Large-scale thermodynamical and dynamical controls on subtropical cloud variability in observations and CMIP3 and CMIP5 models

Uncertainty in radiative feedbacks associated with marine boundary cloud changes over the eastern subtropical oceans was a dominant contributor to the spread of climate sensitivity estimates among models of phase 3 of the Coupled Model Intercomparison Project (CMIP3). The present study compares the interannual sensitivity of boundary layer clouds and overlying high-level clouds to large-scale thermodynamical and dynamical variations in observations, CMIP3 models, and CMIP5 models. In observations, greater outgoing shortwave radiation due to clouds is associated with cooler sea-surface temperature, a stronger temperature inversion above cloud top, a moister free troposphere (thermodynamics), faster surface wind speed, and weaker subsidence (dynamics). More shortwave reflection, generally caused by increased low-level cloud fraction, acts to cool the climate system. One third of CMIP3 models simulate within the range of observational uncertainty each of the observed linear regression coefficients between top-of-atmosphere shortwave cloud radiative effect and the large-scale thermodynamics and dynamics, while only one fourth of CMIP5 models do so. Fewer than half of the simulated regression coefficients of the shortwave cloud radiative effect with respect to variations in surface wind speed are within the range of observational uncertainty in both CMIP3 and CMIP5 models. Interestingly, the regression coefficients for shortwave cloud radiative effect on sea-surface temperature and inversion strength are more poorly simulated in CMIP5 than CMIP3. On this basis, the simulation of subtropical marine boundary layer cloud variability has deteriorated from CMIP3 to CMIP5. In observations, reduced outgoing longwave radiation due to clouds is associated with a moister free troposphere and weaker subsidence. Less outgoing radiation, caused by increased high-level cloud fraction, acts to warm the climate system. These regression coefficients are well simulated by CMIP3 and CMIP5 models. The observed and modeled longwave warming due to weaker subsidence and a moister free troposphere is of similar magnitude to the shortwave cooling due to those effects, yielding a smaller change in net cloud radiative effect compared to the change in shortwave or longwave cloud radiative effect alone.