Atmosphere, Vol. 15, Pages 499: Data-Driven Prediction of Severe Convection at Deutscher Wetterdienst (DWD): A Brief Overview of Recent Developments

Atmosphere, Vol. 15, Pages 499: Data-Driven Prediction of Severe Convection at Deutscher Wetterdienst (DWD): A Brief Overview of Recent Developments

Atmosphere doi: 10.3390/atmos15040499

Authors: Richard Müller Axel Barleben

Thunderstorms endanger life and infrastructure. The accurate and precise prediction of thunderstorms is therefore helpful to enable protection measures and to reduce the risks. This manuscript presents the latest developments to improve thunderstorm forecasting in the first few hours. This includes the description and discussion of a new Julia-based method (JuliaTSnow) for the temporal extrapolation of thunderstorms and the blending of this method with the numerical weather prediction model (NWP) ICON. The combination of ICON and JuliaTSnow attempts to overcome the limitations associated with the pure extrapolation of observations with atmospheric motion vectors (AMVs) and thus increase the prediction horizon. For the blending, the operational ICON-D2 is used, but also the experimental ICON-RUC, which is implemented with a faster data assimilation update cycle. The blended products are evaluated against lightning data. The critical success index (CSI) for the blended RUC product is higher for all forecast time steps. This is mainly due to the higher resolution of the AMVs (prediction hours 0–2) and the rapid update cycle of ICON-RUC (prediction hours 2–6). The results demonstrate the potential of the rapid update cycle to improve the short-term forecasts of thunderstorms. Moreover, the transition between AMV-driven nowcasting to NWP is much smoother in the blended RUC product, which points to the advantages of fast data assimilation for seamless predictions. The CSI is well above the critical value of 0.5 for the 0–2 h forecasts. Values below 0.5 mean that the number of hits (correct informations) is lower than the number of failures, which results from the missed cells plus false alarms. The product is then no longer useful in forecasting thunderstorms with a spatial accuracy of 0.3 degrees. Unfortunately, with RUC, the CSI also drops below 0.5 when the last forecast is more than 3 h away from the last data assimilation, indicating the lack of model physics to accurately predict thunderstorms. This lack is simply a result of chaos theory. Within this context, the role of NWP in comparison with artificial intelligence (AI) is discussed, and it is concluded that AI could replace physical short-term forecasts in the near future.