The same calendar week, as called by each of the last daily runs — newest on the left. Read a row across: where the numbers hold steady the signal has been there all week; where the newest run jumps, the model has just changed its mind. Rows marked WK 3 and WK 4 are the fortnight where a forecast stops being weather.
| Week of | Decade | 15 Oct | 14 Oct | 13 Oct | 12 Oct | 11 Oct | 10 Oct |
|---|---|---|---|---|---|---|---|
| 19 Oct WK 1 | +0.7 | +1.3 | +1.3 | +1.8 | +2.0 | +1.4 | +1.3 |
| 26 Oct WK 2 | +0.7 | +1.1 | +1.0 | +0.6 | +1.0 | +0.8 | +0.9 |
| 2 Nov WK 3 | +0.7 | +0.7 | +0.6 | +0.6 | +0.7 | +0.9 | +1.0 |
| 9 Nov WK 4 | +0.7 | +0.8 | +0.8 | +0.6 | +0.5 | +1.0 | +1.1 |
| 16 Nov WK 5 | +0.7 | +0.8 | +0.8 | +0.5 | +0.7 | +0.8 | +0.7 |
| 23 Nov WK 6 | +0.7 | +0.7 | · | · | · | · | · |
| Week of | Decade | 15 Oct | 14 Oct | 13 Oct | 12 Oct | 11 Oct | 10 Oct |
|---|---|---|---|---|---|---|---|
| 19 Oct WK 1 | +0.8 | +1.0 | +1.0 | +0.2 | -0.2 | +0.1 | +0.2 |
| 26 Oct WK 2 | +0.8 | +1.4 | +0.7 | -0.6 | -0.3 | +0.1 | +0.3 |
| 2 Nov WK 3 | +0.8 | +0.8 | +0.6 | -0.2 | +0.2 | +0.6 | +0.5 |
| 9 Nov WK 4 | +0.8 | +0.6 | +0.7 | +0.3 | +0.4 | +0.7 | +0.3 |
| 16 Nov WK 5 | +0.8 | +0.3 | +0.4 | +0.1 | +0.5 | +0.6 | +0.5 |
| 23 Nov WK 6 | +0.8 | +0.3 | · | · | · | · | · |
| Week of | Decade | 15 Oct | 14 Oct | 13 Oct | 12 Oct | 11 Oct | 10 Oct |
|---|---|---|---|---|---|---|---|
| 19 Oct WK 1 | +0.2 | +0.2 | +0.1 | -0.6 | -0.1 | -0.4 | +0.3 |
| 26 Oct WK 2 | +0.3 | +0.1 | +0.1 | +0.4 | +0.8 | +0.4 | +0.8 |
| 2 Nov WK 3 | +0.7 | +0.2 | +0.1 | +0.5 | +0.3 | +0.4 | +0.4 |
| 9 Nov WK 4 | +0.7 | +0.8 | +0.7 | +0.3 | +0.7 | +0.5 | +0.8 |
| 16 Nov WK 5 | +0.7 | +0.2 | +0.6 | +0.4 | +0.8 | +0.4 | +0.3 |
| 23 Nov WK 6 | +0.7 | +0.8 | · | · | · | · | · |
| Week of | Decade | 15 Oct | 14 Oct | 13 Oct | 12 Oct | 11 Oct | 10 Oct |
|---|---|---|---|---|---|---|---|
| 19 Oct WK 1 | +0.7 | +1.0 | +1.0 | +1.1 | +1.1 | +1.0 | +1.0 |
| 26 Oct WK 2 | +0.7 | +0.9 | +0.9 | +1.0 | +1.0 | +1.0 | +1.0 |
| 2 Nov WK 3 | +0.7 | +0.9 | +0.9 | +0.9 | +1.0 | +1.0 | +0.9 |
| 9 Nov WK 4 | +0.7 | +0.9 | +0.9 | +0.8 | +0.9 | +1.0 | +0.8 |
| 16 Nov WK 5 | +0.7 | +0.9 | +0.9 | +0.8 | +1.0 | +1.0 | +0.7 |
| 23 Nov WK 6 | +0.7 | +0.9 | · | · | · | · | · |
°C against ERA5 1991–2020, averaged over land (Antarctica excluded). Decade is what the last ten years’ climate alone would contribute — ERA5 2016–2025 minus 1991–2020 for the same weeks. A cell no warmer than that says the model expects an ordinary recent week, not a warm one. The numbers are published as runs.json.
Two weeks out, the atmosphere has forgotten its starting point; two months out, the ocean is still remembering. The subseasonal range sits between them, and it was long called the predictability desert for that reason — too far for the weather’s memory, too near for the ocean’s to dominate. What skill there is comes from a handful of slow processes, and a week-3 map is best read as a statement about which of them is active:
A pulse of tropical rainfall that travels east around the equator in 30 to 60 days. It is the largest source of skill at this range: its heating sets off wave trains that reach the mid-latitudes two to three weeks later, and a strong, well-forecast MJO is the clearest reason to trust week 3.
A sudden stratospheric warming takes one to three weeks to reach the ground, where it shifts the storm track south and favours cold outbreaks over northern Europe and North America. A forecast issued after the warming has the stratosphere’s memory on its side.
A block — a persistent high that diverts the storm track — can hold for weeks, and the regime it belongs to is more forecastable than any day inside it. The pressure anomaly map here is where a block shows up first.
Dry soil heats faster and rains less; wet soil does the opposite. Soil moisture and snow cover carry an anomaly forward by weeks, most of all in late spring and summer over the continents.
The slowest memory on the map. El Niño and La Niña shift tropical rainfall and with it the extratropical jets; at six weeks the ocean looks much like it does at week 1, and that constancy is exactly what the rest of the forecast leans on.
That is also why this page averages. A daily map at day 25 is one member’s weather smeared across a hundred others; a weekly mean keeps the part of the signal those processes hold in place, and a fortnight keeps more of it. For the seasonal view of the same drivers, see the SEAS5 Explorer; for the day-by-day members out to seven months, the Daily Seasonal Forecast.
ECMWF’s extended-range ensemble (EC46) is the medium-range system run on past the point where any single day can be forecast: 101 members, every day at 00 UTC, out to 46 days. This page draws its mean and its spread as Open-Meteo redistributes them, regridded to 0.25°, averaged into calendar weeks, Monday to Sunday UTC — ECMWF’s own convention for weekly means. This run covers 19–25 Oct to 23–29 Nov. Weeks rather than “days 1–7 of the run” because the runs are daily: counted from each run’s start, a week would move under the reader every morning and no two runs would ever forecast the same one. The fortnight axis is every rolling pair of weeks; weeks 3–4 is the one CPC issues and the range at which a forecast stops being weather.
Anomaly or amount. Temperature, rainfall, pressure and SST are drawn as anomalies against ERA5 1991–2020 — the WMO normal the rest of the site’s ERA5 work uses, turned into a smooth daily curve from the monthly normals so a week in late October is compared with late October and not with the October average. Two fields are the value itself: Absolute temperature, and Rain total — the millimetres the ensemble mean expects in the week, not the departure. A mean of 101 members smooths the heaviest falls away, so the total is the expected amount, not the wettest plausible one.
What the anomaly carries that ECMWF’s own charts do not. EC46 is verified against its own 20-year reforecast; that reforecast is not in these files, so this page cannot remove the model’s bias the way ECMWF does. Two things follow. The anomaly includes the model’s own bias for the place and season — largest in steep terrain, where the model’s mountains are not ERA5’s. And it includes the warming since 1991–2020, which a reforecast of recent years mostly would not. Measured for these weeks: the last decade alone runs +0.7 °C warmer than 1991–2020 over the world’s land, so a week coloured faintly warm is, to that extent, an ordinary week of the present climate. The run-to-run table prints that figure beside every region.
Does it drift? No reforecast means no way to remove the model’s drift — but the daily runs allow one check. The same calendar week is forecast by up to 21 runs at leads from zero to five weeks, and real weather cannot depend on how long ago the forecast was made. At a fixed target week the global-mean temperature anomaly changes by +0.011 °C per week of lead (median over the target weeks): no measurable drift. Sea-surface temperature does creep, +0.024 °C per week of lead — small, and in the global mean, so read the far weeks’ ocean warmth with that much salt. Published as drift.json.
The spread is the standard deviation across the members at each six-hourly step, averaged over the window. It is an upper bound on the spread of the weekly mean: a member that runs warm on Monday may run cool on Thursday, and those departures partly cancel in a week’s average, by an amount a mean and a spread cannot reveal. Read it as how much the members disagree at any given moment of that week. Rainfall’s spread stays in millimetres per six hours for the same reason — adding six-hourly spreads up to a weekly figure would state the bound as if it were the number.
No skill score. Every other route that can publish one does. This one cannot, for the same reason it cannot remove the bias: no reforecast, and six weeks of daily runs are nowhere near enough to verify against. Published skill for extended-range forecasts falls off steeply after week 2 and is concentrated where the drivers below are active — which is the honest way to read weeks 3 to 6 here: as a statement about those drivers, not about a given day.