28
HPC for better understanding of the tropical meteorology Y. Kajikawa and H. Tomita Oct 11 th , 2013, @GMCL, PNU 1 Necessity of …

HPC for better understanding of the tropical meteorology

  • Upload
    tomai

  • View
    32

  • Download
    0

Embed Size (px)

DESCRIPTION

Necessity of … . HPC for better understanding of the tropical meteorology . Y . Kajikawa and H. Tomita Oct 11 th , 2013, @GMCL, PNU . History of climate modeling (1). Richardson’s Dream (1910s-1920s). History of climate modeling (2). - PowerPoint PPT Presentation

Citation preview

Page 1: HPC for better understanding  of the tropical meteorology

HPC for better understanding of the tropical meteorology

Y. Kajikawa and H. TomitaOct 11th, 2013, @GMCL, PNU

1

Necessity of …

Page 2: HPC for better understanding  of the tropical meteorology

2

History of climate modeling (1)

Richardson’s Dream (1910s-1920s)

Page 3: HPC for better understanding  of the tropical meteorology

3

History of climate modeling (2)ENIAC (Electronic Numerical Integrator And Computer) was

the first electronic general-purpose computer.

http://en.wikipedia.org/wiki/ENIAC

1947-1955@Maryland

Page 4: HPC for better understanding  of the tropical meteorology

4

[Q] Why does the climate model require the HPC?

http://www.nies.go.jp/kanko/kankyogi/19/04-09.html

1. Increase of resolutionTo know more detail structure!

e.g. horizontal Resolution in usual climate model :100km ( 2000 ) / 20km (2005 )3.5km on K-computer

Page 5: HPC for better understanding  of the tropical meteorology

5

2. Increase of processesTo know more complex interactions.

Atm. Ocn. Lnd. modelsCarbon cycle, aerosol, chemistry, dynamic

vegetation processExternal forcing

Variability of solar constantCO2 emission scenario by IPCC run

-> Earth System model

[Q] Why does the climate model require the HPC?

Page 6: HPC for better understanding  of the tropical meteorology

6

[Q] Why does the climate model require the HPC?

http://www.jma.go.jp/jma/kishou/know/kisetsu_riyou/glossary/ensenble.html

3. Increase of ensemblesTo make our results more reliable.Statistical knowledge is necessary.

Page 7: HPC for better understanding  of the tropical meteorology

7

Importance of the cloud process (1)

1. Engine for general circulation :– Cumulus has an important role

for atmospheric heat transfer over the globe. (latitudinal direction).

2. Hierarchical structure generates many phenomena.» Cloud cluster , super cloud

cluster, tropical cyclone, MJO, … 7

Page 8: HPC for better understanding  of the tropical meteorology

8

• Large impact on the energy balance in climate: • Parasol effect :

reduce the incoming solar incidence.• Green house effect :

cloud emits infrared radiation into the surface and space.– Difficulty :

the interaction with aerosol and chemistry through radiation process• Indirect effect of aerosol : optical

thickness of cloud and cloud life time.• Direct effect of aerosol is also important.

8

Importance of the cloud process (2)

Parasol effect Reflection of solar incident

Greenhouse effect Emission of infrared

Page 9: HPC for better understanding  of the tropical meteorology

99…Very difficult to model the cloud!

cumulus

Shallow cloud

cirrus

Various cloud types exist in our earth!

Page 10: HPC for better understanding  of the tropical meteorology

10Earth diameter :12740km10km

10km

Cloud cluster ~ 100km

1 km

Super cloud cluster ~ 1000km   MJOCloud element: cumulus

Hierarchical structure of clouds

Page 11: HPC for better understanding  of the tropical meteorology

11

11

Example of cloud origination meso-scale cloud

Cloud drop aggregation

Fall as precipitation

Cooling by evaporation

Cold poolGeneration of

new clouds

Generation of new clouds

understanding of cloud dynamics

Page 12: HPC for better understanding  of the tropical meteorology

12

Cloud has many features and large impact on the climate through the

complicated processes.

What should we start to study the cloud

processes by modeling

in the age of HPC ?

Page 13: HPC for better understanding  of the tropical meteorology

13

Cumulus: Each of cumuli cannot be expressed directly due to

too coarse grid The effect of cumulus is taken a count as parameterization

Grid intervals: 100 km

Each of clouds < 10 km

Uncertainty : many methods generate many results!

Expression of the clouds 10 years ago

Page 14: HPC for better understanding  of the tropical meteorology

14

Cumulus (cloud-system ) can be resolved! To avoid the parameterizationHigh reliability / expression of cloud dynamics (w/ cold

pool)

New Approach from 2004

Grid intervals: a few km

Page 15: HPC for better understanding  of the tropical meteorology

Numerical techniques in the new approach

15

• Global cloud-system resolving model– Icosahedral grid

• To get a quasi-homogeneous grid– nonhydrostatic DC

• To resolve cloud scale– explicit cloud expression:

• To avoid the ambiguity of cumulus parameterization.

NICAM ( Tomita & Satoh 2004, Satoh et al. 2008 )

• NICAM project : ~2000 – The first target machine :

Earth Simulator– Now, porting to K computer system Prof. Satoh (AORI, Tokyo univ.) Dr. Tomita (RIKEN AICS)

Page 16: HPC for better understanding  of the tropical meteorology

16

Icosahedral grid system?

Regular Icosahedron = Polyhedron with 20 triangular faces.

By dividing each triangles in to 4 small triangle, we can obtain one-higher resolution.

e.g. a -> b -> c-> d …

Page 17: HPC for better understanding  of the tropical meteorology

17

DynamicsGoverning equations Fully compressible non-hydrostatic system Spatial discretizationHorizontal grid configuration

Vertical grid configurationTopography

Finite Volume Method

Icosahedral grid with spring dynamics smoothing  (Tomita et al. 2001/2002)Lorenz gridTerrain-following coordinate

Conservation Total mass, total energy   (Satoh 2002, 2003)Temporal scheme Slow mode  -  explicit scheme  ( RK2, RK3 )

Fast mode -  Horizontal Explicit Vertical Implicit scheme Physics Turbulence/shallow clouds MYNN 2.0,2.5(Nakanishi and Niino 2004) modified by Noda(2009)Surface flux Louis (1979), Uno et al. (1995)Radiation MSTRNX (Sekiguchi and Nakajima, 2005)Cloud microphysics NSW6 (Tomita 2008) --- 6 caegories of water ( 1moment-bulk)Cloud parameterization NONESurface process MATSIRO(Takata et al.)

Ref. Satoh et al. 2008 J. Comput. Phys. / Tomita & Satoh 2004 Fluid Dyn. Res.

Recent DC description paper : Tomita et al. 2011, ECMWF workshop proceeding

NICAM current implementation

Page 18: HPC for better understanding  of the tropical meteorology

18

• NICAM high resolution run:– 14km, 7km, 3.5km, – 1.8km, 800m, 400m

• Many terms :– Cloud permitting?– Cloud resolving?– Cloud system-resolving? (GCRM)– Meso-scale resolving?

• In the terms of methodology,– To avoid the ambiguity of cumulus parameterizationMethodological cloud-system resolving!

Objection : Cloud resolving model? (Grey zone problem)

The examination of impact

without Cumulus Parameterization is the most important!

Page 19: HPC for better understanding  of the tropical meteorology

19

What can the GCRM perform?Explicit expression of cloud clusters from the basic dynamical

mechanism ( Cold pool dynamics )Explicit expression of lifecycle of typhoon (onset &development)

   e.g .NICAM 7-km simulation

筑波大・田中博教授 (2010, vol.29-1, NAGARE)

Page 20: HPC for better understanding  of the tropical meteorology

20NICAM 7km-mesh, one-month simulation: initial = 15 Dec. 2006

23 Dec. 2006

報道発表資料 図2

31 Dec. 2006 8 Jan. 2007

Miura et al. 2007 Science

We can capture MJO realistically by GCRM

Page 21: HPC for better understanding  of the tropical meteorology

21

We are now in the K-computer, 10 Peta-FLOPS, era !!

Page 22: HPC for better understanding  of the tropical meteorology

Earth Simulator Now, we can run such simulations of several decades with “K”, and make a

breakthrough from the case study

Case study(Miura et al 2007)

Several weeks and monthAthena Project: (Sato et al 2012)

  Athena Cray XT-4

From the demonstration to scientific knowledge

Page 23: HPC for better understanding  of the tropical meteorology

10000km 1000km 100km 10km 1km 100m 10m

cumulusBlocking

Low-pressure Cloud cluster stratus

Tropical cyclone

Grand Challenge project:

GL13 (800m)

GL12 (1.7km)

GL11 (3.5km)

GL10 (7km)

GL09 (14km)

GL08 (30km)

Page 24: HPC for better understanding  of the tropical meteorology

24

Successfully conducted the GL13(870m) simulation

Page 25: HPC for better understanding  of the tropical meteorology

25

Essential change of convection statistics

The convection structure, number of convective cells, and distance to the nearest convective cell dramatically changed around 2.0km

Page 26: HPC for better understanding  of the tropical meteorology

26

Future direction of climate modelingIncreases of resolution, model component, ensemble

A key factor to sophisticate the atmospheric modelCloud modelingA new method is to express explicitly each of clouds

A main topics of climate research using K computerCumulus, cloud organization, tropical cyclone, MJOHigh resolution ( less than 2.0km grid spacing) can resolve

convection core using multiple grid.

Summary

Page 27: HPC for better understanding  of the tropical meteorology

27

감사합니다

Page 28: HPC for better understanding  of the tropical meteorology

28