Climate change detection and attribution in extreme precipitation using compact representations

In the subproject CoDEx of the BMBF climXtreme project , we are investigating different data compression techniques to detect and attribute changes in the frequency and severity of extreme weather events in a changing climate. Especially for local processes on the atmospheric mesoscale, climate change signals are often masked by additional variability, resulting in poor signal-to-noise ratios. Therefore, these only become visible when the data are analyzed in compressed form. Our focus is on unsupervised learning approaches such as principal component analysis developed specifically for extremes. We focus on extreme precipitation over Germany and analyze how different data compression techniques can be used in a detection and attribution (D&A)study. Besides others, we use the approach proposed by Cooley and Thibaud (2019) on the decomposition of the tail pairwise dependence matrix, as an analogue to the covariance matrix for extremal dependence. Furthermore, we use a dualtree wavelet transform to study changes in extreme precipitation at different scales and different orientations. A D&A study will provide deeper insight into the effects of climate change on extreme precipitation events.