Compact Description for High Resolution Spatial Weather Extremes

Detecting local climate change signals, particularly within the context of extreme weather events, is challenging due to the significant internal variability of the climate system. This thesis, part of the BMBF-funded ClimXtreme Phase I project, aims to improve the signal-to-noise ratio of climate change signals during extreme weather events through advanced data compression methods. Our approach is twofold, emphasizing data-adaptive techniques and spectral decomposition. These methods are evaluated for their effectiveness in describing heat waves, droughts, and precipitation extremes. <br> We first analyze Principal Component Analysis (PCA). The focus on extremes is achieved using the tail pairwise dependence matrix (TPDM) proposed by Cooley and Thibaud (2019). Applying this method to daily temperature maxima and meteorological droughts of varying durations, we identify an effective technique for analyzing and compactly describing large-scale multivariate weather extremes. Additionally, we introduce the cross-TPDM to identify patterns of concurrent extremes across two variables. We propose an extreme pattern index (EPI) that provides a pattern-based spatial aggregation of extremes and demonstrate that a heat wave definition based on EPI effectively detects major heat waves across Europe. For addressing simultaneous extremes in two variables, we extend this approach by introducing the threshold-based EPI (TEPI). Using the European heat waves of 2003 and 2010 as examples, we show that TEPI describes the large-scale compound character of heat waves and droughts. <br> Next, we examine wavelet transformation, specifically utilizing the complex dual-tree wavelet, which has proven efficient for precipitation fields (see Brune et al., 2021). Focusing on extreme precipitation, we find that the wavelet transform accurately represents these extremes without requiring specific adaptations to method or data. Comparing two European reanalyses (COSMO-REA6, CERRA) for their representation of hourly precipitation, we discover that CERRA cannot accurately resolve small-scale convective events. A scale-aware detection study for three regions in northern, southern and western Germany reveals consistent trends of increasing intensity in small-scale events during summer and in large-scale events during winter.