The SEAS5 Explorer
ECMWF's seasonal system, seven months out. Pick a variable, a part of the world and a month — then look at how the same forecast has changed from one run to the next, which is the part that tells you what it is worth.
Initialised
Members
Leads
Frames
The map plate
9 variables · 10 views · 7 leads
How much warmer or cooler than the model's own climate each month is expected to run. The single most-read seasonal field, and the one the warming trend shows up in most plainly.
* The first lead is the initialisation's own month, so part of it has already happened. It is kept because the model produced it, not because it is a forecast for days that are over.
Run to run
4 initialisationsThe same target month, forecast from four successive initialisations. Where the four runs sit on top of each other the model is telling you something it has been sure of for months; where they scatter, the newest map is one opinion among four.
How to read this plate
Four things it is notBaseline
Not an anomaly from observed normals
Every anomaly here is a departure from SEAS5's own 1993–2016 hindcast climatology — the model re-run over past years. That is deliberate: a coupled model drifts, and measuring it against itself cancels the drift so the departure is like-for-like. It also means these numbers are not comparable with an ERA5 or 1991–2020 anomaly.
Ensemble
A mean, not a forecast
The gridded monthly product carries the ensemble mean of the 51 members and no spread. An ensemble mean is smoother and weaker than any member: a +1 °C mean anomaly does not mean +1 °C is the likely outcome, it means the balance of 51 possible seasons leans warm.
Niño 3.4
This is not the ONI
The Niño-3.4 value on the evolution board is SEAS5's SST anomaly against its own model climate. The Oceanic Niño Index is a different quantity — observed ERSSTv5, in a 30-year centred base period, as a three-month running mean. Reading one as the other will mislead you about the size of an event. Read the trajectory across initialisations, not the absolute figure.
Skill
The map does not tell you where it works
These fields are drawn everywhere with equal confidence and they should not be read that way. Tropical SST is the skilful part; mid-latitude summer rainfall is close to unforecastable at this range. Route 06 exists to put the verification next to the map, and until it is built that judgement is yours to make.
Provenance
Every step, namedECMWF SEAS5 (System 5) is the reference seasonal system: 51 members, seven monthly leads, a ~36 km atmosphere coupled to a quarter-degree NEMO ocean, released on the 5th of each month at 12 UTC. The monthly ensemble-mean product is redistributed by Open-Meteo under CC-BY-4.0 on the O320 octahedral reduced-Gaussian grid, both as means and as pre-computed anomalies against the system's own 1993–2016 hindcast.
We keep our own copy, because Open-Meteo retains only the most recent few initialisations and the run-to-run comparison on this page needs the history. Each init is regridded O320 → 0.25° by cached nearest-neighbour on the unit sphere and appended to a Zarr store chunked one whole field per (init, lead) — so a map is a single object read.
The maps on this page are rendered from that store at build time, not on request. Area means on the evolution board are cosine-latitude weighted and surface-masked, so a land domain never quietly averages in the sea.
Plate built from the initialisation of 1 August 2026; store attribution: ECMWF SEAS5 via Open-Meteo · CC-BY-4.0.