Fig 1 -

Constrained versus unconstrained annual mean changes in a) global mean surface air temperature (GSAT in K), b) global mean land surface relative humidity (LSRH in %), c) global land precipitation (mm/day), and d) Amazonian precipitation (mm/day). Mean (solid lines) and 5–95% range (shading) of the prior (unconstrained projections) and posterior (projections constrained by the observed global warming) distributions of the forced response to natural and anthropogenic radiative forcings in historical simulations and SSP5-8.5 projections from thirty-two CMIP6 models. Although based on a blending of sea surface temperature over ocean and near-surface air temperature over land, the HadCRUT observations are here used as a surrogate for observed GSAT to constrain all projections using the Kriging for Climate Change method developed by [14]. Compared to the prior, the posterior 5–95% interval at the end of the 21st century is reduced by 33% for GSAT projections, but only by 12% for LSRH, 20% for global land precipitation and 6% for Amazonian precipitation, respectively. The observed anomalies are shown as black (gray) filled circles when they are (not) used for constraining the model response. Note that the global warming observational constraint does not necessarily shift the ensemble mean response in the same direction as the observed water cycle changes in panels b and d. Source: inspired from [15].