The frequency of extreme weather events depends relatively more on climate variability than on average changes. This makes variability a crucial element to consider in future projections. Stable water isotopes such as δ18O extracted from climate archives, including ice-cores, have been used to reconstruct regional climate and evaluate climate simulations. These archives have shown that variability in the Holocene is much lower than that at the Last Glacial Maximum (LGM, 21 kyr ago). However, state-of-the-art climate models still fail to simulate this shift. Comparison is difficult, since paleoclimate equilibrium simulations are typically run for few centuries and do not yet incorporate water isotope tracers. Volcanic eruptions offer a unique testbed to analyse the link between regional archives and global climate since well reconstructed aerosol data from 800 CE onward allow the investigation of small and large-scale effects in time and space on the climate. Here, millenial simulations from the isotope-enabled version of HadCM3 forced by solar and volcanic reconstructions in pre-industrial, LGM and past-millenium scenarios were evaluated. We then analysed the influence of volcanic eruptions on climate and δ18O values in polar and alpine regions. This allowed us to test the dependency of isotope values on regional shifts in climatology as well as global anomalies using composite analysis of volcanic eruptions. We finally discuss the impact of these results on the climatic representation of polar and alpine ice-cores representing changes in global climate variability.

Global changes in climate not only affect its mean, but also its variability, which mainly impacts society. For better projections of future climate changes it is crucial to improve the understanding of changes in both the mean, the variability and their relationship. Model-Data comparison between climate simulations and speleothem paleoclimate archives can test and validate the capability of different general circulation models (GCMs) to simulate changes in variability. However, the d18O values measured in climate archives don’t directly represent temperature or precipitation but result from multivariate, non-linear processes on top of the dominant atmospheric controls on precipitation d18O. We aim to assess a model’s capability to simulate climate variability on timescales longer than those observable. Our strategy combines a Proxy System Model (PSM) for the relevant processes with isotope-enabled GCMs. We focus on speleothems, as they are precisely date-able and provide well preserved (semi-)continuous climate signals in the lower and mid-latitudes. We evaluate trends, correlations between different records and power spectral densities across a speleothem database, focusing on the past millennium. We compare proxy results to those obtained by forward models based on isotope-enabled HadCM3 simulations and PSM approaches of increasing complexity. We evaluate the sensitivity of results to parameter choices, and test options to constrain them. We find that some parameters, e.g. transit times of water from the surface to the speleothem’s cave, strongly influences the slope of the spectra in the PSM. Based on the ample proxy and model evidence for the past 1000ys, we test for realistic parameter ranges and the sufficient complexity of speleothem PSM for global application. Given a successful application on this more recent period we envisage application on longer, millennial to orbital timescales, to provide estimates of low-latitude changes in climate variability.

Projected changes in climate are likely to affect not only its mean state but also its variability. As such, improving our understanding of the spectrum of climate variability and how different feedbacks in the climate system influence it is of vital importance. We perform a process-based examination of variability with respect to changing orbital insulation, ice coverage, and land/sea distribution during the Last Glacial Maximum and the Holocene. To this end, we adapt a two-dimensional energy balance model [Zhuang, North & Stevens, 2017] to run transient simulations. The model is forced by carbon dioxide and solar insolation changes for the last Glacial cycle. We evaluate the model’s ability to reproduce changes in local to global, seasonal to millennial temperature distributions during the Last Glacial Maximum and the Holocene. We compare the simulated states and the transient evolution to those obtained by comprehensive coupled climate models. Finally, we test the mean-state dependence of temperature variability over a large range of model configurations and discuss implications for future climate.

Globally consistent natural evidence on past climate evolution is indispensable for climate model evaluations and forecasts. However, it has rarely been investigated quantitatively whether large sets of globally distributed pollen records with limited dating resolution can be statistically linked. This could facilitate the identification of global in contrast to regional climate change signals on millennial to orbital time scales. We consider a global set of time-irregular pollen records for a joint analysis of spatial similarity on different time scales during the last glacial. Making use of measures suitable for irregular time series and by application of a spatio-temporal stochastic model, we examine significant commonality between pollen records. We quantitatively assess the resulting paleo-climate networks while respecting the spatially heterogeneous and sparse proxy archive layout. The network configurations are compared to synthetic proxy networks, which mimic different real-world record impairments. We find strong commonalities of well resolved Chilean, North Pacific and European records on orbital to millennial time scales. They reveal partly inverted deglaciation signals for westward exposed coastal tree vegetation. Such signals are consistently observable for several mid-latitude records, probably indicating equatorward shifts of westerly circulation structures during the last glacial. Surrogate data suggests that a notable part of total records might be insufficiently resolved to detect statistically significant record similarity at least when classical correlation-based measures are utilised. We compare the results to temperature and precipitation signals in PMIP3 models.