The stratospheric polar vortex proxy at 10 hPa, in m/s. Each model is shown over its own total lead days, against the ERA5 1991–2020 climatology. Below it is the polar-cap geopotential height section, the same atmosphere seen from the other side.
A time-height section of the polar cap: the area-weighted mean geopotential height from 60°N to the pole, on a set of pressure levels, as a normalized anomaly — in standard deviations of the CORe reanalysis’s own 1991–2020 climate. Normalising puts every level on the same footing, so the stratosphere and the troposphere can be read on one scale. The pressure axis is logarithmic, so the stratosphere at the top has the room it needs, and the section runs down to the surface at the bottom.
The left of the dashed line is CORe reanalysis — NOAA’s Conventional Observation Reanalysis, which is built on the FV3-GFS model, and it is what actually happened, up to the model’s own run. The right is the forecast — what might happen, running to the end of that model’s own run. The reanalysis and the forecast are on the same scale and the same colour bar, so the two halves can be read against each other.
A seasonal model’s own mean state is not the observed climate: by the far end of a seven-month run the ECMWF SEAS5 sits several thousand m²/s² below it, and the small summer spread then draws that as several sigma of a cold stratosphere when it is nothing of the kind. The forecast half is therefore measured against the model’s own hindcast climate, which removes that offset; the reanalysis half stays on the CORe climate, and both are in standard deviations of it. The GEFS’s 35-day run drifts very little, so it is measured against the CORe climate directly.