islands (18.0N;61.5W).
www.eol.ucar.edu/ projects/rico/
![Page 16: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/16.jpg)
0
1
2
3
4
-15 -10 -5 0
U, V[m/s]
z [k
m]
U
V
surface fluxes •Heat flux: wθl=-Ch*|U|*(θl-SST*(p0/p)(Rd/cp)) •Water vapor flux: wqt=-Cq*|U|*(qt-qsat(Tsfc)) •Momentum flux: uw=-u*Cm*|U| vw=-v*Cm*|U| where,Cm=0.001229, Ch =0.001094, Cq =0.001133
12.8km 12.8km
4km
24h simulation withΔx=Δy=100m, Δz=40m
0
1
2
3
4
0 10 20
qv[g/kg]
z [k
m]
0
1
2
3
4
290 300 310 320
θ[K]
z [k
m]
init. vert. prof.
periodic B.C.
![Page 17: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/17.jpg)
1-hour-simulation with MSSG-Bulk method (for visualization)
Δx=100m
![Page 18: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/18.jpg)
Bulk scheme Bin scheme
(r>40μm droplets are recognized as rain drops)
Clouds distribute uniformly in Bulk simulation (w/o CCN), while rather patchily in Bin simulation (with CCN).
Δx=100m
![Page 19: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/19.jpg)
www.eol.ucar.edu/ projects/rico/
![Page 20: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/20.jpg)
20
Earth Simulator Center
![Page 21: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/21.jpg)
Cloud simulation using MSSG-Bin scheme
ΔH=25m, ΔV=20m 512x512x200 grids 33 bins 3D Mie scattering
Earth Simulator Center
![Page 22: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/22.jpg)
Earth Simulator Center
![Page 23: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/23.jpg)
hydr&turb_case1-4.x33
MSSG-Bin with Turb. Kc
Significant impact of turbulent collisions
![Page 24: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/24.jpg)
Applicability of the turbulent collision model has not been confirmed. Reλ=103~4 in cumulus clouds, while available DNS data is for Reλ<102.
We need DNS data for high Reλ.
![Page 25: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/25.jpg)
>Our numerical schemes >Validation of turbulent
collision models
![Page 26: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/26.jpg)
St=0 St=0.2
St=4 St=1
R : collision radius (=r1+r2)
|wr| : radial relative velocity at contact
g(R) : radial distribution function at contact
preferential distribution
St
![Page 27: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/27.jpg)
27
Paper name fluid
calculation
cell-index method
paralleliza-tion
Reλ,max
Wang et al.(2000)JFM PSM w - 75.4
Reade&Collins(2000)PF PSM - - 82.5
Zhou et al.(2001)JFM PSM w - 58
Ayala et al.(2007)JCP PSM w shared 72.4
Ayala et al.(2008)NJP PSM w shared 72.4
Wang et al.(2008)NJP PSM w shared 72.4
Woittiez et al.(2009)JAS FDM - - 84.9
Onishi et al.(2009)PF PSM w/o shared 68.4
Table 1 Review of recent studies on collision frequencies of inertia particles in stationary isotropic turbulence.
Flow: Euler Particle: Lagrange
Pseudo-Spectral Model (PSM) has been used for flow!
![Page 28: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/28.jpg)
Steady isotropic air turbulence: Finite-Difference Model (FDM)
4th-order conservative scheme (Morinishi et al. (1998)JCP)
MPI, i.e., distributed-memory parallelization, for 3D domain decomposition
RCF(Reduced-Communication Forcing) to attain a stationary state (Onishi et al. (2011)JCP)
Particle motion:
Lagrangian tracking method Cell-index method for efficient
collision detection MPI for 3D domain decomposition
Cell-index method (Allen & Tildesley (1987))
3D domain decomposition
![Page 29: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/29.jpg)
100 101 102 103102
103
104
105
106
107
Present code (N256) Peak Performance 10% of Peak Performance
MFL
OPS
Number of cores
F =∂u1 ∂x1( )4
∂u1 ∂x1( )2 2Flatness factor:
Flow simulation Flow & Particle simulation
Performance on ALTIX for 260,000 droplets in 2563 grids
Good parallel performance Reliable & Fast
Present FDM is estimated to be x3.8 faster than conventional PSM (not shown here).
REF:Onishi et al. (2011)JCP
![Page 30: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/30.jpg)
newly obtained data appears in the contribution paper
? Collision data for Reλ~540 (20003 grids, billion particles) is obtainable on ES2
Reλ=340 10003 grids, 50million particles
![Page 31: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/31.jpg)
31
0.1 1 100.1
1
10
100
g(R)
- 1
St
CASE-1 (Reλ=36) CASE-2 (Reλ=44) CASE-3 (Reλ=54) CASE-4 (Reλ=83)
St2
St<<1 0.1 1 10 100
0.01
0.1
1
10
(g(R
) - 1
) / (R
e λ St-2
)
St
CASE-1 (Reλ=36) CASE-2 (Reλ=44) CASE-3 (Reλ=54) CASE-4 (Reλ=83)
St~1
1 10 1000.01
0.1
1
10
(g(R
) - 1
) / (R
e λ St-1
)1/2
St
CASE-1 (Reλ=36) CASE-2 (Reλ=44) CASE-3 (Reλ=54) CASE-4 (Reλ=83)
St >>1
Empirical, but with some physical support
Problem: g(2r) for St=1 is in between St<<1 and St~1, and parameterized fully-empirically.
![Page 32: Towards Multiscale Simulations of Cloud Turbulencemetstroem.mi.fu-berlin.de/wp/wp-content/uploads/2012/06/Ryo-Onishi.pdf3 Earth Simulator Mar. 2002 – Sept. 2008 . Processors : 5,120](https://reader036.vdocument.in/reader036/viewer/2022071404/60f80f73b9a6d847552dac27/html5/thumbnails/32.jpg)
MSSG (Multi-Scale Simulator for the Geoenvironment) Non-hydrostatic atmos-ocean coupled model Yin-Yang grid
High-resolution cloud simulations (LES) with MSSG-Bin scheme useful to see the turbulence role in clouds Δ=25m simulations are feasible on the ES2
DNS for colliding droplets in high-Re flows useful to check the applicability of turbulent collision
models to cloud turbulence with Reλ=103~4 Data for up to Reλ=340 has been obtained, and one for Reλ=540 is coming.
-These atmos. flows have energy scales L of O(1m).
Turbulence models validated by the DNS for L=O(10m) will strengthen the Δ=O(10m) cloud simulations, on O(10 PFLOPS) supercomputers!