Extended Data Fig. 1: Framework performance depending on regression settings.
From: Response of stratospheric water vapour to warming constrained by satellite observations
As Fig. 2a, that is red circles show abrupt-4xCO2 simulation results (’actual’) regressed against predicted changes in qstrat (here abbreviated as qs), both normalized by Tg, for 27 CMIP models. The multi-model-mean is indicated as a black square; the one-to-one line in solid black. Dashed lines show the least squares regression fit (black) and the 5 to 95% prediction intervals (red). The one-at-a-time differences are that in a no lagged temperature data was considered as predictors; in b one additional time lag (\({\tau }_{\max }=3\)) was considered; in c we did not take the natural logarithm of qstrat; in d temperature predictors at all latitudes were considered; in e temperature predictors only within 30∘N - 30∘S were considered; and in f 444 samples (months covering all years from 1984 to 2020) were used for training the CMIP functions, instead of the 315 months used in the main paper. In g, temperature data at seven pressure levels (300, 250, 200, 150, 100, 70, 50 hPa) were considered as predictors, whereas in h only three levels (200, 150, 100 hPa) and in i temperature only at 100 hPa was considered.
