ECMWF vs UK Met Office seasonal forecasts
How ECMWF SEAS5 and UK Met Office GloSea6 differ, how their current seasonal outlooks compare over Europe, and how to open both on one shared map.
Two different ensemble designs
ECMWF’s SEAS5 and the Met Office’s GloSea6 are the two most widely quoted seasonal systems in Europe, and they are built differently. SEAS5 couples the ECMWF Integrated Forecasting System atmosphere, at roughly 36 km, to the NEMO ocean at a quarter degree and an interactive sea-ice model. All 51 members start on the first of the month, a so-called burst ensemble, and the model climate comes from a 25-member hindcast. GloSea6 is built on the HadGEM3 coupled model, with an atmosphere of roughly 60 km in mid-latitudes and the same quarter-degree ocean family. Instead of one large start it runs a handful of members every day and assembles a lagged ensemble from the most recent start dates, so its members are spread across the preceding weeks.
On this site SEAS5 is C3S system 51 and GloSea6 is system 610. See the ECMWF and UK Met Office model pages for member counts and the hindcast used for probabilities.
What the comparison holds fixed
- Both systems are shown on the same 1° C3S grid, so no regridding is done here.
- Both anomalies are departures from each system’s own 1993–2016 hindcast climatology, which removes each model’s mean bias. See why the reference period matters.
- Both panels share one valid period and one colour scale. For a lagged ensemble, the valid month is aligned on the nominal initialization month that C3S assigns.
- Probabilities for both are derived with the same empirical-quantile method from members and hindcasts.
Europe temperature: the September 2026 runs side by side
| Valid month | Lead month | ECMWF (system 51) | UK Met Office (system 610) | UK Met Office − ECMWF |
|---|---|---|---|---|
| September 2026 | 1 | +1.22 | +0.99 | −0.23 |
| October 2026 | 2 | +0.66 | +1.06 | +0.40 |
| November 2026 | 3 | +0.74 | +1.24 | +0.50 |
| December 2026 | 4 | +0.99 | +1.21 | +0.22 |
| January 2027 | 5 | +0.93 | +1.30 | +0.37 |
| February 2027 | 6 | +0.65 | +1.21 | +0.56 |
Europe average: 35°N–71°N, 11°W–40°E, land and sea cells, area-weighted. Values are ensemble-mean anomalies relative to each system’s own 1993–2016 hindcast climatology, computed from the same 1° grids the maps draw. Data imported from the C3S catalogue on 2026-09-12.
Reading the table
- An area average hides the pattern. Two systems can agree on the Europe mean while placing the warmest anomaly in different places; the maps are where that shows.
- A difference smaller than 0.25 °C sits inside one colour band on the map and should not be over-read. Differences in sign are more informative than differences in magnitude.
- The ensemble mean says nothing about spread. Switch the forecast type to a tercile probability to see how many members back each signal; see what an ensemble-mean anomaly shows.
- Against a 1993–2016 baseline, positive temperature anomalies are the expected background over Europe. Compare their strength between the two systems rather than treating any positive value as a signal.
- Neither value is a verified forecast. Both centres publish skill maps; European seasonal skill is modest, especially beyond lead 3. See why skill is low over Europe.