This study investigates how climate sensitivity depends upon the spatial pattern of radiative forcing. Sensitivity experiments using a coupled ocean-atmosphere model were conducted by adding anomalous incoming solar radiation over the entire globe, Northern Hemisphere mid-latitudes, Southern Ocean, and tropics, respectively, with both positive and negative perturbation considered. The varied forcing patterns led to highly divergent climate sensitivities, with extratropical forcing inducing significantly more global-mean temperature change compared to tropical forcing. This dependence is particularly strong over the Southern Hemisphere, where the climate is nearly twice as sensitive to Southern Ocean forcing as tropical forcing. This dependence of climate sensitivity on the location of radiative forcing stems from covariations between lapse rate feedback, cloud feedback and tropospheric stability. These results contrast with the conventional SST-pattern effect in which tropical surface temperature changes regulate the climate sensitivity, and has important implications for geoengineering and understanding the mechanisms of paleoclimate change.

The most recent generation of climate models exhibits an alarming increase in high climate sensitivity models compared to previous generations. Because the calculation of equilibrium climate sensitivity (ECS) requires simulations of a thousand years or more, most studies estimate ECS using shorter model integrations. However, the most widely used method for estimating ECS from shorter simulations underestimates ECS. Previous studies attributed this underestimate to the time-dependence of climate feedbacks. Here we demonstrate that it actually arises from an underestimate of the radiative forcing. We present a modified method that corrects for this underestimate and is shown to better agree with ECS calculated from “long run”, millennium-scale simulations. This method reveals that the actual number of “too hot” models is roughly double that previously diagnosed, with one out of every three CMIP6 climate models having an ECS greater than 5K - the “very likely” upper bound on ECS.