• Presently performed by SWECLIM (the Swedish regional climate modelling programme) that

• involves the Rossby Centre of the SMHI and two Swedish universities

• SWECLIM terminates in June 2003. The Rossby Centre activity continues. Activities also at Swedish universities.

Climate modeling in SwedenClimate modeling in Sweden

Markku Rummukainen, SMHI/Rossby Centre and SWECLIMErland Källén, Stockholm University and SWECLIM

AA scientific goal of climate modeling is the scientific goal of climate modeling is the prediction of climate change prediction of climate change

Ref: IPCC, 2001

Regional modeling for details of global Regional modeling for details of global change change

Typical present-day regional model < 50 km

Typical present-day global model 200-300 km

A typical Swedish regional climate A typical Swedish regional climate model simulationmodel simulation

• Limited area domain, nested to a global domain.

• Atmosphere:@50 km resolution: 106 102 24 grid points (259 488)@20 km resolution: 206 202 24 grid points (998 688)

• Baltic Sea:@6 nm res. 107 120 41 grid points, 10% active (52 773)@2 nm res. 320 362 41 grid points, 10% active (447 899)

• 30 years/simulation, with a 10-30 min timestep.

• Est. # of flops/atmospheric grid point/timestep > 3103.

Planned during the next 3 years:Atmosphere @ 10 km, up to 200-year simulationsRegional modelling for the Arctic. Also global modelling

Climate modeling needs more computing Climate modeling needs more computing power power

(ENES, Bryssel, 2001)(ENES, Bryssel, 2001)

Purpose est. need of increase

• Global model resolution 300 50 km 100

• Feedback (e.g. clouds, carbon cycle) ??

• Short-term internal variability of climate 10

• Emission scenarios, parameterizations 100

• Long-term internal variability of climate 10 -------------------------------------------------------------------------------

• Sum > 1 000 000 I.e., internationally. This is hardly realistic. Priorities needto be defined and work conducted in international cooperation.Swedish climate modelling participates in international networksfor both climate science, model developmnet and climate computing infrastructure.A possible spearhead contribution to international networks could be the use of PC-clusters for climate modelling.

On the seven requirements...On the seven requirements...

1) Comp. Resources: now ~500 000 CPU-hrs/year (Xeon 2.2 GHz-equiv.).

2) Inter- vs. intra-PE comm. speed: Crucial!

3) Primary memory: 1-2 Gb/PE

4) Secondary storage: now ~2Tb operat., ~4Tb archive.

5) Reaccess of stored data: weekly/monthly.

6) Access to databases: National+international. Infrequent exchange of large data sets.

7) Software: exists for T3E, SGI3k, Linux-clusters

Planned during the next 1-3 years:Computing resources requirement ~factor 2-10Secondary storage requirement ~factor 10.

An example of inter-PE vs. intra-PE aspects:An example of inter-PE vs. intra-PE aspects:

• For Xeon 2.2 GHz (4.8 Gflop/s peak), to use more than 16 PEs requires • latency < 8 s

• bandwidth > 200 Mb/s

• For Xeon 5 GHz, similar inter-PE requirements apply already if #PEs > 8.

• Interconnect! If Gigabit - max # of PEs ~30. If Scali -max # of PEs ~120.

• A generalization:• more PEs = faster/more simulations

• faster network = more PEs can meaningfully be used for a simulation

Swedish climate modeling and SweGridSwedish climate modeling and SweGrid

• SweGrid of use?• Data storage -aspect• Additional computing power• Software development intresting• ”Ensembles” rather than ”high

resolution”• Arctic modelling: an example of a

modelling subtask on the scale of a realistic share of SweGrids for climate modelling. Eventually grounds for a Nordic Virtual Organization, on regional Arctic modelling.

• Linux-cluster ok

• Interconnect important!

Possible DJF precipitation changeby 2071-2100, in %