Determining the capabilities of global climate models is crucial and has long been of primary importance, for instance through dedicated chapters on model evaluation in reports from the International Panel on Climate Change (IPCC 2013; chapter 9, Flato et al. 2013). However, there is a lack of analyses specifically assessing how precipitation extremes are simulated at the global scale and in particular in a multi-model framework.

This work provides an evaluation of six global climate models (generally the latest CMIP6 version) on their representation of precipitation extremes. Models are compared to observations from different sources (i.e in situ-based, satellite-based and blended products) to enable consideration of observational uncertainty when evaluating their performance. Furthermore, we assess the influence of increasing atmospheric horizontal resolution on model performance by evaluating pairs of high and low spatial resolution simulations (hundreds vs tens of kilometers). For each model, parametrization of the convective processes from the low-resolution simulation is kept in the high-resolution simulation, which allows us to solely assess the influence of increasing resolution. Finally, we address regional aspects of model performance based on four indices characterizing different aspects of precipitation extremes.

We find that the spread among observations is substantial and can be as large as inter-model differences. Despite large observational uncertainty, simulated precipitation extremes tend to be significantly different than observations, in particular in the tropics. Our findings highlight a good agreement between the models with an intensification of precipitation extremes with increasing resolution. Interestingly, models generally show lower skill in the high-resolution compared to low-resolution simulations. This suggests that increasing spatial resolution solely in the current versions of global climate models might not induce a systematic improvement in the simulation of precipitation extremes.