Data-model comparisons of Holocene (11,700 years ago to present) climate provide an ideal basis for evaluating climate model performance outside the range of modern climate variability. The Holocene is recent enough so that boundary conditions of the underlying physics and forcing are well known, while paleoenvironmental archives are abundant and dated with enough precision to comprehensively reconstruct climate. To date, efforts to reconstruct the spatial patterns of Holocene climate change have been mainly focused on the mid-Holocene (about 6'000 years ago), but significant discrepancies have already been identified in data-model comparisons.
These data-model discrepancies can be investigated using instrumental datasets covering continental or hemispheric scales which allow us to reconstruct large-scale climatic features, such as atmospheric dynamics or latitudinal temperature gradients. The generation of these datasets for times prior to the 19th century however faces considerable challenge because there are very few direct measurements of climatic variables. We rely on climate proxies as indirect measurements of the paleo climate. The most abundant one is fossil pollen data, i.e. pollen that are produced by vegetation and can be preserved over thousands of years in terrestrial (or coastal) archives (e.g. lake sediments). This proxy is available from all non-glaciated continents over the world in many different climate regimes, and the primary data is becoming increasingly accessible through large publicly available and community-driven relational databases. Our ability to use this proxy for continental-scale climate reconstructions, however, depends on our ability to analyze, explore and find patterns in these rich and heterogeneous databases. In particular, this requires a proper understanding of the uncertainties that are related to the indirect measurement of climate.
In the first part of this thesis, I present three new software tools that tackle the challenge to make this large amount of data accessible, and to build and develop a continental-scale pollen database. These tools cover a wide range of possible applications to leverage our work with site-based proxy data to a continental scale. The first tool I present is a web framework that is built around a map-based interactive database viewer, developed primarily for the Eurasian Modern Pollen Database, EMPD. This new tool makes the database accessible to other researchers and to the general public, and it allows a continuous and stable development of the community-driven database. In addition to the EMPD, I present an extension of this viewer that makes a large northern-hemispheric fossil pollen database accessible and allows its visual exploration.
The second tool tackles the challenge to fill the gaps in certain geographic areas in the pollen database. straditize is a digitization software for stratigraphic diagrams, and pollen diagrams in particular. It can be used to generate new data for the pollen database from publications of the pre-digital era, i.e. from publications where the primary pollen data is not accessible anymore but through the visualization in form of a pollen diagram in a peer-reviewed publication.
Finally, I present the generic python visualization framework psyplot, that bridges the gap between visualization, computation and publication in the day-to-day work of scientists, and that has been used in multiple parts of the thesis. This flexible software can be integrated and enhanced by a variety of applications and already contains multiple convenient visualization methods useful for climate science, particularly the visualization of geo-referenced data and it handles data that is too large to fit into memory or lives on different structured or unstructured grids.
The second part of my thesis contains two new statistical methods to estimate large-scale paleo climatic environments based on modern day relationships. The first one, pyleogrid, uses a large pollen database and turns it into a gridded climate reconstruction that can cover continental, hemispheric or even global scales. This software focuses a lot on the integration of the intrinsic uncertainties in the proxy data. The outcome of this gridding procedure allows a comparison of computational climate models with an independent observational database that comes with reliable estimates of uncertainty.
The last chapter of this thesis applies a converse strategy and uses modern statistical relations within climate variables to inform a computational model. The global weather generator (GWGEN) has been parameterized with thousands of global weather stations and provides a statistical tool that downscales monthly to daily climatology on a global scale. This tool can be embedded in a global paleo vegetation model where it efficiently simulates the necessary daily meteorology.