Cb‑Fusion – forecasting thunderstorm cells up to 6 hours

Abstract

As a severe weather phenomenon, thunderstorms can cause casualties and economic loss to the human society. A reliable forecast of these weather events would help to avoid or at least mitigate this damage. To date, the forecasting of thunderstorms however is still a challenge, especially for lead times of one hour and beyond. In this study we present a methodology to forecast deep-convection for several hours lead time: Cb‑Fusion. Cb‑Fusion estimates the likelihood of thunderstorm occurrence for up to 6 hours in advance over a part of Central Europe, using a data fusion technique that blends data of multiple sources from observations, nowcasts, and numerical weather predictions with a high update rate. The Cb‑Fusion is set up to operate in near real time. The skill of Cb‑Fusion is evaluated based on 1743 hours of thunderstorm observations collected during the months April to October, 2019. Three categories of thunderstorm size have been distinguished: ‘large’ for a coverage area larger than 5000 km2, ‘medium’ for a coverage area between 5000 km2 and 500 km2, and ‘small’ for a coverage area smaller than 500 km2. Compared to thunderstorm forecasts from numerical models alone, the combination of data from various sources by Cb‑Fusion results in a significantly better forecast skill. The study reveals that the forecast is reliable for up to 3 hours lead time for ‘medium’ and ‘large’ scale thunderstorms (median POD of 0.6–0.9) but little skill is found for ‘small’ scale thunderstorms and lead times between 3 and 6 hours (median POD of 0.05). It is argued that Cb‑Fusion provides meaningful improvements in forecasting thunderstorms for various users.

Keywords

• thunderstorm
• real-time
• nowcasting
• forecasting
• data fusion