Compact Description and Statistical Modeling for Non-Stationary Spatial Weather Extremes
Detection and attribution (D&A) of climate change requires a compact description of the spatio-temporal climate state. While, in high dimensional spaces, particularly on small scales, internal climate variability is typically too large for a climate change signal to be significant, a suitable reduction of degrees of freedom can improve the signal-to-noise ratio and, thus, increase the potential to detect less strong signals in this setting. Therefore, in our project, we analyze and develop methods for information compression of high-resolution spatial weather variables and integrate them into the D&A framework for extreme weather related to moist deep convection. So far, most of the available strategies aim at the description of the bulk of the distribution while our analysis focuses on the tail behavior. Thus, we compare different methods and assess their assets and drawbacks with respect to extreme weather. The methods investigated comprise data adaptive decomposition such as principal component analysis, filter approaches using wavelet decomposition, or dynamical decomposition based on reduced dynamical models. Furthermore, we consider various spatial extreme value models. In particular, we develop new statistical models describing the spatial dependence structure of single extreme events based on Pareto processes.
Institutions: University Bonn, University Stuttgart
Contact: Petra Friederichs, Marco Oesting