Abstract:
Fluctuations in Earth’s surface air temperature determine the climatic conditions for life on our planet and influence, among other factors, the occurrence of extreme climatic events. Therefore, a precise understanding of this variability in temperature is paramount to society, politics, and the economy. However, many characteristics and impacts of temperature variability are still uncertain, especially at the local spatial scale and on decadal to multicentennial timescales. Moreover, the relative contribution of climate system-inherent and external drivers of variability needs to be better constrained. Finally, how temperature variability changes with the mean climate state and, thus, under current and future anthropogenic warming remains to be clarified. These uncertainties affect long-term planning, for example, concerning mitigation and adaptation strategies. In three publications, this dissertation examines the statistical properties of temperature variability as a function of the underlying spatiotemporal scales, external drivers, and mean climate state. Spectral methods are combined with time series analysis, conceptual modeling, and Bayesian inference to quantify temperature variability from climate model simulations and paleoclimate records. The results confirm overall confidence in the simulated global temperature variability. Climate models, however, respond more strongly to external forcing and show fewer internal fluctuations at decadal scales than paleoclimate reconstructions of global temperature. At the local level, there are significant data-model mismatches in temperature variance and correlation properties over decadal and longer timescales. Improved representation of natural forcing in climate model simulations can partially offset these differences. By integrating the response to volcanic eruptions, sea ice dynamics plays an essential role in amplifying local multidecadal variability. Decreasing sea ice extent attenuates local variability under warming, especially in high latitudes. Overall, global variability tends to be dominated by external forcing, while local variability is primarily caused by state-dependent internal variations and remains subject to substantial uncertainties. This work extends these findings with complementary investigations. A new estimate of the power spectral density for global temperature beyond the last millennia complements the analyses. The comparison of transient climate model experiments covering the last 27 thousand years using prescribed or interactively coupled ice sheets reveals benefits from dynamic simulation of ice sheet feedbacks for representing millennial-scale variability. Further studies analyze the potential impacts of uncertainties in the simulated variability on projections of extreme climate events, attribution studies, and risk assessments. Missing interannual to millennial temperature variability in simulations could lead to underestimating future climate impacts. Limitations of the presented analyses arise from uncertainties of the data and the assumption that weakly-stationary stochastic processes sufficiently describe the considered temperature time series. Since this work focuses on past and present climates, follow-up studies are needed to test the robustness of the findings in future projections, particularly against the influence of non-linear processes. Altogether, the insights gained on the timescale-, forcing-, and state-dependent statistics of local to global temperatures open new possibilities for improving the analysis and understanding of temperature variations, as well as the ability of models to simulate climate variability. Despite remaining uncertainties in model simulations and reconstructions, the reliable simulation of future local temperature variability requires improved representations of ice sheet feedbacks and natural forcing in climate model experiments. This will contribute to a reliable assessment of future climate-related risks and help inform long-term planning, mitigation, and adaptation efforts.