In the high-dimensional space of the climate system, especially on small scales, the internal variability is usually too large for a climate change signal to be statistically significant. A suitable reduction of the degrees of freedom can improve the signal-to-noise ratio and thusallow the identification and classification of less strong climate change signals.

In phase I of ClimXtreme, we analysed and developed methods for information compression of high-resolution spatial weather extremes. The methods investigated include data-adaptive decomposition strategies such as principal component analysis and filter approaches with wavelet decomposition. Furthermore, we studied various spatial extreme value models. In particular, we developed new statistical models with non-stationary dependence structure to describe the spatial dependence structure of extreme weather events based on extremal-t and Brown-Resnick processes.

Now, in the phase II, we aim at including the temporal component in the compact description framework and the statistical model development. In a first step, we will therefore investigate how the temporal component can be integrated into the methods and statistical models from Phase I. In particular, we will use our wavelet approach to evaluate existing statistical models for their spatio-temporal representation of precipitation extremes. We will also extend the statistical models developed in Phase I by a non-stationary spatio-temporal point process model that jointly models the time and location of extreme events. Finally, we will use these statistical models to derive conditional simulations on different spatial and temporal scales. These results will be merged into a study of extreme event detection & attribution and also be used to improve stochastic weather generators.

Website: CODEX
Institutions: Institute of Geosciences, University of Bonn¹  Institute for Stochastics and Applications, University of Stuttgart²
Contact: Priv. Doz. Dr. Petra Friederichs¹, Svenja Szemkus¹,  Prof. Dr. Marco Oesting², Carolin Forster²