Learning of weather-type transitions for risk assessment

Persistence or sequences of critical weather patterns over Europe can trigger seasonally extreme hydroclimatic conditions in certain regions. In order to better estimate return periods of extremes across Europe, existing time series of sequences of weather-types over Europe were used to train monthly rules for the transition from one situation to another and their duration behaviour. This can be efficiently realized and tested by setting up decision trees and generating up to 10,000 year time series of weather-type sequences.

In an experiment carried out, large-scale weather situation types according to Hess/Brezowsky available from 1961 to 2020 were divided into two time periods and rules for the transition were derived for both by training decision trees. Based on the trained rules of transistions for the periods 1961-1990 and 1991-2020, 10,000-year weather-type sequences were then generated and analysed.

The comparison of the probability density functions of persistence for the 30 different large-scale weather situation types show that omega-like circultion patterns over Europe have a higher tendency to persist in the present time period. In connection with this, the risks of prolonged dry phases in Central Europe have increased. For the translation of different weather-types into local weather-type characteristics, long-term monthly mean daily precipitation values per weather-type was assigned from ERA5 reanalysis data and rearranged in a post-processing step according to the generated weather-type sequences. The analysis of the maximum duration of consecutive dry and wet months in Europe was the main focus and the identified long-term changes in hydroclimatic quantities can be thus exclusively attributed to dynamic factors.