Impact-based temporal clustering of multiple meteorological hazard types in southwestern Germany

Abstract

A series of multiple meteorological extreme events in close succession can lead to a substantial increase in total losses compared to randomly distributed events. In this study, different temporal clustering methods are applied to insurance loss data from southwestern Germany from 1986 to 2023 for the following hazards: windstorms, convective gusts, and hail, as well as pluvial, fluvial, and mixed flood events. We assess the timing and significance of seasonal clustering of single hazard types as well as their serial combination by use of both a simple counting algorithm and the clustering metric Ripley's K. Results show that clustering is significant only for certain hazard types compared to a random time series. However, clustering is robust for a combination of multiple hazard types, namely hail, mixed or pluvial floods, and storms. This particular combination of hazard types is also associated with higher losses compared to their isolated occurrence. Clusters of damaging hazards occur mainly during May-August and depend on the method of defining independent events (peaks-over-threshold method with flexible lengths vs. hours clause method with fixed lengths) and their resulting duration. This study demonstrates the relevance of considering multiple hazard types when evaluating clustering of meteorological hazards.