Evaluating the clustering of Central European rain-on-snow events with flood-inducing potential

Rain-on-snow (ROS) floods are responsible for the overwhelming majority of floods affecting multiple major river basins simultaneously in Europe during the last century. These widespread floods have serious negative economical, social and ecological effects, and knowledge about their rate of occurrence is critical for future projections in the face of climate change.

Recent studies have shown that ROS events (with flood-inducing potential) in Europe increase and decrease based on the elevation range considered since 1950 and there appears to be a clustering pattern of flood-poor and flood-rich periods since 1900. Our goal is to analyze if these changes in frequency can be realistically described by a stationary process (or a combination thereof) or if there must be hidden time-dependent driving factors to explain the observed clustering. To test this theory we analyze a simulation for the time period 1901-2010 based on ERA-20C dynamically downscaled using a coupled RCM. We apply a method from scan statistics and confirm the existence of significant periods poor and rich in ROS events with regards to the reference condition of independent and identically distributed random events and present their position in time. The same procedure is applied to the ROS event constituents (rainfall and snowmelt), where we identify such periods in the rainfall, but not in the snowmelt time series. We construct a stochastic ROS model by modelling precipitation and snowmelt via stationary gamma distributions fitted to our data but are unable to reproduce the observed clustering behaviour using the combined signal.

This study confirms that the observed ROS floods in Central Europe are unlikely to be the result of stationary processes which hints at climate drivers for the compound rain-on-snow process in Europe.