Daily from the ECMWF AIFS ensemble, against MERRA-2 and ERA5; the history from MERRA-2, 1980–2026
Stratosphere and the polar vortex
Whether the winter polar vortex is about to weaken, and what that has meant at the surface before. The forecast side follows the vortex winds, the planetary waves driving them and the heat flux carrying those waves upward, member by member through the 15-day AIFS ensemble. The history side is every winter and every major sudden warming since 1980 from MERRA-2, with only the statistically significant results shown. Background on each diagnostic is in the sudden stratospheric warming, E–P flux and eddy heat flux explainers. The tropospheric circulation is on Global circulation and jets.
← → step through the options · ↑ ↓ change figure · click a still to enlarge
| Central date | Shape | Depth | Lowest wind m/s, days 0–30 | Easterly spell days from day 0 | Cap warming 10 hPa, K | AO before → after days −20…−1, +1…+45 | ENSO | QBO |
|---|
Strong-vortex events (the opposite extreme)
| Date | Peak 10 hPa index | Strongest wind m/s, days 0–30 | AO before → after |
|---|
The explainer: what sudden stratospheric warmings measures, the equations and how to read the chart.
The stratospheric vortex is the slow variable that sets the odds for the surface weeks later, so it is worth watching before it moves. Top: the zonal-mean zonal wind at 10 hPa, 60°N — the standard sudden-stratospheric-warming diagnostic. A reversal to easterly here in winter is a major SSW (the original WMO definition also asked for a reversed 60–90° temperature gradient), which is why the zero line is drawn heavy and the member fraction crossing it is called out. Middle: the same wind at 100 hPa. Anomalies that reach this far down are the ones with a path to the troposphere — a 10 hPa event that never appears at 100 hPa usually never reaches the surface. Bottom: the 100 hPa polar-cap (65–90°N) height anomaly, positive for a weak or displaced vortex; this is the field that actually leads the AO/NAO response. Blue traces are AIFS-ENS members, orange their mean, dashed member 0. The 100 hPa height is member-0-only because ECMWF open data publishes no perturbed geopotential at that level.
ECMWF AIFS-ENS zonal wind and geopotential at 10/100 hPa (open data, CC BY 4.0) · reference: MERRA-2 1980–2026 day-of-year 10th–90th percentiles (NASA GMAO) · analysis tail: ERA5 · updated daily
The vortex as it actually looks, rather than as an index. Wind speed is
shaded with streamlines over it at 10 hPa. This is where the vortex lives and where an SSW is declared — the level the WMO diagnostic is evaluated on. The red contour
is u = 0: where the westerlies end in the zonal sense (not the PV-gradient
vortex edge), and the outer bound of the corridor $0<\bar{u} ECMWF AIFS-ENS member 0 (open data, CC BY 4.0) · 10 hPa speed shaded, streamlines overlaid, red contour u = 0 · fixed colour scale · analysis plus 12-hourly steps to day 15 · updated each 00/12Z cycle
The explainer: what the E–P flux and its divergence measures, the equations and how to read the chart.
Where the torque budget shows how the Earth exchanges angular momentum with the atmosphere at the surface, the Eliassen–Palm flux shows how the waves redistribute it internally — the same quasi-geostrophic diagnostic used to diagnose sudden stratospheric warmings, computed live from AIFS-ENS and stepped through the full 15-day forecast.
ECMWF AIFS-ENS ensemble — member 0 + 25 perturbed members, per-member fluxes averaged (open data, CC BY 4.0), 14 levels, day 0–15 · QG E–P flux, wavenumbers 1–3, global-mean static stability · poleward of 82° masked · rendered in Julia (CairoMakie) · updated each 00/12Z cycle
The zonal wavenumber-1 component of geopotential height, 100 hPa beside 500 hPa for one hemisphere at a time — the Hemisphere selector switches between them, and the loop opens on whichever one currently carries the larger wave. The two levels share a frame because that is the comparison that carries the physics: 100 hPa is what the vortex feels, 500 hPa is the tropospheric source, and a ridge that leans westward with height is actively driving the vortex while one sitting over the same longitude at both levels is not. Each level keeps its own colourbar — wave amplitude grows with height, so a shared scale would flatten the lower one. Because only k = 1 is retained the field is an anomaly by construction: red is a ridge, blue a trough, the zonal mean already removed. Thin contours are the full height field; values inside the deadband are left white. The map is the ensemble mean — the part of the wave the forecast agrees on, which is the only part whose phase is worth reading.
ECMWF AIFS-ENS ensemble mean (open data, CC BY 4.0), 100 and 500 hPa, analysis to day 15 · zonal wavenumber-1 Fourier component of geopotential height · superposition index vs ERA5 1991–2020 (WeatherBench2) · polar stereographic · updated each 00/12Z cycle
The explainer: what the 100 hPa eddy heat flux measures, the equations and how to read the chart.
The wave-1 maps show how large the planetary wave is and where its ridge sits; they do not show whether it propagates upward. That is the vertical component of the Eliassen–Palm flux, which at 100 hPa is proportional to the zonal-mean poleward eddy heat flux [v′T′]. A large wave that is vertically stacked carries almost none; one that tilts westward with height carries a lot. Positive is poleward in both hemispheres, which is wave activity entering the stratosphere.
The flux is quadratic, so it is computed per member and then averaged: the flux of the ensemble-mean fields fades with lead as the members’ phases decorrelate. Top left: the daily flux, members and the analysed record against the climatological 10–90 % range. Top right: the wave-1 and wave-2 parts. Bottom left is the panel to read for the vortex: the trailing 40-day mean, standardised. A few strong days do little; weeks above +1σ weaken the vortex, weeks below −1σ let it strengthen. In early autumn radiative cooling dominates, so a burst slows the vortex’s seasonal spin-up rather than reversing it. Bottom right: the ensemble-mean flux by latitude through the forecast.
ECMWF AIFS-ENS, 25 perturbed members + control, v and T at 100 hPa, daily to day 15 (open data, CC BY 4.0) · zonal wavenumbers 1–72 to match the climatology’s resolution · climatology: NCEP/NCAR Reanalysis 1 (NOAA PSL), 1991–2020, ±15-day window · analysed tail: the AIFS control’s step-0 fields, 00 and 12 UTC averaged (PSL’s R1 stopped in March 2026) · updated each 00/12Z cycle
Each grey line is one northern winter, 1980/81 to 2025/26, October to May, from NASA’s MERRA-2 reanalysis; the band is the 10–90 % range of those 46 winters and the dashed line the median winter. Three views:
- Wind at 60°N, 10 hPa, the textbook measure of the polar vortex. Below zero is a major sudden warming.
- Polar-cap temperature at 10 hPa, the warming itself: often 30–50 K in a week.
- Polar-cap height at 100 hPa, standardised, positive = weak vortex. This is the lower stratosphere, the level that couples to the troposphere: an event that never shows here rarely reaches the ground.
The highlight buttons draw a set of winters in colour: the five strong El Niño winters that are the usual analogs for 2026/27 (1982/83, 1997/98, 2009/10, 2015/16, 2023/24), all El Niño or La Niña winters (CPC’s RONI in DJF at or beyond ±0.5), or the QBO phase at 50 hPa in October–November. Dots mark each sudden warming’s central date (a diamond for a split). The menu draws any single winter in black. Three of the five strong El Niño analogs had no sudden warming at all (a description of those winters, not a tested result: five winters cannot establish a rate).
Our MERRA-2 pull is missing 299 of 16,979 days, scattered and at most nine in a row (one of them a corrupted all-zero record). The 10 hPa wind is filled from NCEP R1 plus the local MERRA-2 minus R1 offset (the two correlate at 0.998); everything else is interpolated.
NASA GMAO MERRA-2 (M2I3NPASM; daily means of the 0/6/12/18Z analyses, zonal means at 42 levels to 0.1 hPa, polar cap 65–90°N cos-weighted) · NCEP/NCAR R1 10 hPa height maps (NOAA PSL) · ERA5 2 m temperature and 500 hPa height (WeatherBench2, 1.5°) · CPC daily AO, RONI and QBO 50 hPa index · SILSO sunspot number · static reference, built 2026-09-25
The explainer: what sudden stratospheric warmings measures, the equations and how to read the chart.
A major sudden stratospheric warming (SSW) follows Charlton and Polvani (2007): the first day the zonal-mean wind at 60°N and 10 hPa turns easterly, November to March. A second event in the same winter needs 20 straight westerly days in between, and a reversal that never recovers (at least 10 westerly days before 30 April) is the spring final warming, not an SSW. On MERRA-2 that finds 30 reversals since 1980. Two of them, 17 February 2002 and 28 November 2025, were one day of easterlies weaker than 0.2 m/s that NCEP R1 never shows; they are listed as unconfirmed and left out of every composite and rate. That leaves 28 events in 24 of 46 winters.
Deep or shallow: the mean standardised polar-cap height at 100–150 hPa over the 30 days after, split at the median. It was the best single predictor of the surface response in our earlier tests, better than how warm the stratosphere got or how far down the warming reached. After the 14 deep events the AO averaged 1.42 below the same dates in other years over days 1–30 and 0.69 below over days 31–60, both significant (t-test, false-discovery rate 10 %). After the 14 shallow events there is no significant change. Against dates in the same season with the same AO over the previous 20 days and no warming, deep events ran 1.2 lower over the next 45 days, also significant (95 % interval 0.7–1.7); shallow events were no different. Blocking before an event does not predict its depth (r = 0.03). Depth is measured over the same weeks as the surface response, so this is association, not proof of direction. Depth is also only known two to three weeks in: it tells you what an event is doing once it has happened, not whether one is coming.
Split or displacement, from the 10 hPa height maps (MERRA-2 here is zonal-mean only, so the maps are NCEP R1): a split if, on any day from −3 to +7, two separate low centres lie below the vortex edge (the DJFM mean 10 hPa height at 60°N, minus 150 m), the smaller at least a quarter the size of the larger; otherwise a displacement. It matches the type usually given in the literature for all 20 events that have an established one (Charlton and Polvani 1981–2002, plus the well-studied later events), but it was tuned on them. The moment method’s aspect-ratio threshold of 2.4 (Seviour et al. 2013) got 17 of 20 on this 2.5° grid. 10 of the 28 are splits.
Strong-vortex events are the opposite extreme: the 10 hPa annular index (the standardised polar-cap height anomaly with its sign flipped) first crossing +1.5, November to March, at least 60 days apart, after Baldwin and Dunkerton (2001). There were 27. The AO averaged 0.76 above normal over the 30 days after them, which is significant, and was back to normal by days 31–60.
NASA GMAO MERRA-2 (M2I3NPASM; daily means of the 0/6/12/18Z analyses, zonal means at 42 levels to 0.1 hPa, polar cap 65–90°N cos-weighted) · NCEP/NCAR R1 10 hPa height maps (NOAA PSL) · ERA5 2 m temperature and 500 hPa height (WeatherBench2, 1.5°) · CPC daily AO, RONI and QBO 50 hPa index · SILSO sunspot number · static reference, built 2026-09-25
Baldwin and Dunkerton (2001) showed that large stratospheric anomalies tend to work their way down over a few weeks and then hold the surface in the same state for a month or two. Drawn against time and height, the composite looks like paint dripping down a wall. This is that chart from MERRA-2: the polar-cap (65–90°N) geopotential height anomaly at 42 levels from 1000 to 1 hPa, standardised for each level and day of year, with each level’s linear trend removed. Above 5 hPa a step at August 2004 is removed too, when MERRA-2 began assimilating Aura MLS; that is a data change, not climate. The standard deviation is floored at a third of its winter peak, so late-spring days do not blow up.
Colour appears only where the composite passes a significance test: a t-test across events at every level and day, with the false-discovery rate held at 10 % over the whole chart. With spatially correlated fields that keeps the chance of any false signal near 5 % (Wilks 2016). Grey is not significant. Read it for three things. First, how fast the red reaches 100 hPa. Second, whether it lingers there: significantly, to about day 60 after the deep events and the splits, to day 39 after displacements, and never after the shallow events. Third, whether the surface AO below turns negative. Day by day that passes only after the deep events (days 5–26) and, with the opposite sign, after strong-vortex events (days 3–9). The box gives 30-day means, a more powerful test that also passes for all warmings pooled. The blue that follows aloft is the vortex re-forming. The dashed line is the same calendar days averaged over every other year, the baseline every test is against.
NASA GMAO MERRA-2 (M2I3NPASM; daily means of the 0/6/12/18Z analyses, zonal means at 42 levels to 0.1 hPa, polar cap 65–90°N cos-weighted) · NCEP/NCAR R1 10 hPa height maps (NOAA PSL) · ERA5 2 m temperature and 500 hPa height (WeatherBench2, 1.5°) · CPC daily AO, RONI and QBO 50 hPa index · SILSO sunspot number · static reference, built 2026-09-25
What happened at the surface after each kind of event, averaged over every event since 1980, showing only what is statistically significant. Temperatures and heights are ERA5 anomalies against each grid point’s own seasonal cycle and linear trend (1979–2026), so the warming climate does not count as a signal.
The map shades 2 m temperature, and contours 500 hPa height, only where the mean over days 1–30 or 31–60 passes a t-test across events with the false-discovery rate held at 10 % over the map (Wilks 2016). The count of significant cells is printed under each map. Shallow events have none in either window, and neither do days 31–60 after splits or displacements; those maps say so. After deep events the significant part in days 1–30 is the familiar negative-AO pattern: warm over Greenland and north-east Canada, cold over northern Siberia, high heights over Greenland.
The Arctic Oscillation view compares 30-day means with the same calendar days in other years, with the false-discovery rate held at 10 % over the 12 tests. Significant: after all warmings −0.60 (days 1–30) and −0.34 (31–60); after deep ones −1.42 and −0.69; after splits −0.98 (days 1–30 only); after strong-vortex events +0.76 (days 1–30 only). Not significant: shallow events and displacements in either window.
The by region table uses land-only box means: Eastern US 32–45°N 95–70°W, Central Canada 48–60°N 105–85°W, NW Europe 47–60°N 10°W–15°E, Scandinavia 58–70°N 5–30°E, Siberia 50–65°N 70–120°E and East Asia 30–45°N 105–135°E. Each cell tests the share of events that ran cold against the region’s normal chance of a cold 30-day window (binomial test, false-discovery rate 10 % over all 72 cells). One cell passes: Scandinavia, days 1–30 after deep events, cold 13 times in 14 (−1.7 K) against a normal 44 %. Nothing else is significant, including the eastern US and northwest Europe.
NASA GMAO MERRA-2 (M2I3NPASM; daily means of the 0/6/12/18Z analyses, zonal means at 42 levels to 0.1 hPa, polar cap 65–90°N cos-weighted) · NCEP/NCAR R1 10 hPa height maps (NOAA PSL) · ERA5 2 m temperature and 500 hPa height (WeatherBench2, 1.5°) · CPC daily AO, RONI and QBO 50 hPa index · SILSO sunspot number · static reference, built 2026-09-25