The Polar Vortex
Twenty kilometres up, one ring of wind decides how much of the Arctic's cold the mid-latitudes actually get. When it breaks down, the surface follows within a fortnight and stays followed for a month. This route asks whether a seasonal model can see that coming — and publishes the answer either way.
Major warming
This model's normal
What actually happens
Provisional
What SEAS5 says, initialised October 2026
When, if it happens
| Month | Share of members whose warming falls here | |
|---|---|---|
| December 2026 | 20% | |
| January 2027 | 53% | |
| February 2027 | 29% | |
| March 2027 | 16% |
The QBO underneath
| Level | DJF ensemble mean | Member range | Westerly in |
|---|---|---|---|
| 10 hPa | -21.9 | -24.5 to -20.0 | 0% |
| 30 hPa | +7.8 | +5.5 to +10.4 | 100% |
| 50 hPa | +4.9 | +3.5 to +7.1 | 100% |
| 100 hPa | +2.4 | +0.2 to +4.2 | 100% |
At 30 hPa the ensemble puts the coming winter in the westerly phase, at the 70th percentile of the 67 winters in the ERA5 record. Note the agreement: every member on the same side of zero. The QBO is a slow, nearly deterministic oscillation, and it is the one thing in this whole route a seasonal model genuinely knows seven months out.
Which matters because of Holton–Tan: warmings are more common when the equatorial wind is easterly. In the observed record, westerly winters had a major warming in 12 of 28 (43%), against 54% overall — so this phase leans toward fewer warmings. Lean gently: a two-sided Fisher exact test on that split gives p = 0.15. The direction is textbook; the size, from sixty-odd winters, is not a number to plan around. The catalogue behind it →
What it delivers at the ground
These are raw model counts, and they are not calibrated. There is no reforecast behind the surface fields, so there are no tercile probabilities, no correction for the model's own bias, and no skill score — which is why this is a section of route 13 and not a route of its own. Read the numbers against each other, not against normal: the spread between members is real information, the absolute level carries whatever bias SEAS5 has and nothing here removes it. Every threshold is an absolute physical one for the same reason — a percentile would need a climatology this route does not have, and would hide the bias inside the definition instead of leaving it visible.
| Region | Nov 2026 | Dec 2026 | Jan 2027 | Feb 2027 | Mar 2027 | |
|---|---|---|---|---|---|---|
| US Northeast | Nights below −10 °C | 0 | 0 0–1 | 0 0–5 | 1 0–4 | 0 0–1 |
| Nights below 0 °C | 1 0–4 | 13 5–19 | 22 14–28 | 21 14–26 | 13 7–20 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 0–2 | 1 0–3 | 1 0–2 | 1 0–2 | |
| Snowfall, mm water equivalent | 6 2–13 | 20 13–35 | 29 16–50 | 25 16–42 | 20 11–32 | |
| US Midwest | Nights below −10 °C | 0 | 2 0–8 | 10 2–20 | 7 1–16 | 0 0–6 |
| Nights below 0 °C | 10 3–17 | 26 20–31 | 31 27–31 | 27 23–28 | 20 13–27 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 0–1 | 0 0–1 | 0 0–2 | 0 0–1 | |
| Snowfall, mm water equivalent | 10 4–21 | 22 12–34 | 26 14–37 | 25 12–36 | 16 10–28 | |
| US Southeast | Nights below −10 °C | 0 | 0 | 0 | 0 | 0 |
| Nights below 0 °C | 0 | 0 | 0 | 0 | 0 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 | 0 | 0 | 0 | |
| Snowfall, mm water equivalent | 0 0–0 | 1 0–3 | 3 0–7 | 2 0–5 | 0 0–2 | |
| US West | Nights below −10 °C | 0 | 0 | 0 0–3 | 0 | 0 |
| Nights below 0 °C | 5 0–12 | 18 12–28 | 25 18–30 | 23 15–28 | 15 7–20 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 | 0 0–1 | 0 | 0 | |
| Snowfall, mm water equivalent | 12 6–24 | 23 16–32 | 31 21–48 | 28 21–39 | 26 17–35 | |
| UK & Ireland | Nights below −10 °C | 0 | 0 | 0 | 0 | 0 |
| Nights below 0 °C | 0 | 0 | 0 0–3 | 0 0–3 | 0 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 | 0 | 0 | 0 | |
| Snowfall, mm water equivalent | 0 0–2 | 2 1–5 | 5 1–11 | 4 2–13 | 3 1–9 | |
| NW Europe | Nights below −10 °C | 0 | 0 | 0 0–5 | 0 0–5 | 0 |
| Nights below 0 °C | 5 1–14 | 13 2–21 | 20 8–30 | 19 9–27 | 14 6–22 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 | 0 0–1 | 0 0–1 | 0 | |
| Snowfall, mm water equivalent | 10 4–20 | 16 7–25 | 21 10–41 | 21 12–37 | 18 8–28 | |
| Scandinavia | Nights below −10 °C | 0 | 0 0–5 | 6 0–18 | 6 0–18 | 0 0–6 |
| Nights below 0 °C | 18 7–25 | 27 19–30 | 31 27–31 | 28 26–28 | 30 27–31 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 | 0 | 0 | 0 | |
| Snowfall, mm water equivalent | 24 14–32 | 36 27–43 | 41 32–50 | 37 26–46 | 33 24–44 | |
| Eastern Europe | Nights below −10 °C | 0 | 0 0–7 | 5 0–17 | 5 0–16 | 0 0–4 |
| Nights below 0 °C | 10 3–20 | 22 12–29 | 28 23–31 | 27 23–28 | 22 15–29 | |
| Days with ≥5 mm w.e. of snow | 0 | 0 | 0 | 0 | 0 | |
| Snowfall, mm water equivalent | 11 4–21 | 19 10–30 | 31 17–41 | 26 18–38 | 20 9–29 | |
| Nor'easter corridor (US East Coast and the western Atlantic) | Days below 980 hPa | 0 0–1 | 1 0–3 | 2 0–4 | 2 0–4 | 2 0–4 |
| North Atlantic storm track | Days below 980 hPa | 11 5–17 | 14 7–25 | 15 8–20 | 10 5–15 | 9 3–15 |
| North Pacific storm track | Days below 980 hPa | 17 11–23 | 26 19–29 | 27 21–31 | 23 14–26 | 21 14–26 |
| US West Coast (127.5°W) | Days with IVT ≥ 500 kg m⁻¹ s⁻¹ | 5 2–9 | 6 4–10 | 7 3–12 | 5 1–8 | 3 1–6 |
| Iberia to Britain (12.5°W) | Days with IVT ≥ 500 kg m⁻¹ s⁻¹ | 10 5–15 | 11 5–15 | 9 5–13 | 7 3–11 | 5 2–8 |
Median of the 51 members, with the 10th–90th percentile beside it. Cold and snow counts are taken on the area mean of each box, so a night below −10 °C is a regional outbreak rather than one cold grid cell; storms are the box minimum pressure, because a deep low is a local object and averaging it away would be exactly wrong. Atmospheric rivers are integrated vapour transport across a landfall gate, computed through five pressure levels rather than from column moisture and a surface wind.
Every winter it has been asked about
One winter cannot separate the model was wrong about that winter from the model is always like this. This is the same measurement made over 33 winters and 99 initialisations, which can.
| Initialised | Winters | Any reversal in the window | Confirmed warmings only | DJF U60 | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Obs. | Model P | BSS | ROC | Obs. | Model P | BSS | ROC | ACC | Bias | RMSE | ||
| 1 August | 33 | 52% | 68% | -0.059 | 0.52 | 33% | 50% | -0.093 | 0.500.28–0.71 · p 0.50 | +0.25 | -4.2 | 11.3 |
| 1 September | 33 | 76% | 85% | +0.038 | 0.62 | 55% | 71% | -0.030 | 0.580.38–0.78 · p 0.22 | +0.29 | -4.9 | 11.5 |
| 1 October | 33 | 55% | 77% | -0.173 | 0.53 | 55% | 77% | -0.173 | 0.530.33–0.73 · p 0.37 | +0.13 | -4.9 | 11.7 |
ROC area is the one to read first. It asks whether the model put a higher probability on the winters that had a warming than on the winters that did not — 0.50 is no discrimination at all, and a forecast can look confident and score 0.50 by saying the same thing every year. BSS below zero means the probability was worse than issuing the climatological rate. Dimmed cells are the ones that failed. Bias and RMSE are on DJF-mean U60 in m s⁻¹, and a large negative bias means the model runs a weaker vortex than the atmosphere did.
The two halves are the same winters, counted twice, and the difference is a warning rather than a detail. The left half counts every reversal of the 10 hPa wind at 60°N inside the initialisation’s own window. A later initialisation runs that window further into spring, and a reversal in late March may be the vortex’s final warming for the year rather than a mid-winter one. Telling those apart needs ten westerly days back before 30 April, which a window ending in early April cannot supply, so they are marked provisional — 16 of them here. The right half drops them from both sides, observed and modelled.
1 October is the control. Its window runs to early May, past the 30 April recovery test, so every reversal it sees can be classified and it has nothing marked provisional — and its two halves come out identical, which is what says the reclassification above is arithmetic rather than opinion. Its fully resolved rate also matches what 1 September finds once the unconfirmable events are dropped, which is the second thing that would not line up if the flag were wrong.
Read the ranges, not the ranking. The best confirmed score on this page is 1 September at 0.58, whose bootstrap range over these winters is 0.38–0.78 and whose p against shuffled outcomes is 0.22 — a model carrying no information at all would score at least that high about 22 times in a hundred. No initialisation month clears the no-discrimination line. The gaps between the months are smaller than any one month’s own uncertainty, so this page does not rank them — an earlier version did, and was wrong twice.
| 1993/94 | 1 Aug | reforecast · 25 | 72% | 52% | ○ no | ✗ wrong side | +23.2 | +28.4 | -5.2 | 5/25 |
|---|---|---|---|---|---|---|---|---|---|---|
| 1993/94 | 1 Sep | reforecast · 25 | 100% | 88% | ○ no | ✗ wrong side | +21.2 | +28.4 | -7.2 | 4/25 |
| 1993/94 | 1 Oct | reforecast · 25 | 76% | 76% | ○ no | ✗ wrong side | +21.3 | +28.4 | -7.1 | 5/25 |
| 1994/95 | 1 Aug | reforecast · 25 | 68% | 36% | ○ no | ✗ wrong side | +27.1 | +31.0 | -3.9 | 8/25 |
| 1994/95 | 1 Sep | reforecast · 25 | 72% | 48% | ○ no | ✗ wrong side | +28.7 | +31.0 | -2.2 | 10/25 |
| 1994/95 | 1 Oct | reforecast · 25 | 68% | 68% | ○ no | ✗ wrong side | +28.7 | +31.0 | -2.3 | 9/25 |
| 1995/96 | 1 Aug | reforecast · 25 | 48% | 32% | ○ no | ✓ right side | +26.2 | +42.0 | -15.8 | 0/25 ⚠ |
| 1995/96 | 1 Sep | reforecast · 25 | 88% | 72% | ○ no | ✗ wrong side | +23.5 | +42.0 | -18.5 | 0/25 ⚠ |
| 1995/96 | 1 Oct | reforecast · 25 | 76% | 76% | ○ no | ✗ wrong side | +23.9 | +42.0 | -18.1 | 0/25 ⚠ |
| 1996/97 | 1 Aug | reforecast · 25 | 64% | 44% | ○ no | ✗ wrong side | +26.4 | +35.9 | -9.5 | 4/25 |
| 1996/97 | 1 Sep | reforecast · 25 | 80% | 68% | ○ no | ✗ wrong side | +23.8 | +35.9 | -12.0 | 1/25 |
| 1996/97 | 1 Oct | reforecast · 25 | 76% | 76% | ○ no | ✗ wrong side | +25.0 | +35.9 | -10.9 | 3/25 |
| 1997/98 | 1 Aug | reforecast · 25 | 84% | 52% | ○ no | ✗ wrong side | +23.1 | +20.2 | +2.9 | 17/25 |
| 1997/98 | 1 Sep | reforecast · 25 | 88% | 76% | ◐ yes* | ✓ right side | +22.9 | +20.2 | +2.7 | 17/25 |
| 1997/98 | 1 Oct | reforecast · 25 | 84% | 84% | ○ no | ✗ wrong side | +21.5 | +20.2 | +1.3 | 11/25 |
| 1998/99 | 1 Aug | reforecast · 25 | 80% | 80% | ◐ yes* | ✓ right side | +21.9 | +19.4 | +2.5 | 17/25 |
| 1998/99 | 1 Sep | reforecast · 25 | 92% | 80% | ● yes | ✓ right side | +22.9 | +19.4 | +3.5 | 16/25 |
| 1998/99 | 1 Oct | reforecast · 25 | 68% | 68% | ● yes | ✓ right side | +23.1 | +19.4 | +3.7 | 14/25 |
| 1999/00 | 1 Aug | reforecast · 25 | 60% | 44% | ○ no | ✗ wrong side | +26.3 | +43.0 | -16.7 | 0/25 ⚠ |
| 1999/00 | 1 Sep | reforecast · 25 | 72% | 60% | ● yes | ✓ right side | +25.7 | +43.0 | -17.3 | 1/25 |
| 1999/00 | 1 Oct | reforecast · 25 | 76% | 76% | ● yes | ✓ right side | +24.9 | +43.0 | -18.1 | 1/25 |
| 2000/01 | 1 Aug | reforecast · 25 | 80% | 48% | ◐ yes* | ✓ right side | +22.5 | +16.5 | +6.0 | 21/25 |
| 2000/01 | 1 Sep | reforecast · 25 | 88% | 72% | ● yes | ✓ right side | +22.4 | +16.5 | +5.8 | 20/25 |
| 2000/01 | 1 Oct | reforecast · 25 | 68% | 68% | ● yes | ✓ right side | +25.2 | +16.5 | +8.7 | 22/25 |
| 2001/02 | 1 Aug | reforecast · 25 | 64% | 48% | ● yes | ✓ right side | +26.5 | +15.3 | +11.2 | 24/25 |
| 2001/02 | 1 Sep | reforecast · 25 | 84% | 72% | ● yes | ✓ right side | +24.5 | +15.3 | +9.2 | 24/25 |
| 2001/02 | 1 Oct | reforecast · 25 | 56% | 56% | ● yes | ✓ right side | +24.0 | +15.3 | +8.7 | 19/25 |
| 2002/03 | 1 Aug | reforecast · 25 | 56% | 32% | ● yes | ✓ right side | +27.6 | +24.0 | +3.5 | 17/25 |
| 2002/03 | 1 Sep | reforecast · 25 | 80% | 56% | ● yes | ✓ right side | +26.0 | +24.0 | +1.9 | 16/25 |
| 2002/03 | 1 Oct | reforecast · 25 | 76% | 76% | ● yes | ✓ right side | +25.4 | +24.0 | +1.4 | 10/25 |
| 2003/04 | 1 Aug | reforecast · 25 | 56% | 40% | ● yes | ✓ right side | +25.6 | +14.2 | +11.4 | 22/25 |
| 2003/04 | 1 Sep | reforecast · 25 | 84% | 68% | ● yes | ✓ right side | +21.9 | +14.2 | +7.7 | 22/25 |
| 2003/04 | 1 Oct | reforecast · 25 | 56% | 56% | ● yes | ✓ right side | +26.4 | +14.2 | +12.1 | 24/25 |
| 2004/05 | 1 Aug | reforecast · 25 | 72% | 56% | ○ no | ✗ wrong side | +24.8 | +43.7 | -18.8 | 0/25 ⚠ |
| 2004/05 | 1 Sep | reforecast · 25 | 76% | 60% | ◐ yes* | ✓ right side | +23.7 | +43.7 | -19.9 | 0/25 ⚠ |
| 2004/05 | 1 Oct | reforecast · 25 | 80% | 80% | ○ no | ✗ wrong side | +21.8 | +43.7 | -21.9 | 0/25 ⚠ |
| 2005/06 | 1 Aug | reforecast · 25 | 88% | 64% | ● yes | ✓ right side | +22.6 | +12.7 | +9.9 | 25/25 |
| 2005/06 | 1 Sep | reforecast · 25 | 92% | 72% | ● yes | ✓ right side | +22.7 | +12.7 | +10.0 | 23/25 |
| 2005/06 | 1 Oct | reforecast · 25 | 88% | 88% | ● yes | ✓ right side | +23.4 | +12.7 | +10.7 | 24/25 |
| 2006/07 | 1 Aug | reforecast · 25 | 68% | 36% | ◐ yes* | ✓ right side | +25.4 | +34.5 | -9.2 | 4/25 |
| 2006/07 | 1 Sep | reforecast · 25 | 84% | 76% | ● yes | ✓ right side | +26.2 | +34.5 | -8.4 | 5/25 |
| 2006/07 | 1 Oct | reforecast · 25 | 76% | 76% | ● yes | ✓ right side | +23.6 | +34.5 | -11.0 | 0/25 ⚠ |
| 2007/08 | 1 Aug | reforecast · 25 | 80% | 48% | ◐ yes* | ✓ right side | +24.3 | +32.6 | -8.3 | 3/25 |
| 2007/08 | 1 Sep | reforecast · 25 | 76% | 64% | ● yes | ✓ right side | +26.7 | +32.6 | -5.9 | 5/25 |
| 2007/08 | 1 Oct | reforecast · 25 | 76% | 76% | ● yes | ✓ right side | +25.9 | +32.6 | -6.7 | 4/25 |
| 2008/09 | 1 Aug | reforecast · 25 | 56% | 48% | ● yes | ✓ right side | +25.2 | +20.5 | +4.8 | 16/25 |
| 2008/09 | 1 Sep | reforecast · 25 | 80% | 68% | ● yes | ✓ right side | +22.8 | +20.5 | +2.4 | 14/25 |
| 2008/09 | 1 Oct | reforecast · 25 | 80% | 80% | ● yes | ✓ right side | +22.4 | +20.5 | +2.0 | 15/25 |
| 2009/10 | 1 Aug | reforecast · 25 | 64% | 52% | ◐ yes* | ✓ right side | +24.0 | +23.7 | +0.3 | 12/25 |
| 2009/10 | 1 Sep | reforecast · 25 | 96% | 92% | ◐ yes* | ✓ right side | +21.3 | +23.7 | -2.5 | 9/25 |
| 2009/10 | 1 Oct | reforecast · 25 | 88% | 88% | ● yes | ✓ right side | +20.0 | +23.7 | -3.7 | 7/25 |
| 2010/11 | 1 Aug | reforecast · 25 | 60% | 40% | ○ no | ✗ wrong side | +28.7 | +39.5 | -10.8 | 5/25 |
| 2010/11 | 1 Sep | reforecast · 25 | 72% | 60% | ○ no | ✗ wrong side | +26.9 | +39.5 | -12.7 | 1/25 |
| 2010/11 | 1 Oct | reforecast · 25 | 68% | 68% | ○ no | ✗ wrong side | +24.1 | +39.5 | -15.4 | 2/25 |
| 2011/12 | 1 Aug | reforecast · 25 | 44% | 40% | ○ no | ✓ right side | +26.0 | +24.8 | +1.3 | 16/25 |
| 2011/12 | 1 Sep | reforecast · 25 | 80% | 64% | ○ no | ✗ wrong side | +25.3 | +24.8 | +0.5 | 11/25 |
| 2011/12 | 1 Oct | reforecast · 25 | 72% | 72% | ○ no | ✗ wrong side | +25.0 | +24.8 | +0.2 | 11/25 |
| 2012/13 | 1 Aug | reforecast · 25 | 64% | 48% | ● yes | ✓ right side | +27.1 | +12.6 | +14.5 | 25/25 |
| 2012/13 | 1 Sep | reforecast · 25 | 92% | 76% | ● yes | ✓ right side | +25.0 | +12.6 | +12.4 | 25/25 |
| 2012/13 | 1 Oct | reforecast · 25 | 80% | 80% | ● yes | ✓ right side | +23.0 | +12.6 | +10.4 | 23/25 |
| 2013/14 | 1 Aug | reforecast · 25 | 84% | 64% | ○ no | ✗ wrong side | +24.0 | +39.8 | -15.9 | 0/25 ⚠ |
| 2013/14 | 1 Sep | reforecast · 25 | 88% | 76% | ◐ yes* | ✓ right side | +23.7 | +39.8 | -16.1 | 0/25 ⚠ |
| 2013/14 | 1 Oct | reforecast · 25 | 68% | 68% | ○ no | ✗ wrong side | +27.0 | +39.8 | -12.9 | 2/25 |
| 2014/15 | 1 Aug | reforecast · 25 | 80% | 52% | ○ no | ✗ wrong side | +24.0 | +29.1 | -5.0 | 8/25 |
| 2014/15 | 1 Sep | reforecast · 25 | 92% | 84% | ◐ yes* | ✓ right side | +20.7 | +29.1 | -8.4 | 1/25 |
| 2014/15 | 1 Oct | reforecast · 25 | 100% | 100% | ○ no | ✗ wrong side | +20.3 | +29.1 | -8.8 | 3/25 |
| 2015/16 | 1 Aug | reforecast · 25 | 80% | 60% | ○ no | ✗ wrong side | +23.8 | +45.2 | -21.5 | 0/25 ⚠ |
| 2015/16 | 1 Sep | reforecast · 25 | 96% | 84% | ◐ yes* | ✓ right side | +22.4 | +45.2 | -22.8 | 0/25 ⚠ |
| 2015/16 | 1 Oct | reforecast · 25 | 88% | 88% | ○ no | ✗ wrong side | +23.9 | +45.2 | -21.3 | 0/25 ⚠ |
| 2016/17 | 1 Aug | reforecast · 25 | 72% | 68% | ○ no | ✗ wrong side | +22.5 | +29.7 | -7.2 | 4/25 |
| 2016/17 | 1 Sep | reforecast · 25 | 92% | 84% | ○ no | ✗ wrong side | +20.3 | +29.7 | -9.4 | 5/25 |
| 2016/17 | 1 Oct | reforecast · 25 | 76% | 76% | ○ no | ✗ wrong side | +22.3 | +29.7 | -7.3 | 5/25 |
| 2017/18 | 1 Aug | real-time · 51 | 80% | 67% | ◐ yes* | ✓ right side | +21.7 | +24.0 | -2.3 | 18/51 |
| 2017/18 | 1 Sep | real-time · 51 | 92% | 82% | ● yes | ✓ right side | +20.9 | +24.0 | -3.0 | 12/51 |
| 2017/18 | 1 Oct | real-time · 51 | 84% | 84% | ● yes | ✓ right side | +20.4 | +24.0 | -3.6 | 15/51 |
| 2018/19 | 1 Aug | real-time · 51 | 65% | 49% | ● yes | ✓ right side | +25.6 | +15.5 | +10.1 | 45/51 |
| 2018/19 | 1 Sep | real-time · 51 | 90% | 76% | ● yes | ✓ right side | +23.3 | +15.5 | +7.8 | 44/51 |
| 2018/19 | 1 Oct | real-time · 51 | 75% | 75% | ● yes | ✓ right side | +26.3 | +15.5 | +10.8 | 47/51 |
| 2019/20 | 1 Aug | real-time · 51 | 59% | 39% | ○ no | ✗ wrong side | +26.4 | +41.9 | -15.4 | 1/51 |
| 2019/20 | 1 Sep | real-time · 51 | 76% | 55% | ○ no | ✗ wrong side | +28.2 | +41.9 | -13.7 | 3/51 |
| 2019/20 | 1 Oct | real-time · 51 | 80% | 80% | ○ no | ✗ wrong side | +26.9 | +41.9 | -15.0 | 2/51 |
| 2020/21 | 1 Aug | real-time · 51 | 63% | 47% | ● yes | ✓ right side | +22.8 | +18.8 | +4.0 | 36/51 |
| 2020/21 | 1 Sep | real-time · 51 | 80% | 61% | ● yes | ✓ right side | +25.2 | +18.8 | +6.4 | 36/51 |
| 2020/21 | 1 Oct | real-time · 51 | 76% | 76% | ● yes | ✓ right side | +24.1 | +18.8 | +5.3 | 33/51 |
| 2021/22 | 1 Aug | real-time · 51 | 71% | 65% | ○ no | ✗ wrong side | +22.9 | +45.9 | -23.0 | 0/51 ⚠ |
| 2021/22 | 1 Sep | real-time · 51 | 88% | 65% | ◐ yes* | ✓ right side | +21.7 | +45.9 | -24.2 | 0/51 ⚠ |
| 2021/22 | 1 Oct | real-time · 51 | 82% | 82% | ○ no | ✗ wrong side | +22.4 | +45.9 | -23.5 | 0/51 ⚠ |
| 2022/23 | 1 Aug | real-time · 51 | 69% | 57% | ◐ yes* | ✓ right side | +23.7 | +30.7 | -7.0 | 5/51 |
| 2022/23 | 1 Sep | real-time · 51 | 73% | 65% | ● yes | ✓ right side | +24.9 | +30.7 | -5.8 | 14/51 |
| 2022/23 | 1 Oct | real-time · 51 | 80% | 80% | ● yes | ✓ right side | +25.5 | +30.7 | -5.2 | 11/51 |
| 2023/24 | 1 Aug | real-time · 51 | 75% | 55% | ● yes | ✓ right side | +20.3 | +21.4 | -1.1 | 21/51 |
| 2023/24 | 1 Sep | real-time · 51 | 96% | 84% | ● yes | ✓ right side | +22.8 | +21.4 | +1.4 | 27/51 |
| 2023/24 | 1 Oct | real-time · 51 | 84% | 84% | ● yes | ✓ right side | +24.8 | +21.4 | +3.4 | 29/51 |
| 2024/25 | 1 Aug | real-time · 51 | 63% | 51% | ○ no | ✗ wrong side | +26.4 | +49.0 | -22.6 | 0/51 ⚠ |
| 2024/25 | 1 Sep | real-time · 51 | 69% | 59% | ◐ yes* | ✓ right side | +26.7 | +49.0 | -22.3 | 0/51 ⚠ |
| 2024/25 | 1 Oct | real-time · 51 | 63% | 63% | ○ no | ✗ wrong side | +25.7 | +49.0 | -23.3 | 0/51 ⚠ |
| 2025/26 | 1 Aug | real-time · 51 | 67% | 53% | ◐ yes* | ✓ right side | +23.2 | +24.8 | -1.6 | 21/51 |
| 2025/26 | 1 Sep | real-time · 51 | 90% | 80% | ● yes | ✓ right side | +22.8 | +24.8 | -2.0 | 20/51 |
| 2025/26 | 1 Oct | real-time · 51 | 88% | 88% | ● yes | ✓ right side | +21.7 | +24.8 | -3.1 | 14/51 |
One row per initialisation per winter; click any column heading to sort by it, and again to reverse. yes* marks an observed reversal the window could not resolve — it may be the vortex's annual final warming rather than a mid-winter one. Call is whether the any-reversal probability fell on the right side of a half. Every colour has its number printed in the cell: on the vortex columns warm means weaker — the direction of the physics, where a weak vortex is a warm polar cap. ⚠ marks a winter in which no member was as strong as the atmosphere. Reforecast rows carry 25 members and real-time rows 51: scored the same way, but not the same instrument.
How this is built
A major sudden stratospheric warming is defined here exactly as Charlton and Polvani (2007) define it, which is the definition the field settled on: the zonal-mean zonal wind at 10 hPa and 60°N turns easterly between 1 November and 31 March; two events must be separated by twenty westerly days; and the wind must return westerly for ten consecutive days before 30 April, which is what separates a midwinter warming from the vortex's annual spring death. That last rule is quietly load-bearing — without it the record shows a warming in every single winter and the quantity stops meaning anything.
The indices are not computed here. They come from polarvortex.earth's strat_indices, which defines them once for whatever produced the field — ERA5 out of a cloud archive, ECMWF's analysis out of a GRIB, and SEAS5 out of the Copernicus seasonal originals. Three roads, one U60.
Anomalies are against the model's own lead-day climatology, never against ERA5. SEAS5's stratosphere drifts with lead time, and measured against a reanalysis that drift would appear in every forecast as a slow trend in vortex strength that is not in the atmosphere at all.
Model and observations are counted over the same window. An August initialisation runs out in early March and physically cannot see a late-March warming; charging it for that miss would be scoring it on a question it was never asked. Where a forecast ends before an event can demonstrate its recovery, the event is reported as provisional rather than dropped or silently counted.
Data: ECMWF SEAS5 (system 51) daily geopotential, temperature and wind at 10, 50 and 100 hPa from the Copernicus seasonal originals, 51 members forward and 25 × 24 years of reforecast behind; ERA5 for verification. Not an official forecast product.