21/03/2000 9th ORAP Forum ECMWF/DM-1 Simulation et prevision du temps Dominique Marbouty Head, Operations Department European Centre for Medium-range Weather.

21/03/2000 9th ORAP Forum ECMWF/DM-1

Simulation et prevision du temps

Dominique Marbouty

Head, Operations Department

European Centre for Medium-range Weather Forecast

[email protected]

http://www.ecmwf.int


Agenda

• The current situation

– What is ECMWF?: activity, requirements

– Past evolution of ECMWF’s computer means

– Comparison with other NWP centres

• Future evolution

– Requirements

– Schedule

– The main issues


Member States

Belgium The NetherlandsDenmark NorwayFederal Republic of Germany AustriaSpain PortugalFrance SwitzerlandGreece FinlandIreland SwedenItaly TurkeyYugoslavia United Kingdom

Co-operation agreements or working arrangements with:Croatia Iceland WMO ACMADHungary Slovenia EUMETSAT


European Centre for Medium-Range Weather Forecasts

Background 1967 European Council of Ministers propose co-operation in science and technology

1969 Expert group in meteorology propose ‘European Meteorological Computing Centre’

1971 Report on ‘EMCC’ : Net benefit of £100m per annum at 1971 prices

Establishment 1975 ECMWF convention in force

1978 Headquarters building completed

Start of operational activities 1978 Installation of first computer system (CRAY 1-A)

1979 Start of operations (N48 grid point model)

1983 T63 / L16 spectral model

1983 CRAY X-MP/22, IBM data storage system

1985 T106/L19 spectral model1985 CRAY X-MP, DEC telecommunications system (VAX)1990 CRAY Y-MP8/8-641991 T213 / L31 spectral model

1992 CRAY C90 - ocean wave forecasting

1993 Ensemble prediction system1996 Fujitsu VPP300-C and VPP700-461997 Fujitsu VPP700-1161999 Enhanced Fujitsu VPP system


ECMWF in a few figures

• Participants: 20 States

• Age: 25 years

• Staffing: 200 persons

• Budget: 25m £

• Model resolution: 60 km

• EPS: 50 elements


ECMWF’s principal objectives

• development of numerical models for weather forecasting

• daily distribution of forecasts to its Member States

• development of seasonal forecasting

• collection and storage of appropriate meteorological data

• allowing research staff to access these archives from remote sites

• providing computer resources to Member States




IFS/ARPEGE

• NWP package developed jointly by ECMWF and Météo-France

• includes forecast model, 4D-Var assimilation system, ensemble prediction system

• Medium range forecast: TL319 L60 (60 km)

• 4D-Var: TL319/T63 (60/300km)

• EPS: TL159 L40 (120km) 51 members

• SF: atm TL95 (200km) ocean 2ox2ox20


a

12345 12345

FUJITSU VPP700 (116 PEs)FUJITSU VPP700E (48 PEs)

FUJITSUVPP300 (9 PEs)

FUJITSUVPP300 (4 PEs)

SGIOrigin2000

SGI Challenge

VAX

HP 9000

Router

Router

Router Router DECFirewall

E T H E R N E T

IBM SP

FDDI

IBM RS/6000

HIPPI

PCs

SGIs

SGIs

TCP/IP DECNET

MEMBER STATES

WWW

JANET/INTERNET


ECMWF Supercomputer Growth

0.01

0.10

1.00

10.00

100.00

1,000.00

10,000.00

79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 00 01 02

Peak Performance (Gflops) Sustained Performance (Gflops)

CRAY-1XMP/2

X-MP/4Y-MP/8

C90/16

VPP700/46

VPP5000/100 + VPP700


The current ECMWF’s computers

• Fujitsu VPP 700/116 peak 255 Gflops sustained 80 Gflops

• Fujitsu VPP 700E/48 peak 115 Gflops sustained 35 Gflops

• Fujitsu VPP 5000/38 peak 365 Gflops sustained 110 Gflops ____ _____ 735 ~225

• From next August

• Fujitsu VPP 5000/100 peak 960 Gflops sustained 288 Gflops ____ _____

1330 ~400


Other NWP Centres

• UKMO SGI T3E900 /876 pk 788 Gf st ~79 GfSGI T3E1200/636 pk 763 Gf st ~76 Gf

_________ ______ 1550 Gf ~155? Gf

• DWD SGI T3E1200/812 pk 974 Gf st ~97? Gf

• Météo-France Fujitsu VPP5000/31 pk 298 Gf st ~90? Gf

• NCEP IBM SP/768 pk 614 Gf st ~45?Gf


The targets setfor period 1999 to 2008

• an extension of the skill of the deterministic forecast, as measured by the NH 500 hPa height field reaching 60%, by one day;

• a gain of one day at D+6 in the Brier skill-score for EPS-based probabilistic forecasts of moderate 850 hPa temperature anomalies (4K or larger) over Europe;

• preparation by 2003 of an assessment of seasonal forecast skill over the last 40 years;

• improving timeliness and reliability of product dissemination;

• provision of good forecasts of severe weather towards day 4 or day 5 - this requires the development of a suitable performance evaluation relating to severe weather.


Approach

• the development of an increasingly accurate earth-system model, using efficient and economical numerical methods together with a comprehensive and exhaustively-validated physical parameterisation package;

• the development of improved data assimilation systems employing the most advanced mathematical methods to extract information from in-situ data and from operational and research satellites;

• the development of improved EPS techniques for medium-range, extended-range and seasonal forecasting.


Development of the operational system-2000 to 2001

• a 60-level TL511 (40km horizontal resolution) 4D-Var data assimilation system, with analysis increments corresponding to the inner loop resolution of TL255 (80km);

• a Reduced Rank Kalman Filter (RRKF) to provide estimates of flow-dependent background and analysis errors;

• a 60-level TL511 (40km horizontal resolution) deterministic model;

• a 60-level TL255 (80km horizontal resolution),100 elements, EPS model;

• a 60-level TL95 (120km horizontal resolution) atmospheric/ocean model for seasonal prediction.


Next developments

• By 2001 - the effective utilisation of METEOSAT Second Generation geostationary satellite data will require a move towards a resolution of 40 km for the global analysis/forecast system employing 90 levels

• By 2003 - full use of a 30 km 90 level system to enable effective use of data from Europe’s polar orbiting METOP satellite

• By 2008 - doubling of the horizontal resolution to 20 km and increased vertical resolution will improve the exploitation of data from the high resolution satellite-borne sounders such as IASI and AIRS and other high resolution data over land


aa

ECMWF Computer requirements

0.1

1

10

100

Year

Frontier (Japan) peak performanceASCI peak performance

Contracted Required

* ASCI = Accelerated Strategic Computing Initiative (USA)

* Frontier = Earth Simulator Project (Japan)

19991998 2001 2003 2005 20072000 2002 2004 2006 2008


Draft schedule for the next computer

• December 2000: specifications decided by ECMWF’s Council

• 2001: run Invitation To Tender (ITT)

• December 2001: decision

• 2002: installation, parallel run, move operations to new machine


The preparation of the ITT

• Facilitate competition: RAPS

– RAPS6 to be issued this May (includes 4D-Var)

• Vector / Scalar : open choice

– code adaptation is being prepared


Growth in the ECMWF archives

0

2

4

6

8

10

12

14

16

Archive

Pet

abyt

es

Storage growth in petabytes


Conclusions

• Exploitation of the current Earth System Modelling capabilities and of the full information content of new satellite data will require substantial computer resources - the benefit will vastly improved forecast from a few days to a few seasons ahead

• The HPC could be acquired provided budgets are maintained in real-terms.

• Managing petabyte archives will be a major challenge.

21/03/2000 9th ORAP Forum ECMWF/DM-1 Simulation et prevision du temps Dominique Marbouty Head, Operations Department European Centre for Medium-range Weather.

Documents

th orap forum ecmwfdm

pes fujitsu vpp300

computer system cray

pes fujitsu vpp700e

forecast model

fujitsu vpp300c

gflops fujitsu vpp 700e48peak

o x2 o x20 slide