Modelling the occurrence of convective hazards using ERA5 reanalysis data

Convective hazards such as large hail, severe wind gusts, tornadoes, and heavy rainfall cause high economic damages, fatalities, and injuries across Europe. There are insufficient observations to determine whether trends in such local phenomena exist, however recent studies suggest that the conditions supporting such hazards have become more frequent across large parts of Europe in recent decades.

We model the occurrence of these hazards using Generalized Additive Models (GAM) to investigate the existence of such long-term trends, and to enable objective probabilistic forecasts of these hazards. The models are trained with storm reports from the European Severe Weather Database (ESWD), lightning observations from the EUCLID network, and predictor parameters derived from the ERA5 reanalysis. Our work is based on the framework AR-CHaMo (Additive Regression Convective Hazard Models), previously developed at ESSL.

Preliminary results include a spatial depiction of the environmental conditions giving rise to convective hazards at a higher resolution than was possible before. The skill of hail models developed using AR-CHaMo has been shown to be superior to composite parameters used by weather forecasters for the occurrence of large hail, such as the Supercell Composite Parameter (SCP) and the Significant Hail Parameter (SHP). Likewise, for tornadoes, more skillful models can be constructed using the AR-CHaMo framework than predictors such as the Significant Tornado Parameter (STP).

The developed models have use both in climate studies and in medium-range severe weather forecasting. We will report on initial efforts to detect long term (1979-2019) trends of convective hazards and present how these models can be used to support severe weather forecasting using medium-range ensemble forecasts.