Correlation Between Cloud Adjustments and Cloud Feedbacks Responsible for Larger Range of Climate Sensitivities in CMIP6

While the higher mean Equilibrium Climate Sensitivity (ECS) in CMIP6 has been attributed to more positive cloud feedbacks, it is unclear what causes the greater range of ECS values across CMIP6 models compared to CMIP5. Here, we investigate the relationship between radiative forcing and cloud feedbacks across the two model generations to explain the very high ECS values in some CMIP6 models. The relationship is sensitive to the definition of the forcing, particularly in CMIP6, but fixed-sea surface temperature simulations suggest the shortwave cloud feedback (lambda(SW,CL)) is anticorrelated with the forcing in CMIP5 and either uncorrelated or weakly positively correlated with the forcing in CMIP6. These relationships reflect the cloud adjustment to the forcing, which is anticorrelated with lambda(SW,CL) in CMIP5 and positively correlated in CMIP6. Although we are unable to identify a systematic change across the model generations, we do show that the difference is not due to land effects, and that cloud adjustments are generally driven by low and, especially, midlevel clouds. Furthermore, models derived from a small number modeling centers seem to be responsible for much of the difference between the CMIP5 and CMIP6 ensembles. Our analysis is severely limited by the available simulations, highlighting the need for targeted experiments to better understand cloud adjustments and to probe the sources of intermodel differences. Plain Language Summary This study investigates what caused several of the latest generation of climate models to have very high Equilibrium Climate Sensitivities (ECSs) of over 5 degrees C. We show that the explanation depends on how the radiative forcing from CO2 (F) is calculated. Using the best available estimates of F, we find that in previous model generations F is anticorrelated with models' climate feedback, damping the range of ECS, but in the newest generation of models F is positively correlated with the feedback, increasing the range of ECS. These relationships are driven by correlations between F and the cloud adjustments to the forcing (changes in clouds that are independent of surface temperature), which also change sign between the model generations. Few models provided the necessary data to calculate cloud adjustments, making it difficult to identify systematic differences between the generations, but we are able to trace the difference to models' atmospheric physics, rather than changes in land properties. Most of the intergenerational differences also seem to come from models originally derived from a small number of modeling centers, so that better understanding the cloud adjustments in a few models may explain the greater spread of ECS in the latest generation of climate models.