Differences between simulated and observed surface temperature variability in the tropics versus extratropics

Projections of future climate and especially extreme events depend on the reliable simulation of climate variability. On timescales inaccessible to direct observations, the evaluation of simulated variability requires comparison to proxy-based reconstructions. However, simulated and proxy-based estimates of surface temperature variability disagree locally on decadal and longer timescales.

Here, we expand comparisons of surface temperature variability in simulations and observations with data covering up to the last 2 Million years. For the comparison, we compile a multi-proxy database, as well as use direct observations and an ensemble of transient and equilibrium simulations of varying complexity, including an ESM with a dynamically-coupled ice-sheet-solid earth model. Based on these, we create global and regional spectra of observed and simulated surface temperature covering daily to multi-millennial timescales. We evaluate the variability with respect to differences between land and sea, proxy type, model complexity, employed forcings and model properties such as resolution.

Our results confirm that global agreement between reconstructions and models extends to the past 2 Million years. The global composite spectrum follows a power law scaling with a break at multi-millennial timescales. The results further demonstrate that a range of models can simulate the continuum of global mean surface temperature. Regionally, we find substantial differences between simulations and observations in the tropics and subtropics, where reconstructed temperature variability surpasses simulated variability. On the model-side, the complexity of the atmosphere and representation of cloud processes seem particularly relevant for the magnitude of simulated tropical variability, however, the relationship between tropical dynamics and local temperature variability requires further investigation. In the mid and high latitudes, differences between simulations and observations are smaller, especially for complex models that include volcanic forcing. The choice between dynamic and prescribed ice sheets affects temperature variability in particular in the southern polar region, where dynamically coupled ice sheets tend to lead to better agreement with proxy-based reconstructions. We further find that forcings affect simulated variability not only on their characteristic timescales, but on both longer and shorter timescales. This highlights the importance of including long-term feedbacks and volcanic forcing to simulate the spectrum of temperature variability across timescales, a necessity for reliable projections, attribution studies and assessments of the risks of extreme events.