27
Computing at Météo- France CAS 2003, Annecy 1.News from computers 2.News from models

Computing at Météo-France CAS 2003, Annecy 1.News from computers 2.News from models

Embed Size (px)

Citation preview

Computing at Météo-

France

CAS 2003, Annecy

1. News from computers

2. News from models

Fujitsu VPP 5000/31

dec 1999 – june 2003

31 PE

297.6 Gflops

208 Gbytes main memory

Crossbar 2x1,6 GB/s

CAS 2003, Annecy

VPP 5000/64 and VPP

5000/60 june 2003 – sept 2006

Research system : 64 PE

614.4 Gflops

300 GBytes main memory

Operations system : 60 PE

576 Gflops

280 GBytes main memory

CAS 2003, Annecy

Final configuration

SYS-OP (60)SYS-RD (64)

3400 GB 2680 GB

File server

HIPPI switch

CAS 2003, Annecy

HIPPI switch

2 HIPPI links

Météo-France computing

centre

CAS 2003, Annecy

HIPPI 800 Mbit/s

Switching bandwith 64 Gbit/s

STK silos 150 TB

SGI O2K

H-P servers

Operations & development

Workstations

Fujitsu VPP’s

Historical point of view

1

10

100

1000

1000019

91

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

GF

lop

s

puissance crête puissance soutenue

Cray 2

C98-4

Cray 2

C98-8

J916

VPP700E

VPP5000-31

VPP5000-64

VPP5000-60

CAS 2003, Annecy

Performance and the Moore

law

1

10

100

1000

1000019

91

1992

1993

1994

1995

1996

1997

1998

1999

2000

2001

2002

2003

2004

2005

GF

lop

s

puissance crête puissance soutenue loi de Moore

CAS 2003, Annecy

Archived data volume and the

Moore law 1

99

1

19

92

19

93

19

94

19

95

19

96

19

97

19

98

19

99

20

00

20

01

20

02

20

03

Te

rao

cte

ts

volume loi de Moore

CAS 2003, Annecy

Archiving system main requirements

CAS 2003, Annecy

Phase A (march 2004) :

230 TB, 8.5 M files, I/O : 2.5 –5.2 TB/day

Phase B (sept 2004) :

360 TB, 12 M files, I/O : 3.4 – 7.1 TB/day

Phase C (sept 2005) :

560 TB, 17 M files, I/O : 4.5 –9.4 TB/day

During all three phases :

4 days on disk cache

75 % of data must be accessed in less than 45 seconds

100 % of data must be accessed in less than 340 seconds

Computing at Météo-France

part 2 : News from models

CAS 2003, Annecy

CAS 2003, Annecy

IntegratedForecastingSystem

A numerical weather prediction system developed and supported by Météo-France and the European Centre (ECMWF)Includes all contituents needed for global numerical prediction : A global spectral model (and associated

tangent linear and adjoint models) 3D et 4D-VAR global assimilation etc

Action deRecherchePetiteEchelleGrandeEchelle

CAS 2003, Annecy

ECMWF (European Centre forMedium-range Weather Forecast)

Reading – United KingdomGlobal spectral model T511, 60 levels (up to 0.1 hPa). « Linear » (lat/lon) grid, 35 to 40 km. « Semi-lagrangian, semi-implicit » temporal integration, time step 20’.Forecasts up to 10 days, starting at 12 UTC every day4D-VAR assimilation on a 12h window, through 2 minimisations at T159 (increment resolution)Ensemble forecast (EPS - Ensemble Prediction System) through 51 integrations of the same model, with resolution T255 / 40 levels

CAS 2003, Annecy

Grid for the ARPEGE model

The ARPEGE global model

Global spectral model

TL358 C2.4, 41 levelsAssociated grid: 23 km (France)

133 km (antipodes)

Representation on the globe with stretching and turning of the pole over the interest zone

Collaboration with ECMWF

CAS 2003, Annecy

The ARPEGE-Tropiques model

• Uniform resolution

- horizontal grid without stretching (TL359 grid)

- vertical levels unchanged (41 levels)

• Run once a day

from 00 UTC up to 72h range

CAS 2003, Annecy

The ALADIN project

Genesis

• Project started by Météo-France in 1990• A mutually beneficial collaboration with National

Meteorological Services of Central and Eastern Europe concerning numerical prediction

• Acronym: Aire Limitée, Adaptation dynamique, Développement InterNational = Limited Area, dynamical Adaptation, InterNational Development

CAS 2003, Annecy

The ALADIN project

Different operational domains

CAS 2003, Annecy

The Aladin-France limited area model

• A spectral model

• Domain: – a square 2740 km in side, centered on the point of maximal

resolution of Arpège

• Vertical levels: – same as Arpège (41)

• Horizontal resolution – (9 km) ~ 2.5 × max resolution in Arpège

• Coupling: – applied every three hours to the global model Arpège

Principles of 4D-VAR assimilation

9h 12h 15h

Assimilation window

Jb

Jo

Jo

Jo

obs

obs

obs

analysis

xa

xbcorrectedforecast

previousforecast

CAS 2003, Annecy

A short description of the 4D-VAR used at Météo-France (Toulouse) since June 2000

Minimization window 6h (maximum)INCREMENTAL technique: increments are evaluated by 2 minimizations at T107 and T149 (c=1) for the T358/c2.4 modelStructure functions are not separableOBSERVATIONS WE USE: conventional, satellite winds, ATOVS radiances (no diffusiometer data)Weak constraint based on digital filters (also used to smoothen the trajectory after the last minimization)

CAS 2003, Annecy

Operational use of models at Météo-France (June 2002)

• ARPEGE model (variable mesh), routine run 4 times a day– starting 00h UTC, until 96h– starting 06h UTC, until 42h– starting 12h UTC, until 72h– starting 18h UTC, until 30h

• Uniform ARPEGE model, routine run once a day– starting 00h UTC, until 72h

• ALADIN/France model, routine run 4 times a day– coupled to the corresponding ARPEGE model that gives boundary

conditions and initial conditions, until the same forecast ranges

• Initial conditions are given by a 4D-VAR assimilation – run on 6h long time windows, centered on each of 00, 06, 12 and 18h

UTC.– ARPEGE/Tropiques has its own 4D-Var. – Digital Filter Initialisation (DFI) is used for ARPEGE and ALADIN.

CAS 2003, Annecy

Other operational models

• Oceanography : the MERCATOR project– North-Atlantic & Mediterranean sea : high

resolution model (1/15°)– Global ocean : low resolution model (2°)– 6 PEs (8 GB), 6 hours, 1week(http://www.mercator.com.fr/html/science/concept_en.html)

• Ocean wave models

• Pollutant dispersion models

CAS 2003, Annecy

Research models

• Climate model : ARPEGE-Climat– TL63 or TL159C2.5

– 31 or 60 levels

– TL63 L31 used for seasonal forecasting

• Mesoscale model : MESO-NH– Mesoscale non-hydrostatic model

– Joint development with Laboratoire d’Aérologie (Université de Toulouse Paul Sabatier)

(http://www.aero.obs-mip.fr/mesonh/index2.html)

CAS 2003, Annecy

Météo-France plans for 2000-2010 concerning Numerical Weather Prediction

• Optimization of the ARPEGE-ALADIN system• mainly for physics, observation use and assimilation algorithms

• AROME project (Application de la Recherche à l’Opérationnel à Méso-Échelle = Applying Research to Operational use at Meso Scale) • NWP system with a 2-3 km horizontal mesh over France in about 2008-2010

• Target • more concern for short-range prediction of dangerous phenomena (eg.

precipitations), together with a greater coordination with other organizations using such predictions (eg. hydrologist)

CAS 2003, Annecy

Météo-France cooperations concerning NWP

ECMWFALADIN communityHIRLAM

EUMETSAT (SAF)Spatial Agencies

CAS 2003, Annecy

AROME PROJECT

• NH model based on existing dynamics in ALADIN.

• Largely derived from Meso-NH for physical parameterizations.

• Specific meso-scale data assimilation system (inspired from existing ALADIN 3D-VAR to start with).

CAS 2003, Annecy

Regional use of observations for AROME

• Priority on infra-red radiances and imagery from new satellites, and on radar observations.

• Use of all surface regional and national networks

• Use of other satellite data: micro-waves, GPS (especially ground-based), lidar winds.

Computing at M-F

THANKS

[email protected]://www.meteo.fr

CAS 2003, Annecy