Dependence of simulated variability of surface climate on model complexity – insights from an ensemble of transient simulations of the Last Deglaciation

Climate variability is crucial to our understanding of future climate change and its impacts on societies and the natural world. However, the climate records of the observational era are too short to explore long-term variability. Conversely, an exploration of long transient simulations from state-of-the-art Earth System Models (ESMs) poses high computational demands. It is therefore pertinent to identify the level of complexity sufficient to simulate the variability of surface climate from annual to centennial and longer timescales.

To this end, we use an ensemble of transient simulations of the Last Deglaciation, the last period of significant global warming. The ensemble covers an energy balance model (EBM), models of intermediate complexity (EMICs), general circulation models (GCMs) and ESMs. This constitutes a hierarchy that we categorize based on employed atmosphere and ocean components and their resolution, as well as implemented radiation, land hydrology, vegetation and aerosol schemes.

To investigate the simulated variability of surface temperature and precipitation, we analyze changes in the shapes of their distributions as characterized by their higher order moments – variance, skewness, kurtosis – with warming. These higher order moments relate the tails to the extremes of the distributions. We identify spatial and temporal patterns and how they depend on model complexity. The EMICs can generally match the global and latitudinal changes in temperature variability found in more complex models. However, they lack in precipitation variability. We further find that the EMICs fail to simulate the tails of the precipitation distributions. We observe dependency of variability on the background state, generally increasing with model complexity. However, there is still a large spread between models of similar complexity, some of which can be related to differences in forcings. Furthermore, questions remain on the abilities of models of any complexity to simulate a magnitude of long-term variability similar to that found regionally in proxy reconstructions. Our analysis offers implications as to the complexity needed and sufficient for capturing the full picture of climate change and we offer some first insights into how the findings translate to future projections of climate change.