Efficient usage of information for modeling precipitation extremes and large scale influence

Extreme precipitation is currently the biggest climate-change-related thread in middle Europe with flooding events leading to high death tolls and huge existential and financial losses. Evaluating the probabilities of these extremes can help preventing casualties and reducing impact consequences. Our analysis is based on Intensity-Duration-Frequency (IDF) curves which describe the major statistical characteristics of extreme precipitation events. They provide information on the probability of exceedance of certain precipitation intensities for a range of durations and can help to visualize how extreme the event for different durations is.  A popular way of evaluating return periods of extremes is to model the underlying distribution of block maxima with the Generalized Extreme Value (GEV) distribution. A core problem when modeling extremes is the scarce availability of data. This can be addressed by using the available data more efficiently. Therefore, including maxima from different measurement durations is useful for (1) gathering more information from the data and (2) estimating return periods for different time scales with a consistent modeling approach. Duration-dependence is implemented directly into the parameter estimation (Koutsoyiannis et al., 1998) and enables a consistent model, i.e. without quantile-crossing. We present a new parameterization of the duration-dependent GEV (d-GEV) that is more flexible with respect to long ranges of durations and is considering the different time scales on which extremes occur in winter and summer (Fauer et al., 2021). Applying the new model to the extreme rain event on 14 July 2021 in Ahrtal, Germany reveals that the event was most extreme on a time scale of 20-30 hours.

Investigating the impact of large-scale atmospheric flow on extremes will help to learn how extremes changed in the past and make projections about their change in the future. Large scale variables are incorporated into the model as covariates in generalized linear models for the d-GEV parameters. An ongoing study tests for the inclusion of NAO, a blocking index, monthly mean temperature, etc., as predictor variables. First results show a significant correlation (5%-level) between monthly precipitation maxima and NAO/blocking for some durations and some seasons. It will be analyzed whether this connection can be useful for modelling d-GEV parameters with large-scale variables.