COST Action CA15211 Training School
Advanced Data Analysis Methods for Identifying and Characterizing Atmospheric Electricity Variations, their Causes and Impacts

When it comes to analyzing observations or model output data in Earth and environmental sciences, many researchers resort to basic statistical tools, potentially missing a whole world of information on underlying processes that these simple methods cannot resolve by their construction. This training school provides an introduction into the world beyond these classical statistical methods and the variety of problems in atmospheric electricity research that could be tackled by more sophisticated time series analysis techniques. The school will provide extended lectures in the mornings supplemented by hands-on training in the afternoons in smaller groups. All participants are kindly requested to indicate their preferred programming language and/or statistical software (R, Matlab, Python, etc.) with their application, and are encouraged to bring their own data to the school to allow working on actually scientifically relevant problems.

Course topics

The first part of the course will start by introducing basic time series analysis tools like correlation functions and spectral analysis concepts, highlighting their common problems when dealing with real-world time series (like handling nonstationarity, trends, periodic components and stochastic persistence). On this basis, sophisticated methods for proper spectral estimation, time series decomposition and timefrequency analysis will be introduced, which may provide solutions to these common challenges.

Part 2 will draw upon the methods introduced in the first part and thoroughly extend them to the analysis of interdependencies among time series. Subsequently, the idea of conditioning on third variables will be introduced and implemented into the previously discussed methodological frameworks. Going beyond the common paradigm of linear interrelationships, it will be discussed how concepts from information theory and statistical mechanics can be employed to generalize correlation based methods in a way that also general (nonlinear) statistical relationships are captured. Finally, it will be demonstrated how phase information can be used (complementarily to amplitude information commonly built upon by spectral methods) to uncover weak relationships between observables and across time scales.

Part 3 will be devoted to a selection of novel time series analysis methods rooted in the theory of complex dynamical systems, including state space reconstruction from observed time series, intuitive visualization and quantification tools based on recurrences in this state space, and complex network approaches for studying spatio-temporal data sets.

Besides discussing the underlying concepts of all methods and their respective limitations, a particular focus will be on providing particular examples highlighting how to interpret the thus obtained results.

Trainer

Reik Donner holds a professorship for Mathematics with a focus on Data Science and Stochastic Modeling at the Magdeburg-Stendal University of Applied Sciences. In addition, he is leading a research group on developing and applying complex systems approaches for analyzing time series in Earth and environmental sciences at the Potsdam Institute for Climate Impact Research. Being trained as physicist and mathematician, he is recognized expert in applied statistics and time series analysis. He serves as Division Science Officer for Time Series Analysis at the European Geosciences Union and has been organizer of various topical sessions, international conferences and summer schools. He is representative of Germany in the COST Action ELECTRONET.

Practical training during the school will be supported by several experienced PhD students.

Organization

COST Action CA15211

Local organizing committee