A thesis for the degree of Doctor of Philosophy defended in November 2018.
The PhD School of Science, Faculty of Science, Physics of Ice Climate and Earth, Niels Bohr Institute, University of Copenhagen
Dynamics of abrupt glacial climate change
The pre-industrial interglacial climate has been characterized by warm and stable temperatures, without major abrupt changes. This has been different during the last glacial period, which was much colder, but has been frequently interrupted by abrupt climate changes, the so-called Dansgaard-Oeschger (DO) events. These events are most pronounced in Greenland ice core records where the temperature proxy suggests jumps in between a cold (stadial) and a warm (interstadial) state with temperature increases of up to 16 K within few decades. Their cause is unknown and most realistic climate models do not produce the observed behavior. Uncovering how these models need to be changed in order to reproduce past climate changes will greatly increase faith in their projections of present day anthropogenic climate change. Nevertheless, a small number of realistic models, and a larger number of conceptual ones, have been proposed that show variability resembling DO events.
They represent different dynamical mechanisms, ranging from transitions in between multiple stable climate states, due to stochastic perturbations or changes of a control parameter, to periodic oscillations. Most comparisons of models to proxy data are qualitative, and do not address the complex temporal pattern of DO events.
This thesis presents a characterization of DO events by robust statistical analysis of the ice core record, which can serve as a basis for comparisons of model output and data. Different simple statistical and dynamical models are tested against the data, using statistical hypothesis tests and Bayesian model comparison, with the aim to distinguish different dynamical mechanisms underlying DO events. The temporal variability in the DO warming and cooling event series is too pronounced to originate from two independent random processes that are stationary and memo-ryless. Instead, there is evidence for external modulation of DO events, with distinct factors influencing stadials and interstadials. However, external modulation, such as changes in solar radiation, is not sufficient to explain the entire variability. Additionally, the stadial and interstadial durations have distinct statistics, indicating different mechanisms that cause the respective transitions. Bayesian model comparison on the basis of summary statistics shows that different stochastic dynamical systems can equally well explain the data, because the dynamics are dominated by high intensity noise in order to reproduce the large temporal variability of DO events. The corresponding dynamical regimes include noise-induced transitions in a bi-stable potential, excitability and relaxation oscillations. This thesis provides new insights into the dynamical mechanisms underlying DO events, and constitutes a statistical basis for future quantitative comparisons of proxy records and realistic climate models that exhibit DO-type variability.