Lightning‐Fast Convective Outlooks: Predicting Severe Convective Environments With Global AI‐Based Weather Models.

Severe convective storms are among the most dangerous weather phenomena and accurate forecasts mitigate their impacts. The recently released suite of AI‐based weather models produces medium‐range forecasts within seconds, with a skill similar to state‐of‐the‐art operational forecasts for variables on single levels. However, predicting severe thunderstorm environments requires accurate combinations of dynamic and thermodynamic variables and the vertical structure of the atmosphere. Advancing the assessment of AI‐models toward process‐based evaluations lays the foundation for hazard‐driven applications. We assess the forecast skill of the top‐performing AI‐models GraphCast, Pangu‐Weather and FourCastNet for convective parameters at lead‐times up to 10 days against reanalysis and ECMWF's operational numerical weather prediction model IFS. In a case study and seasonal analyses, we see the best performance by GraphCast and Pangu‐Weather: these models match or even exceed the performance of IFS for instability and shear. This opens opportunities for fast and inexpensive predictions of severe weather environments. Plain Language Summary: Over the past year, several global AI‐based weather models were released and produce a similar quality of forecasts as traditional weather models. AI‐models are very fast and computationally cheap to produce forecasts. The evaluation of AI‐models has largely focused on single atmospheric variables at certain heights. To forecast specific phenomena, such as thunderstorms, a combination of variables must be accurate at multiple heights. Here we use the output of AI‐models to derive thunderstorm‐related ingredients. We compare 10‐day‐forecasts between the AI‐models GraphCast, Pangu‐Weather and FourCastNet and a state‐of‐the‐art traditional weather model while using a reanalysis data set as the reference. The example of a tornado outbreak in the southern United States shows that all models are capable of forecasting thunderstorm ingredients multiple days in advance. To obtain a robust assessment, we evaluate the entire thunderstorm season in 2020 in North America, Europe, Argentina, and Australia, where severe thunderstorms occur frequently. Two of the three AI‐models achieve similar or even better results than the traditional weather model while being much cheaper to operate computationally. Forecasting thunderstorm parameters directly, instead of calculating them afterward, is likely to produce even better results. This opens opportunities for rapid and accessible forecasts for severe thunderstorm phenomena. Key Points: AI‐based global weather models produce forecasts with sufficient accuracy to derive instability and shear metrics skillfullyThe best AI‐based weather models are capable of competing with state‐of‐the‐art numerical weather predictions of instability and shearThis is a major step toward computationally inexpensive and fast convective outlooks [ABSTRACT FROM AUTHOR]