Revealing the Structure of Precipitation Extremes: A Spatio-Temporal Wavelet Approach

We assess the potential of wavelet filtering to detect non-stationarities in extreme precipitation related to systematic changes in spatio-temporal precipitation process. We find a noticeable increase in precipitation intensity in recent decades in Summer months over Germany, which is likely related to increased energy at the convective scales – indicating potential signals of ongoing climate change.

To this end, we apply a space-time wavelet filter to gain deeper insights into the spatio-temporal characteristics of precipitation extremes. Our three-dimensional wavelet approach allows to simultaneously resolve features in both space and time, making it a valuable tool for investigating scale-dependent behaviour and structural changes in extreme events. Our analysis is based on hourly precipitation data from the RadKlim dataset provided by the German Weather Service. This homogenized, high-resolution precipitation data offers 23 years of continuous observational data to date and serves as a robust foundation for analyzing the dynamics of extreme precipitation across different spatial and temporal scales.

We also analyse and compare the spatio-temporal characteristics of historical heavy rainfall events over Germany.  Our key findings reveal two dominant types of extreme precipitation events: (1) long-lasting events with low propagation speed and (2) events marked by recurring convective activity over the same region, leading to localized accumulation and potential flash flooding.

This research is conducted within the BMBF-funded ClimXtreme CoDEx project, which aims to advance data compression techniques for the analysis of high-dimensional spatio-temporal weather extremes. By reducing the degrees of freedom in the data, we enhance the signal-to-noise ratio, enabling a more precise and detailed characterisation of extreme events. It also allows us to better isolate relevant physical signals from background variability, especially in complex and noisy data sets, as is typical for climate observation data.