Fig. 4: Interpretation of the statistical-learning results.
From: Response of stratospheric water vapour to warming constrained by satellite observations
a, Ridge coefficients Θ at 100 hPa and for lag τ = 2 months averaged over all 150 observational functions. b, The same for the multi-model mean of the 27 CMIP models. The predictor temperature data were standard scaled for each grid point (units log(ppmv) σ−1). The Θ magnitudes are therefore also directly comparable, that is, larger positive coefficients imply a greater humidifying effect for a typical local temperature increase. c,d, Zonal mean Θ for observations (c) and for the CMIP multi-model mean (d). The latitude–height cross sections illustrate an upward propagation of the sources of predictive information over time, reflective of the slow ascent of air through the tropical UTLS.
