Detecting Severe Weather Trends Using an Additive Regressive Convective Hazard Model (AR-CHaMo)

Abstract

A statistical model for the occurrence of convective hazards was developed and applied to reanalysis data to detect multidecadal trends in hazard frequency. The modeling framework is based on an additive logistic regression for observed hazards that exploits predictors derived from numerical model data. The regression predicts the probability of a severe hazard, which is considered as a product of two components: the probability that a storm occurs and the probability of the severe hazard, given the presence of a storm [P(severe) 5 P(storm) 3 P(severejstorm)]. The model was developed using lightning data as an indication of thunderstorm occurrence and hazard reports across central Europe. Although it uses only two predictors per component, it is capable of reproducing the observed spatial distribution of lightning and yields realistic annual cycles of lightning, hail, and wind fairly accurately. The model was applied to ERA-Interim (1979–2016) across Europe to detect any changes in lightning, hail, and wind hazard occurrence. The frequency of conditions favoring lightning, wind, and large hail has increased across large parts of Europe, with the exception of the southwest. The resulting predicted occurrence of 6-hourly periods with lightning, wind, and large hail has increased by 16%, 29%, and 41%, respectively, across western and central Europe and by 23%, 56%, and 86% across Germany and the Alps during the period considered. It is shown that these changes are caused by increased instability in the reanalysis rather than by changes in midtropospheric moisture or wind shear.