The Model Garage
Twelve operational long-range systems, what is actually under each hood, and where to find the specification that supersedes this page.
The systems
Coupled ocean–atmosphere · monthly cycle| Centre | System | Members | Lead | Model core & notes |
|---|---|---|---|---|
| ECMWF | SEAS5 (System 5) | 51 | 7 months (13 from the February start) | Tco319 atmosphere (~36 km) coupled to NEMO ORCA0.25The reference system for much of the world, and the one C3S is built around. Released on the 5th of each month at 12 UTC. |
| Met Office (UK) | GloSea6 | lagged daily | 6 months | N216 atmosphere coupled to NEMO/CICEBuilt on the same HadGEM3 family as the Met Office's climate model, run as a continuously lagged ensemble rather than one monthly burst. |
| Météo-France | System 8 | 51 | 7 months | ARPEGE atmosphere coupled to NEMOThe French contribution to C3S, and one of the longest continuous seasonal records in Europe. |
| DWD (Germany) | GCFS2.1 | 50 | 6 months | ECHAM6 atmosphere coupled to MPIOMThe German Climate Forecast System, run with the Max Planck Institute's model core. |
| CMCC (Italy) | SPS3.5 | 50 | 6 months | CAM5.3 atmosphere coupled to NEMOThe Euro-Mediterranean Centre's seasonal prediction system. |
| NCEP (USA) | CFSv2 | 4 runs a day, lagged | 9 months | GFS atmosphere coupled to MOM4The operational American system since 2011 — the backbone of CPC's guidance and an NMME member in its own right. |
| ECCC (Canada) | CanSIPS-IC4 | 2 models × 10 | 12 months | CanESM5.1 and GEM5.2-NEMOCanada's system is itself a two-model ensemble, and it also flies as an NMME member. |
| JMA (Japan) | CPS3 | lagged | 7 months | JMA-GSM atmosphere coupled to MRI.COMJapan's operational seasonal system, and a C3S contributor. |
| BoM (Australia) | ACCESS-S2 | 99 | 9 months | UM atmosphere coupled to NEMO/CICEShares the Met Office's model core; the reference system for the Southern Hemisphere. |
| NASA GMAO | GEOS-S2S | ~4 a week | 9 months | GEOS atmosphere coupled to MOM5An NMME member, with a strong emphasis on assimilating the ocean it starts from. |
| NOAA GFDL | SPEAR | ~15–30 | 12 months | AM4 atmosphere coupled to MOM6GFDL's seasonal-to-decadal system, and the model behind much of the NMME's hurricane-season signal. |
| NCAR / Miami | CESM1 · CCSM4 | 10–20 | 12 months | CAM5 atmosphere coupled to POP2The academic members of the NMME — the reason the ensemble spans more than the operational centres' own model heritage. |
What they all have in common
The shape of a seasonal systemCoupled
Ocean and atmosphere together
Every system here runs an atmospheric model coupled to a dynamic ocean, because at a season's range the ocean is not a boundary condition — it is half the forecast. Most add sea ice and an interactive land surface for the same reason.
Hindcast
Decades of re-runs
Each system is re-run over 20 to 36 past years, several members per year. That hindcast set is what defines the model's own climatology, what the live run is expressed as an anomaly against, and what all published skill scores are measured on.
Monthly
A cadence, not a stream
Unlike medium-range NWP, seasonal systems publish on a monthly calendar. The interesting quantity is often not the newest run but the change from the previous one for the same target month.
The two great ensembles
How the multi-model archives are builtNMME — North American Multi-Model Ensemble
A research-to-operations collaboration between NOAA, NASA, GFDL, NCAR, the University of Miami and Environment and Climate Change Canada. Each participating model contributes a fixed monthly ensemble, all on a common 1° grid, all with a matched hindcast period. Around a hundred members between them, published near the start of each month and archived openly by NOAA CPC and the IRI Data Library. It is the backbone of the American seasonal outlook and the reason CPC can talk about model agreement rather than model output.
C3S — the Copernicus multi-system
The European equivalent, and administratively the more ambitious one: eight centres from three continents, contractually committed to delivering a fixed set of variables, on a fixed grid, on the 5th of every month, with a hindcast period they all share. It is distributed through the Climate Data Store, free, with the graphical products published at the same moment. If you want to know whether an outlook is a signal or one centre's opinion, this is the archive that answers it.
The two overlap — NCEP's CFSv2 and Canada's CanSIPS fly in both — which is convenient, because it means the American and European views of a season can be compared on genuinely common ground.