Norbert Marwan

Niklas Boers

Dominik Traxl

Niklas Boers

The Chapman Chair 2017 seminars is all about where complex systems science meets Arctic science, and the intersection is a large one. Complex systems is a type of science that has real world applications and can apply to a wide range of disciplines like weather patterns, climate, geoscience and biology

This year we have three presenters from France and Germany representing the Potsdam Institute for Climate Impact Research, the École Normale Supérieure in Paris and Humboldt University of Berlin. Please join our guest speakers for their illuminating seminars.

Norbert Marwan

Potsdam Institute for Climate Impact Research

Recurrence is a fundamental property of dynamical systems, which can be exploited to characterize the system’s behavior in phase space. A powerful tool for their visualization and analysis are recurrence plots and related quantification methods. The lecture provides an introductory overview covering recurrence-based methods and their applications with an emphasis on recent developments. This includes the quantification of recurrence plots by complex networks, which is highly effective to detect, e.g., transitions in the dynamics of systems from time series. As the respective phase spaces of two systems change due to coupling, recurrence plots also allow studying and quantifying their interaction, e.g., for the detection of synchronization. The different topics are illustrated by real world examples, mainly covering selected geological and palaeoclimate research questions.

Niklas Boers

École Normale Supérieure in Paris

We will show how the theory of complex networks can be applied to analyze the spatial structure of co-variability between climatic observables at possibly remote locations. Possible choices of variables include temperature, pressure, wind, or rainfall. After an introduction to the mathematical preliminaries, I first review some of the most successful applications of climate networks. Thereafter, I will focus on regional and global synchronization patterns of extreme rainfall. It will be shown how network approaches can be used for statistical prediction of extreme rainfall events, but also how global-scale teleconnections can be inferred along these lines. In particular in the global context, it becomes apparent that the phenomenon of multiple comparisons generally imposes serious problems for data-driven interdependency analyses in the climate sciences. Possible biases caused be the resulting, spurious relationships are finally discussed, and techniques to correct for them will be proposed.

Norbert Marwan

Potsdam Institute for Climate Impact Research

Caves as scientific archives

Dominik Traxl

Humboldt University of Berlin

In this talk I will present my recent work on the representation and analysis of complex systems, with a focus on geoscientific data. Although climate networks have already provided valuable new insights to the earth sciences, the combination of heterogeneous data (e.g., rainfall, lightning and fire data) and the representation of properties and interactions throughout multiple scales are still posing challenges. To overcome them, we developed a theoretical framework – Deep Graphs – which is based on network theory. As opposed to the previous most general form of network representation (i.e. multilayer networks), our framework incorporates groups of objects (supernodes) and their respective interrelations (superedges) into a self-contained network representation. Furthermore, potentially unstructured and diverse information is explicitly associated with the different (super) nodes and (super) edges. For these reasons, our framework is capable of acting as a go-between, joining a unified and generalized network representation of systems with the statistical tools of traditional fields, as well as the methods developed in the rising field of machine learning. I will first introduce this framework along with its software implementation (written in Python). After pointing out the theoretical implications of Deep Graphs, I will demonstrate its utility by applying it to rainfall, fire and land use data.

Niklas Boers

École Normale Supérieure in Paris

Low-dimensional dynamical models have a long history in climate science. In many cases, the physical mechanisms at play are so poorly understood that one has to retreat to stochastic approaches, where the model structure is not or only partially dictated by our physical understanding, but rather proposed in general form and then tested against observations. The values of the (few) parameters of such models are obtained by training them on observed time series. I will first introduce the necessary concepts from random dynamical system theory and Bayesian model inference. Thereafter, several examples of climatic subsystems for which a convincing modeling in terms of empirical stochastic models could be achieved, such as the El Niño Southern Oscillation, are summarized. I will then present an approach to obtain a stochastic model from the North Greenland Ice Core Project (NGRIP) records of isotope ratios and dust concentrations. After discussing technical challenges due to dating uncertainties in these records, I will show that the proposed model, which takes the form of a stochastic delay differential equation with cubic drift terms, is indeed capable of reproducing the statistical and dynamical properties of the NGRIP records during the last 60’000 years, including the so-called Dansgaard-Oeschger events.