Probabilistic Modeling of the European Severe Thunderstorm Climate

Severe thunderstorms with phenomena like large hail, damaging winds, tornadoes and flash floods are among the most important weather-related hazards in Europe. Despite their high societal impact and the big economic losses they inflict every year, relatively little is still known about the behavior of this risk on climatological timescales. Constraints are on the one hand the limited knowledge of historic severe weather occurrences and on the other hand the necessity of a sufficiently high resolution of reanalyses and climate models for a realistic reproduction of environments which are conducive to severe convection.

In our study, state-of-the-art datasets of historic severe weather observations, reanalysis fields, and decadal climate hindcasts and predictions were combined in order to undertake a comprehensive assessment of the past, present and future risk of severe thunderstorms in Europe. Quality-controlled information on historic occurrences of large hail, severe wind gusts, tornadoes and excessive precipitation events is collected in the European Severe Weather Database (ESWD). A 35-year dataset of ERA-Interim reanalyses (1979-2013) provides information on the underlying meteorological conditions and can be used as a homogeneous grid of "synthetic proximity soundings", from which various predictors representing the ingredients for severe convection (instability, moisture, sources of lift, and vertical wind shear) were computed.

A comparison of these observations and reanalyses for a subset in which the reporting rate was found to be sufficiently high (Central Europe since 2006) allows deriving probabilities for each type of severe weather under certain combinations of predictors. Various methods of smoothing and extrapolation of these probabilities into so far unprecedented combinations of predictor values were tested, including logistic regression (a particular case of a generalized linear model) and the corresponding generalized additive model as well as heuristic extrapolation by kernel smoothing and binning. These different methods, each of them conducted under the constraint that differences between observed and smoothed probabilities should be insignificant at the 95% confidence level, were used to include some uncertainty into the translation of environmental conditions into an expected number of severe weather events.

Applying these probabilities from the Central European training data set to the entirety of ERA-Interim reanalyses offers a more comprehensive perspective onto the pan-European severe thunderstorm climate, as it provides an estimate of the expected number of severe weather events also for those areas where the reporting rate of such events is low, in particular large parts of Southern and Eastern Europe. Selected findings will be showcased in this presentation.

Even more powerful is the application of this method to climate forecasts in order to estimate the future behavior of the severe thunderstorm risk. For that purpose, we used the MiKlip system ("medium-range climate predictions") which is operated at the German Climate Computing Center (Deutsches Klimarechenzentrum, DKRZ) in Hamburg. MiKlip provides 15 decadal climate hindcasts and predictions initialized yearly from 1960 to 2011. Every target year is thereby covered by 150 individual realizations, adding further uncertainty by the ensemble spread to the methodological uncertainties outlines above. This implementation allows decadal projections of the future severe thunderstorm risk in Europe on a probabilistic basis.