Quantitative paleoclimate reconstructions in the European region based on multiple proxies

Abstract

The Earth’s climate system is highly dimensional and has to be described with the help of probabilities. Such probability descriptions can be estimated from measured data. Relatively large-scale instrumental records, however, date back only about 150 years. Paleoclimatology studies the climate history of the Earth before the period of instrumental measurements. For this purpose, it uses so-called climate proxies, which indirectly provide information about past climate conditions. In order to produce quantitative paleoclimate reconstructions based on these data, several problems must be analyzed and solved in the course of this thesis. <br /> First, we present a new age-depth/distance transformation in a Bayesian formulation by determining the uncertainty information of depths in sediments and distances in speleothems at a given age. This allows us to perform a data-driven transformation to past ages that behaves like a convolution with different kernel smoothers and avoids too much certainty in the statements about these time periods. Another result of this technique is the determination of the age resolution and its projection onto a regular grid. Thus, multiple proxies can be linked in time and spectral analyses can be performed. <br /> Furthermore, we introduce a new way to establish transfer functions that map climate variables to plant distributions. This includes consideration of various machine learning algorithms for solving the classification problem of taxa absence and presence, taking into account uncertainties in the proxy-climate relationship. For the models and plant distributions used in this work, a simple feedforward neural network with one hidden layer wins in 70 % of the cases. <br /> Based on our age-depth/distance transformation and transfer functions, we formulate a new Bayesian Hierarchical Model that produces local paleoclimate reconstructions. This considers various proxy sources such as plant data from lake and mire sediments, isotopic information from speleothems, marine sediments, and ice cores. These are studied not only in temporal space, but also in spectral space using wavelet power spectra. Such a comprehensive use of the spectral behavior of proxy information is possible due to the new age-depth/distance transformation and has therefore not been performed before. In addition, a priori information on the actual climate distribution in specific time periods are incorporated as further constraints. To solve the local reconstruction model, we use two different Markov chain Monte Carlo sampling methods called Metropolis-within-Gibbs and random walk Metropolis-Hastings. During the inference processes, our new method generates taxa weights that provide information about their importance to each site. As a result, over 600 sites in Europe, Northwest Africa, Anatolia, and the Levant are being processed, resulting in final 186 accepted local paleoclimate reconstructions. They show not only small-scale climate changes, which can be identified as Bond, Heinrich, and Dansgaard-Oeschger events, but also large-scale variations such as the last deglacialization and various glacial-interglacial cycles. Human influence on plant information in the lake and mire sediments studied affects our local reconstructions, which can be minimized to some extent with our new method by paying more attention to isotope-based proxies during the inference process. <br /> Finally, these local paleodata are summarized using spatial reconstruction methods over the European region. In this context, Earth System Models from CMIP5/PMIP3 experiments are linked to our local information, producing assimilated spatial climate fields. The results compared to present-day conditions for the Middle Holocene show warming in northeastern Europe and cooling in southern Europe. The reconstructed annual precipitation indicates an increase in northeastern Europe, a decrease in western Europe, and an increase in the Eastern Mediterranean and Levant. Compared to present-day conditions, the spatial reconstruction of the Last Glacial Maximum reveals a general cooling and an increase in precipitation in the west of the Iberian Peninsula and in northwest Africa, Anatolia, and the Levant. The latter shows a dipole structure with higher precipitation in the southern Levant.