Clouds have a significant impact on climate. They contribute to controlling the planetary energy balance, and the precipitation distribution. Global Climate Models (GCMs) designed to reproduce the state of the climate system, however, have difficulties representing clouds. The use of parametrization methods to face those difficulties, and their varying accuracy has led to large model uncertainties regarding climate change. This doctoral thesis aims to contribute to reducing climate change uncertainties, particularly those related to the sensitivity of tropical extreme precipitation to warming, and to climate feedbacks controlling the global temperature response to increased atmospheric carbon dioxide concentrations. 

An evaluation of the representation of tropical precipitation across phases 3, 5, and 6 of the Coupled Model Intercomparison Project (CMIP) points at continuous improvements in the number of consecutive dry days, the modes of variability, and the twentieth-century trend in dry months. On the other hand, there is little change in the representation of the summer monsoons, the double-ITCZ bias, and the diurnal cycle of precipitation, as well as biases in the trend in extremely wet months, and the precipitation frequency. These issues warrant alternative approaches, such as high-resolution storm-resolving models, which may be able to provide insights into how tropical precipitation might change as a result of anthropogenic warming.

Taking this approach, using a global-scale non-hydrostatic model in aquaplanet configuration and varying resolutions, we find that the sensitivity of tropical extreme precipitation to warming is greater than the given Clausius-Clapeyron rate (~7%/K). This can be ascribed to strengthening updrafts where extreme precipitation occurs. Here, similar sensitivity is found both at convection-resolving and at the resolution at which the parametrization scheme was tuned for. And somewhat surprisingly, there is no apparent relation to the degree of convective organization.

To evaluate the representation of internal variability feedbacks in models, radiation balance changes during natural variations in temperature are compared between the latest generation climate models (CMIP6) and observations. Biases are found particularly in the tropics, subtropics, and the Southern Ocean. We identify relationships between simulated longwave and shortwave internal variability feedbacks and those where atmospheric carbon dioxide is abruptly quadrupled. Comparing those relations with observations indicates that models with moderately negative longwave feedback and weak shortwave feedback are more realistic in that respect. However, uncertainty in observations and model estimates of internal variability feedback makes it challenging to use observations to constrain models total forced climate feedbacks.

Finally, we explore relationships between CMIP6 internal variability and forced climate feedbacks using observations and reanalysis data. We find evidence supporting the idea that models with moderately negative longwave and moderately positive shortwave internal variability feedback are more realistic. In addition, a relationship is emerging between net forced climate and net internal variability feedbacks over a period of at least 60 years,  so continuous satellite records are needed for at least 24 more years to constrain net forced climate feedback using observations.